Predicting battery health for energy storage and electric vehicles systems by integrating EMD signal processing and machine learning

Accurate prognostics of battery State-of-Health (SOH) and Remaining Useful Life (RUL) are paramount for the operational safety and economic feasibility of sustainable energy systems, yet are frequently hindered by noise-corrupted sensor data. This study introduces and validates a novel hybrid framework that integrates Empirical Mode Decomposition (EMD) as an adaptive signal pre-processing technique with advanced machine learning models to overcome this critical limitation. Utilizing the NASA Ames prognostic dataset with synthetically introduced Gaussian noise to simulate real-world conditions, we demonstrate that EMD-based filtering effectively denoises battery discharge profiles, revealing a more coherent degradation trajectory. A comparative analysis of the resulting hybrid models SVM_EMD, LSTM_EMD, and GRU_EMD conclusively shows that the SVM_EMD model delivers superior performance, consistently achieving the lowest Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), and providing the most accurate RUL predictions across all tested battery units. This research establishes the two-stage SVM_EMD framework as a robust, low-complexity, and highly effective solution for enhancing the reliability and longevity of batteries in real-world applications, underscoring the vital importance of dedicated signal pre-processing in battery prognostics.