摘要：AbstractModern engine is a typical multi-objective control system. This paper proposed a multi-objective Bayesian optimization strategy to deal with the performance optimization for diesel engines. Since the objective functions of diesel engines are complicated and computationally expensive, Gaussian processes(GPs) are constructed by using the data collected from the diesel engine to approximate the real objective functions. Non-dominated sorting genetic algorithm II(NSGAII) leverages the Gaussian process to generate the Pareto-optimal solutions. The Gaussian process will be updated iteratively by Bayesian posterior information, which increases the reliability of the models. The acquisition function Expected Hyper Volume Improvement(EHVI), which can balance the trade-off between exploration and exploitation throughout the optimization process, is used to select the solutions for real computationally expensive multi-objective evaluation. The proposed algorithm is applied on a diesel engine, which shows its reliability and high efficiency. The metrics hypervolume(HV) and the control results demonstrate that the proposed algorithm has outstanding effects for performance optimization of diesel engine airpath.
