A Smart Energy Management System for Battery-Supercapacitor in Electric Vehicles based on the Discrete Wavelet Transform and Deep Learning

The Discrete Wavelet Transform (DWT) is used to distribute power between the battery and supercapacitor in an electric vehicle so that the fast dynamic power demand is met by the supercapacitor and the slow dynamic is met by the battery. This results in a decline in battery ageing as the supercapacitor absorbs the high charge and discharge stress that would otherwise be imposed on the battery. However, implementing DWT introduces a time delay that increases as the level of decomposition increases. This time delay makes real time implementation difficult. This paper proposes the use of Deep Learning Recurrent Neural Networks with Long-Short Term Memory (LSTM) units to predict the power demand from raw data and compensate for the time delay so that DWT based energy management strategy can be implemented in real time. To compensate for the delay introduced by a second level DWT, the LSTM obtained a prediction root mean squared error of 3.69KW for the federal test procedure 72 (FTP72) driving cycle. Simulation results are presented to validate the design.