Machine learning-enabled software-defined wireless sensor network architecture for battery monitoring in electric vehicles

For safe, optimal operation of electric vehicles, timely battery fault diagnosis and accurate determination of the battery’s state of health (SoH) are essential. To address
this challenge, this paper presents a long short-term memory (LSTM)–feature-based diagnostic–prognostic framework that incorporates software-defined wireless sensor networks, deep learning, and ensemble regression methods to intelligently
monitor lithium-ion batteries. A controlled synthetic dataset was generated based on IEEE Standards 1188 and 1679 to represent charge–discharge behavior, operating
variations, and battery degradation-related conditions. In the diagnostic module, an LSTM neural network uses voltage, current, and temperature data to diagnose multiple battery faults with 92% classification accuracy, an average area under the receiver operating characteristic curve of 0.997, and an average equal error rate of 0.007. In the prognostic module, a bagged and boosted decision tree model based on
battery SoH predicts its value, with a root mean square error of 0.146, mean absolute error of 0.034, and R2 of 0.999. The proposed framework enables comprehensive real
time diagnostics and prognostics for predictive maintenance, thereby improving vehicle operating safety and extending battery service life in next-generation electric
vehicles.