Improved BP neural network algorithm to SOC estimation of mine power supply

Abstract

Abstract Since Ni MH battery has the characteristics of high energy density and good stability, it is mainly used as a backup power supply for mining at present, but the accuracy of estimation the SOC of the power supply is poor. To enhance the precision of estimation and extend the lifespan of the battery, based on a single nickel-hydrogen battery, this paper applies a new BP neural network to rise the self-adaption and momentum terms. At the same time, the discharged power is included in the Input feature quantity of the estimation network. The temperature, current and terminal voltage are also included in the input characteristic quantity of the estimation network. There is output during training, and then the momentum term is adjusted based on the Variance of variance error by comparing the actual output. The error of the estimated network should be back propagated from the output to the input to correct the weight. The direction of change in the error determines the adaptive learning rate of the improved BP neural network. The speed of the algorithm is increased by 72% and the error rate is less than 5%, compared to the traditional BP neural network algorithm.