1. Introduction

2. Related Work

3. Methods

3.1. Energy Harvesting System

The battery is generally used to supply energy for nodes in WSN, but it is necessary that the battery limited energy storage is timely replaced and supplied to ensure the continuous normal operation of the node. In practical application, the real environment is often very complex, which brings great difficulties to lay the power wire or replace the battery for each node, and therefore, it largely limits the universality and flexibility of WSN applications. To solve this problem, we use the energy harvesting technology to realize the autonomous power supply of WSN.

At present, the sources of energy harvesting technology generally include solar energy, electromagnetic radiation energy, mechanical vibration energy and thermal energy. The solar harvesters can obtain several milliwatts of power per square centimeter, which can meet the application demand of wireless sensor nodes. However, the solar cells can convert energy output by the panel only if the light is sufficient, and a single solar energy cannot guarantee a sustainable and reliable power supply for nodes. The energy harvester based on mechanical vibration has been the attention of scholars at home and abroad, because the vibration energy is widely found in the natural environment. The main ways of vibration-to-electric energy conversion are as follows: piezoelectric, electromagnetic and electrostatic. Among them, the basic principle of piezoelectric vibration energy harvesting is: under the excitation of the external vibration source, the mechanical deformation of the piezoelectric material causes the movement of its internal electrons to generate electric energy. Piezoelectric has the advantages of simple conversion structure, fast response speed and high energy density, which is more and more favored by research institutions and high-tech enterprises. The piezoelectric power generation mechanism relies on the d₃₁ and d₃₃ effects of piezoelectric materials. Among them, the cantilever type piezoelectric generator utilizes the d₃₁ effect, which demonstrates a remarkable power output [30-31]. Additionally, the cantilever structure is characterized by its simplicity, high flexibility, and ability to achieve a low natural frequency. Currently, numerous studies [32-34] have been conducted on the vibration generation of piezoelectric cantilever beams, with a primary focus on investigating the impact of piezoelectric structure, size, and parameters on power generation. Considering the cost and reliability of the energy harvesting system, this paper organically combines the piezoelectric technology and the solar technology to ensure the energy supply to the sensor nodes.

3.1.1. Overall Scheme

Energy harvesting technology can collect the ubiquitous energy such as thermal energy, solar energy, wind energy and electromagnetic energy into electrical energy through the appropriate power electronic equipment and the voltage conversion circuit. At present, there are many more developed energy harvesting technologies, and the main source of its energy is generally mechanical vibration energy, electromagnetic radiation energy, solar energy and heat energy and other energy. In order to realize the self-power supply demand of each sensor node in WSN, we use solar energy and mechanical energy as the energy source of energy harvesting technology to realize the conversion to electric energy. The framework diagram of node energy supply in WSN is as shown in Fig. 2.

3.1.2. Design of Power Supply Circuit

The power supply circuit is mainly divided into three parts. The first is the solar power generation circuit, as shown in Fig. 3(A). The solar panel can output about 5V DC voltage, and the large capacitance is used to maintain the stability of the voltage. The second is piezoelectric ceramic power generation circuit, as shown in Fig. 3(B). We use the MP1541 chip to design voltage boost conversion circuit with a fixed frequency and peak current mode. The third is electric energy storage circuit, as shown in Fig. 3(C). We use the TP4056 chip to design charging circuit completed the power supply to the lithium battery.

Figure 2. Framework diagram of node energy supply in WSN.

Figure 2. Framework diagram of node energy supply in WSN.

Figure 3. Diagram of power supply module (A): solar power generation circuit and (B): piezoelectric ceramic power generation circuit and (C): electric energy storage circuit.

Figure 3. Diagram of power supply module (A): solar power generation circuit and (B): piezoelectric ceramic power generation circuit and (C): electric energy storage circuit.

