Modelling and Prediction of Spark-ignition Engine Power Performance Using Incremental Least Squares Support Vector Machines

Abstract

Modern automotive spark‐ignition (SI) power performance usually refers to output power and torque, and they are significantly affected by the setup of control parameters in the engine management system (EMS). EMS calibration is done empirically through tests on the dynamometer (dyno) because no exact mathematical engine model is yet available. With an emerging nonlinear function estimation technique of Least squares support vector machines (LS‐SVM), the approximate power performance model of a SI engine can be determined by training the sample data acquired from the dyno. A novel incremental algorithm based on typical LS‐SVM is also proposed in this paper, so the power performance models built from the incremental LS‐SVM can be updated whenever new training data arrives. With updating the models, the model accuracies can be continuously increased. The predicted results using the estimated models from the incremental LS‐SVM are good agreement with the actual test results and with the almost same average accuracy of retraining the models from scratch, but the incremental algorithm can significantly shorten the model construction time when new training data arrives.