Complicated Optimization Problem Oriented Multi-objective Evolutionary Algorithm

Multi-objective optimization problems(MOPs)aimed at the optimization of multiple conflicting optimization objectives at the same time and have been recognized as a category of important yet difficult problems.Two aspects have seen the complexity of MOP,namely the complexity in the decision space and the objective space.In the decision space,the decision vector is chain-correlated with dimension explosion,whereas,in objective space,the complexity is brought about by the complicated-shaped Pareto front(PF)with the imbalance between each optimization objective.All those factors lead to the poor convergence and diversity of the final-obtained solutions.Decomposition based multiobjective evolutionary algorithm(MOEA)decompose the optimization problem by employing a set of pre-defined weight vectors,which is significant for the maneuverability of searching direction.This way,decomposition-based MOEA is a promising method for overcoming the abovementioned difficulties.This work proposing several ideas in both the decision space and the objective space at the same time on the basis of decomposition-based MOEA to enhance the optimization quality.Furthermore,the reservoir flood control operation problem for the Ankang reservoir in Shaanxi province is introduced and optimized.Experimental results indicate the proposed methods in this work to be effective for both benchmark problems and the real-world problems.The main contributions are as follows: 1.An adaptive population uniformity adjusting method is proposed to improve the uniformity of the obtained population.This method can be assembled to the MOEA/D framework to improve the quality of the obtained population by adjusting the correspondent weight vectors in terms of the triangulation mesh generated by Delaunay triangulation.2.A user-preference based weight vectors generation method is proposed in the weight space of decomposition-based Multi-objective optimization problem to generate an arbitrary number of weight vectors specified by the user in the given Region of interest in any area.3.The user-preference based weight vector generation method is assembled to the MOEA/D framework to convert it to its user-preference based version to obtain a solution in the region of interest.Meanwhile,Achievement scalarizing function is also assembled into the MOEA/D framework to further improve the convergence speed.Finally,the userpreference based version of MOEA/D is applied to the RFCO problem to obtain scheduling schemes of RFCO problems with particular preference.4.In the decision space,a data-driven population initialization method is proposed.This method boosts the convergence speed of the RFCO problem by transferring existing solutions of history optimization results of history flood-optimization instance to the given new instance so that the quality of the initial population can be improved.This way knowledge of solving the RFCO problem can also be conveyed to the new problem inexplicitly and the convergence speed can thus be improved,and finally,the difficulty of low searching efficiency brought about by the large-scale decision vector can also be alleviated.