A Novel Control Method Combining Multi-Level Dimension Reduction and NSGA-II for Octane Number

Octane number is an important indicator reflecting the combustion performance of gasoline. To accomplish octane number modeling and its control variables optimization effectively, this paper proposes a novel control method MDR-NSGA-II which combines the multi-level dimension reduction (MDR) and non-dominated sorting genetic algorithm II (NSGA-II). Firstly, based on the chemical data, the MDR method is used to select the control variables to obtain the main control variables. The MDR integrates the advantages of the distance correlation coefficient, the sparse representation method, and a machine learning fusion algorithm RF-XGBoost combining random forest (RF) and eXtreme gradient boosting (XGBoost). Secondly, the main control variables are controlled and optimized using NSGA-II. Specially, an octane number loss model based on Elman neural network is established as a prediction model for the solution of NSGA-II. Finally, the results of MDR-NSGA-II are analyzed and evaluated, and it can be concluded that the proposed method can accomplish the octane modeling and its optimal control well.