Research On Dynamic Multi-Objective Optimization Algorithm Based On Classification Of Decision Variables

There are many dynamic multi-objective optimization problems in engineering practice and scientific research,such as industrial production planning and scheduling,vehicle path planning and portfolio optimization,etc.These practical problems ha ve important research significance.Solving dynamic multi-objective optimization problems requires that the population can quickly adapt to the new environment after environmental changes and obtain good diversity and convergence.Since dynamic multi-objective optimization problems contain many uncertainties,it is a major challenge to obtain good convergence and diversity.In order to quickly respond to the environmental changes and improve the convergence and diversity of the algorithm,the dynamic multi-objective optimization algorithm is studied as follows:(1)A decision variable classification method based on k-means clustering is proposed for the problem of inaccurate classification of decision variables in the decision variable classification method based on Pareto dominance relationship.The strategy measures the type of decision variables based on the angle between the fitted line and the hyperplane normal,and uses a hybrid optimization strategy to increase population diversity and convergence.Since post-environmental change information is extremely helpful in guiding population evolution,a way to calculate the severity of environmental change is proposed by combining post-environmental change information with historical information to help populations adapt to new environments quickly.Compared with three advanced algorithms on 16 test functions,the experimental results show that the proposed algorithm has good performance and achieves good convergence and diversity on different dynamic multi-objective optimization benchmark test sets.(2)For the problem of slow convergence in the k-means clustering-based decision variable classification method,the objective monotonicity-based decision variable classification method is proposed.In order to spee d up the algorithm operation and reduce the algorithm running time,the decision variable classification method based on objective monotonicity uses only the monotonicity of the objective to classify the decision variables without using other additional an gles or Pareto dominance relations.In addition,a multi-step prediction method and a diversity maintenance strategy are used in the optimization process to help the population cope with complex environmental changes and improve prediction results.The alg orithm was tested experimentally in comparison with three advanced algorithms on 16 benchmark test functions with different characteristics,and the experimental results show that the algorithm can quickly adapt to new environments and effectively solve dynamic multi-objective optimization problems.