A novel optimal support vector machine ensemble model for NOX emissions prediction of a diesel engine. (October 2016)

• Record Type: Journal Article
• Authors: Liu, Bo; Hu, Jie; Yan, Fuwu; Turkson, Richard Fiifi; Lin, Feng
• Is Part Of: Measurement. Volume 92(2016)
• Page Start: 183
• Page End: 192
• Publication Date: 2016-10
• Subjects: NOX emissions -- Principal component analysis -- Genetic algorithm -- Support vector machine -- Prediction model; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2016.06.015
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A novel optimal SVM ensemble model for NOX emissions prediction is proposed. A combination of normalization and PCA was employed for data preprocessing. A combined method of GA and grid search was used for optimizing model parameters. K-fold CV showed better performance on prediction accuracy than LGOCV. The model was based on data from testing a YC6L-42 diesel engine. Abstract: A novel ensemble method based on principal component analysis (PCA), genetic algorithm (GA) and support vector machine (SVM) implemented in MATLAB® is presented for establishing the NOX emissions prediction model for a diesel engine for both steady and transient operating states. The different stages of data preprocessing, modeling, optimization and prediction were discussed in detail. Normalization and PCA were used to reduce differences and redundancy of the datasets respectively. Subsequently, the SVM model was trained with 1/3 of the equi-spaced data samples (a simple DoE) selected after preprocessing. A grid search and GA were then applied as the combination strategy with the fitness function being the cross-validated root mean square error ( RMSE ) for optimizing the model parameters to improve the prediction accuracy. The optimal model was finally tested using the rest 2/3 data samples. Compared with other three methods, the proposed model exhibited superior accuracy both on training and testing datasets.