1. Introduction

1.1. Background and Theoretical Foundations

Ensuring reliable and affordable access to electricity is paramount for households and communities to attain fundamental capabilities [1]. The useful energy services associated with adequate access to electricity are a cornerstone for economic development [2,3] but are also indispensable for advancement across diverse dimensions, i.e., education, nutrition, sanitation, and health [4,5,6]. Furthermore, access to electricity is a sociotechnical imperative, fostering social innovation that is pivotal in facilitating a low-carbon energy transition, promoting civic empowerment, and addressing overarching social objectives [7].

On the global political stage, the acknowledgement of the critical role of universal access to electricity is reflected in target 7.1 of the Sustainable Development Goals (SDGs), adopted by the United Nations in 2015 [8]. While notable progress has been documented in the preceding decade, the realization of universal electrification by 2030, as stipulated within SDG 7.1, remains at a considerable distance. Globally, more than 675 million people lacked access to electricity in 2021 [9]. Rural regions are disproportionally affected, with eight out of ten people without access living in rural areas [9]. In Sub-Saharan Africa, where 567 million people lacked access to electricity in 2021, the disparity in electricity access rates between urban and rural areas has risen in recent years [9]. This trend can be attributed to the financial and technical challenges of reaching the rural population. In rural areas that are only sparsely populated and where purchase power and electricity consumption can be low [10], extending the national power grid to supply electricity is economically often not viable [11]. Here, isolated renewable-based off-grid systems can be a cost-efficient and sustainable solution to enable electricity access to communities (e.g., [11,12,13,14]) and provide communities a basis for conducting activities that enhance development across the various dimensions interlinked with electricity (see, e.g. [2,6,10,15]).

More granular research on the interlinkage of access to electricity and development on the micro-level (see, for example, [16]) has shown that while access to electricity via off-grid energy systems can stimulate development in rural communities, access to electricity alone does not guarantee development. One must note that literature assessing a correlation between electricity access and development often ascribes economic metrics as a central effect measure for development, e.g., household income [e.g. [16,17] ]. Given the significance of household income for causally related household activities that may lead to changes evoking development in other dimensions [17], this is a meaningful indication. Nevertheless, evidence of projects, in which access to electricity in rural locations was enabled and no direct effect on income or well-being was observed, is abundant (e.g. [18]). In fact, the impact of electrification projects on enhanced development (i.e., increasing economic activities or household income) seems to crucially depend on the community's choice of how to use the electricity within the scope of action that is facilitated by the local energy system. Thus, literature evidence is strong that the outcomes and impact of electrification projects (note that in this paper, we understand as per the logical framework theory ‘outcomes’ as the project´s effects at the target-group level as opposed to ‘impact’ as project´s effects at the societal or regional level) depend on i) the ability and choice of the community to use the electricity for productive activities [19,20] and ii) external factors supporting the community in their capacity to utilize electricity for productive activities, e.g., finance, training, awareness, etc. [15,17], and iii) the degree to which the energy system design facilitates the community´s choices.

The use of electricity for productive uses is commonly referred to as productive use of electricity (PUE) as opposed to consumptive use of electricity in households [19]. Such PUE commonly is electrically powered machinery used by the community, according to their operational preferences, and may be directly integrated into the electricity supply system that serves residential loads of the community. Thus, the PUE appliances and the user of the PUE asset directly influence the operational requirements of the electricity supply system and its financial viability. Energy system planners (note that we use the term "energy system" instead of "electricity system" to account for potential additional energy vectors in the system) pay increasing attention to PUE system integration and PUE user behavior. In addition to supporting the stimulation of the economic and social development of the community or individual user [15,17], PUE appliances can benefit the financial viability of the energy supply system. PUE appliances typically consume more energy than residential appliances in rural villages [21,22] and therefore may pose a reliable (and relatively larger) source of income for the system operator compared to domestic loads (see relevant discussions on ‘anchor loads’ as relatively large non-domestic loads in [23]). Prominently, the financially viable operation of off-grid energy supply systems that serve residential customers with a low electricity consumption poses significant challenges. Including PUE assets as anchor loads can increase the energy system utilization rate and provide a predictable high off-take guarantee, which in turn improves the projects' bankability [21] and de-risks electrification projects for the private sector [10]. One must note that the prevailing narrative in the relevant literature that PUEs are a reliable and relatively higher source of income for rural electrification operators ('anchor load') is not universally applicable and depends on the respective context. In rural businesses that are often operated by single informal entrepreneurs and may not be well organized, the operation of PUE may in fact be erratic. In addition, the continuous electricity demand of the PUE asset depends on the economic success of the associated business. The dependency poses a financial risk to the energy system operator. This is especially relevant in contexts that are characterized by short lifetimes of businesses. Nevertheless, in communities with limited financial capacity to invest in stand-alone energy systems that power PUE asset, the systematic planning of integrated energy system serving both PUE assets and residential loads is imperative for the utilization of electricity for productive uses and the development of associated capacities within the community.

The essential aim of off-grid energy system planning is to design a system that adequately addresses the electricity-related needs of the community it serves. In this, the system must be financially viable to sustainably be operated and maintained to ensure it´s proper function. Figure 1 describes basic dynamics and interactions between local parties involved in and relevant for the description of the considerations that guide the planning of an off-grid energy system that integrates domestic household loads and PUE loads.

Figure 1. Conceptual considerations for the integration of PUE and household electricity demands in off-grid energy system planning.

Figure 1. Conceptual considerations for the integration of PUE and household electricity demands in off-grid energy system planning.

The community is composed of individual households using the energy system. These households have individual time-variable preferences and capabilities (i.e., assets) to utilize electricity. Within the community there are needs for services, some of which can be supplied by specific PUE that can be integrated in the local electricity supply system. The electricity and service needs are highly context specific. The electricity utilization preferences of the individual households and the operational preferences of the PUE operator determine the principal design requirements for the integrated energy system. For the energy system and the PUE two roles are relevant, namely the role of operation and the role of financing and ownership. These roles may be taken by the community, a single member of the community or an external entity, including the energy system planner.

We make use of Figure 1 to both guide the discussion of existing literature and its respective underlying perspectives, and to describe the deliberate assumptions we made in our analysis. In research and in the practical implementation of project the planning rationale deviates from a generalized perspective. For context-specific relevance it is useful to deviate from a generalized consideration, presume system design choice and make context-specific assumptions. The assumptions made may be a result of limitations of available data, e.g. the household electricity demand, or may result from the intention to study certain underlying dynamics or the consequences of specific system choices e.g. applicable business models. In practice the basis for decision-making is in many cases the financial viability of the energy system, as rural electrification efforts are often driven by the private sector that requires a cost recovery business model. It is important to note that this constitutes a specific perspective, namely system financing and operation. Accordingly, the current literature investigating the integration of PUE in off-grid systems supplying electricity to household loads often and usefully takes the financial perspective and evaluates the integration of PUE in off-grid systems based on it’s expected financial impact. Given the complexity of and the various possible constellations of interaction and behavior of parties involved in the local energy system as depictured in Figure 1 it is no surprise that the existing literature finds contradictory results regarding the financial benefits of integration of PUE in off-grid energy system. For example, Booth et al. [24], in a hypothetical community microgrid scenario (peak load of 5.7 kW), find that integrating a single 10-kW maize mill could either decrease the cost of electricity provided the system by 14% or increase it by 7% compared to a system only serving domestic loads, depending on the mill's daily and seasonal operational parameters (notably, the authors exclude the costs of the mill from their calculation, assuming any community member or external party taking the role of PUE financing). Specifically, the authors find that the economic impact of integrating the maize mill varies across ‘operating scenarios’, which denote different usage patterns of the mill across days of the week or seasons of the year. Similarly, van Hove et al. [21], studying the economic impact of integrating various PUE in mini-grids serving household loads, find that the impact is determined by the usage patterns of the PUE. Especially seasonally used PUE may offer only little improvements of the system costs, as they require additional energy system assets to meet peak demand in the high season that under-utilized during the low season [21]. These two examples (for other similar examples see [25,26]) support our suggestion that to ensure economic improvements of off-grid electricity supply via integration of PUE the energy consumption patterns of the PUE appliances (notably determined via usage patterns and community preferences) need to fit into the household residential electricity consumption patterns to avoid costly additional production (and storage) devices being required to power the PUE aside from the residential loads.

