Adaptive Memetic Algorithm For Data Clustering

Clustering analysis is a basic form of data mining technology,which can be used to discover useful knowledge and information from vast amounts of irregular data.K-means clustering algorithm is favored by many scientific researcher enthusiasts,because of its simple and efficient.But it is easily affected by the initial clustering centers and isolated points,at the same time,it has a strong local search ability,so it is easy to fall into local optimum when solving the optimal solution.Evolutionary algorithm is a stochastic optimization bionic algorithm based on Darwin's theory of biological evolution,which simulates natural evolution and biological evolution.It is based on population parallel search,with a strong global search ability,especially suitable for solving nonlinear optimization problems with complex solution space such as clustering.However,due to the randomness of the search process so that the convergence speed is slow when solving complex clustering problems.Hybrid the k-means algorithm and the evolutionary algorithm,which is known as memetic algorithm in the field of evolutionary computation,can effectively balance the advantages and disadvantages of both in the search process,greatly improve the performance of the algorithm in solving complex clustering problems.So the study of clustering analysis algorithm based on memetic algorithm can not only promote the development of the research field better and faster,but also provide a more effective solution for practical engineering application.The optimization search algorithm is a dynamic search process,so in the memetic algorithm,the set of local search intensity has a great impact on the efficiency and accuracy of the algorithm.Based on the k-means algorithm and evolutionary algorithm(we used genetic algorithm in this paper),according to k-means algorithm in the existing literature as a local search algorithm,with one-step intensity when mixing with the global search algorithm and all individuals of each generation are using the same intensity,an adaptive intensity control strategy based on Q-Learning reinforcement learning is proposed in this paper.The maximum k-means iterative intensity is chosen in an adaptive way using the adaptive intensity control strategy,then use the chosen intensity as the maximum iterative intensity of the k-means local search when combining it with the global search of the evolutionary algorithm.At the same time,the localsearch result will be saved in order to guide the subsequent search process,thus in the search process it can dynamically according to the current characteristics of the solution individual and learning feedback adaptively adjusting the global search and local search.The proposed adaptive k-means intensity control strategy is tested on a large number of simulated and real data sets,and compared with the related methods,experimental results show that the proposed method has high robustness in large data sets,and for different types of data sets it shows high universality.