3.2. Neighbor Discovery Mechanism

3.2.1. Low Duty Cycle

In WSN, nodes must be in a listening state to receive the data from the neighbor node. However, nodes are in idle listening state for most of the time, and the time used for data transmission is often very short, while idle listening causes a lot of invalid energy consumption. Therefore, in many practical applications, nodes adopt low-power communication chips with alternating listening and hibernation modes. It can reduce the unnecessary energy consumption of nodes by replacing idle listening through dormancy.

Considering the application in practice, the low duty cycle mechanism is introduced into WSN. Where, duty cycle is the ratio of periodic dormancy to activity time, and low duty cycle is that duty cycle is not greater than 10%. When the node works in the duty cycle mode, the time will be divided into the time slot of the same size. At a certain number of time slots, the RF will enter the active state that can send and receive data, and the RF will be turned off and enter the dormant state after the time slot ends. The low duty cycle mechanism greatly extends the life cycle of the network and reduces the energy consumption caused by idle listening.

However, under the low duty cycle mechanism, nodes are in sleep state for most of the time. But, nodes are required to open RF as much as possible to communicate with other nodes in order to quickly complete the mutual discovery between nodes. Therefore, saving energy consumption and reducing the detection delay are contradictory, and it is very necessary to study the neighbor discovery problem in the low duty cycle wireless sensor networks.

3.2.2. Sending and Receiving Separation Beacon Mechanism Based on Reply to ACK

The neighbor discovery protocol focuses on optimizing the scheduling period when the node works, and it determines the frequency of active time-slot encounters between neighboring nodes. While the beacon mechanism determines the probability of achieving mutual discovery when neighbor nodes meet in an active time slots. At present, most of the studies focus on how to optimize the work scheduling cycle of nodes. There are few studies around the beacon mechanism. In fact, the beacon mechanism also plays a very important role in the discovery of neighbor nodes.

Most of the existing beacon mechanisms in WSN are used for sending and receiving beacons in a single time slot. The sending of the beacon will have a certain constraints on the variation range of the time slot size and the listening time in the time slot, which brings serious effects on the mutual discovery between nodes, especially when the time slot is small. Therefore, we will propose a sending and receiving separation beacon mechanism based on reply to ACK, which only listens to the beacon in the active time slot, and sends the beacon at the end of its previous time slot. In addition, the node immediately replies to ACK after receiving the beacon sent by the adjacent node.

The sending and receiving separation beacon mechanism based on reply ACK is as shown in Fig. 4. Where, nth Slot represents the nth time slot, and it is active. At this point, the node will turn on the RF at the end of the (n-1)th slot and transfer the RF from the launch completion state to the send beacon state. Then, the beacon is sent before the end of the (n-1)th slot and the RF is converted to the listening beacon state, and the node will only listen to the beacon in the active time slot. When node A meets node B in an active time slot, node A receives the beacon sent by node B, and it immediately enters the stage of replying to ACK after analysis and verification processing. The bi-directional discovery between nodes can be completed after node B successfully receives ACK.

Figure 4. Time slot model of sending and receiving separation beacon mechanism based on reply ACK.

Figure 4. Time slot model of sending and receiving separation beacon mechanism based on reply ACK.

Discovery probability is an important indicator to measure the performance of the discovery beacon mechanism of neighbor nodes. Improving the discovery probability between nodes can reduce the detection delay and realize the rapid discovery between nodes. When the response ACK mechanism is adopted, the bidirectional discovery can be completed by dynamically extending the time slot after realizing one-way discovery. The bidirectional discovery probability P_d is:

$$P_d=\frac{t_{slot}-t_{PR}}{t_{slot}-t_{TX}}$$

(1)

where, t_slot is the time of an active time slot for a node. t_PR is the time that the node takes to send the synchronization packet. t_TX is the time interval between sending the beacon and listening the beacon.

3.3. Data Fusion Mechanism Based on Integer Wavelet Transform

Data fusion is a very important technology in WSN and is a research hotspot. This technology can process a large amount of raw data collected by sensor nodes in various networks through a certain algorithms, remove the redundant information, and transmit only a small number of meaningful processing results to the convergence node. The use of data fusion technology can greatly reduce the amount of data needed to be transmitted in WSN, reduce data conflicts, reduce network congestion, and thus effectively save energy costs, and prolong the life of the network.

