A dynamic ensemble extreme learning machine model for aircraft engine health condition prediction

A dynamic ensemble extreme learning machine (ELM) model was proposed for aircraft engine health condition prediction. The AdaBoost.RT algorithm was used to integrate ELM to construct the ensemble model. During the training process, the neighboring samples of every training sample were employed to evaluate the local performance of ELM. In the prediction process, the neighboring samples of new samples in the training sample set were selected firstly, then the combined weights of ELM were determined by the performance on the neighboring samples, implementing the dynamic ensemble of the weak learning machine. Fuel flow was utilized as a health index for aircraft engine health condition prediction. For short term prediction, the mean absolute percentage error (MAPE) of the dynamic ensemble ELM model was 3.688%, less than the MAPE of the single ELM model and the static ensemble ELM model, which were 3.830% and 3.719%, respectively. And for long term prediction, the MAPE of the dynamic ensemble ELM model was 3.075%, also less than that of the single ELM model of 4.355% and the static ensemble ELM model of 3.884%. Thus, the dynamic ensemble ELM model is better for the aircraft engine health condition prediction.