基于适应度景观与雅可比矩阵引导梯度下降的改进花授粉算法OA

To address the defects of flower pollination algorithm(FPA),such as its susceptibility to local optima,inadequate optimization accuracy,and tendency toward premature convergence,this paper proposed an improved FPA by adopting the fit-ness landscape and Jacobian matrix-guided gradient descent mechanism,namely FJFPA.Firstly,FJFPA dynamically divided the population into an elite swarm and a common swarm based on individuals' fitness values and their relative distances,ba-lancing between exploration and exploitation.Secondly,FJFPA introduced the fitness landscape and Jacobian matrix for gradi-ent descent mechanism.This bolstered vanilla FPA's single-mechanism update and insufficient adaptability.Thirdly,FJFPA also incorporated a refined full opposition-based learning strategy.It generated opposite solutions by leveraging the weighted center and performed the stochastic dimension-mixture operation to further increase the randomness of the opposite solutions,and then retained high-quality solutions through simulated annealing mechanism.This strategy could further mitigate the likeli-hood of being trapped in local optima.The CEC2022 test suit was used as the benchmark to evaluate the performance of FJFPA with the vanilla FPA and 6 other state-of-the-art improved algorithms:DMEFPA,HASMFP,NGFPA,PMFPA,DMSSA and MSNSSA.Friedman tests based on the experimental results indicate that FJFPA can achieve the supreme performance among all co-evaluated algorithms.The results of ablation experiment also demonstrates that FJFPA can attain the outstanding per-formance when all 3 improvement strategies are synergistically combined.