Nowadays, automotive engine compensation control iscarried out electronically by utilizing many compensationmaps in engine management systems; such that the enginecan sustain its performance under the variations in engineoperating conditions and environmental parameters. Intraditional engine compensation map calibration, theparameters are normally set by a trial and error method because the exact mathematical engine model has not beenderived. In this paper, a new framework, namely multi-input/output least-squares support vector committeemachine, is proposed to construct the engine compensationcontrol system (ECCS) models based on experimental data.As the number of adjustable parameters involved in theECCS is very huge, the model accuracy and training time areusually degraded. Nonlinear regression is thereforeemployed to perform dimension reduction before modelling.The ECCS models are then embedded in an objectivefunction for parameter optimization. Two widely-usedevolutionary optimization algorithms, Genetic algorithm(GA) and particle swarm optimization (PSO), are applied tothe objective function to determine the optimal calibrationmaps automatically. Experimental results show that theproposed modelling and optimization framework is effectiveand PSO is superior to the GA in compensation mapcalibration.