Enhanced Gradient-based Optimizer And Application Research

Gradient-based optimizer(GBO)is a novel,gradient-based,meta-heuristic optimization algorithm whose search kernel is Newton's method.The algorithm algorithm structure is simple,easy to implement,strong exploration and development ability,can effectively avoid falling into local optimum and other characteristics,and has been successfully applied in many fields.However,the algorithm also has some shortcomings,such as slow convergence speed and low accuracy in the late stage of the algorithm,and the large-scale problem is easy to fall into local optimum.In this thesis,three different Gradient-based optimizers are proposed for the analysis and improvement of the shortcomings of the gradient optimizer algorithm and applied to some practical optimization problems,with the aim of further improving the performance of the GBO algorithm and expanding its application scope.The main research work of this paper is as follows.(1)In order to enhance the diversity of populations and broaden the application scenarios of the algorithm,a binary encoding strategy is introduced,and a class of Binary Gradient-based optimizer(BGBO)is proposed,which is applied to the feature selection problem and compared with other metaheuristic algorithms through simulation experiments on 18 basic and 10 high-dimensional datasets,and the experimental results show that the BGBO-V3 optimizer is the best.(2)In order to balance the exploration and exploitation capability of GBO and improve the solution accuracy of the algorithm,an improved gradient-based optimizer(IGBO)is proposed by introducing adaptive weighting and chaotic operator strategies.The experimental results of different models are compared and analyzed with other algorithms by applying IGBO to the parameter optimization problems of photovoltaic models,and the experimental results show that IGBO is more effective than other metaheuristic algorithms in solving this kind of problems.(3)An enhanced gradient-based optimizer(EGBO)is proposed to address the shortcomings of poor algorithm population diversity and the tendency to fall into local optimality by introducing dynamic contrastive learning and differential evolution strategies.It increases the population diversity of the algorithm,improves the algorithm's ability to solve large-scale and high-dimensional optimization problems,and apply it to three trpes of engineering optimization problems.