2. Materials and Methods

2.2. Methods

As part of this study, a novel methodological workflow was developed to evaluate different PUE's technical and economic fit with residential household energy consumption patterns based on the residential community's socioeconomic and demographic composition. Figure 2 illustrates the proposed methodological workflow. The workflow is divided into five steps, for each of which the applied methods and integrated data are described in detail in dedicated Subsections (Subsections 2.2.1 – 2.2.5).

Figure 2. Methodological workflow of the analysis.

Figure 2. Methodological workflow of the analysis.

2.2.1. Consumption Data Analysis

Our methodology relies on thoroughly assessing historical electricity consumption patterns of a representative residential community connected to a nanogrid. The consumption data analysis aims to identify i) time-dependent determinants of residential electricity consumption patterns (based on power and energy demand on hourly resolution) and ii) socio-economic and demographic determinants of electricity consumption patterns. 107 village residents currently connected to nanogrids were chosen as the sample size for the study due to the consistency of electricity consumption data and socio-economic data. The following data was used:

• Electricity consumption data: to reconstruct historic hourly electricity consumption patterns, the sample's electricity current measurements (10 minutes resolution) between January 2018 and December 2021 (Earliest data point: 10.02.2018. Latest data point: 01.12.2021) were multiplied with measured voltage (hourly resolution) and interpolated. The data was cleaned to cover for eventual reboot events of the electricity consumption logging system or other missing values and passed to a Python-capable environment for further processing.
• Socio-economic and demographic data: we used socio-economic data from irregularly conducted household surveys conducted by Nanoé in 2018 – 2021 for the purpose of assessing potential nanogrid clients. As the surveys at that time were not intended to be used for a thorough statistical treatment to identify socio-economic predictors of energy consumption patterns, only a few useful characteristics were assessed (this poses a major aspect to be improved in future work within this research). However, to develop the methodology, we relied on this survey data. The surveys were conducted with any household resident available with the option to reject the answer to any question. Characteristics assessed (indicating the descriptive statistics of only valid answers in brackets behind) include housing occupant status (74,7% owner, 5% tenant), number of adults (median (Md): 2), number of children (Md: 2), monthly income (Md: 150,000 Ar ~ 30 €), housing wall type (Ravinala wood (40%), wood-concrete structure (18%), concrete/stone (22%), tin (2%)), housing roof type (tin (73%), leaves (9%), concrete (1%), floor type (concrete (77%), board (4%), appliance ownership (LED bulb, LED spot, TV, USB phone charger, 12 V plug), profession of the client (grouped into a trader (22.2%), farmer (31.6%), employee (6.8%), other (6.8%), public lighting (32.5%)). Notably, “public lighting” was included as a ‘profession’ by the stated purpose of electricity use in the client data. In addition to the socio-economic and demographic data, the historical tariff subscription option of residentials was identified from records. Notably, to mitigate the fact that data availability limitations within tariff records, we applied a machine leaning algorithm to calculate tariff subscriptions based on a multi-class problem. A detailed description of the method and it´s application in our analysis is available in the public project report [31]. For a description of the tariffs see Appendix Table A1.

We use advanced statistical analysis to identify socio-economic and demographic determinants of electricity consumption patterns, including preferred tariffs. First, we apply cluster analysis to historical electricity consumption data to identify common clusters of representative annual electricity consumption patterns. K-means clustering was used as the clustering method. K-means clustering is a machine learning algorithm used to partition a given dataset into k clusters based on the similarity of the data points [33]. It effectively identifies similarities between numerical data, defining distinct groups of patterns [33]. K-means clustering does not require uniform cluster densities and allows for multiple dimensional data [34]. Before applying the sensitive-to-outliers k-means algorithm, the data must be cleaned from outliers and re-scaled. Common min-max normalization was applied: Compared to other kinds of cluster analysis, the main shortcoming of the method is the challenge of pre-determining the appropriate number of clusters [34]. However, this drawback can be overcome by calculating similarity measures (silhouette score) [35]. Given that the data comprises numerical data, we use Gower as a dissimilarity measure (for mathematical equations see [36]).

Having established representative clusters of annual electricity consumption patterns within the representative sample group, we aim to identify socio-economic and demographic characteristics that can be used to predict the electricity consumption behavior of a specific community resident. Thus, we search for statistically significant predictors for cluster membership. Therefore, we use the Chi-square test or Fisher exact probability test (depending on the type of underlying variable) to identify socio-economic characteristics that significantly often occur within the distinct load profiles. The Chi-square test of independence, a nonparametric method, assesses the potential association between two categorical variables in a contingency table [37]. The test involves organizing variables into rows (variable $i$) and columns (variable $j$), with cells containing the total count of cases for each category pair. By comparing observed counts ($o_{ij}$) to expected counts ($e_{ij}$) for the sample size, the significant difference between expected and observed counts can be calculated (for mathematical equations see [38]). If the resulting $X^2$ is greater than the redefined critical $X^2$, the null hypothesis of independent variables may be rejected [37].

Like the Chi-square test, the Fisher exact test can assess the significance of the relationship between two categorical variables. The test calculates the probability of obtaining the observed distribution of frequencies or more extreme ones, assuming that the row and column marginal totals are fixed (i.e., the marginal totals are the same as those in the observed data). The Fisher's exact test provides an exact p-value, making it suitable for situations where the chi-square approximation might be unreliable due to small sample sizes [39].

The Chi-square test and the Fisher exact test assess the significance of the relationship between two variables. However, to quantify the strength of the significance, we calculate Cramer´s V, which is a normalized version of the Chi-square statistic. Cramer´s V effect size takes values between 0 and 1 with increasing relationship strength. Values up to 0.1 indicate weak association, values around 0.3 indicate moderate association, and values around 0.5 or higher indicate strong association [36]. In addition, it is determined whether the variables are characteristic of a single group, thus allowing it to be distinguished from the other two.

As a final output of the consumption data analysis (Step 1), we have a set of representative annual residential electricity consumption load profiles, which can be predicted by the residential community's socio-economic and demographic characteristics (including preferred tariffs).

2.2.2. Productive Use of Electricity Analysis

As a critical challenge when aiming to integrate PUE in off-grid energy systems, services needed by the local community and the respective PUE asset delivering the service must be identified. On the one hand, this ensures that the PUE asset will be operated, and electricity will be consumed sustainedly. On the other hand, the identification of needed services is required to ensure value creation for and improved development of the local community. Thus, PUE services and assets, that are relevant for the given context, i.e., significantly intertwined within existing value streams must be identified and characterized. Further, potential usage patterns of the PUE must be assessed, including factors influencing possible alterations to usage patterns.

In the case under investigation, a market assessment revealed rice hulling and ice production as relevant activities to be targeted with PUE as of their current dominance and importance in the local value chains. A DC rice huller and DC freezer were identified as the respective relevant PUE, technically feasible to be integrated into nanogrids. The required characteristics of the technical assets were identified during semi-structured interviews with key informants in the study area that already own respective assets, or the fossil alternatives (i.e., diesel-based rice huller), local market analysis, and previous market assessments conducted by Nanoé. Semi-structured interviews with open-ended questions based on the guidelines of Witzel [40] were introduced as a tool for information acquisition to understand the context of the study and the complex correlation of local issues. The interviews were used to capture i) the status of PUE, ii) prospects of PUE, iii) local value chains, iv) community structures, and – essentially – v) time-dependent usage patterns of PUE. A detailed evaluation of the interviews can be found in the publicly available project report of the ENERGICA project [31].

