STATE OF CHARGE ESTIMATION FOR LFP BATTERY USING FUZZY NEURAL NETWORK

Transcription

1 International Journal of Electrical and Electronics Engineering Research (IJEEER) ISSN(P): X; ISSN(E): X Vol. 6, Issue 5, Oct 2016, TJPRC Pvt. Ltd STATE OF CHARGE ESTIMATION FOR LFP BATTERY USING FUZZY NEURAL NETWORK ABSTRACT WEN-YEAU CHANG Department of Electrical Engineering, St. John s University, Taiwan This paper proposes a fuzzy neural network based state of charge (SOC) estimation method using for LiFePO 4 (LFP) battery, because the fuzzy neural network is one of the best tools applied to state estimate. The proposed SOC estimation method uses three input data, the terminal voltage, discharging current, and temperature of battery to estimate the SOC for LFP battery under different discharging conditions. To demonstrate the effectiveness of the proposed estimation method, the prototype SOC estimator has been tested on 3.2V, 10AH LFP batteries under several different discharging conditions. The experimental data are found to be in close agreement. The test results show that the proposed method is efficient and reliable. KEYWORDS: State of Charge, LFP Battery, Fuzzy Neural Network Received: Aug 29, 2016; Accepted: Sep 19, 2016; Published: Sep 22, 2016; Paper Id.: IJEEEROCT INTRODUCTION The worldwide awareness of environmental issues have resulted in the development of energy storage system. The battery is one of the most attractive energy storage systems because of its small size, relatively low weight, low pollution, and high conversion efficiency. Batteries are used commonly for portable utilities and industrial applications (Rao, et al 2012). LFP battery is among the most advanced types of batteries (Chang, 2013). Original Article The SOC of a battery, which is used to describe its remaining capacity, is a very important parameter for a control strategy of battery management system (He, et al 2012). As the SOC is an important parameter, which reflects the battery performance, so accurate estimation of the SOC can not only protect battery, prevent over discharge and improve the battery life, but also allow the application to make rational control strategies to save energy (Cai, et al 2012). However, a battery is a chemical energy storage source and this chemical energy cannot be directly accessed. This issue makes the estimation of the SOC of a battery difficult (Watrin, et al 2012). Accurate estimation of the SOC remains very complex and is difficult to implement, because battery models are limited and there are parametric uncertainties (Elgammal, et al 2012). Many examples of poor accuracy and reliability of the estimation of the SOC are found in practice (Prajapati, et al 2011). The aim of this paper is at proposing the fuzzy neural network based SOC estimating methods. The rest of this paper is organized as follows: Section 2 discusses the existing SOC estimating methods. The fuzzy neural network based SOC estimating method is presented in Section 3. The experimental results with the proposed estimating method are discussed in Section 4. Finally, conclusion is drawn in Section 5. editor@tjprc.org

