Nowadays, automotive engine compensation control is\ncarried out electronically by utilizing many compensation\nmaps in engine management systems; such that the engine\ncan sustain its performance under the variations in engine\noperating conditions and environmental parameters. In\ntraditional engine compensation map calibration, the\nparameters are normally set by a trial and error method\nbecause the exact mathematical engine model has not been\nderived. In this paper, a new framework, namely multiinput/\noutput least-squares support vector committee\nmachine, is proposed to construct the engine compensation\ncontrol system (ECCS) models based on experimental data.\nAs the number of adjustable parameters involved in the\nECCS is very huge, the model accuracy and training time are\nusually degraded. Nonlinear regression is therefore\nemployed to perform dimension reduction before modelling.\nThe ECCS models are then embedded in an objective\nfunction for parameter optimization. Two widely-used\nevolutionary optimization algorithms, Genetic algorithm\n(GA) and particle swarm optimization (PSO), are applied to\nthe objective function to determine the optimal calibration\nmaps automatically. Experimental results show that the\nproposed modelling and optimization framework is effective\nand PSO is superior to the GA in compensation map\ncalibration.
Loading....