Research On Battery SOC Prediction Algorithm Based On Improved FCM And ANFIS

As the core of the intelligent battery management System, the prediction of the state of charge (SOC) of battery can be used to control battery charge and discharge situation and extend its life. However, due to the strong non-linearity of the battery system and the complexity of the influence factors of the SOC, it is difficult to accurately predict the remaining capacity in real time.For this challenging program, the Adaptive Network-based Fuzzy Inference System(ANFIS) is used to model the surplus capacity of the battery after analyzing and comparing the traditional and intelligent prediction algorithms. ANFIS combines fuzzy logic and neural networks, so it has talent logical reasoning ability, self-organization and self-learning ability and generalization ability.However, ANFIS uses the grid method to linearly divide each of the inputs into fuzzy subsets. If the inputs are complex and non-linear, the number of fuzzy rules will inevitably increase exponentially. In order to solve this problem, the fuzzy C-means clustering (FCM)algorithm is used to classify the inputs nonlinearly to achieve the purpose of reducing the number of fuzzy rules and the complexity of system. However, whether FCM algorithm or ANFIS learning algorithm, there are some shortcomings. So the subtractive clustering and conjugate gradient method are respectively used to improve them to establish the battery surplus capacity prediction model.The work of the thesis are as follows:(1) Subtractive clustering is used to improve the performance of traditional FCM algorithms. Firstly, the improved FCM algorithm is initialized by the clustering centers,which are obtained by subtractive clustering. Then the density functions are used to give weight to the samples. The improved clustering algorithm solves the problem of FCM,such as needs of setting the number of clusters, the poor stability and the sensitivity to the noise or the wrong samples. In addition, it also speeds up the iteration rate.(2) In order to accelerate the convergence rate and avoid falling into local optimum,the Fletcher-Reeves conjugate gradient method is choosed to improve the Back Propagation (BP) learning algorithm of ANFIS. The results show that the new algorithm has faster convergence speed and higher prediction accuracy.(3) After analyzing the various factors of the battery SOC, static resistance, voltage,current and temperature are choosed as the inputs of the model. Then, the SOC prediction model is established based on the improved algorithms, and a large number of training samples are used to correct the model. In order to check the validity of the model, on the one hand, the performance of the traditional and the improved algorithms are compared based on the test set. On the other hand, BP neural network prediction model is established to compare with the ANFIS model.(4) The real-time forecasting platform for surplus capacity of battery is built, which realizes real-time forecasting, dynamic display, data storage, failure warning and other functions based on Lab VIEW, MATLAB, MySQL, PHP and other technologies.