2 26 Wen-Yeau Chang 2. EXISTING SOC ESTIMATING METHODS Recently, several methods have been employed for the SOC estimating including: Coulomb integration method, open circuit voltage method, artificial neural network method, impedance spectroscopy method, and Kalman filter method (Chang, 2013). However, none of these estimation methods can accurately predict the SOC compared with the actual battery residual capacity. The Coulomb integration method measures the current flowing into and out of a battery and integrates the current over time in order to estimate SOC (Ng, et al 2009). But there are several factors affect the accuracy of Coulomb integration method include temperature, battery history, discharge current, and cycle life. The open circuit voltage method is based on the open circuit voltage of batteries are linearly proportional to the SOC when they are disconnected from the loads for a period longer than two hours. However, such a long disconnection time may be too long to be implemented (Ng, et al 2008). Artificial neural network could deal with non-linear and complex problems in terms of estimation or classification (Babaei, et al 2011, Karimi, et al 2011). As the problem defined, the relationship between the input and target is non-linear and very complicated in SOC estimation (Weigert, et al 2011). The artificial neural network based SOC indicator predicts the current SOC using the recent history of voltage, current and the ambient temperature of a battery (Linda, et al 2009). Impedance spectroscopy method measures battery impedances over a wide range of ac frequencies at different charge and discharge currents. The values of the model impedances are found by least-squares fitting to measured impedance values. SOC may be indirectly inferred by measuring present battery impedances and correlating them with known impedances at various SOC levels (Li, et al 2010). Using real-time measurement road data to estimate the SOC of battery would normally be difficult or expensive to measure. Application of the Kalman filter method is shown to provide verifiable estimations of SOC of the battery via the real-time state estimation (Xu, et al 2012). 3. FUZZY NEURAL NETWORK BASED SOC ESTIMATION METHOD The proposed fuzzy neural network based SOC estimation method has been successfully implemented using a micro-processor based control core for the SOC estimating. The system configuration of the proposed SOC estimation system is illustrated in Figure 1. A voltage detecting interface measures the terminal voltage of the battery; a temperature detecting circuit detects the temperature of the battery; and a current detection interface measures the discharge current. The micro-processor is the control core of the system. The fuzzy neural network estimates SOC, the discharge control program decides whether the over discharge condition is met. Impact Factor (JCC): NAAS Rating: 2.40

3 State of Charge Estimation for LFP Battery using Fuzzy Neural Network 27 Figure 1: The System Configuration of the Proposed SOC Estimation System Fuzzy logic is a powerful tool for modeling human thinking and perception. Fuzzy systems store rules and estimate sampled functions from linguistic input to linguistic output (Kwan, et al 1994). On the other hand, neural networks were widely applied in optimization, pattern recognition, forecasting. The main property of neural network that it can learn from examples is the key fact of preferring neural network in many nonlinear and complex problems. The fuzzy neural network pattern recognition approach was proposed via taking full advantage of processing fuzzy information of the fuzzy pattern recognition and self-learning of the neural network pattern recognition (Xu, et al 2009). This paper uses the four-layer fuzzy neural network to estimate the SOC of LFP battery. The architecture of the four-layer fuzzy neural network based SOC estimation system is shown in Figure 2. As shown in Figure 2, the architecture of fuzzy neural network used in this study contains four layers. First layer is the input layer which accepts pattern into the network. The input layer has 3 neurons for the terminal voltage, discharging current, and temperature of battery. The second layer is the maximum fuzzy neuron layer which has 3 neurons to fuzzify the input patterns through the weight function. The third layer is the minimum fuzzy neuron layer. In the minimum fuzzy neuron layer each minimum fuzzy neuron represents one learned pattern. The fourth layer is the output layer. In the output layer each competitive fuzzy neuron represents one learned pattern. The output layer provides nonfuzzy outputs (Kwan, et al 1994). The proposed pattern recognition approach is described briefly in the following steps: Step 1: Creating data base of the SOC vectors of battery models. Step 2: Normalize all of the SOC vectors data. editor@tjprc.org

4 28 Wen-Yeau Chang Step 4: Prepare the training set for fuzzy neural network. Step 5: Using the training set to train the fuzzy neural network for SOC estimation. Step 6: Save the trained fuzzy neural network, as training procedure is finished. Step 7: Use trained fuzzy neural network to estimate the SOC of LFP battery. Figure 2: Architecture of the Fuzzy Neural Network Based SOC Estimating System 4. EXPERIMENTAL RESULTS Experiments were conducted to verify the effectiveness of the proposed SOC estimation approach. The fuzzy neural network is constructed by a supervised training process as described in Section 3. Empirically, an input vector with three elements was constructed. The three measurements of voltage and current combined with the temperature constitute the input attributes. The output is the desired value of SOC. Hence, the training dataset is composed of input vectors and the desired response SOC. Owing to prevent the simulated neurons from being driven too far into saturation, all of the training data needs to be normalized after acquisition. Each input and target data are required to be divided by the maximum absolute value in corresponding factor. Each value of the normalized data is within the range between 0 and +1 so that the artificial neural network could recognize the data easily. After training set is created, training procedure of artificial neural network is started. Several tests with different discharging currents are used to verify the accuracy of the proposed estimation method, which are: (a) fixed discharging current of 0.3C, (b) fixed discharging current of 0.7C, (c) fixed discharging current of 1C, and (d) varied discharging current. In the fixed current discharged tests, batteries are full charged by 0.1 C fixed current mode. The fixed current discharge experimental results are shown in Table 1. The SOC are estimated by three methods: Coulomb integration method, back propagation (BP) neural network method, and fuzzy neural network method. The estimation errors comparisons of three methods are shown in Table 2. Table 2 shows that both the maximum and the average absolute errors of fuzzy neural network method are better than the BP neural network method and Coulomb integration method. Impact Factor (JCC): NAAS Rating: 2.40

