Battery Fault Prognosis for Electric Vehicles Based on AOM-ARIMA-LSTM in Real Time

In order to ensure the safety of drivers and passengers, the voltage prediction and fault diagnosis of the power batteries in electric vehicles are very critical issues. The AOM-ARIMA-LSTM model are proposed to study the inconsistency of voltage, current, temperature and other parameters which can detect the potential safety hazards of batterys in time and take corresponding measures to reduce the occurrence of thermal runaway. First, the approximation optimization method (AOM) are adopted to optimize the parameters of the Autoregressive Integrated Moving Average model (ARIMA), which realizes single-factor real-time prediction of battery voltage. At the same time, the genetic algorithm-based LSTM neural network are also adopted to carry out multi-factor prediction of battery voltage, from which the information of battery current, temperature, SOC, etc is adopted as the model input. Finally, the least square method is adopted to fuse the predicted results of the ARIMA and LSTM models, and the inconsistency judgment of single batterys is carried out to find the battery failure in time. This paper adopts the real-world vehicle data of two electric vehicles to prove the accuracy of battery voltage prediction and the effectiveness of the proposed fault diagnosis method.