Research And Application Of Multi-objective Genetic Algorithm For Many-objective Optimization Problems

Genetic Algorithm (GA) is a kind of global adaptive optimization technique, and it is widely used mainly because of its ability in solving highly complex nonlinear problems. The research of Multi-Objective Genetic Algorithm (MOGA) mainly focuses on improving two important performances:algorithm’s convergence ability and Pareto Front’s distribution. The algorithm’s convergence ability means the ability of the MOGA to converge to the optimal Pareto Front. The Pareto Front’s distribution is the solutions’distribution in the objective space, which is an important part in assessing Pareto Front. Recent years, many effective MOGAs have been proposed in the field of GA research including the famous Non-dominated Sorting Genetic Algorithm (NSGA) proposed by Deb.After studying the classic MOGAs. we find their shortcomings in solving high-dimensional optimization problems, and we note that the classic MOGAs only applies to two or three dimensional optimization problems successfully, but real-life optimization problems will probably be more than three dimensions (i.e. high-dimensional optimization problems). In view of this, this paper proposed a novel MOGA for high-dimensional optimization problems. Some testing problems (such as DTLZ1) and several evaluation indexes (such as general distance) have been used to test and evaluate the new algorithm. After comparing with two classic MOGAs we can obtain the following conclusions:For high-dimensional optimization problems.the new algorithm is not only significantly better than NSGA-II and SPEA2, but also performs well, which implies the proposed algorithm in this paper is effective.Course Scheduling Problem (CSP) is a high-dimensional optimization problem with several constraints. It is also an NP-complete problem (This conclusion has been verified in the field of computer applications). So far, using genetic algorithm to solve CSP is divided into two categories:First, considering only two or three of the optimization objectives in the CSP, so which can directly use classical MOGAs; Second, integrate different objectives with integration weights, this can simplify the CSP into a single objective optimization problem. The first category directly ignored certain optimization objectives to obtain the Pareto optimization results, which may be difficult to meet the requirements of the decision makers. In the second category, how to determine the weight values itself is also a complex optimization problem. Because the algorithm proposed in this paper can deal with high-dimensional optimization problems, so it can overcome the two shortcomings discussed above successfully at one time in solving CSP. And this paper gives a detail procedure in solving CSP with the algorithm proposed in this paper.