In WSN, the perceptual data collected by nodes from the environment is a time-series data, as a strong time correlation and periodicity because of the short interval, while the data set generated by the whole network has a certain spatial correlation due to the close distance. For spatial correlation between data, integer wavelet transform can also be used to remove redundant parts.

3.3.1. Integer Wavelet Transform

Spatiotemporal data have strong linear features, which consists of multiple curves and is well-suited for applying the wavelet transform. After wavelet transformation, most of the energy of the data is stored in the low frequency component, and the detail information is stored in the high frequency component. It does well in removing the redundant information from the original data.

The integer wavelet transform is a new wavelet construction method, and has the following advantages: (1) It has only the integer shift and addition and subtraction operation, processes data quickly, has low hardware requirements, and is easy to implement. (2) It is completely reversible, which can perform both lossy coding and lossless coding. There are 12 commonly used integer wavelet filters, which have different computational complexity and compression performance. Among them, 5/3 wavelet has a small calculation amount and a good compression effect, which is often used to do lossless compression of spatiotemporal data [35]. Its transformation formula is as follows:

$$\begin{array}{l}{d}_{1,j}=s_{0,2i+\_1}-⌊\frac{s_{0,2i}+s_{0,2i+2}}{2}⌋ \\ s_{1,j}=s_{0,2i}+⌊\frac{d_{1,j-1}+d_{1,j+2}}{4}⌋ \end{array}$$

(2)

Where, s_i,j is the j-th approximation of the k-level transformation decomposition, d_i,j is the j-th detail value of the k-level transformation decomposition, ⌊·⌋ means rounding down. In practice, we need to use symmetric periodic extension for boundary data.

3.3.2. Select Sink Node and Construct Hierarchical Structure

(1) Select Sink node

The Sink node undertakes a lot of operations of data processing and forwarding within the group. Often, the node with strong ability is selected as the Sink to complete the communication function of the group and extend the life cycle of network at the same time. When the Sink node is selected, three factors need to be considered: residual energy, link quality and degree. Therefore, we set a weight w, and then the Sink is finally determined based on the calculated w value. The calculation of w can be defined as:

$$w_i=\alpha \cdot \frac{E_i}{\frac{1}{n}{\displaystyle \sum _{j=1}^n{E}_j}}\cdot \frac{E_i}{E_0}+\beta \cdot \frac{{\displaystyle \sum _{j=1}^n{L}_{ij}}}{L_{\max }}+\gamma \cdot \frac{D_i}{D_a}$$

(3)

where, E_iis the residual energy of the node v_i, E₀ is the initial energy of nodes, E_jis the residual energy of the neighbor node, L_ij is the link quality between the node v_i and the neighbor node, L_max is the maximum value of link quality in the group, D_i is the degree of node v_i, D_a is the average of degrees of nodes in the group, ɑ, β, γ is the adjust parameters, and satisfy: ɑ + β + γ = 1.

(2) Construct hierarchical network structure

The nodes in the group transfer the data to the convergence nodes by a hop transmission, and then the convergence nodes should forward these data to the Sink node. To achieve this submission mode, it is required to build a hierarchical network structure.

Figure 5. Diagram of hierarchical network structure in WSN.

Figure 5. Diagram of hierarchical network structure in WSN.

As shown in Fig. 5, the white nodes are the ordinary nodes in the network, and they can be layered by the message flooding of Sink nodes. In this process, the Sink node broadcasts a request message to all adjacent nodes, and these nodes that have received the message need to set their layer value to 1 and continue to broadcast the message to the surrounding nodes. After receiving the message, the surrounding nodes should first check whether their layer value has been set. If not, their own layer value adds 1 on the basis of the broadcast layer value, that is, their layer value is 2. These nodes with level 2 continue to broadcast this message, and thus the third layer node will be formed, and then all nodes will finally have their own layer value.