5 State of Charge Estimation for LFP Battery using Fuzzy Neural Network 29 In the varied current discharged test, battery is full charged by 0.1 C fixed current mode. The varied current discharge experimental results are shown in Table 3. The SOC are estimated by three methods: Coulomb integration method, BP neural network method, and fuzzy neural network method. The estimation errors comparisons of two methods are shown in Table 4. Table 4 shows that the average absolute error of fuzzy neural network method is better than the BP neural network method and Coulomb integration method. The maximum absolute error of proposed fuzzy neural network method is worse than the Coulomb integration method. Discharge Current Table 1: The Fixed Current Discharge Experimental Results Discharge Current Discharge Time Discharge Capacity 0.3C min Ah 0.7C min Ah 1C 61.6 min Ah Table 2: The Estimation Errors Comparisons of Three Methods Error Type Coulomb Integration method BP Neural Network Method 0.3C Average absolute error 5.01% 0.091% 0.03% 0.3C Maximum absolute error 5.53% 14.28% 1.3% 0.7C Average absolute error 3.068% 0.127% 0.09% 0.7C Maximum absolute error 7.66% 23.52% 2.4% 1C Average absolute error 2.537% 0.064% 0.04% 1C Maximum absolute error 2.60% 3.70% 1.23% 5. CONCLUSIONS Table 3: The Fixed Current Discharge Experimental Results Discharge Current Discharge Time Discharge Capacity 0.3C 30 min Ah 1C 20 min 3.3 Ah 0.7C 47 min Ah Total 97 min Ah Table 4: The Estimation Errors Comparisons of Three Methods Fuzzy Neural Network Method Error Type Coulomb BP Neural Fuzzy Neural Integration Method Network Method Network Method Average absolute error 1.58% 0.188% 0.145% Maximum absolute error 2.09% 12.50% 4.25% This paper proposes a fuzzy neural network based estimating method using for battery SOC estimating. The performance of the proposed method to SOC estimating is effective. An evaluation of the estimation methods is performed, using the practical information of SOC of a battery. The results demonstrate the effectiveness of the proposed estimating method and this method provided improved accuracy in the battery SOC estimating. 6. ACKNOWLEDGMENTS The author would like to express his acknowledgements to the Ministry of Science and Technology of ROC for the financial support under Grant MOST E MY3. editor@tjprc.org

