Multi-objective Optimization Algorithm For Large-scale Reservoir Flood Control Operation

Reservoir flood control operation(RFCO)is a complex multi-objective optimization problem.During flood season,it is expected for decision-makers to obtain a set of accurate scheduling schemes.However,in order to accelerate the convergence rate of the algorithm,most of the existing multi-objective optimization algorithms for the reservoir flood control operation problem take a specific scheduling time interval to reduce the number of decision variables.These algorithms can only obtain a rough approximate scheduling schemes with scheduling time intervals.For meet the accuracy requirements of decision makers,this paper aims to obtain more elaborate hourly scheduling schemes,but this will cause a sharp increase in the dimension of decision space of RFCO problem.And due to chain-like correlation between the decision variables,the RFCO problem raises big challenge to existing large-scale multi-objective optimization algorithms which were generally designed based on the variable grouping strategy.Therefore,in order to meet these challenges,this paper proposes a multi-objective optimization algorithm to directly solve the large-scale RFCO problem.At the same time,it is expected by decision makers to quickly obtain a set of scheduling schemes in the flood season,so the proposed algorithm in this paper should also meet the real-time requirements of decision makers.The main research results and innovations of this paper are as follows.1.A new divide-and-conquer strategy,namely coarse to fine decomposition method is proposed to overcome the defect of the existing multi-objective optimization algorithms that the solution obtained is a set of rough approximate scheduling schemes.Following the algorithmic framework of multi-objective evolutionary algorithm based on decomposition(MOEA/D),a multi-objective optimization algorithm based on the coarse to fine decomposition method is developed for large-scale RFCO problem,named CFD-MOEA/D.The CFD-MOEA/D algorithm decomposes the original RFCO problem into a sub-problems sequence with different levels of detail and then optimizes them following the order from a coarse scale sub-problem to fine scale sub-problem.By optimizing each sub-problem step by step,the finest solution is finally obtained.Experimental results on six typical flood at Ankang reservoir have validated the effectiveness of the proposed CFD-MOEA/D algorithm.By solving the sub-problems successively from coarse to fine scale,CFD-MOEA/D algorithm can progressively converges onto the target Pareto front,and successfully obtains more elaborate hourly scheduling schemes for the large-scale RFCO problem.In addition,the determination of sub-problem sequence with different scheduling time interval is a key point of the coarse to fine decomposition method in this paper.So analysis on the relationship between the scheduling time interval and the difficulty of the sub-problem is made and the characteristics of the reservoir flood control operation problem are found.Combined with this characteristics,we give suggestions for setting the number of sub-problems and the setting method for the scheduling time interval of the sub-problems in the sub-problem sequence.On this basis,the sub-problem sequence of CFD-MOEA/D algorithm can be determined.2.In order to obtain the elaborate optimal hour schedules in a reasonable time,the proposed CFD-MOEA/D algorithm combined with an island parallel model is developed,named pCFD-MOEA/D.Among them,each sub-problem in the CFD-MOEA/D algorithm is assigned to a processor for independent optimization.The individual migration occurs within a certain interval,and the communication between sub-problems is based on an island parallel model of unidirectional topology proposed in this paper.As these sub-problems are optimized simultaneously,the final hourly scheduling schemes can be obtained in a reasonable time.Experimental results on six typical flood at Ankang reservoir have demonstrated the pCFD-MOEA/D algorithm effectively improves the efficiency of the CFD-MOEA/D algorithm,and increases the speed to more than three times compared with the original serial CFD-MOEA/D algorithm.