An Improved Strength Pareto Evolutionary Algorithm 2 Based on Markov Networks

As an excellent representative of MOEAs, the strength Pareto evolutionary algorithm 2(SPEA2) becomes an important method for solving MOPs and has been widely applied. However, traditional genetic operators in SPEA2 have strong randomness, which inevitably break the discovered linkage relationship between variables. Therefore, it is more possible for SPEA2 to premature or non-convergence. Fortunately, Markov networks with good linkage learning ability can help SPEA2 improve its performance. This paper presents an improved SPEA2 (MN-SPEA2) by building and sampling Markov network to generate new population. Machine learning in MN-SPEA2 can provide an intelligent search towards global optimal Pareto front. Experimental results also show that MN-SPEA2 has better global convergence than SPEA2.