Forecasting Maximum Mechanism Temperature in Advanced Technology Microwave Sounder (ATMS) Data Using a Long Short-Term Memory (LSTM) Neural Network

Abstract

Among the monitored telemetry raw data record (RDR) parameters with the STAR Integrated/Validation System (ICVS), the Advanced Technology Microwave Sounder (ATMS) scan motor mechanism temperature is especially important because the instrument might be unavoidably damaged if the mechanism temperature exceeds 50 °C. In the current operational flight processing software, the instrument automatically enters safe mode and stops collecting scientific data whenever the mechanism temperature exceeds 40 °C. This approach inevitably leads to the instrument entering safe mode unnecessarily at a premature time, causing the loss of scientific data before the mechanism temperature reaches 50 °C. This study seeks to leverage the influence the main motor current, compensation motor current, and main motor loop integral error have on mechanism temperature to forecast the maximum mechanism temperature over the upcoming 6 min. A long short-term memory (LSTM) neural network predicts maximum mechanism temperature using ATMS RDR telemetry data as the input. The performance of the LSTM is compared with observed maximum mechanism temperatures by applying the LSTM coefficients to several cases. In all cases studied, the mean average error (MAE) of the forecast remained under 1.1 °C, and the correlation between forecasts and measurements remained above 0.96. These forecasts of maximum mechanism temperature are expected to be able to provide information on when the ATMS instrument should enter safe mode without needlessly losing valuable data for the ATMS flight operational team.

Keywords

• Joint Polar Satellite System (JPSS)
• Advanced Technology Microwave Sounder (ATMS)
• long short-term memory (LSTM) networks
• machine learning (ML)
• anomaly forecasting
• scan drive mechanism temperature