Dynamic Parameterization of Metaheuristics Using a Multi-agent System for the Optimization of Electricity Market Participation

Abstract

Metaheuristic optimization algorithms are increasingly used to reach near-optimal solutions for complex and large-scale problems that cannot be solved in due time by exact methods. Metaheuristics’ performance is, however, deeply dependent on their effective configuration and fine-tuning to align the algorithm’s search process with the specific characteristics of the problem that is being solved. Although the literature already offers some solutions for automatic algorithm configuration, these are usually either algorithm-specific or problem-specific, thus lacking the capability of being used for diverse metaheuristic models or diverse optimization problems. This work proposes a new approach for the automatic optimization of metaheuristic algorithms’ parameters based on a multi-agent system approach. The proposed model includes an automated fine-tuning process, which is used to optimize a given function in an algorithm- and problem-agnostic manner. Results show that the proposed model is able to achieve better optimization results than standard metaheuristic algorithms, with a negligible increase in the required execution time.