Satellite battery sensor values prediction using Bayesian ridge regression models

Abstract

Proper mission control plays a key role in the lifetime of space mission operation, as it ensures that all resources are efficiently utilized when achieving mission goals. Ground control station operation mainly depends on received telemetry together with models simulating spacecraft's subsystems. Created models help in raising the level of autonomy of MCC (Mission Control Center). Data driven models describe the actual state of the subsystem in real operation situations rather than theoretical costly physical models. This paper proposes data driven models for satellite battery subsystem based on Bayesian ridge regression algorithm. The ridge coefficients minimize a penalized residual sum of squares Thirty models of all thirty battery variables (capacitance, voltage, pressure and temperature) are built from normal operation data. Sensor reading value can be predicted from an observation of all other 29 values. Faults present in sensors or in system can be detected if predicted values are not equal to actual downloaded data from satellite. Bayesian ridge regression models are validated in terms of slope, intercept, R2-value, Q2-value P-value and standard error.