شبکه های عصبی مصنوعی کاربردی در بهینه سازی Artificial neural networks used in optimization problems

Introduction Optimization problems are an important part of soft computing, and have been applied to different fields such as smart grids [1], logistics [2] [3] resources [4] or sensor networks [5]. Such problems are characterized by the presence of one or more objective maximizing or minimizing functions [5] and various restrictions that must be met so that the solution is valid. The problems are easy to resolve when we are working with linear restrictions and objective functions because there are methods to obtain the optimal solution However, in the case of non linear restrictions or objective functions it may be necessary to use heuristics [2] [5] to obtain a pseudo optimal solution. The management of heuristic solutions is continually evolving, which is precisely why we are looking for alternatives to problems in which it is not feasible to find an optimal solution. hen working with linear restrictions and objective functions, optimization problems can be resolved with algorithms such as the Simplex [6] which limits the study of this type of problem. Certain non linear problems can be optimally resolved by using algorithms such as Lagrange multipliers or Kuhn–Tucker conditions [7]. In many cases, it is no possible to resolve a problem with Lagrange multipliers because the generated system of equations cannot be resolved without resorting to numerical methods, which would prevent a direct approach to resolving the problem. n other cases the Kuhn Tucker condition are not met There is a broad range of opportunities to study optimization problems that cannot be solved with an exact algorithm. These problems are usually solved by applying a heuristics and metaheuristics solution such as genetic algorithms [8], particle swarm optimization [9], Simulated annealing [10], ant colony optimization [12] etc.