Research On Differential Evolution-based Optimization Algorithms And Applications

Differential evolution algorithm is a new kind of evolutionary computation approach, which has the merits including excellent performance in global optimization, simple structure and easy to implemention. Because of its huge application potential, differential evolution algorithm has received widespread attentions of researchers. At present, differential evolution algorithm has been applied to many fields and its research achievements include a number of relative disciplines.Scheduling optimization, continuous optimization and constrained optimization problems are very common in mathematics and engineering practice. Therefore, how to solve them has important theoretical and practical significance. This dissertaion addresses solving the above three classes complex optimization problems with differential evolution algorithm. The main contributions of this work can be summarized as follows:1. To solve scheduling problem effectively, two improved differential evolution algorithms, permutation-based differential evolution (PDE) and hybrid PDE (HPDE), are proposed. Permutations are adopted to represent the solutions of scheduling problems, according to the characteristics of the permutations, a new location-based subtraction and addition operations are proposed and used to construct new mutation operator. And the crossover operator based on permutation is introduced to generate new individuals. These operators can guarantee the feasibility of solutions. In addition, in order to speed up the evaluation of solutions and improve the quality of new individuals, with the help of idle time increment matrix, the zero-wait scheduling problem is formulated as an asymmetric traveling salesman problem (ATSP), and a new local search method based on fast complex heuristics (FCH) is proposed. HPDE is used to solve large-scale zero-wait batch scheduling problems with setup times, results show that the proposed algorithm can quickly find high-quality solutions, and its performance is superior to genetic algorithm, tabu search and FCH approcah. In addition, HPDE has fewer parameters.2. In order to increase the diversity of the population, avoid unnecessary search and escape from local minima, HPDETL, which is the approach hybrid PDE with Tabu search, is proposed. In HPDETL, Ulam distance is utilized to measure the similarity between two individuals. The minimum Ulam distance between the new individual and members in tabu list is set in order to ensure the diversity of population, which can improve the global search capability. Furthermore, FCH local search strategy is used to find better solutions. To demostrate the performance of the proposed approach, HPDETL is used to solve no-wait flowshop scheduling problems and compare with several representive intelligient methods reported in literature, the results show that HPDETL is effective.3. To improve the performance of composite differential evolution (CoDE), a modified CoDE (MCoDE) algorithm is proposed based on improved trial vector generation strategies, and its performance is tested with benchmark functions. Furthermore, adaptive network based fuzzy inference system (ANFIS) modeling approach with MCoDE and Leave-one-out cross-validation (LOO-CV) is presented. In this approach, LOO-CV is used to reduce ANFIS fuzzy rule set and MCoDE is employed to learn the parameters of ANFIS. This modeling approach is used to predict the surface roughness in the milling process, the results show that it can obtain high-quality model and get satisfied prediction performance with small-scale dataset.4. To solve some engineering optimization problems with complex constraints, MOCoDE, which is the combination of the excellent constraint-handling capability of Oracle penalty function and good search capability of CoDE, is proposed. In order to meet the standards of constrained optimization problems, an improved Oracle penalty function method is presented to adaptively handle complex constraints. Furthmore, to deal with mixed variables, a generic continuous representation method for discrete variables is introduced. Therefore, complex constrainted problems with mixed variables are converted into unconstrained optimization problems with only continuous variables, and then CoDE is used to solve them. The computional results of constrained benchmark problems verify the feasible of MOCoDE. In addtion. MOCoDE is used to solve constrained optimization problems from the practical engineering. The compared results demonstrate that it can effectively solve optimization problems with complex constraints. Moreover, compared with the reported methods, MOCoDE has fewer parameters and better stability.