State prediction for the actuators of civil aircraft based on a fusion framework of relevance vector machine and autoregressive integrated moving average

Abstract

Relevance vector machine is a newly proposed and effective state prediction algorithm proved by practical applications; however, the accuracy of the single relevance vector machine model for the long-term prediction is unable to achieve satisfactory results with time goes by. Then, an autoregressive integrated moving average model is introduced to correct the prediction error caused by the single relevance vector machine, and a fusion framework based on the combination of relevance vector machine and autoregressive integrated moving average model is adopted to improve the accuracy of long-term prediction. In addition, a targeted approach for retraining the old model is put forward so that the state prediction model can be updated in time and suits the actual situation better. The effectiveness of the proposed fusion framework is illustrated via an aircraft actuator, and the experiments based on a model of civil aircraft actuator data set show that the proposed method yields a satisfied performance in state prediction of aircraft actuators.