Título del libro: Proceedings Of Gecco 2007: Genetic And Evolutionary Computation Conference
Título del capítulo: Incremental refinement of solutions for multiple objective optimization problems
VICTOR HUGO ALCOCER YAMANAKA;
Autores externos:
Idioma:
Año de publicación:
2007
Palabras clave:
Machine learning; Multiobjective optimization
Resumen:
MDQL is an algorithm, based on reinforcement learning, for solving multiple objective optimization problems, that has been tested on several applications with promising results [1]. MDQL discretizes the decision variables into a set of states, each associated with actions to move agents to contiguous states. A group of agents explore this state space and are able to find Pareto sets applying a distributed reinforcement learning algorithm. The precision of the Pareto solutions depends on the chosen granularity of the states. A finer granularity on the states creates more precise solutions but at the expense of a larger search space, and consequently the need for more computational resources. An important improvement is presented. The new algorithm, called IMDQL, starts with a coarse granularity to find an initial Pareto set. A vicinity for each of the Pareto solutions in refined and a new Pareto set is founded in this refined state space. It is shown that IMDQL not only improves the solutions found by MDQL, but also converges faster and is capable to approximate dynamic Pareto fronts. The main consideration in the application of IMDQL to dynamic environments is that the agents in the algorithm start from the Pareto solutions obtained. Agents start with a deterministic environment constructed with fixed values for the value functions for the first dynamic parameters; when convergence is reached and a Pareto set is obtained, a new cycle is started, changing to the next value for the dynamic parameters. Agents start searching (adapting solutions) from the existing environments which correspond to the Pareto solutions obtained for the previous value for the dynamic parameters. Searching for new solutions, from the last Pareto set, given the new values for the dynamic parameters, significantly reduces the convergence time. IMDQL is tested on a real water distribution network design involving water-reusing treatment plants and different contaminants concentrations [1]. In this problem, the concentration of contaminants can change over time so the search for optimal solutions becomes a continuous process. It is shown that IMDQL improves on the solutions found by MDQL with fixed concentrations and, that due to its incremental nature, it is able to adequately adjust its Pareto set solutions with dy
Entidades citadas de la UNAM:
ISBN:
1595936971
Editorial:
