A Differential Evolution Algorithm for Dynamic Multi-Objective Optimization

Dynamic multi-objective optimization problems are widespread in real life. Artificial equivalents of the problems are abound in literature. Evolutionary algorithms are nature-inspired search algorithms used in finding optimal trade-off solutions of dynamic multi-objective optimization problems. This study proposes a new differential evolution algorithm, namely the Dynamic Differential Evolution Vector Evaluated Non-Dominated Sorting or 2DEVENS. 2DEVENS draws from the non-dominated sorting and the vector-evaluated procedures which are typical of search algorithms such as Dynamic Non-Dominated Sort Genetic Algorithm (DNSGA-D) and Dynamic Vector Evaluated Particle Swarm Optimization Algorithm (DVEPSO) respectively. Computing the target vector is a key operation in a differential evolution algorithm. The proposed algorithm, 2DEVENS, employs the vector-evaluated procedure of DVEPSO in computing the target vector, while the next generation vectors are computed using the non-dominated sorting procedure of DNSGA-D. The proposed algorithm is compared with DNSGA-D and DVEPSO on different measures of performance, benchmark functions and for different experimental configurations. The ability of the proposed algorithm to track the changing Pareto Optimal Front, i.e. finding the various optimal trade-off solutions in objective space, is compared with DNSGA-II and DVEPSO. The results show that the proposed 2DEVENS algorithm compares very well with the established algorithms, such as DNSGAD and DVEPSO, in computing the optimal tradeoff solutions for the various dynamic multi-objective optimization problems studied in this paper. 2DEVENS performs best for the two importance measures of accuracy and stability, making it a favourite in situations where accuracy and stability measures are critical in making choice of algorithms when solving dynamic multi-objective optimization problems.