A Least Squares Support Vector Machine for Condition Time Series Prediction Based on Bayesian Evidence Framework

A method using Bayesian evidence framework(BEF)and least squares support vector machine(LSSVM)is proposed to predict electronic system condition time series accurately.A LSSVM model for prediction is trained with all the current condition time series data.Then,the LSSVM model is iteratively updated by adopting the latest data and pruning the oldest data.Matrix transform is applied to reduce the computational cost of retraining the LSSVM model.Finally,the updated LSSVM model is dynamically optimized by BEF.Numerical experiments on radar transmitter condition time series prediction are carried out to test the effectiveness of the proposed method.The experimental results and comparisons with the conventional adaptive grey model show that the proposed method has better performance in prediction accuracy,prediction stability and computational efficiency,and that the prediction accuracy and the computational efficiency for electronic system condition time series prediction are raised by 9.52%and 73.26%respectively.