The load curve of the DC freezer was obtained from historical consumption data of a freezer (type: Steca PF166-H [41]) operated in the Ambohimena village, used to produce ice and conserve juice. The load curve of the rice huller was estimated by observing usage patterns and product flows of currently used diesel-based rice hullers. The interviews suggest a substantial seasonal variation in the use of the rice huller, ranging from 1 h to 2 hours per day in the rainy season up to 9 hours a day in June, which is the peak month of the harvest season. Based on the monthly production of rice hullers assessed via a survey of Nanoé, we interpolated the required rice to be processed in every respective hour of the year. We further assumed the operation of the rice huller would start at 6 am, as suggested by one interviewee, and finalize once the calculated output of the specific day has been reached (with a three-hour break between 11 am and 2 pm). Fitting the load curve in our hourly-based model, the rice huller would, for example, produce 84 kg paddy rice per hour between 6 am and 8 am in January but 71 kg/h per hour between 6 am and 11 am and again from 2 pm until 6 pm in June. The PUE's average daily electricity load patterns are illustrated in Figure A1, Figure A2 and Figure A3.

Due to time restrictions and the language barrier, it was only possible to record to a limited extent which factors, in addition to external influences, determine user behavior and to what extent user behavior is therefore variable. The answers received suggested that the user behavior of the rice huller is largely determined by the seasonal weather and vegetation cycles, but otherwise offers little flexibility.

We used Microsoft excel to transfer the information into numerical data and process the data. As final output of the PUE analysis, we established i) time series of the assumed electricity consumption over the year (8760 timesteps) and ii) technical and economic characteristics of the PUE assets as numerical data.

2.2.3. Scenario Formulation

In the scenario formulation, we perform the matchmaking of residential electricity consumption compositions and PUE asset electricity consumption pattern, as illustrated in Figure 3.

Figure 3. Scheme of the matchmaking performed, matching different residential electricity demand profile compositions with different PUE assets.

Figure 3. Scheme of the matchmaking performed, matching different residential electricity demand profile compositions with different PUE assets.

We first develop five distinct sets of residential nanogrids reflecting distinct residential community compositions as explained in Chapter 3. These differ in the constellation of the five residential load average profiles representing the respective electricity consumption pattern profile cluster identified during cluster analysis.

The five residential load sets will be matched with different PUE and respective load profiles. With this, we can derive valuable information on the fit of a PUE in specific residential energy systems, reflecting residents' distinct socioeconomic and demographic characteristics. Therefore, we match four PUE integration scenarios with the five distinct residential compositions. The PUE integration scenarios are (1) Base case residential nano grid without any integrated PUE according to the five distinct sets of residential load compositions ( C1-C5 in Figure 3.), (2) integration of a rice huller in a nano grid with residential load profiles C1 – C5, (4) integration of a freezer, and (3) integration of an unconstrained and flexible in operation rice huller in a nano grid. While the interviews insufficiently assessed the flexibility of the users in adopting energy system beneficial usage patterns, we additionally introduce a scenario which offers ‘total operational freedom’ to the rice huller (4). In this “flexible operation” scenario, we only constrain for a minimum required yearly rice output. Notably, this scenario poses an unrealistic extreme. However, it reflects (extreme) adoption of the user behavior to respect energy system constraints.

2.2.4. Energy System Modeling

We apply computational energy system modeling to derive a quantitative basis to evaluate the fit of PUE and different residential electricity consumption profiles. Via energy system modeling we obtain technical (i.e., component capacity and dispatch) and economic (cost) information of how a nanogrid serving the included residential and productive loads would ideally be designed.

We rely on the open energy modeling framework (oemof). For a detailed description of the framework, see Hilpert et al. [42]. Oemof is based on a graph-based approach, setting components and buses into a mathematical relationship, holding both technical and economic numerical data. With this, we can establish a mathematical representation of the energy systems and underlying economic characteristics to perform an economic optimization. We establish a linear problem to be optimized. For this analysis, we apply the minimization of the total annualized energy system costs (including capital and operational expenditures) as objective function. We estimate economic and technical characteristics of energy system asset based on local market data and key-informant interviews as summarized in Table 1 and Table 2. The project lifetime was assumed to be 10 years, while the weighted average costs of capital (WACC) are estimated with 10%. Note that the WACC may significantly differ depending on the entity investing in the energy system, i.e., in a corridor between close to 0% for local private companies receiving funding grants and up to 30% for local individuals. As input data for a time series of PV irradiation in hourly resolution, we rely on the MERRA-2 dataset with the reference year 2019. Data was accessed via [43].

Table 1. Economic parameters of energy system assets assumed in the analyses. *According to a market available product [41]. **While re-fitting diesel-based rice hullers with a DC motor is tested within the ENERGICA project, the assumed costs are in line with commercially available DC products [44].

Table 1. Economic parameters of energy system assets assumed in the analyses. *According to a market available product [41]. **While re-fitting diesel-based rice hullers with a DC motor is tested within the ENERGICA project, the assumed costs are in line with commercially available DC products [44].

Component	CAPEXfix	CAPEXvariable	OPEX
PV	101 €	540 €/kW	14 €/kW/yr
Battery	26 €	246 €/kWh	14 €/kW/yr
Supplementary components	306 €	-	9.2 €/yr
DC freezer*	1220 €	-
DC rice huller**	-	607 €/kW	28 €/kW/yr

Table 2. Technical characteristics of energy system assets assumed in the analyses.

Table 2. Technical characteristics of energy system assets assumed in the analyses.

Component	Parameter	Value
PV	Lifetime	10 y
	Optimal tilt	-29° [45]
	Loss fraction	10% [46]
Battery	Lifetime	3.5 y
	Efficiency	0.8
	SOC min	0.3
	C-rate	C/10
DC rice huller	Lifetime	5 y
	Conversion rate electricity to rice flour	70 kg/kWh
DC freezer	Lifetime	10

2.2.5. Evaluation

To evaluate the modeled scenarios, we further compute technical and economic measures of the optimized energy systems. As economic measures, we calculate the levelized costs of the entire energy system (LCOS), levelized costs of residential electricity consumption (LCOE_Residential), and levelized costs of providing the PUE service (LCOE_Service). The LCOS reflect the average costs per kWh of useful electricity the system generates. We calculate the LCOS by dividing the total annualized costs ($TAC$) by the amount of electricity served $Electricit{y}_{served}$.

$$LCOS=\frac{TAC}{Electricit{y}_{Served}}$$

(1)

The terminus $Electricit{y}_{served}$ includes the total energy delivered, including residential and PUE loads. In contrast, the levelized costs of electricity for residential loads $LCO{E}_{Resdidential}$ account for the average cost per kWh of useful electricity energy produced by the system to serve residential electric loads only. We divide the annualized costs of producing electricity (notably excluding any cost associated with the potential PUE loads) by the total electric load served.

$$LC{OE}_{Residential}=\frac{(TAC-{\mathrm{T}\mathrm{A}\mathrm{C}}_{\mathrm{a}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{t}}*\frac{Electricit{y}_{PUE}}{Electricit{y}_{Residential}}-TA{C}_{PUE})}{{Electricity}_{Residential}}$$

(2)

With ${\mathrm{T}\mathrm{A}\mathrm{C}}_{\mathrm{a}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{t}}$as total annualized costs of a specific energy system asset (i.e., PV, battery) and $TA{C}_{PUE}$ as the costs of the PUE asset itself and ${Electricity}_{PUE}$ [kWh/yr], and ${Electricity}_{Residential}$ as total electric power served to residential electric loads[kWh/yr]. Notably, as we assume simultaneously developing the energy system for residential load and PUE load, we include the costs for the PUE in the calculations, keeping an approach to optimize the entire energy system without taking a specific perspective (see Section 4.1 for a related discussion of alternative calculation methods). Vice versa, we compute the levelized costs of electricity for service of the PUE ($LC{OE}_{Service}$) accounting for the costs and energy share associated with the PUE load.

