Sub-population evolutionary particle swarm optimization with dynamic fitness-distance balance and elite reverse learning for engineering design problems

Abstract

The main idea of particle swarm optimization (PSO) is to imitate the social behavior of birds, and the individuals in the algorithm can quickly converge to the optimal region of the solution space, to effectively find the optimal solution to the problem. When faced with high and complex optimization problems, it is easy to fall into local optimal solutions. To this end, this paper improved and optimized the original particle swarm optimization algorithm by introducing four stages, namely, the introduction of elite reverse learning strategy, dynamic fitment-distance balance (dFDB) strategy, subpopulation evolution strategy, factor adjustment and random position update strategy, and named the improved algorithm dFDBMPSO. Firstly, the diversity of the population is enhanced by introducing the elite reverse learning strategy in the initial stage. Secondly, in order to prevent the algorithm from identifying the optimal solution more effectively, dFDB policies are added to the algorithm to provide better guidance for the birds during the search process. In addition, subpopulation evolution strategy and random location update are introduced into the algorithm as new individual update methods, which makes the search method of birds in the algorithm more flexible. Finally, the formula factor of bird location update in the algorithm is adjusted to better balance the ability of global search and local search. In this paper, to test the performance of the proposed algorithm, the proposed algorithm is compared with a variety of comparison algorithms in different dimensions of the CEC2020 test set, four engineering design problems, two other truss topology optimization problems, and parameter optimization problems based on VCCS-FOPID vehicle control system. The experimental results show that the newly developed algorithm is superior to the previous algorithm. In addition, a new grey prediction model RC_TDGM (1,1,r,ξ, Csz) based on triangular residual correction and dFDBMPSO is proposed and applied to the forecast of global biofuel production. The predicted results are closer to the trend of the original data, which further verifies the competitiveness of the algorithm.