Research On Gravity - Moving Algorithm Based On Weight

In recent years, many experts conduct inspired by nature, we made a lot of heuristic optimization algorithms to solve complex computational problems, such as: ant colony algorithm, particle swarm optimization, simulated annealing algorithm, genetic algorithm, and so on. While these algorithms to solve some problems it provides an effective way, but there is in solving optimization problems in high dimensional space easily fall into local optimum result in slow convergence and low precision. Therefore, continue to explore new heuristic optimization algorithm is still necessary.With the development of the times, it is widely used in many fields of data mining, clustering analysis as an important research direction in data mining, but also more and more attention and discussion of research staff. Traditional clustering algorithm for high dimensional data at cluster appears sensitive to initial data, clustering performance and other shortcomings can not even complete the clustering task.In this paper, the following aspects for the above mentioned problems:(1) To improve the searching performance of gravitation move algorithm, in accordance with problems of bad performance in search accuracy and slow convergence speed in the high dimensional space optimization, weighted GMA is proposed by introducing a weighted value to inertia mass of every individuality in each iteration process. The induction is made to updating individuality’s position by the new arithmetic in order to improve the ability of exploitation. Thirteen benchmarks function are tested and show that new algorithm is better than GMA with both a steady convergence and a better accuracy of solution.(2) The fuzzy C-means clustering algorithm is a local search algorithm, which is an iterative optimization process gradient approach to achieve a gradual decline, resulting algorithm is easy to fall into local optimum value; and the clustering algorithm to initialize the cluster center position and other sensitive data, however, the improved algorithm is based on gravitational movement group behavior, we do so with a certain probability of an initial position groups uniformly distributed in the solution space, compared to some other algorithms of its relatively strong global search capability, convergence speed fast, we combine these two algorithms to overcome the shortcomings of fuzzy C-means algorithm, resulting in an ideal clustering effect.