Notably, to calculate the share of costs of the PUE sub-system and residential electricity supply sub-system respectively (analogous for $LC{OE}_{Service}$and $LCO{E}_{residential}$) we take an objective technical perspective, sharing the costs of installation and use of the total system based on share of energy consumption (bottom-up). We therefore calculate the fraction of asset costs, e.g., PV investment costs, which are required to feed the PUE or residential electricity supply sub-systems respectively by relying on the share of PV electricity flows through each sub-system. While this technical approach is useful to evaluate the performance of the entire system it may differ from an approach taken by current off-grid system operators to calculate tariffs; see Section 4.1 for a related discussion.

3. Results

3.1. Influencers of Electricity Consumption

We observe the time-dependent evolution and variation of electricity consumption patterns (subsection 3.1.1) and time-independent variation of consumption patterns based on socio-economic characteristics (subsection 3.1.2).

3.1.1. Time-Dependent Influences of Electricity Consumption

We study the evolution of electricity consumption across 107 residential electricity users over three years (earliest data log: 10.02.2018; latest data log: 01.12.2021). It is important to note that the applied payment scheme foresees optional (daily, weekly, or monthly) prepayments, in which the user can choose between different credit options reflecting daily power and energy limits. When exceeding the daily energy limit, the user is remotely cut off and connected again when credits are left the next day. When exceeding the power limits, the user is cut off only shortly and reconnected if the power load is reduced. Hence, the studied electricity consumption patterns can be constrained by the tariff chosen and credit management of the household and may not reflect an unconstrained evolution. Table 3 presents the evolution of the average energy and power consumption per capita. Whereas the maximum average power demand per household remained more stable over the years, the average daily energy demand increased significantly. While a granular analysis confirms that the average energy demand of connected households increases over time, we also observed that more recently connected households tend to have higher average energy demands than clients connected several years ago (see also Figure A4 of the Appendix). This is explained by an increasing set of available DC appliances offered to local residents, and higher share of high consuming public lights integrated in the nanogrids.

Table 3. Evolution of the daily average energy consumption and peak power load per household.

Table 3. Evolution of the daily average energy consumption and peak power load per household.

Year	Average daily household electricity consumption [Wh]	Annual change of average daily electricity consumption [%]	Average daily maximum household power demand [W]	Annual change of maximum average power consumption [%]
2018	8.16	-	2.26	-
2019	21.88	168.27	2.25	-0.75
2020	35.47	62.11	2.27	0.92
2021	50.62	42.72	2.99	32.11

Studying the average monthly energy consumption and peak load per household reveals a seasonal variation of electricity consumption, see Figure 4. Both significantly increase after the rainy season (January – April). This is explained by the seasonality of crops, potentially increased liquidity of the households during these months, and less hours of sunshine.

Figure 4. Seasonal variation of the average maximum peak power demand per household and average daily electricity consumption.

Figure 4. Seasonal variation of the average maximum peak power demand per household and average daily electricity consumption.

3.1.2. Socio-Economic Predictors of Electricity Consumption

Based on the cluster analysis (see Section 2.1), we identify three distinct representative annual electricity residential consumption profiles (silhouette score s = 0.53). Figure 5 illustrates the average daily load profile of each household of the sample group within the distinct clusters, with the aggregated average load profile of all households belonging to a cluster highlighted in bold. Cluster 0, including 17% of the residential sample group, exhibits a significant evening peak demand, peaking at 15W around 8 pm, accompanied by a baseload demand of approximately 3 W throughout the rest of the day. Due to its comparatively high demand (ca. 40 Wh per day), we label the profile as “high-consumption”. The majority (66%) of residentials belong to Cluster 1 – “low-consumption” – in which we observe minimal daytime consumption, with a low night demand of around 1 W and a small evening peak of about 4 W. Cluster 2 – “nighttime consumption”- (17%) displays a moderate-sized evening peak of 10 W, a night consumption around 8 W, and no daytime consumption. Within these profiles, two extremes emerge: a low-demand consumer in Cluster 1, characterized by a consumption not exceeding 4 W and consistently lower than other groups, and a high-demand consumer in Cluster 0, with an evening peak demand four times higher than the low demand consumer and twice that of Cluster 2. Throughout the day, the high demand profile maintains the highest overall demand, dominated by a nearly constant medium baseload at 3 W.

Figure 5. Representative daily electricity consumption profile of each household in a cluster and the average cluster profile, highlighted in bold.

Figure 5. Representative daily electricity consumption profile of each household in a cluster and the average cluster profile, highlighted in bold.

Based on the series of Chi-square and Fisher tests, we detect a significant correlation between socioeconomic characteristics and cluster membership of residents in the community. Table 4 presents the socio-economic and demographic variables detected as significant to predict a distinct cluster membership (for an overview of the statistical analysis results, see Table A2. of the Appendix). Based on the results, we can characterize the distinct clusters and use the variables to predict a particular energy consumption pattern. For instance, a high share of “Eco” tariffs within the sample group significantly predicts a Cluster 1 membership – expressing a low load consumption load profile. Similarly, the public lighting tariff predicts the membership to a high consumption Cluster 2. While predicting energy consumption behavior based on tariffs is trivial, we find other more complex correlations. For instance, we see a high share of residents working as traders, which significantly correlates with an electricity consumption profile as reflected in Cluster 0, with a high peak during the evening. Most of the samples included in the low-consumption profile (Cluster 1) instead are stated they are farmers (47%). Cluster 0 households also report the highest number of LED bulbs (2.2). In addition, the presence of a phone charger (83%) and 12V plugs (56%) correlate with the membership in Cluster 0. LED spot ownership, however, predicts the membership to Cluster 2.

Table 4. Socio-economic correlation with cluster groups. EC = Expected count, C = Count. Statistically significant at p-value confidence level = 0.05. * Statistically significant at p-value confidence level = 0.1.

Table 4. Socio-economic correlation with cluster groups. EC = Expected count, C = Count. Statistically significant at p-value confidence level = 0.05. * Statistically significant at p-value confidence level = 0.1.

Variables		Cluster 0		Cluster 1		Cluster 2		Chi2	df	Fisher Exact	p-value	Cramer´s V
		C	EC	C	EC	C	EC
Tariff Groups
Eco	Yes	0	5.7	33	22.6	1	5.7	21.172	2		<0,001	0.445
	No	18	12.3	38	48.4	17	12.3
Eclairage Plus	Yes	9	4.7	17	18.6	2	4.7	7.585	2	6.98	0.025	0.275
	No	9	13.3	54	52.4	16	13.3
Multimedia	Yes	7	2	4	8	1	2.2	16.654	2	12.37	0.001	0.395
	No	11	16	67	63	17	16
Public Lighting	Yes	0	2	1	8	11	2	54.137	2	37.199	<0,001	0.711
	No	18	16	70	63	7	16
Tariff Switch	Yes	11	5	17	19.9	2	5	12.9	2		0.002	0.347
	No	7	13	52	51.1	16	13
Appliances Ownership
Led Bulb	Yes	16	14.1	61	55.7	7	14.1	20.201	2	16.635	<0,001	0.435
	No	2	3.9	10	15.3	11	3.9
Led Spot	Yes	0	2	1	8	11	2	54.137	2	37.199	<0,001	0.711
	No	18	16	70	63	7	16
USB Phone Charger	Yes	15	8.9	31	35.2	7	8.9	10.021	2	10.199	0.006	0.306
	No	3	9.1	40	35.8	11	9.1
12V Plug	Yes	10	5.7	21	22.6	3	5.7	6.749	2		0.034	0.251
	No	8	12.3	50	48.4	15	12.3
LED Bulb Quantity	0	2	3.9	10	15.3	11	3.9	40.883	12	35.687	<0,001	0.437
	1	6	7.9	40	31.2	1	7.9
	2	4	4	14	15.9	6	4
	3	3	1.2	4	4.6	0	1.2
	4	1	0.7	3	2.7	0	0.7
	5	1	0.2	0	0.7	0	0.2
	8	1	0.2	0	0.7	0	0.2
Demographic variables
Number of children*	0	1	2.9	13	10.9	1	1.3	16.175	8	13.283	0.065	0.31
	1	8	4.6	16	17.4	0	2
	2	3	4.4	19	16.7	1	1.9
	3	3	3.6	11	13.8	5	1.6
	4	1	0.6	2	2.2	0	0.3
Job Groups
Trader	Yes	9	5.4	17	14.8	0	5.8	13.263	2		0.001	0.429
	No	6	9.6	24	26.2	16	10.2
Employee*	Yes	4	1.7	4	4.6	0	1.8	5.751	2	4.959	0.056	0.283
	No	11	13.3	37	36.4	16	14.2
Public Lighting	Yes	0	2.5	1	6.8	11	2.7	40.226	2	31.89	<0,001	0.747
	No	15	12.5	40	34.2	5	13.3

3.2. Techno-Economic Evaluation of PUE

According to the findings of the statistical analysis in the previous section, we establish scenarios to compare the integration of PUE in residential energy systems with different compositions of residential consumption patterns as explained in Subsection 2.1.3. Hence, we match four PUE integration scenarios with five distinct residential load profile sets, each composing five households posing the average daily load profile of a certain cluster (see Figure 5). With this, we can derive valuable information on the fit of a PUE in specific residential energy systems, reflecting residents' distinct socioeconomic and demographic characteristics.

