Power state prediction of battery based on BP neural network

The demand of electrical vehicles for batteries with higher power has increased rapidly. The existed prediction method for power state involves composite pulse method, dynamic fusion state prediction algorithm based on multiparameters and SOC method based on battery, while the adaptation of the former two methods is competent for dynamic working conditions, and the last one may require high level hardware and huge amount of calculation when involved in batteries pack. So the research on power state of batteries is of significance in extending the service life, regenerative braking and energy recycling fields. BP neural network has an advantage of application and feasibility in tackling with the severe nonlinear characteristic of power state of battery at dynamic working conditions without any high level hardware. It has been approved by vehicle test that the factors that affect the power state of battery at dynamic working conditions involve road conditions, natural environments, and human factor, among which the main factor is the current and voltage value. Then this paper has analyzed the main factors that affect the power state of battery, established a prediction model for power state of Li-ion power battery based on BP neural network and probed into the influence on the precision of the network prediction model with training functions: trainscg, trainlm and traincgb, which proved that the output of function trainlm has the least relative error and rapid constringency, which highlights the incomparable superiority of trainlm training function in the middle-scale BP neural network in rapid approaching.