Research And Application Of Dynamic Multi-objective Optimization Based On Multi-mechanism Evolution

Dynamic multi-objective optimization problems widely exist in scientific research and engineering applications.Since the objective function and the constraint function in the dynamic multi-objective optimization problem change with the environment,when solving the dynamic multi-objective optimization problem,the algorithm must be able to quickly and accurately trace the changed optimal solution.Especially when dealing with complex practical engineering problems,the performance requirements of the algorithm are higher.In this thesis,we study the dynamic multi-objective optimization problem.The traditional dynamic multi-objective optimization algorithm is improved b y combining multiple mechanisms,and the improved algorithm is applied to traveling salesman problem and PID controller parameter optimization problem.The content is summarized as follows:Aiming at the constant change of environmental variables in the dynamic multi-objective optimization problem,this thesis improves the algorithm NSGA2 and proposes a dynamic multi-objective optimization algorithm based on amplitude variation.The algorithm uses the amount of angle change to express the influence of frequent changes in the magnitude of the change on the optimal solution.At the same time,individuals are divided according to the amount of angle change,and the divided individuals are optimized by different prediction strategies to obtain the optimal solution after the change.Experimental comparisons show that the improved algorithm has obvious advantages in terms of convergence and distribution.Building predictive models is a research direction to solve dynamic multi-objective optimization problems.Most existing methods ignore the non-independent and identically distributed nature of data to construct the prediction model.Aiming at this problem,this thesis proposes a dynamic multi-objective optimization algorithm based on transfer learning,which combines the transfer learning algorithm with the dynamic multi-objective optimization algorithm to optimize.The optimized optimal solution is saved.When the magnitude of the change is similar,the execution efficiency of the algorithm is further improved by using the history information of optimal solution.The experimental results show that compared with the traditional dynamic optimization algorithm,the optimal solution obtained by the improved algorithm has better convergence and distribution.In this thesis,the improved algorithm is applied to traveling salesman problem and PID controller parameter optimization problems.For the traveling salesman problem,the encoding mechanism of the algorithm is adjusted under the premise of the shortest travel time and the minimum cost.When the road conditions change,the algorithm optimizes the path.The experimental results prove the effectiveness of the algorithm.For the parameter optimization problem of PID controller,this thesis analyzes the situation that the controlled object changes during the running process.In this case,the algorithm optimizes the PID parameters of changed system and selects individuals that meet the user's preference.The good experimental results demonstrate the feasibility of the improved algorithm.