3.3.3. Data Fusion Based on Integer Wavelet Transform

The data collected by a single node have strong temporal correlation due to the short interval, while the data generated by multiple nodes in the network have some spatial correlation due to their close distance. We used integer wavelet transform techniques to remove spatio-temporal dependent redundancy of these data.

(1) Time-dimensional data fusion of single node

The operation process of time-dimensional data fusion using integer wavelet transform is as follows: Step 1, Data preprocessing. It mainly carries out data cleaning and data integer transformation on the raw data collected by the nodes. Step 2, Integer wavelet transform. The time dimensional data series is decomposed into low and high frequency coefficients based on integer wavelet transform. Step 3, Coefficients quantification. The wavelet coefficients were quantified by using scalar quantification techniques. Step 4, Encoding. The wavelet coefficients were encoded using the encoding algorithm, such as deflate algorithm.

(2) Spatial-dimensional data fusion of nodes

The operation process of spatial-dimensional data fusion using integer wavelet transform is as follows: Step 1, Integer wavelet transform of data for the underlying node. the nodes at the underlying of network are divided into the even nodes s_0,i and the odd nodes d_0,i in the order, and then the even node s_0,1 transmits its data to the adjacent odd node d_0,1 and calculate the integer wavelet coefficients s_1,1 and d_1,1. Correspondingly, we can be obtained all the coefficients of the 1-level wavelet decomposition (d_1,1, d_1,2, …, d_1,n) and (s_1,1, s_1,2, …, s_1,n) . Step 2, Integer wavelet transform of data for the convergence node. The underlying odd nodes send the low frequency coefficient (s_1,1, s_1,2, …, s_1,n) to the convergence nodes of the upper layer and form a new data sequences. By repeating the Step 1 operation, we can be obtained the 2-level wavelet transform coefficients (d_2,1, d_2,2, …, d_2,n) and (s_2,1, s_2,2, …, s_2,n). Step 3, Integer wavelet transform of data for the Sink node. The Sink node obtains the k-level wavelet transform coefficients (d_k,1, d_k,2, …, d_k,n) and (s_k,1, s_k,2, …, s_k,n). The diagram of spatial-dimensional data fusion of nodes is as shown in Fig. 6.

Figure 6. Diagram of spatial-dimensional data fusion of nodes

Figure 6. Diagram of spatial-dimensional data fusion of nodes

4. Simulations and Results

4.1. Energy Harvesting System

4.1.1. Performance of the Solar Energy Harvester

The test data of the energy harvesting system that we have designed are shown in Table 1, and it records the test data of this system on a sunny day in Yiyang, Hunan Province. The elements of the test mainly include: voltage regulator circuit input voltage, node power supply voltage and node power supply current. The test results show that the energy harvesting system is stable in power supply to the node, maintains about 3.3 V, less than 1.3% in voltage deviation, and it has high accuracy, strong robustness, and can provide reliable energy guarantee for WSN.

4.1.2. Performance of the Bistable Piezoelectric Cantilever Oscillator

(1) Simpleandharmonic incentive

In this experiment, we examinedthe power generation performance of a bistable piezoelectric cantilever oscillator composed of two magnets under simple harmonicincentive, with varying magnetization strength and load resistance. In Figure 7(A), the voltage amplitude frequency curve is compared for different magnetization intensities. The system's voltage amplitude solution exhibits jumps as the excitation frequency changes. Furthermore, an increase in the nonlinear term coefficient results in a rightward shift of the curve, widening the response band of large motion and increasing the amplitude. Figure 7(B) displays the power generated at different impedances. It is noteworthy that the jump phenomenon's corresponding excitation frequency in the system's amplitude and frequency curve remains unchanged regardless of the resistance value. In the system, there is an issue with resistance matching. Specifically, the resistance value R=30k needs to be addressed and optimized.

