The Multi-Objective Evolutionary Algorithm Based On CUDA And Its Application In The Process Of Chemical Engineering

Intelligent optimization algorithm has a very important position in solving optimization problems, especially in NP-hard fields. With the rapid development of intelligent optimization algorithm in theory and application, the complexity and data density of problems are increasing. The structure and implementation of the original algorithm makes the solving efficiency relatively low. This phenomenon is particularly normal in the multi-objective problem. The time complexity of calculating Pareto curve is high, and the efficiency of algorithm is reduced.By this problem of the multi-objective evolutionary algorithm, a solution based on CUDA is proposed in this paper. It accelerates the algorithm, making its application prospect broadly. Firstly, this paper summarizes the multi-objective problem, the multi-objective evolutionary algorithm, the CUDA parallel programming architecture and programming model from the macro-level. And then details the development of multi-objective evolutionary algorithm as well as related concepts. New CUDA-based parallel multi-objective evolutionary algorithms are proposed, which use niching technology based on fitness-sharing model and its improvement density entropy scheme to keep the distribution of population. Experiments show that the new algorithms are more efficiency in dealing with multi-objective optimization problem of the ZDT series.At the last of this paper, to the optimization problem for chemical process, the proposed parallel algorithm based on niche is applied to constrained multi-objective problem of gasoline blending optimization. The proposed parallel algorithm based on density entropy is applied to dynamic multi-objective problem of feeding batch bioreactors. The application examples show that the proposed algorithms are much more efficient.