Research On Method Of Prediction SOC Of Bus Battery Based On FNN

With the development of new energy technology,purely electric bus will play a more and important role in public transportation.State of Charge(abbreviated as SOC)is a key parameter of the power battery of purely electric bus.SOC can guide the battery management system and the prediction of remainder range and it is related to the life and safe operation of the battery.However,SOC cannot be measured directly and its prediction is affected by many factors among which there are complicate relationships.As a result,the accuracy of SOC prediction is low.Thus,the realization of a model with high prediction accuracy of SOC is significant and challenging.Adaptive Network-based Fuzzy Inference System(abbreviated as ANFIS)combines the fuzzy logic and neural network effectively.It is good at reasoning and selflearning and can infinitely approximate a nonlinear system.This paper researches on the way of improving the accuracy of bus battery SOC prediction.The specific research work is as follows:1)The factors that affect SOC of battery are analyzed firstly,and the voltage,current,average temperature of battery and the maximum of voltage difference between different single batteries are selected as the model’s input.Then,the model of ANFIS is built separately by the method of grid partition and the Fuzzy C-means(abbreviated as FCM)algorithm which is optimized based on subtractive clustering method.Both models are trained with hybrid learning method on data set.The experimental results show that the model built by the latter method has a better performance than the former.However,the maximum error rate of the both model is high and the training efficiency is low.2)To solve the problem of the large maximum error,the objective function of FCM and the function to calculate the clusters of FCM is optimized through a way of weighting on sample with density,which reduce the sensitivity of FCM to the samples which are away from cluster centers.Then,FCM is optimized by simulated annealing and genetic algorithm,which reduces the dependence of FCM on initial cluster centers.Moreover,aiming at the problem of the low training efficiency,a kind of PSO algorithm which can adjust inertia weight based on the particles’ distribution is used to optimize the model’s parameters.This method balances the global search ability and partial search ability of PSO algorithm,and then the ANFIS model’s prediction accuracy is improved.The improved model is trained and tested on the data set.The experimental results show that the improved ANFIS model effectively reduces the prediction error of the prediction of bus battery SOC.