Figure 7. The power generation performance of a bistable piezoelectric cantilever oscillator (A): Amplitude-frequency curves at different magnetization strengths and (B): Output power at different impedance levels

Figure 7. The power generation performance of a bistable piezoelectric cantilever oscillator (A): Amplitude-frequency curves at different magnetization strengths and (B): Output power at different impedance levels

(2) Random incentive

In this experiment, we examined the impact of various incentive strengths and load resistances on the response of a bistable piezoelectric cantilever subjected to random incentive. The excitation strength was set to 0.01~0.09 G²/Hz and the load resistance was 1k. As shown in Figure 8(A), the output voltage of the system increases proportionally with the excitation strength. The incentive strength is 0.06 G²/Hz, and Figure 8(B) is the average power curve when the impedance changes. The figure clearly demonstrates that at a resistance of 19.1 k, the output power reaches its peak. This signifies that matching the impedance maximizes the output power of the system when subjected to random excitation.

Figure 8. The power generation performance of a bistable piezoelectric cantilever oscillator (A): Average voltage curves for different excitation intensities and (B): Average power curves at the different impedances

Figure 8. The power generation performance of a bistable piezoelectric cantilever oscillator (A): Average voltage curves for different excitation intensities and (B): Average power curves at the different impedances

4.3. Data Fusion

In this experiment, the experimental data are from the Tropical Atmospheric Ocean Project (TAO). Since 1984, the project has deployed about 100 sensors at different depths in 71 mooring in the tropical Pacific to collect the temperature of seawater in the region. We use these data for the lossless compression testing, and the test computer configuration is as follows: Processor: Intel (R) Core (TM) i5-6500 CPU @ 3.20GHz, memory: 16GB, operating system: 64-bit operating system. The test results are shown in Table 2. Compressed by using integer wavelet transform, the spatial and temporal data can obtain a average compression rate of 5.42. Compared with other lossless compression algorithms Huffman, LZSS, LZW and WinRAR, the integer wavelet transform has a higher compression ratio, more than 1 times.

Figure 9. Contrast experiment for neighbor node discovery (A): Symmetry scene and (B): Asymmetric scene

Figure 9. Contrast experiment for neighbor node discovery (A): Symmetry scene and (B): Asymmetric scene

On the other hand, we took the raw data of 96 sensors from the data set of TAO to carry out the three-level integer wavelet transform, and then analyze the data storage of nodes. After integer wavelet transform, the 1-level wavelet coefficients is (s_1,1, s_1,2, …, s_1,52) and (d_1,1, d_1,2, …, d_1,52), the 2-level wavelet coefficients is (s_2,1, s_2,2, …, s_2,30) and (d_2,1, d_2,2, …, d_2,30), and the 3-level wavelet coefficients is (s_3,1, s_3,2, …, s_3,19) and (d_3,1, d_3,2, …, d_3,19). Among these data, the data (d_1,1, d_1,2, …, d_1,52) is stored in the bottom node, the data (d_2,1, d_2,2, …, d_2,30) is stored in the convergence node, and the data (d_3,1, d_3,2, …, d_3,19) is stored in the Sink node. Compared with the traditional communication mode, the amount of data stored in the integer wavelet transform is much smaller, as shown as Table 3, which indicates that the integer wavelet transform has certain advantages in reducing the amount of communication data and transmission energy.

