The Improvement Of Differential Evolutionary Algorithm And Its Application In Clustering

Differential evolution algorithm(DE) is a new evolutionary algorithm used a difference-based mutation strategy, its unique memory function can dynamically track the current search status. Its principle is simple, easy to implement and has strong ability of convergence and robustness. DE algorithm can effectively deal with complex optimization problems and do not depend on the specific characteristics of the problem. At present, the differential evolution algorithm has been widely used in many fields, its research results have involved multiple related disciplines.Mutation strategy plays a decisive role on the success of the differential evolution algorithm. However, the direction information has not been fully exploited in the designing of DE and the balance between the evolution speed and the population diversity cannot be well handled so far. In this paper, we explore a novel direction information which is generated by the selection operation and it’s directive effect on the mutation operation. On this basis, we propose an evolution direction-based mutation strategy "DE/current-to-pbest/ 1/Gvector" and an improved differential evolution algorithm based on adaptive differential evolution algorithm JADE for comparison. We name our algorithm as DVDE and compare it with five state-of-the-art adaptive DE variants (JADE, SaDE, CoDE,jDE, EPSDE), using 25 standard numerical benchmarks taken from the IEEE Congress on Evolutionary Computation 2005.The simulation results indicate that the average performance of the DVDE is better than those of all other competitors, especially for the unimodal functions. The experimental results also illustrate that the using of the evolution direction is helpful to improve the algorithm’s convergence speed, maintain the population, and effectively avoid premature convergence problem.K-means algorithm is a kind of typical partition clustering algorithm, which is very effective for the processing of large data sets. K-means algorithm, however, need to determine the clustering number beforehand. In this paper, we use DVDE algorithm for clustering and propose a new automatic clustering algorithm based on DVDE,named as AC-DVDE. Firstly, we take double crossover strategy for K-means clustering. Specifying to the encoding mode of DE used for clustering, we add a new crossover strategy after the conventional two point crossover operation. This new crossover strategy act directly on two complete clustering centers derived from parent vector and trial vector. Secondly, we make improvements on the drawbacks that the selected clustering centers may deviate from the data set or they are too close result from the randomness of mode for choosing clustering center. Sift clustering centers before choose some of them for clustering results in a better effect. The experimental results carried on 4 UCI datasets verified effectiveness of the proposed algorithm.