Abstract: The input current of the satellite solar array is affected by the earth albedo, satellite albedo, etc., which will produce fluctuations of different frequencies, resulting in insufficient prediction accuracy. To solve this problem, a current data prediction method based on discrete wavelet transform (DWT) and long-term short-term memory (LSTM) is proposed. First, the signal is normalized, and then discrete wavelet transform is used to decompose the telemetry signal, to obtain the multi-layer high and low-frequency wavelet coefficients of the signal to improve the signal data characteristics, and then use dual-channel LSTM is used to perform feature learning to predict each layer of wavelet coefficients, and finally the final prediction signal is obtained by reconstructing and de-normalizing the predicted wavelet coefficients. The model is verified by using the current telemetry data of an on-orbit satellite solar array. The results show that the proposed method has better prediction accuracy than traditional LSTM. MAE is reduced by 16.4%, RMSE is reduced by 29.9%, and R is improved by 3.2%.

Key words: long short-term memory, discrete wavelet transform, telemetry data, solar array, prediction model