Abstract: In order to improve the accuracy of multivariable prediction of solar array and solve cyclical volatility and growth of telemetry parameters of solar array couple with each other, a multivariable prediction algorithm of solar array based on STL Prophet Informer model was proposed. The algorithm firstly uses the seasonal and trend decomposition procedure based on loess to decompose multiple parameters of the solar array into trend components, periodic components and residual components. Then Prophet is used to predict the trend component, and Informer model is used to predict the periodic component and residual component. Finally, the predicted values of the total solar array parameters are obtained by adding the predicted results of each component. Taking the actual telemetry data of a satellite solar array as an example, this paper proposes that the various error evaluation indexes of the algorithm are significantly reduced compared with the single Informer model and LSTM model, etc. Applying the combined prediction model to the multivariable parameter prediction of the satellite battery array can improve the accuracy of parameter prediction and improve the autonomous operation performance of the satellite.

Key words: satellite telemetry data, multivariate prediction, Informer, seasonal and trend decomposition procedure based on loess