A Study On Envolutionary Multi-objective Optimization Method Based On Decomposition And Reference Point Adaption

There are many multi-objective optimization problems in scientific research and industrial engineering practice, this kind of problems have two or many objectives to be optimized simultaneously and normally they are conflicting with each other. Thus we cannot find one solution that optimizing all the objectives to be optimum, only we can do is to find a set of solutions that make a good trade-off between all the objectives, and we call that pareto optimum solutions. Using traditional mathematical programming techniques to handle multi-objective problem, we can only find a single solution in a separate run, and then we have to perform algorithm many times to find several solutions. So, they cannot provide more comprehensive decision support. Recent years, Multi-objective Evolutionary Algorithms (MOEAs) is given more and more attention and widely proposed to optimize scientific and engineering field because of its high efficiency and practicability.In this article, we apply Multi-objective Evolutionary Algorithm base on Decomposition (MOEA/D), the most representative algorithm of MOEAs, to the multi-objective reservoir flood dispatching. In real world, the Reservoir Flood Dispatch problem has great realistic significance on cutting the peak of flood, protecting people’s lives and property. In order to provided more information support to the decision makers. We need to obtain a set of Pareto optimal scheduling schemes with better convergence and wider distribution with the used algorithm. Meanwhile the Reservoir Flood Dispatch problem is a kind of optimization problem with complex multi-objective and multi-constrained. Otherwise, because of the constrain of the problem, in real dispatching, not all of the Pareto optimal scheduling schemes satisfy the ending upstream water level, and the scheduling scheme that given to decision makers to choose may focus on a limited area of PF. In order to improving the solution searching efficiency while reducing the burden of the decision makers, we improve MOEA/D by adding preference information in the algorithm, and give a MOEA/D with preference base on reference point adoption, so that make the algorithm converge to the preferred region that decision makers interested eventually. The main work of our article includes the following aspects:First, we give the abstract mathematical model of the Reservoir Flood Dispatch problem, and analyze the feasibility of solving the problem with our given algorithm combining with characteristic of MOEA/D. And through the experiment of two typical floods using MOEA/D to handle, verify the feasibility and effectiveness by the application of MOEA/D on the Reservoir Flood Dispatch problem.And then, we propose the necessity of adding preference information in our algorithm on the base of the summarization of our experimental results and the actual characteristic of the Reservoir Flood Dispatch problem. Then we describe the influence of the reference point on the searching direction of algorithm. And then we give the method on how to improve MOEA/D by using reference point. At last, we use our improved algorithm to handle six typical floods. The results of our experiments are consistent with our experimental motivation and achieve our desired effect.