5. Discussion

6. Conclusions

References

1. Nurlan, Z.; Zhukabayeva, T.; Othman, M.; et al. Wireless Sensor Network as a Mesh: Vision and Challenges [J]. IEEE Access 2021, 10, 46–67. [Google Scholar] [CrossRef]
2. Ko, J.; Lu, C.; Ma, S.B.; et al. Wireless Sensor Networks for Healthcare [C]. Proc. IEEE 2010, 98, 1947–1960. [Google Scholar] [CrossRef]
3. Tekler, Z.D.; Low, R.; Yuen, C.; Blessing, L. Plug-Mate: An IoT-based occupancy-driven plug load management system in smart buildings [J]. Build. Environ. 2022, 223, 109472. [Google Scholar] [CrossRef]
4. Low, R.; Cheah, L.; You, L. Commercial Vehicle Activity Prediction With Imbalanced Class Distribution Using a Hybrid Sampling and Gradient Boosting Approach [C]. IEEE Trans. Intell. Transp. Syst. 2021, 22, 1401–1410. [Google Scholar] [CrossRef]
5. Tekler, Z.D.; Lei, Y.; Peng, Y.; et al. A hybrid active learning framework for personal thermal comfort models[J]. Build. Environ. 2023, 234, 110148. [Google Scholar] [CrossRef]
6. Tekler, Z.D.; Low, R.; Zhou, Y.; et al. Near-real-time plug load identification using low-frequency power data in office spaces: Experiments and applications [J]. Appl. Energy 2020, 275, 15391. [Google Scholar] [CrossRef]
7. Sam, M.; Mahmoud, S.; Arash, D. Microwave wireless power transfer system using a dual- dielectric resonator oscillator[J]. AEUE-Int. J. Electron. Commun. 2023, 170, 154836. [Google Scholar]
8. Martha, G.; Lukas, Z.; Christian, D.; et al. Self-powered elementary hybrid magnetoelectric sensor[J]. Nano Energy 2023, 115, 108720. [Google Scholar]
9. Ilam, P.P.V.; et al. A hybrid ANFIS reptile optimization algorithm for energy-efficient inter-cluster routing in internet of things-enabled wireless sensor networks [J]. Peer--Peer Netw. Appl. 2023, 16, 1049–1068. [Google Scholar]
10. Lu, X.; Wang, P.; Niyato, D.; et al. Wireless Charging Technologies: Fundamentals, Standards, and Network Applications [J]. IEEE Commun. Surv. Tutor. 2017, 18, 1413–1452. [Google Scholar] [CrossRef]
11. Ren, J.; Hu, J.; Zhang, D.; et al. Rf energy harvesting and transfer in cognitive radio sensor networks: Opportunities and challenges [J]. IEEE Commun. Mag. 2018, 56, 104–110. [Google Scholar] [CrossRef]
12. Ma, T.; Jia, S.; Ding, Y.; et al. Research on characteristics of piezoelectric vibration energy harvester of trapezoidal cantilever beam[J]. Instrum. Tech. Sens. 2021, 11, 43–47. [Google Scholar]
13. Yang, J.; Zhao, X.; Zhou, X. Correction of Lumped-parameter Model for Piezoelectric Energy Harvesters with Variable Cross-section Cantilever Beams [J]. Noise Vib. Control 2020, 40, 42–46. [Google Scholar]
14. Zou, H.; Wang, Z.; Song, J. Piezoelect ric Energy Harvesting Based on Nozzle - resonator System[J]. Sci. Technol. Vis. 2018, 7, 30–33. [Google Scholar]
15. Sun, X.; Feng, W.; Zhang, Y. Study on the vibration power generation of piezoelectric ceramic composites[J]. Mach. Des. Manuf. Eng. 2019, 48, 88–92. [Google Scholar]
16. Wang, S. Wake and energy harvesting based on the piezoelectric cantilever beam model in fluid; Zhejiang University: Hangzhou, China, 2014. [Google Scholar]
17. Pradeesh, E.L. Effect of placement of piezoelectric material and proof mass on the performance of piezoelectric energy harvester [J]. Mech. Syst. Signal Process. 2019, 130, 664–676. [Google Scholar] [CrossRef]
18. Qiu, Y.; Li, S.; Xu, X.; et al. Talk more listen less: Energy-efficient neighbor discovery in wireless sensor networks [C]. IEEE INFOCOM 2016-The 35th Annual IEEE International Conference on Computer Communications. IEEE, 2016; pp. 1–9.
