The Application of Machine learning to Amazonia-1 satellite power subsystem telemetry prediction

Abstract

Abstract This article presents the data acquisition, exploratory data analysis, model training, evaluation, and use of hyperparameters in a machine learning model that will be used to predict telemetry data from the Amazonia-1 satellite. The Amazonia-1 satellite was launched in 2021, it uses the Multi-Mission Platform as a service module and has a Wide Field Imager imaging camera. Its power subsystem has 715 telemetries with distinct data types that will be used as dependent and independent variables. The amount of telemetry data generated daily is large, making manual analysis of this data unfeasible. The ensemble XGBoost machine learning algorithm is used to predict the values of the dependent variable D008 “Battery Module 1 Voltage” that belongs to the electric power subsystem. For the evaluation and performance Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R2 are used. The final learning model resulted in the coefficient of determination (R2) with 99.99%, MAE of 0.005749, and RMSE of 0.007727. After the cross-validation step, RMSE reached 0.006888. The execution time was 57 minutes and 32 seconds. Based on these numbers, we can consider that the machine learning model built reached a good result, especially when used with cross-validation.