The residential load profile sets are labeled as

•

“Representative demand”: five residential loads, including three Cluster 1 loads (“low-consumption”) as the most common cluster, one Cluster 0 load (“high-consumption), and one Cluster 2 load (“nighttime consumption”). This set reflects the overall percentage distribution of all samples. Annual residential demand: 101 kWh.

•

“Low demand”: five residential loads of Cluster 1 (“low-consumption”). Annual residential demand: 43 kWh.

•

“High demand”: five residential loads of Cluster 0 (“high consumption”). Annual residential demand: 202 kWh.

•

“Low demand with nighttime load”: four residential loads of Cluster 1 (“low-consumption”) and one load with Cluster 2 profile representing a nighttime load (public lighting). Annual residential demand: 70 kWh.

•

“High demand with nighttime load”: four residential loads of Cluster 0 (“high consumption”) and one load with Cluster 2 profile representing a nighttime load (public lighting). Annual residential demand: 196 kWh.

We evaluate the integration of PUE within the different compositing residential energy systems on technical metrics in Figure 6 - taking the example of a representative residential electricity consumption patterns (see Table A3 of the Appendix for all scenarios and compositions) and economic metrics in Figure 7.

Figure 6. Annual energy generation (kWh/yr) (left) and share of energy consumption of residential loads, PUE loads, and excess generation across the different PUE integration scenarios for the example of the representative demand residential load profile set.

Figure 6. Annual energy generation (kWh/yr) (left) and share of energy consumption of residential loads, PUE loads, and excess generation across the different PUE integration scenarios for the example of the representative demand residential load profile set.

For interpreting the economic results illustrated in Figure 7 we may compare

i) the difference of $LCOS$within one PUE integration scenario across the different residential load profile sets to understand the suitability of the specific PUE for different communities

ii) the difference of $LCOS$ across different PUE integration scenarios within one specific residential load profile set to understand the best fitting PUE for the respective socio-economic character of the community

iii) the distribution of $LCOS$, $LCO{E}_{Resdidential}$ and $LCO{E}_{Service}$ within each combination of PUE integration and residential cluster composition to understand the share of costs associated with supplying electricity to the residential users or the PUE appliance.

Below, we highlight some of the key results derivable from Figure 7 per PUE integration scenario:

Figure 7. Economic results of the PUE integration scenarios across the different residential load composition sets.

Figure 7. Economic results of the PUE integration scenarios across the different residential load composition sets.

Scenario: Base case residential nanogrid: Supplying only residential loads, the optimized energy system considering a representative demand residential load profile set consists of 160 Wp PV and 560 Wh battery storage. The costs of the system are dominated by battery storage (43% of the TAC); see Figure 8. Considering a low demand profile set characteristic for a community significantly consisting of residents subscribing for the “Eco tariff” and characterized by owning only a few appliances (light bulbs) with a high share of farmers results in lower total system costs (110 €/yr. vs. 205 €/yr.). However, the relative costs of supplying electricity compared to an opposed high demand consumption expected in communities with a significantly increased share of traders, many change-in-tariffs towards multimedia tariff, and many LED bulbs identified are higher. This tendency can increase when considering public lighting as a nighttime load (high demand with nighttime load) to be included in the residential energy system with relatively high demand.

Scenario: Integrated rice huller: Integrating a DC rice huller following current consumption patterns of fossil alternatives into a residential energy system increases the TAC of the system by ten times (high demand residential load profile set) or up to 20 time (low demand residential load profile set). The costs of the rice huller – with an optimized capacity of 1.57 kW_el – account for 15% of the total system costs. However, with the PUE load dominating the share of energy consumed in the nanogrid (sixfold the residential consumption in a high demand residential load composition set, see Table A3), it also dominates the system costs. Hence, the benefits of integrating the rice huller towards reducing the $LCOS$depend on the residential load composition set, reflecting a different socio-economic and demographic community composition. With the rice huller relatively increasing the costs of electricity production (as seen by the $LCO{E}_{Service}$ exceeding the $LCO{E}_{Resdidential}$ under any residential load composition set), we observe increasing $LCOS$when including a rice huller in a high demand load profile composition set – while we observe beneficial effects to the $LCOS$when increasing the utilization of low demand residential load compositions. With splitting the costs across residential and PUE via energy shares (see Section 2.1), the $LCO{E}_{Resdidential}$ may decrease by 1.35 €/kWh (see a related discussion on cost distribution in Section 4). However, we must therefore carefully note that the integration of PUE can actually result in higher cost for residential electricity consumption (based on the applied calculation of electricity cost), if the two load profiles conflict.

Scenario: Integrated flexible rice huller: Assuming the DC rice huller to have total ‘operational freedom’ and only requiring a minimum throughput within one year is an unrealistic extreme (assuming abundant rice resources, and ubiquitous storage opportunities). However, some increased flexibility in the operation – ergo, some changes in the usage patterns of the asset users compared to current usage patterns of fossil counterparts – can arguably be expected to a certain degree to facilitate operational constraints imposed by the DC system (see Section 4 for a related discussion). Comparing the results with the ones of the constrained rice huller scenario shows the significant cost reduction achievable when increasing the flexibility of PUEs, which is unlocked by shifting the load of the PUE towards the peak PV irradiation hours with least conflicting residential loads; thus, avoiding costly energy storage (see Figure A2 for a representative load profile). The required power of the PUE asset is reduced to 0.6 kW_el (compared to 1.57 kW_el). Further, the additional amount of PV and battery to be installed is reduced compared to a residential system to satisfy the PUE load. While for example, in a representative demand residential household composition set, the optimal size and associated share of TAC of the PV and battery exclusively feeding residential loads are 160 Wp and 0.56 kWh, the size and costs (see Figure 8) increase to 3.43 kWp (20% of the TAC), and 13.28 kWh (60% of the TAC) respectively when integrating a DC rice huller following the load profile of fossil alternatives currently used (scenario: integrated rice huller). However, when maximizing the flexibility of the huller (scenario: integrated flexible rice huller), the required PV size increases to ca. 900 Wp compared to the residential system, while the battery size remains the same compared to feeding only residential loads. Hence, significant battery costs can be saved when increasing the flexibility of the PUE to be operated at peak irradiation times. Consequentially, with the amount of energy consumed in the nanogrid being dominated by the rice huller (see Table A3), achieving low costs of supplying the rice huller with electricity due to its operation harnessing excess electricity of the residential grid only, the $LCOS$ can be reduced to less than 0.3€/kWh. Significantly, by smoothening the load curve in low-demand consumption scenarios, increasing the total system utilization, and reducing the excess electricity share (see Table A3), the cost-efficiency measures are improved compared to a high demand residential load profile set (see Figure 7).

