Multi-objective genetic local search algorithm using Kohonen's neural map

摘要: Genetic Algorithms (GAs) are population based global search methods that can escape from local optima traps and find the regions. However, near optimum set their intensification process is often inaccurate. This because strategy of GAs completely probabilistic. With a random sets, there small probability to improve current solution. Another drawback genetic drift. The black box no one knows which region being searched by algorithm it possible only in feasible space. On other hand, usually do not use existing information about optimality regions past iterations. In this paper, new method called SOM-Based Multi-Objective GA (SBMOGA) proposed diversity. SBMOGA, grid neurons concept learning rule Self-Organizing Map (SOM) supporting Variable Neighborhood Search (VNS) learn improving both search. SOM neural network capable efficiency data processing algorithms. VNS developed enhance Evolutionary (EAs). uses multi-objective based-on Pareto dominance train its neurons. gradually move toward better fitness areas some trajectories knowledge front generations saved form trajectories. final state determines solutions be regarded as density distribution function high potentially overall efficiency. last section applicability examined developing optimal policies for real world multi-reservoir system non-linear, non-convex, optimization problem.