Accurate RUL Prediction Based on Sliding Window with Sparse Sampling

Accurate prediction of the Remaining-Useful-Life (RUL) of a battery, sufficiently long in advance from the End-of-Life (EOL), is essential for the safety and maintenance of its many applications. The problem is that such a long-term prediction of RUL suffers from the growing uncertainty in predicting the State-of-Health (SOH) of a battery as the terms or cycles of prediction increases. Conventional approaches are yet to address an effective solution to this problem. This paper presents an approach to the accurate long-term prediction of battery RUL based on the sliding window combined with sparse sampling. The proposed approach attains the accuracy by defining the window of a limited number of cycles at which SOHs are predicted single-shot and sliding the window further into the future to consecutively predict SOHs for the subsequent window of cycles. Furthermore, sparse sampling is to be incorporated into the selection of an optimal size of prediction window in such a way as to maximize the accuracy involved in a long-term prediction. A Stacked-LSTM network is adopted to carry out the prediction of a window of SOH cycles. Experiments are conducted based on the Center for Advanced Life Cycle Engineering (CALCE) dataset. The result verifies that the proposed approach tops the conventional approaches in terms of the accuracy of long-term RUL prediction.