Research On Intelligent Optimization Algorithms Base On Agile And Dynamic Strategies

Optimization problems are widely used in the fields of manufacturing and vehicle scheduling.With the increasing complexity of optimization problems in various fields,it is very important to find intelligent optimization algorithms that can better solve these complex problems.this paper studies the high-dimensional real number optimization problem in continuous optimization and the dynamic vehicle routing problem in discrete optimization.Considering the high-dimensional real number optimization problem,this paper proposes a sardine optimization algorithm(SOA)inspired by the survival mechanism of sardines.SOA has the following characteristics:(1)Agile locality and globality strategies.The local and global search behaviors of sardine schools are controlled by using the economic indicators of adjacent period ratio and corresponding period ratio.adjacent period ratio is used to decide whether sardine schools transform or not.corresponding period ratio is used to control the reproduction behaviors of schools.The agile locality and globality strategies are a new technical road to balance exploration and exploitation efforts.(2)Use unique search operators such as transformation,migration,reproduction and elimination.These operators are based on the center movement of schools.These search operators are different from most metaheuristic algorithms.(3)The SOA algorithm is strictly tested on the 50 dimensions using the benchmark provided by CEC2013 competition.and SOA is compared with eight efficient intelligent optimization algorithms,including the champion algorithms in 2013(ICMAES-ILS)and 2014(LSHADE).The results show that the Friedman test of SOA ranks first in unconstrained problems.And Wilcoxon results between SOA and two highly competitive algorithms are 17:8 and 15:11,respectively.Focusing on the dynamic vehicle routing problem,this paper proposes a parallel variable neighborhood search algorithm based on Q-learning.Different from other algorithms used in DVRP,the algorithm proposed in this paper has the following characteristics:(1)This paper proposes a new parallel strategy,which executes the neighborhood structure of shake in parallel.This strategy can not only reduce the overall running time of the algorithm,but also increase the search space.(2)Uses Q-learning to independently select the optimal neighborhood structure to perform local search,avoiding repeated traversal of meaningless neighborhood structures.At the same time,the probability of receiving the poor solution is selected through Q-learning to avoid the algorithm falling into local optimization effectively.(3)This paper adopts the classical benchmarks testing the proposed algorithm.The results show that the dynamic client insertion time of the proposed algorithm is the shortest.At the same time,the proposed algorithm to obtain the minimum total distance on most test problems.