Based on MLP and BiLSTM battery capacity forecast method research

Batteries are integral to a wide range of applications, including electronics, automotive systems, and smart grids. The accurate prediction of remaining battery capacity remains a key challenge and focal point of contemporary research. However, temperature, voltage, and current data generated during battery operation are often underutilized in existing capacity prediction models. To address this limitation, this paper introduces an advanced method that combines multi-layer perceptron (MLP) and bidirectional long short-term memory networks (BiLSTM) for battery capacity forecasting. Initially, parameters such as temperature, voltage, and current are input into the MLP model to derive weights, which are subsequently fed into the BiLSTM network for further processing. This approach enables the prediction of the battery’s remaining capacity, while a GPU-based parallel acceleration module is incorporated to optimize performance when handling large volumes of data. Experimental results show that the proposed model achieves a worst-case RMSE of 0.0302 and MAPE of 0.0484, demonstrating superior prediction accuracy compared to traditional models such as CNN and LSTM.