19. Zhang, Y. ; Bian K; Chen, L.; et al. Dynamic slot-length control for reducing neighbor discovery latency in wireless sensor networks [C]. GLOBECOM, 2017. [Google Scholar]
20. Bian, K.; Zhang, Y.; Qia, P.; et al. Fine-grained collision mitigation control for neighbor discovery in wireless sensor networks [C]. 2017 IEEE/CIC International Conference on Communications in China (ICCC). IEEE, 2017; pp. 1–6.
21. Jin, S.; Meng, X.; Wong, D.L.K.; et al. Improving neighbor discovery by operating at the quantum scale [C]. 2018 IEEE 15th International Conference on Mobile Ad Hoc and Sensor Systems (MASS). IEEE, 2018; pp. 202–210.
22. Galluzzi, V.; Herman, T. Survey: Discovery in wireless sensor networks [J]. Int. J. Distrib. Sens. Netw. 2012, 8, 271860. [Google Scholar] [CrossRef]
23. Saeedi, I.D.I.; Al-Qurabat, A.K.M. A systematic review of data aggregation techniques in wireless sensor networks [J]. J. Phys. Conf. Ser. 2021, 1818, 012194. [Google Scholar] [CrossRef]
24. Zhang, W.; Cao, G. DCTC: Dynamic convoy tree-based collaboration for target tracking in sensor networks [J]. IEEE Trans. Wirel. Commun. 2004, 3, 1689–1701. [Google Scholar] [CrossRef]
25. Luo, H.; Liu, Y.; Sajal, K.D. Routing correlated data with fusion cost in wireless sensor networks [J]. IEEE Trans. Mob. Comput. 2006, 5, 1620–1632. [Google Scholar]
26. Sharaf, M.A.; Beaver, J.; Labrinidis, A.; et al. TiNA: A scheme for temporal coherency aware in network aggregation [C]. In The 3rd ACM International Workshop on Data Engineering for Wireless and Mobile Access; ACM Press: San Diego, CA, USA, 2003; pp. 69–76. [Google Scholar]
27. Vuran, M.C.; Akan, O.B.; Akyildiz, I.F. Spatio-temporal correlation: Theory and applications for wireless sensor networks [J]. Comput. Netw. 2004, 45, 245–259. [Google Scholar] [CrossRef]
28. Das, S.N.; Misra, S.; Wolfinger, B.E.; et al. Temporal-Correlation-Aware Dynamic Self- Management of Wireless Sensor Networks [J]. IEEE Trans. Ind. Inform. 2016, 12, 2127–2138. [Google Scholar] [CrossRef]
29. Kandukuri, S.; Murad, N.; Lorion, R.; et al. Energy-Efficient Data Aggregation Techniques for Exploiting Spatio-Temporal Correlations in Wireless Sensor Networks [C]. In Wireless Telecommunications Symposium; IEEE: London, UK, 2016; pp. 1–6. [Google Scholar]
30. Du, X. Micro piezoelectric power generator driven by ambient vibration [D]. Dalian University of Technology: Dalian, China, 2008. [Google Scholar]
31. He, X.; Du, Z.; Zhao, X.; et al. Modeling and experimental verification for cantilevered piezoelectric vibration energy harvester [J]. Opt. Precis. Eng. 2011, 19, 1771–1778. [Google Scholar]
32. Johnson, T.J.; Charnegie, D.; Clark, W.W.; et al. Energy harvesting from mechanical vibrations using piezoelectric cantilever beams [J]. Int. Soc. Opt. Eng. 2006, 6169, 61690D–12. [Google Scholar]
33. Roundy, S.; Wright, P.K. A piezoelectric vibration based generator for wireless electronic[J]. Smart Mater. Struct. 2004, 13, 1131–1142. [Google Scholar] [CrossRef]
34. Erturk, A.; Inman, D.J. On Mechanical Modeling of Cantilevered Piezoelectric Vibration Energy Harvesters [J]. J. Intell. Mater. Syst. Struct. 2008, 19, 1311–1325. [Google Scholar] [CrossRef]
35. Guttorp, P.; Fuentes, M.; Sampson, P. Using Transforms to Analyze Space-Time Processes [M]. Stat. Methods Spatio-Temporal Syst.