Research On Parameter Adaptive Differential Evolution Algorithm And Parallelization

Differential Evolution(DE) is one of the excellent algorithms in evolutionary optimization, with features such as self-searching, self-adapting, parallelization, which has been successfully used to solve various engineering and scientific problems. Differential evolution algorithm is sort of parallel iterative optimization algorithm based on population. Compared with other evolutionary algorithms, DE outperforms better searching ability and convergence. And the performance of the algorithm is determined by the control parameters like F and CR. Classic differential evolution algorithm shows slow convergence rate, and has more possibility to fall into the local optimal problem in solving complex optimization problems. Therefore, there is a lot left to be improved in differential evolution algorithm. With the explosive growth of data, stand-alone operation of differential evolution algorithm in high-dimensional operations failed to meet the actual computing demands, which makes cloud computing platform with distributed architecture the ideal choice.Differential evolution algorithm parameter selection is studied in this paper, a novel type of dynamic guide parameter selection algorithm and parallel implementation of algorithms in cloud platform is proposed, which optimized the performance of the algorithm and improved the ability of the algorithm to deal with high-dimensional data. Work of this paper stated as follows:(1) To solve the problem of slow convergence speed of Differential Evolution algorithm, and easy to fall into local optimum, this paper proposed a new Dynamic Guide Parameters Selection Differential Evolution algorithm(DGPSDE), accelerated the convergence speed and enhanced global searching ability of the algorithm. The main idea of DGPSDE is to use the historical value of population information to guide the optimal parameter choice of the next generation. First, Establishing two ordered resource pools to preserve the scaling factor and crossover rate which generated by the Cauchy distribution. And algorithm detects convengence according to historical optimum during evolution process. By utilizing feedback of parameter control, DGPSDE is capable of guiding the distribution parameter evolution of the new generation dynamiclly, which accelerating the convergence speed and bettering the global searching ability of the algorithm.(2) Aimming at inefficiency of classic DE in processing high-dimensional problems, DGPSDE adopts parallel structure based on cloud computing platform to speed up the funtion efficiency. Main idea listed as follows: By realizing DGPSDE on cloud distributed computing platform, new algrithm employs the Map module to parallel processing the time-comsuming and complex evaluation function in evaluation progress. Meanwhile, DGPSDE summarizes the output data of the parallel computing results to enter the stage of selection. By repeating steps mentioned above in sequence to achieve global optimum. Parallelization of evaluation process abbreviates iteration time comsumption and accelerates running speed.