Abstract: By adopting ε constraint to handle mixed average Gbest selection,this paper proposes a new multi-objective particle swarm algorithm for improving convergence and diversity,and increasing the applicable scope.The proposed algorithm adopts a new average Gbest selection mechanism,considers Euclidean distance of the particle and non-dominated solutions of archives and corresponding to the target function value such that the convergence and diversity of algorithm can be improved.Besides,an improved constraint handling mechanism with relaxation phase is utilized,in which some excellent infeasible solutions are allowed to join in early stage so as to improve the ability to jump out of local optimum.Moreover,the proposed algorithm blends the discrete variables coding mechanism of Sigmoid function such that this algorithm can handle mixed integer problem to increase the applicable scope of algorithm.Compared with the classical MOPSO,NSGA2 and SNSGA,the proposed algorithm has advantages in convergence and distribution of steam power system in ethylene plant,which further proves the effectiveness of the algorithm.