Directional Evolution Algorithm With Local-directional Reproduction Strategy And Self-learning Mechanism

Evolutionary Algorithms (EAs) is an intelligent swarm optimization algorithm.It provides new ideas to solve complex optimization problems, because of itsintelligence, versatility, robustness, essentially parallelism and global search ability.Therefore, it has been widely used in many optimization fields, including constrainedoptimization, clustering optimization, nonlinear optimal control, and neural networkoptimization. In recent years, some issues has attracted a lot of researchers’ attention,including improving the searching capability of the evolutionary algorithm, the abilityto solve high-dimensional, multi-objective search, accelerating convergence rate andovercome the common premature convergence phenomenon of heuristic algorithm.Now, Directional Evolutionary Algorithm (DEA) becomes a new research field inEvolutionary Computation (EC). DEA is a thought of directional variation which isimposed to deal with the problem that mobilizing individual subjective initiative isnot fully mobilized in conventional evolutionary algorithms. This paper proposes andexams an improved directional evolutionary algorithm for different evolutionarystrategies and operators. Main contributions of this paper are as follows.1． Firstly, based on a simple evolutionary direction operator which is consistentwith the thought of directional variation, a new directional reproduction strategy isproposed. Further, local search technology is utilized to propose a new ActiveEvolutionary Algorithm with Local-directional Reproduction Strategy (LRSEA). Themain processes of local-directional reproduction strategy (LRS) are as follows:initially, fitness differences of individuals are used to standardize the search direction,and local search space is quickly accessed. Then, the offspring are partiallyreproduced based on the optimal evolution direction. Lastly, in order to avoid localoptimum, dynamic adjustment of the direction is made based on fitness comparisonbetween global optimal individuals and local individuals. The above processes are thecore operations of Active Evolutionary Algorithm with Local-directionalReproduction Strategy (LRSEA). The results of experiments show that convergencerate and local search ability of the new algorithm are improved. 2． Designs a new learning operator (Activity Learning Operator) which is basedon learning efficiency, and it can proactively learn the “excellent knowledge”. Thepositivity of this operator is manifested in the relevance between the learningefficiency of the evolutionary individuals and their fitness. Individual with highfitness has high learning efficiency. Therefore, its learning positivity is high. Base onLRSEA and activity learning operator, another new algorithm, LRSEA Base onActivity Learning Operator (LRSEA*), is proposed. LRSEA*combineslocal-direction reproduction strategy and activity learning operator. Experimentalresults show that the new algorithm improves both the effectiveness and the speed ofsolving optimal value.