Scenario: Integrated freezer: Integrating a freezer into the energy system quadruples the TAC of the nanogrid when considering a low demand residential load composition set of the community (552 €/yr. vs. 139€/yr.) and triples the costs assuming a high demand residential community set (607 €/yr. vs. 205€/yr.) (see Figure 8). There is only a slight variance in the $LCO{E}_{Residential}$, suggesting that the applicable consumption composition has little impact on the cost of electricity provision for households. The comparison of the base case residential nanogrid only feeding residential loads and the integrated freezer scenarios reveals that through the integration of the freezer, the $LCO{E}_{Residential}$ can be reduced substantially. For the low demand residential load composition set, the $LCO{E}_{Residential}$ are reduced by 75% and for the high demand consumption composition the $LCO{E}_{Residential}$ are reduced by 43%. Compared to the integrated rice huller, the PV and battery optimized capacity and share of costs of the TAC are 1.1 kWp (23% of TAC) and 1.84 kWh (31% of TAC) – the latter of which is a sixth of the capacity required to satisfy the rice huller. Notably, the freezer device (150 W) constitutes a third of the TAC (198€/yr), see Figure 8.

Figure 8. Share of energy system asset costs of total annualized costs for the different residential load profile sets and PUE integration scenarios.

Figure 8. Share of energy system asset costs of total annualized costs for the different residential load profile sets and PUE integration scenarios.

4. Discussion

4.2. Implications of the Key Findings

This study shows that integrating PUE can reduce the costs of residential energy provision. However, for individuals, the costs of PUE assets may pose a significant barrier to it’s uptake. In the case of a DC freezer, one interviewee, whose family owns a DC freezer, reported paying off the microcredit over a duration of two years with monthly paybacks of 160,000 AR (ca. 33€). In addition, the electricity consumed by the freezer is charged via a dedicated tariff (we may approximate a fictive tariff with the $LCO{E}_{Service}$ to approximately 38,000 AR per week (ca. 8€)). The owner reports daily profits after paying bills of 5000 AR (ca. 1€) from selling a broad spectrum of goods, including iced water, ice for cooling food and beverages, and frozen juice. While the economic benefits for the PUE owner are thus evident, our analysis shows potential for other community members to gain financial benefits from the integration of PUEs when assuming different stacked tariffs are in place. We observed the difference in $LCO{E}_{Residential}$and $LCO{E}_{Service}$, during the analysis. Notably, the LCOE only reflects a part of the costs to be included when determining a tariff but may be seen as indicative for the tariff to be set. From our analysis, we can observe that the $LCO{E}_{residential}$ (significantly) decreases when integrating the PUE (integrated freezer), while the $LCO{E}_{Service}$, exceed the $LCO{E}_{Residential}$. To illustrate the consequences, we conduct a very simplified thought experiment. Considering a low demand residential composition set, the average annual electricity consumption per household is 10.6 kWh. When adopting the $LCO{E}_{Residential}$as a tariff, in a base case nanogrid ($LCO{E}_{Residential}$= 2.5€/kWh) the annual costs of electricity per household would be approximately 27€. We now consider the case of having five low demand loads, and one freezer integrated into the system, the operator of which is charged a tariff in the magnitude of the $LCO{E}_{Service}$,. Adopting the new $LCO{E}_{Residential}$found within the analysis for integrating a freezer ($LCO{E}_{Residential}$= ca. 0.6€/kWh), the annual costs of electricity to pay for each household are reduced by 74 percent to approximately 6.5 €, while the costs for operating the freezer would total to 533€ per year. While this calculation is a simplification, it shows the distributional monetary benefits amongst the community potentially to be unlocked when setting an appropriate tariff scheme. Thus, our analysis supports the suggestion from the literature [15] that the entire local economy may be improved when overcoming initial barriers to investing in the PUE to be integrated into the system. However, one must carefully note that we also reported the exact opposite and increasing $LCO{E}_{Residential}$ when PUE and residential load consumption patterns conflict, causing additional energy system costs (integrated rice huller in a high demand residential load composition set).

Our study highlights the importance of matching the electricity consumption patterns of the residential community and the PUE to minimize the costs of electricity production. The associated cost of electricity production varies significantly between the different scenarios and underlying household load composition sets considered in this case study. Figure 9 shows the $LCOS$and the respective excess electricity percentage share for every scenario and residential load profile composition set considered in this study. Significant deviations in the associated cost of supplying electricity can be observed for both different residential consumption composition, when considering a single selected PUE asset, which is indicated by a uniform colour code but different marker in Figure 9 and between different PUE assets, which are represented in different colours. The high variance among electricity generation costs underlines the potentials that arises from a ‘match making’ approach, as proposed in this study. The spread of costs within a specific PUE integration scenario across the different residential load composition sets can be interpreted as ‘robustness’ of the PUE load curve towards potentially conflicting residential load profiles. The results indicate that the robustness of the associated electricity provision cost can be increased when comparing non-integrated and integrated systems. The maximum spread of $LCOS$in the base case scenarios is 1.5 €/kWh – whereas it is only 0.13 €/kWh for the integrated freezer scenario and the integrated rice huller scenario. The economic impact of integrating a rice huller into a residential energy system in flexible rice huller scenario is even less sensitive towards the load profiles of the residentials than a freezer with a spread of only 0.02 €/kWh. This expresses that integrating a PUE in a residential off-gid energy system decreases the sensitivity of the economic results (measured by $LCOS$) towards underlying residential community electricity consumption patterns.

Figure 9. LCOS and percentage share of excess electricity for scenarios considered in the relevant case study.

Figure 9. LCOS and percentage share of excess electricity for scenarios considered in the relevant case study.

The proposed matchmaking approach necessitates information regarding the locally relevant PUE and the associated operational behaviour. In our study we propose an approach using survey data and statistical analysis to identify consumption determinants and calculate representative consumption profiles. Provided sufficient data can be collected to ensure a high degree of informative value, the identified determinants can be an efficient short-cut in identifying favourable locations of integrated energy systems. Further statistical verification may expand the area of application to a broader geographical scope, beyond the surveyed region. The appropriate selection of PUE for a given location ultimately requires close collaboration with the community that the energy system serves. The identification of existing value streams, the selection of relevant PUE and through understanding of operational preferences, which can be achieved through KIIs, is a key requirement for the efficient integration of households and PUE loads. The long-term success of a deployed energy system is substantially influenced by the accuracy of the assumptions made regarding the energy-related needs in the community while planning the energy system. To improve the accuracy and to increase the overall value of the energy system, it is essential to include the community in energy system planning. We explored the potential of adapted operational behavior for PUE. To showcase the impact on electricity provision cost, we defined a PUE asset integration case in which its operation did not collide with residential patterns and to harness peak PV irradiation. In line with current literature (e.g., [21]), our study underscores the substantial impact of increasing the operational flexibility of PUEs on decreasing energy system costs. This is reflected in the fundamental differences in the percentage share of excess electricity when comparing the rice huller and the flexible rice huller in Figure 9. While the assumed operational adaptation is not realistically achievable, it showcases the theoretical potential that exists. The actual operational preferences and willingness to adapt can only be determined in close collaboration with the PUE operator and may be linked to further product and value streams within the community. To even extend the importance of community participation in energy system planning and a productive, democratic, and constant exchange between energy system investor and it´s users, we may consider the earlier (see Chapter 1) mentioned risk of the PUE associated business failure being earlier than the payback of the energy system. Such financial risks may be mitigated when constantly being in exchange with the user of the energy system and providing supportive measures where relevant.

4.3. Considerations on PUE Impact on Value Streams of the Community and its Environment

Our results highlight the crucial importance of including the local community in the decision making and energy system design processes. First, energy system planners must identify a PUE asset that is relevant to address the service needs of the community to ensure a sustained uptake of the service, and electricity consumption. Hence, a close interaction with the community members is inevitable. Second, identifying preferences in and the degree of flexibility of usage patterns of PUE users can support aligning the PUE load profile with residential load profiles, which is critical to minimize the costs of electricity supply (see Section 3.2). In the qualitative data acquisition and in our analysis, we focused on the local community directly interacting with the energy system. These considerations offer new insights into more relevant and more efficient designs of off-grid energy system. However, we recommend that future studies further systematically integrate considerations regarding the external environment, in which the community is set. The interaction of the community with the external environment further improves the information value of the conducted analysis in several ways. First, the potential for a flexible operation of the PUE asset is constrained by the product and value streams of local markets. Second, the identification of relevant service needs (see Section 2.1.2) is linked to services that may exist outside the scope of the community. Third, the impact of introducing PUE assets, in terms of development, may have implications beyond the community, as it will change existing value and product streams.

