Optimized LSTM based on an improved sparrow search algorithm for power battery fault diagnosis in new energy vehicles

Abstract

Rapidly and accurately diagnosing power battery faults in new energy vehicles can significantly improve battery safety. Aiming at the collected power battery historical fault data information, a power battery fault diagnosis method based on an improved sparrow search algorithm (ISSA) optimized LSTM neural network is proposed. First, typical fault types are screened out through statistical fault sample data, and feature extraction is carried out by using wavelet packet unsupervised learning, solving the problem that long time series signal features are difficult to extract and recognize. Second, to solve the uneven distribution problem in initial population randomization, which can result in slow process of the algorithm, the initial position of the sparrow population is initialized using piecewise chaotic mapping with a homogenized distribution. Then, the population's optimal position in each iteration is perturbed using a variant of Gaussian difference, addressing the issue of the population easily converging to local optima. Finally, the hidden layer's optimal number of neurons of LSTM neural network is optimized by improving the sparrow search algorithm. Solving the problems of randomness and the difficulty in selecting the hyperparameters of the LSTM, a feature matrix is used as the input of the LSTM for model training and fault diagnosis and classification. The effectiveness of this method is verified by comparative experiments. The results indicate that the improved Sparrow search algorithm proposed can improve the capabilities of power battery fault diagnosis.

Keywords

• power battery
• fault diagnosis
• gaussian difference variation
• sparrow search algorithm
• lstm neural network