6 30 Wen-Yeau Chang 7. REFERENCES 1. Babaei, A., Rafiei, S., Mrazeban, A., & Saeidmanesh, M.(2011) Shape and dimension estimation of cracks using neural network in eddy current testing. International Review of Electrical Engineering, 4(1), Cai, Z.H., Liu, G.F., & Luo, J.(2012) Research state of charge estimation tactics of nickel-hydrogen battery. Proceedings of 2010 International Symposium on Intelligence Information Processing and Trusted Computing, doi: /IPTC Chang, W.Y.(2013). Estimation of the state of charge for a LFP battery using a hybrid method that combines a RBF neural network, an OLS algorithm and an adaptive genetic algorithm. International Journal of Electrical Power and Energy Systems, 53, doi: /j.ijepes Chang, W.Y.(2013) The state of charge estimating methods for battery-a review. ISRN Applied Mathematics, 2013, Article ID doi: /2013/ Elgammal, A.A.A., & Sharaf, A.M.(2012) Self-regulating particle swarm optimised controller for (photovoltaic fuel cell) battery charging of hybrid electric vehicles. IET Electrical Systems in Transportation, 2(2), doi: / iet-est He, H.W., Xiong, R., & Guo, H.Q.(2012) Online estimation of model parameters and state-of-charge of LiFePO 4 batteries in electric vehicles. Applied Energy, 89(1), doi: /j.apenergy Karimi, A., & Seyedtabaii, S.(2011) The use of SOM and MLP neural networks in the classification of pulse-echo ultra-sonic signals. International Review of Electrical Engineering, 4(2), Kwan, H.K., & Cai, Y.(1994) A fuzzy neural network and its application to pattern recognition. IEEE Trans. on Fuzzy Systems, 2(3), doi: / Li, R., Wu, J.F., Wang, H.Y., & Li, G.C.(2010) Prediction of state of charge of lithium-ion rechargeable battery with electrochemical impedance spectroscopy theory. Proceedings of 5th IEEE Conference on Industrial Electronics and Applications, doi: /ICIEA Linda, O., William, E.J., Huff, M., Manic, M., Gupta, V., Nance, J., Hess, H., Rufus, F., Thakker, A., & Govar, J.(2009) Intelligent neural network implementation for SOCI development of Li/CF x batteries. Proceedings of 2nd International Symposium on Resilient Control Systems, doi: /ISRCS Ng, K.S., Moo, C.S., Chen, Y.P., & Hsieh, Y.C.(2009) Enhanced Coulomb counting method for estimating state-of-charge and state-of-health of lithium-ion batteries. Applied Energy, 86(9), doi: /j.apenergy Ng, K.S., Moo, C.S., Chen, Y.P., & Hsieh, Y.C.(2008) State-of-charge estimation with open-circuit-voltage for lead-acid batteries. Proceedings of IEEE 2nd International Power and Energy Conference, doi: /PCCON Prajapati, V., Hess, H., William, E.J. Gupta, V., Huff, Manic, M., Rufus, F., Thakker, A., & Govar, J. (2012) A literature review of state-of-charge estimation techniques applicable to lithium poly-carbon monoflouride (LI/CF x ) battery. Proceedings of the India International Conference on Power Electronics, 1-8. doi: /IICPE Rao, Z.H., Wang, S.F., & Zhang, G.Q.(2012) Simulation and experiment of thermal energy management with phase change material for ageing LiFePO 4 power battery. Energy Conversion and Management, 52(12), doi: /j.enconman Impact Factor (JCC): NAAS Rating: 2.40

7 State of Charge Estimation for LFP Battery using Fuzzy Neural Network Watrin, N., Blunier, B., & Miraoui, A.(2012) Review of adaptive systems for lithium batteries state-of-charge and state-of-health estimation. Proceedings of 2012 IEEE Transportation Electrification Conference and Expo, 1-6. doi: /ITEC Weigert, T., Tian, Q., & Lian, K.(2011) State-of-charge prediction of batteries and battery supercapacitor hybrids using artificial neural network. Journal of Power Sources, 196(8), doi: /j.jpowsour Xu, L., Wang, J.P., & Chen, Q.S.(2012) Kalman filtering state of charge estimation for battery management system based on a stochastic fuzzy neural network battery model, Energy Conversion and Management, 53(1), doi: /j.enconman Xu, Y.H., Zhang, Z., Liu, K., & Zhang, G.Y.(2009) Fuzzy neural networks pattern recognition method and its application in ultrasonic detection for bonding defect of thin composite materials. IEEE International Conference on Automation and Logistics, doi: /ICAL editor@tjprc.org

8