A change type-based self-adaptive response strategy for dynamic multi-objective optimization

Abstract

Dynamic multi-objective optimization (DMO) devotes to search the optimal solutions of each environment. Therefore, it is greatly important to respond effectively to changes in environment, but none of the existing response strategies considers the impact of the type of environmental changes on the response strategies. This paper proposes a novel change type-based self-adaptive response strategy (CTSRS) for handling DMO (CTSRS-DMO). CTSRS could detect whether the Pareto-optimal set (PS) changes or not, in which PS changes is considered as a type of change and PS remains unchanged is another change type. Then, different response strategies are adaptively activated to react to environmental changes on the guidance of different change types. For the change with the PS changes over time, a linear prediction strategy wakes up to respond to environmental changes. While for the changes with PS remains unchanged, a dynamic mutational diversity introduction strategy (DMDIS) works to react to the changes. In DMDIS, the proportion of individuals to be mutated is dynamically determined by the difference between the mappings of PS in the objective space about the current environment and that of the new environment. Empirical results bear out that CTSRS-DMO outperforms the other six advanced algorithms. Finally, CTSRS-DMO commendably solves the parameter-tuning problem of proportional integral derivative (PID) controller on dynamic system.