An agent based approach for the implementation of cooperative proactive S-Metaheuristics

• This paper introduces several cooperative proactive S-Metaheuristics.
• The proposal is based on two characteristics of agents: proactivity and cooperation.
• Proactive S-Metaheuristics avoid local optima by adjusting parameters and operators.
• Simple forms of cooperation are used to combine proactive metaheuristics.
• The experiments consider binary problems, knapsack and travelling salesman problems.

This paper introduces several cooperative proactive S-Metaheuristics, i.e. single-solution based metaheuristics, which are implemented taking advantage of two singular characteristics of the agent paradigm: proactivity and cooperation. Proactivity is applied to improve traditional versions of Threshold Accepting and Great Deluge Algorithm metaheuristics. This approach follows previous work for the definition of proactive versions of the Record-to-Record Travel and Local Search metaheuristics. Proactive metaheuristics are implemented as agents that cooperate in the environment of the optimization process with the goal of avoiding stagnation in local optima by adjusting their parameters. Based on the environmental information about previous solutions, the proactive adjustment of the parameters is focused on keeping a minimal level of acceptance for the new solutions. In addition, simple forms of cooperation by competition are used to develop cooperative metaheuristics based on the combination of the four proactive metaheuristics. The proposed metaheuristics have been validated through experimentation with 28 benchmark functions on binary strings, and several instances of knapsack problems and travelling salesman problems.