To showcase the relevance of integrating considerations regarding the environment in which the community is set, in the following, we discuss potential implications. The results of our economic analysis (see Section3.2) suggest that the integration of PUE can improve the economics of off-grid energy systems. As suggested in our discussion, these benefits could further be distributed to residential energy system users. Additionally, we provided evidence of potential improved household income (Section 4.2) when investing in PUE. It is important to note that in this analysis we only consider improved household income via potential savings of energy expenditures. A detailed description of additional income generated by the PUE products is laid out in a report of the associated ENERGICA project [31]. However, we must keep in mind that the integration of PUEs impacts the energy system operator and the PUE user and may disrupt the local value stream patterns of the community and its external environment. The integration of PUE in energy systems, communities, and energy access projects is complex, given the multidimensional and bilateral relation of the PUE with the ecosystem embedded in. Previous research has developed the causal relation of PUE integration in energy access projects, identifying risks, preconditions and external factors impacting the implementation of PUE in projects [15,48]. Riva et al. [17] show the complex causal loops associated with energy access and productive activities. Based on interviews with several community members in our case study, we can simplified illustrate changes in the local community value changes when for example, considering the uptake of a freezer in a village (Figure 10). A corresponding description for the case of an electrical rice huller is documented in the associated ENERGICA project [31].

Figure 10. Simplified schematic representation of community value flows when adopting a freezer in a village. For an analogous description of changes in community dynamics for electrified rice processing machine see [31].

Figure 10. Simplified schematic representation of community value flows when adopting a freezer in a village. For an analogous description of changes in community dynamics for electrified rice processing machine see [31].

Figure 10 depicts the systems and value streams of the environment associated with the product ‘ice’ that is anticipated in a scenario that introduces decentralized ice production via a nanogrid-connected freezer. The value streams that existed prior to the availability of locally produced ice are disrupted. It is expected that ice buyers (i.e., grocery shops and bars) that previously purchased ice from mobile retailers, who are transporting ice from large factories in nearby cities, may now buy ice that is produced locally. Hence, one must note that the introduction of PUE in a decentralized grid and the associated increased local economic activity may lead to other stakeholders experiencing negative consequences (e.g., ice delivery or fuel delivery). Consequently, it is not given per se that all local stakeholders profit from increased uptake of PUE. For the planning of off-grid energy systems, it is therefore relevant to be aware of the entire associated value chains of products and thoroughly establish potential negative consequences. The importance of carefully weighing potential system disruptions is acknowledged in literature. Literature shows that the uptake of PUE may increase inequalities or may have negative impacts the income of some parts of the population. For example, Khandker et al. [49] find comparatively wealthier families that can afford to invest in PUE to increase their income due to PUE usage. In contrast, families that cannot afford a PUE remain comparatively poor. Further, substituting labor-intensive jobs with the PUE appliances threatened the jobs of low-income families. This also applies to eradicating jobs associated with fossil fuels when implementing renewable energy sources, including (i.e. selling candles or kerosene [50]). Prominently, the threat of increasing inequalities associated with PUE applies to gender-based inequalities. Many PUE stimulate activities and professions that either men or women predominantly carry out. Hence, a discrepancy in income stimulation from PUE is evident in the literature [48]. Given the risk of unintended consequences, in particular with regard to reinforcing local inequalities, the type of employment should also be evaluated to determine whether a certain PUE actually contributes to livelihood improvements [51]. In addition, Terrapon-Paff et al. [48] highlight including dedicated activities to raising awareness amongst stakeholders across the entire value chain to prevent from causing inequalities when implementing PUE.

