Differential Evolution Algorithm Based On Accompany Population

Differential evolution(DE)is an intelligent optimization algorithm inspired by biological evolution.Due to its simple structure,fewer parameters,and strong robustness,DE is widely used in many fields such as engineering technology.However,the characteristics of optimization problems are developing towards more complex structures,larger dimensions,and more interference conditions,making the improvement of DE research face unprecedented challenges.Moreover,the performance of DE is very sensitive to mutation strategies and control parameters.Current optimization problems pose dynamic and time-varying requirements for related settings.Therefore,how to adjust evolutionary schemes through optimization feedback has become the focus and difficulty of algorithm design.Based on the analysis of existing research,this paper specifically optimizes the crux of deteriorating DE performance,proposes the accompanying population,and creates a new and efficient working mechanism around it.Numerical experiments verify the competitiveness of the new algorithm.To evaluate and analyze algorithm performance more comprehensively,an evolutionary process visualization curve is proposed,which maps the optimization process of high-dimensional problems into a three-dimensional coordinate system,achieving real-time tracking of optimization progress.The main research work of this paper is as follows:1.Introduce the research significance,basic principle,and operation process of DE algorithm in detail.According to the execution order of DE algorithm,the implementation method of evolutionary operations,how each module affects optimization,what problems are prone to occur in the optimization process,and adjustment measures are introduced.Clarify that population diversity,mutation strategy,and control parameters are the main factors that affect DE performance,and conduct a literature review on these three points,introducing the key directions and main measures of research at home and abroad.2.Aiming at the main problems of optimization under the existing DE framework,such as being prone to falling into local optimum and slowing down the convergence rate with the optimization process,a differential evolution algorithm based on adjoint populations is proposed.As the core of the new algorithm,the accompanying population consists of suboptimal solutions,whose generation and update do not require additional computational resources,and undergoes accompanying evolution at minimal computational cost.The population size gradually decreases according to the evolution stage,which saves computing resources and reduces meaningless computing while compensating for the additional costs caused by the accompanying evolution.The accompanying population provides assistance for optimization by interacting with the main population,participating in the formation of the difference vector,and maintaining diversity.It can also achieve adaptive adjustment of mutation strategies and control parameters in parallel,guiding the algorithm to converge to the global optimum with richer evolutionary information and dynamically adapted optimization settings.To make the accompanying population work more efficiently and always maintain differences from the main population,a strategy based on generalized opposition-based learning is applied to the reinitialization of the accompanying population,achieving simultaneous improvement in population diversity and solution quality.Innovative use of radial spatial projection technology to visualize the evolution process of high-dimensional optimization problems solves the problem that numerical results and convergence curves based on fitness values can only indirectly infer optimization progress.3.The performance of the proposed algorithm is tested on the four sets of benchmark problem suites proposed at the Institute of Electrical and Electronics Engineers Congress on Evolutionary Computation(IEEE CEC)in the past decade,and compared with the award-winning algorithms of CEC and the state-of-the-art intelligent optimization algorithms on multiple test dimensions,confirming the effectiveness and competitiveness of the proposed method.Furthermore,multi-angle experimental analysis was conducted to comprehensively verify the algorithm performance,including optimization performance analysis based on numerical results,convergence effect analysis based on turning points,algorithm contribution analysis based on ablation experiments,and optimization direction analysis based on evolution process visualization curves.