The Modified PNN Prediction Interval for Spacecraft Data

In this paper, a new method is proposed for predicting the future evolution of a time series of spacecraft telemetry data. Because such a time series has usually a non-stationary trend, nonlinear functional relationship between inputs and outputs and other uncertainties, an appropriate prediction model for spacecraft data is needed to set up. To this end, we analyze the characteristics of the Probability Neural Network (PNN) and modify its architecture for fitting the requirements of prediction. We propose a new model: the Modified Probability Neural Network (MPNN) which incorporates the characteristics of statistics into the ANN prediction model. We also analyze the statistical characteritics of the error of MPNN prediction for spacecraft data to compute approximate prediction interval. Finally we construct the interval prediction with the MPNN model and apply it to predicting the spacecraft data.