5. Conclusions

References

1. Frigo, G.; Baumann, M.; Hillerbrand, R. Energy and the Good Life: Capabilities as the Foundation of the Right to Access Energy Services. Journal of Human Development and Capabilities 2021, 22, 218–248. [Google Scholar] [CrossRef]
2. González-Eguino, M. Energy Poverty: An Overview. Renew. Sust. Energ. Rev. 2015, 47, 377–385. [Google Scholar] [CrossRef]
3. Arto, I.; Capellan-Perez, I.; Lago, R.; Bueno, G.; Bermejo, R. The Energy Requirements of a Developed World. Energy Sustain Dev 2016, 33, 1–13. [Google Scholar] [CrossRef]
4. McCollum, D.L.; Echeverri, L.G.; Busch, S.; Pachauri, S.; Parkinson, S.; Rogelj, J.; Krey, V.; Minx, J.C.; Nilsson, M.; Stevance, A.-S.; et al. Connecting the Sustainable Development Goals by Their Energy Inter-Linkages. Environ. Res. Lett. 2018, 13, 033006. [Google Scholar] [CrossRef]
5. Fuso Nerini, F.; Tomei, J.; To, L.S.; Bisaga, I.; Borrion, A.; Spataru, C.; Castán Broto, V.; Mulugetta, Y. Mapping Synergies and Trade-Offs between Energy and the Sustainable Development Goals. Nat. Energy 2017, 3, 10–15. [Google Scholar] [CrossRef]
6. Schöne, N.; Timofeeva, E.; Heinz, B. Sustainable Development Goal Indicators as the Foundation for a Holistic Impact Assessment of Access-to-Energy Projects. J. sustain. dev. energy water environ. syst. 2022, 10, 1–24. [Google Scholar] [CrossRef]
7. Dall-Orsoletta, A.; Cunha, J.; Araújo, M.; Ferreira, P. A Systematic Review of Social Innovation and Community Energy Transitions. Energy Research & Social Science 2022, 88, 102625. [Google Scholar] [CrossRef]
8. United Nations Sustainable Development Goals Knowledge Platform Available online:. Available online: https://sustainabledevelopment.un.org (accessed on 11 July 2022).
9. IEA Tracking SDG7: The Energy Progress Report. 2023.
10. Liu, Y.; Bah, Z. Enabling Development Impact of Solar Mini-Grids through the Community Engagement: Evidence from Rural Sierra Leone. Energy Policy 2021, 154, 112294. [Google Scholar] [CrossRef]
11. Morrissey, J. The Energy Challenge in Sub-Saharan Africa: A Guide for Advocates and Policy Makers: Part 2: Addressing Energy Poverty; Oxfam Research Backgrounder, 2017; pp. 23–44. [Google Scholar]
12. ESMAP Mini Grids for Half a Billion People: Market Outlook and Handbook for Decision Makers; 2019.
13. Azimoh, C.L.; Klintenberg, P.; Mbohwa, C.; Wallin, F. Replicability and Scalability of Mini-Grid Solution to Rural Electrification Programs in Sub-Saharan Africa. Renew. Energy 2017. [Google Scholar] [CrossRef]
14. Agenbroad, J.; Carlin, K.; Ernst, K.; Doig, S. Minigrids in the Money: Six Ways to Reduce Minigrid Costs by 60% for Rural Electrification; Rocky Mountain Institute, 2018. [Google Scholar]
15. Terrapon-Pfaff, J.; Gröne, M.-C.; Dienst, C.; Ortiz, W. Impact Pathways of Small Energy Projects in the Global South – Findings from a Systematic Evaluation. Renew. Sust. Energ. Rev. 2018, 95, 84–94. [Google Scholar] [CrossRef]
16. Kooijman-van Dijk, A.L. The Role of Energy in Creating Opportunities for Income Generation in the Indian Himalayas. Energy Policy 2012, 41, 529–536. [Google Scholar] [CrossRef]
17. Riva, F.; Ahlborg, H.; Hartvigsson, E.; Pachauri, S.; Colombo, E. Electricity Access and Rural Development: Review of Complex Socio-Economic Dynamics and Causal Diagrams for More Appropriate Energy Modelling. Energy for Sustainable Development 2018, 43, 203–223. [Google Scholar] [CrossRef]
18. Madon, R. Energy, Poverty, and Gender: Impacts of Rural Electrification on Poverty and Gender in Indonesia Volume 1: Facts, Analysis, and Recommendations.; ENPOGEN, MARGE France for The World Bank, ASTAE: Washington, DC, 2003. [Google Scholar]
19. ESMAP Productive Use of Energy – PRODUSE Measuring Impacts of Electrification on Small and Micro-Enterprises in Sub-Saharan Africa; 2013.
20. Domegni, K.M.S.; Azouma, Y.O. Productive Uses of Energy: A Solution for Promoting Energy Justice in Rural Areas in West Africa. Renewable and Sustainable Energy Reviews 2022, 160, 112298. [Google Scholar] [CrossRef]
21. van Hove, E.; Johnson, N.G.; Blechinger, P. Evaluating the Impact of Productive Uses of Electricity on Mini-Grid Bankability. Energy for Sustainable Development 2022, 71, 238–250. [Google Scholar] [CrossRef]
22. Little, M.; Blanchard, R. Pre-Feasibility Methodology to Compare Productive Uses of Energy Supplied by Stand-Alone Solar Photovoltaic Systems: A Tanzanian Case Study. Energy for Sustainable Development 2022, 70, 497–510. [Google Scholar] [CrossRef]
23. Beath, H.; Hauser, M.; Sandwell, P.; Gambhir, A.; Few, S.; Chambon, C.L.; Nelson, J. The Cost and Emissions Advantages of Incorporating Anchor Loads into Solar Mini-Grids in India. Renewable and Sustainable Energy Transition 2021, 1, 100003. [Google Scholar] [CrossRef]
24. Booth, S.; Li, X.; Baring-Gould, I.; Kollanyi, D.; Bharadwaj, G.; Weston, P. Productive Use of Energy in African Micro-Grids: Technical and Business Considerations; NREL, Energy 4 Impact; 2018. [Google Scholar] [CrossRef]
25. Kyriakarakos, G.; Balafoutis, A.T.; Bochtis, D. Proposing a Paradigm Shift in Rural Electrification Investments in Sub-Saharan Africa through Agriculture. Sustainability 2020, 12, 3096. [Google Scholar] [CrossRef]
26. Farthing, A.; Rosenlieb, E.; Steward, D.; Reber, T.; Njobvu, C.; Moyo, C. Quantifying Agricultural Productive Use of Energy Load in Sub-Saharan Africa and Its Impact on Microgrid Configurations and Costs. Applied Energy 2023, 343, 121131. [Google Scholar] [CrossRef]
27. Louie, H.; Dauenhauer, P. Effects of Load Estimation Error on Small-Scale off-Grid Photovoltaic System Design, Cost and Reliability. Energy Sustain Dev 2016, 34, 30–43. [Google Scholar] [CrossRef]
28. Hunink, Y.T.A.; Kamp, L.M.; Blom, E.M. Co-Creating Responsible Energy Systems. In ISGW 2018 Compendium of Technical Papers; Pillai, R.K., Ghatikar, G., Seethapathy, R., Sonavane, V.L., Khaparde, S.A., Yemula, P.K., Chaudhuri, S., Eds.; Lecture Notes in Electrical Engineering; Springer Singapore: Singapore, 2020; Vol. 580, pp. 329–342. ISBN 978-981-329-118-8. [Google Scholar] [CrossRef]
29. McGookin, C.; Ó Gallachóir, B.; Byrne, E. Participatory Methods in Energy System Modelling and Planning – A Review. Renewable and Sustainable Energy Reviews 2021, 151, 111504. [Google Scholar] [CrossRef]
30. The World Bank Access to Electricity (% of Population) - Madagascar, Arab World. Available online: https://data.worldbank.org/indicator/EG.ELC.ACCS.ZS?locations=MG-1A.
31. Mwasaru, J.; Schöne, N.; Britton, T.R. D2.4 Energy Flow Patterns for Productive Uses of Energy, ENERGICA Consortium.
32. Grandini, M.; Bagli, E.; Visani, G. Metrics for Multi-Class Classification: An Overview. 2020. [Google Scholar] [CrossRef]
33. Sterling, T.; Anderson, M.; Brodowicz, M. High Performance Computing: Modern Systems and Practices; Elsevier, 2018; ISBN 978-0-12-420158-3. [Google Scholar]
34. Schöne, N.; Greilmeier, K.; Heinz, B. Survey-Based Assessment of the Preferences in Residential Demand Response on the Island of Mayotte. Energies 2022, 15, 1338. [Google Scholar] [CrossRef]
35. Rousseeuw, P.J. Silhouettes: A Graphical Aid to the Interpretation and Validation of Cluster Analysis. Journal of Computational and Applied Mathematics 1987, 20, 53–65. [Google Scholar] [CrossRef]
36. Field, A. Discovering Statistics Using IBM SPSS Statistics; 5; SAGE Publications Ltd: Sussex, 2017; ISBN 978-1-5264-4578-0. [Google Scholar]
37. Siegel, S. Nonparametric Statistics. The American Statistician 1957, 11, 13–19. [Google Scholar] [CrossRef]
38. McHugh, M.L. The Chi-Square Test of Independence. Biochem Med 2013, 143–149. [Google Scholar] [CrossRef]
39. Haschek, W.; Rousseaux, C.; Wallig, M. Handbook of Toxicologic Pathology; Elsevier, 2002; ISBN 978-0-12-330215-1. [Google Scholar]
40. Witzel, A. The Problem-Centered Interview. In Qualitative Forschung in der Psychologie: Grundfragen, Verfahrensweisen, Anwendungsfelder; Beltz: Weinheim; ISBN 3-407-54680-7.
41. Steca PF166-H. Available online: https://www.steca.com/Steca-PF-166-H-PF-240-H.
42. Hilpert, S.; Kaldemeyera, C.; Krien, U.; Günther, S.; Wingenbach, C.; Plessmann, G. The Open Energy Modelling Framework (Oemof) - A New Approach to Facilitate Open Science in Energy System Modelling. Energy Strategy Reviews 2018, 22, 16–25. [Google Scholar] [CrossRef]
43. renewables. Available online: https://www.renewables.ninja (accessed on 11 July 2022).
44. Agsol Micromill; 2023.
45. Jacobson, M.Z.; Jadhav, V. World Estimates of PV Optimal Tilt Angles and Ratios of Sunlight Incident upon Tilted and Tracked PV Panels Relative to Horizontal Panels. Sol. Energy 2018, 169, 55–66. [Google Scholar] [CrossRef]
46. Pfenninger, S.; Staffell, I. Long-Term Patterns of European PV Output Using 30 Years of Validated Hourly Reanalysis and Satellite Data. Energy 2016, 114, 1251–1265. [Google Scholar] [CrossRef]
47. Yilmaz, S.; Xu, X.; Cabrera, D.; Chanez, C.; Cuony, P.; Patel, M.K. Analysis of Demand-Side Response Preferences Regarding Electricity Tariffs and Direct Load Control: Key Findings from a Swiss Survey. Energy 2020, 212, 118217. [Google Scholar] [CrossRef]
48. Terrapon-Pfaff, J.; Gröne, M.-C.; Dienst, C.; Ortiz, W. Productive Use of Energy – Pathway to Development? Reviewing the Outcomes and Impacts of Small-Scale Energy Projects in the Global South. Renewable and Sustainable Energy Reviews 2018, 96, 198–209. [Google Scholar] [CrossRef]
49. Khandker, S.R.; Barnes, D.F.; Samad, H.A. Welfare Impacts of Rural Electrification: A Case Study from Bangladesh. World Bank Policy Research Working Paper 4859. 2009. [Google Scholar]
50. Rao, N.D.; Agarwal, A.; Wood, D. Impacts of Small-Scale Electricity Systems: A Study of Rural Communities in India and Nepal. 2016. [Google Scholar]
51. Poor People’s Energy Outlook 2012. Practical Action Publishing. UK; 2012.; Practical Action, 2012.