Automatic Surrogate-assisted Swarm Intelligence Optimization Method Based On Ensemble Learning

In the real engineering field,numerous complex optimization problems often require high computational cost,and the data-driven surrogate-assisted intelligence algorithm(SAIA)is an effective way to deal with it.Nevertheless,in the process of algorithm development and implementation,it is often accompanied by abundant manual participation and configuration.Thus,learning from automatic machine learning(Auto ML),based on ensemble learning,this dissertation is devoted to exploring advanced heuristic optimization techniques with good generalization ability and adaptive to expensive optimization problems.The purpose is to realize the functions of automatic construction,automatic parameter adjustment,and automatic optimization of algorithms,effectively reduce the subjectivity of artificial design,so that the algorithm can automatically explore the characteristics of the objective to be solved,to effectively improve the accuracy and computational efficiency of the problem.The main contributions of the dissertation are as follows:1.To reduce human trial and error and algorithm configuration to construct an efficient optimization mechanism,firstly,an automatic agent model algorithm framework based on generalized ensemble learning is constructed based on data science,machine learning,and meta-heuristic optimization algorithm.Secondly,focusing on the selection of surrogate models,the accuracy,robustness,and scalability of the model are introduced to evaluate the performance of the commonly used metamodel,which provides a basis for the reasonable combination of surrogate models and the optimizer in the optimization process.Finally,a series of simulation experiments are conducted to analyze the performance of related models and optimizers based on the benchmark problems.Experimental results show that for different optimization problems,the reasonable combination of the optimizer and surrogate model is helpful to improve the performance of algorithms.2.Aiming at the computational cost of complex single-objective optimization,an automatic surrogate-assisted particle swarm optimization algorithm based on ensemble learning is proposed(Auto SAEPSO).Firstly,construct the automatic selection strategy to realize the automatic selection of ensemble particle swarm optimizers and surrogate models.Secondly,utilized the automatic linear adjustment strategy to adaptively adjust the optimizer parameters.Then,adopt the interactive cooperative optimization of the metamodel and objective function to improve the generalization ability and accuracy of algorithms.Finally,the performance of the algorithm is analyzed and verified based on the benchmark problems and a direct methanol fuel cell(DMFC)system.Experimental results show that the proposed algorithm can achieve good performance on different types of optimization problems.3.Aiming at the computational cost of complex multi-objective optimization,an automatic surrogate-assisted multi-objective particle swarm optimization algorithm based on hybrid ensemble learning(HEASMOPSO)is proposed.Firstly,based on the framework of Auto SAEPSO,reinforcement learning(RL)strategy is introduced to carry out the global search,and the state,behavior,3D Q table,and reward function are designed to determine the local generation strategy which is more suitable for the current process characteristics,to better balance the global exploration and local development ability of the algorithm.Secondly,adopt the automatic linear adjustment strategy to adjust the optimizer parameters in the process of optimization.Meanwhile,construct the elite subset and utilize the mixed sampling method based on simulated evolutionary variation to improve the quality of the candidate solution.Then use the experience of elite particles to update the social experience of the population to maintain the balance between convergence and diversity.Finally,the proposed method is effectively verified by benchmark problems and applied to the control optimization of the sewage treatment process(WWTP).Experimental results show that HEASMOPSO is practical and effective.