Physics-Informed BiLSTM for Predicting Battery Remaining Useful Life: A Hybrid Deep Learning Framework

Predicting how long a lithium-ion battery can continue operating is now a make-or-break capability for electric mobility, distributed solar, and any product that depends on dependable energy storage. We introduce Physics-Informed Bidirectional Long Short-Term Memory (PI-BiLSTM), a hybrid framework that blends neural sequence modeling with the first principles that govern electrochemical aging. Convolutional layers distill localized information from charge-discharge cycles, bidirectional LSTMs describe long-range temporal behavior, and physicsinformed heads keep the network's predictions aligned with Arrhenius temperature sensitivity and Peukert's rate-capacity law. Evaluations on the NASA PCoE dataset and auxiliary synthetic profiles report an overall accuracy of 87.5 % with balanced precision and recall above 0.93 for both healthy and degraded cells. Beyond the scores, PI-BiLSTM is packaged for deployment: the loss term is transparent, the model ports cleanly to cloud or edge targets, and the output can be interpreted by battery engineers. The study demonstrates that lightweight physics regularization provides a practical path to humantrustworthy RUL estimation.