GA-EMD-SVR condition prediction for a certain diesel engine

Support vector regression (SVR) is proved to be a good and effective method for machine condition prediction. But prediction results are usually not satisfying for complex machines, e.g., a certain diesel engine. So a hybrid method GA-EMD-SVR is proposed in this paper integrating genetic algorithm (GA), empirical mode decomposition (EMD) and support vector regression (SVR). The main ideal of the method is to select effective parameters combination from condition signal based on GA. Then EMD is employed to decompose the effective parameter sequences into several intrinsic mode functions (IMFs) and a residual. Finally, each IMF and the residue are considered as training samples to train SVR based on space construction, whose model parameters is generated by using grid-search. Experiment data from a certain diesel engine is used to validate the model. Prediction results of single and multiple steps based on GA-EMD-SVR are validated to be feasible and more accuracy than GA-SVR model.