Research On Metaheuristic Algorithm For Large Scale Global Optimization Problems

Optimization problems is important in scientific research and engineering applications,and traditional optimization algorithms have limitations when facing complex optimization problems.To overcome these limitations,heuristic optimization algorithms have emerged,which simulate the principles and mechanisms from the nature world,providing a new approach for solving complex optimization problems.However,with the advent of the big data era,the number of decision variables in optimization problems has been gradually increasing,making the solving of such large-scale optimization problems more complex and challenging.The main problem lies in:with the increase of decision variables,the search space of the optimization problem grows exponentially;the correlation between decision variables introduces irrelevant and redundant dimensional information,leading to significant interference with particle optimization and leading to the problem of "curse of dimensionality."Therefore,aiming at the above problems,this thesis focuses on improving the performance of heuristic optimization algorithms in large-scale optimization problems.The main contents and contributions of this thesis are as follows:Firstly,an analysis is conducted on the entanglement of decision variables in large-scale optimization problems.Then,a scheme of dimension extraction mechanism is proposed specifically for heuristic optimization algorithms.In this scheme,redundant information of optimization particles is filtered out using manifold learning,preserving the effective dimensional information relevant to optimization.A new feature mapping method is then introduced to remap lowdimensional particles back to the high-dimensional space,and through dimension comparison,identify the dimensions that are effective for particle optimization in the original space,In order to visually verify the performance of the dimension extraction scheme,this thesis conducted validation on the constructed dataset.The simulation results demonstrate that the dimension extraction scheme achieves a high accuracy in extracting redundant dimensions and also validate the rationality of each component in the scheme.Secondly,based on the proposed dimension extraction mechanism,an iteration mechanism based on effective dimensions is designed.By combining the iteration mechanism with three classical heuristic optimization algorithms:Particle Swarm Optimization(PSO),Gravitational Search Algorithm(GSA),and Differential Evolution(DE),the particle movement direction guided by different algorithmic mechanisms is adjusted,enabling particles to move towards better positions.Finally,based on the CEC2013LSGO benchmark suite,optimization simulation experiments are conducted to validate the improvement in the optimization ability of the original algorithm by the dimension extraction scheme.By comparing with 4 latest improvement mechanisms and 11 heuristic optimization algorithms specifically designed for LSGO,the advanced nature of the scheme is verified.Furthermore,the proposed combined algorithm shows certain advantages over state-of-the-art algorithms.