Condition trend prediction of aero-generator based on particle swarm optimization and fuzzy integral

Abstract

In order to improve and enhance the prediction accuracy and efficiency of aero-generator running trend, grasp its running condition, and avoid accidents happening, in this paper, auto-regressive and moving average model (ARMA) and least squares support vector machine (LSSVM) which are used to predict its running trend have been optimized using particle swarm optimization (PSO) based on using features found in real aero-generator life test, which lasts a long period of time on specialized test platform and collects mass data that reflects aero-generator characteristics, to build new models of PSO-ARMA and PSO-LSSVM. And we use fuzzy integral methodology to carry out decision fusion of the predicted results of these two new models. The research shows that the prediction accuracy of PSO-ARMA and PSO-LSSVM has been much improved on that of ARMA and LSSVM, and the results of decision fusion based on fuzzy integral methodology show further substantial improvement in accuracy than each particle swarm optimized model. Conclusion can be drawn that the optimized model and the decision fusion method presented in this paper are available in aero-generator condition trend prediction and have great value of engineering application.