Abstract:Large dimensional multi-objective optimization is a common topic in engineering application. The conventional multi-objective optimization algorithms are inefficient in the problem of which the objectives number exceeds four. For the problems of conventional and the current dimensional reduction algorithms, a group divided dimensional reduction evolutionary algorithm is put forward. For this algorithm, the objective functions are divided into several groups, sub non-dominated set of every function group is built separately using SPEA2, then the non-dominated set of all objective functions is obtained basing on the sub non-dominated sets. The feasibility of this method is proved theoretically. For the sake of engineering application, individual fitness of SPEA2 is modified. In the simulation, this method is compared with current large dimensional reduction algorithms in terms of some metrics on two test problems. The test results show us the better performance of the algorithm.