Fusing Driving Behavior Features for Detecting Early Voltage Consistency Anomaly of Battery Pack in Electric Vehicles

To prevent battery thermal runaway incidents in electric vehicles (EVs), it is crucial to detect faults on voltage consistency between cells (VCC) in the battery pack by a timely and accurate manner. This paper conducts a naturalistic driving experiment on 20 EVs to collect high-quality vehicle operation data, and proposes an early VCC anomaly detection method that integrates driving behavior features. For four types of micro-running segments of EVs, the model of genetic-algorithm-optimized back propagation neural network (GA-BPNN) is established to estimate the normal voltage variation coefficient between cells (VVCC) during vehicle running. The input features of models include related parameters of ambient temperature, state of charge (SOC), current, vehicle speed, acceleration, and pedal stroke. The GA-BPNN models are validated to exhibit a good robustness. The VVCC anomaly threshold is obtained using methods of Box-Cox transformation, 3σ rule, and box plot. The validation results indicate that the proposed method can effectively identify early abnormalities before VCC exceeds the conventional threshold. The average recall rate for 12 test EVs is 0.8746. This method overcomes the shortcomings of existing methods that cannot automatically adjust anomaly detection standard due to lacking consideration of driving behavior’s impact on voltage fluctuation.