Fault Detection of Single Cell Battery Inconsistency in Electric Vehicle Based on Fireworks Algorithm Optimized Deep Belief Network

Abstract

Abstract Because the fault characteristics of inconsistent fault single battery are not obvious in the electric vehicle battery pack, it is difficult to identify the inconsistent fault. Therefore, this paper proposes an inconsistent fault detection method based on a fireworks algorithm (FWA) optimized deep belief network (DBN). The method feeds the raw data signal into a deep belief network algorithm for training, which automatically performs feature extraction and intelligent diagnosis of inconsistencies, without requiring the time domain signal to be periodic. The top-level algorithm of the deep belief network adopts error Back Propagation (BP). Using FWA training to optimize DBN-BP, the best DBN-BP-FWA model structure can be obtained. Experimental verification was carried out using real vehicle data from electric vehicles. The inconsistency diagnosis results show that, compared with the traditional inconsistency diagnosis method, the application of this paper's method for electric vehicle single battery fault detection can obtain higher accuracy, with an average accuracy of 96.19%.