Hybrid Bat Search Algorithm And Its Application

Bat search algorithm (Bat Algorithm, BA) is inspired by bat’s echolocation behavior, a new heuristic search algorithm proposed by Cambridge University Professor YANG Xin-she in2010. It has been successfully applied in the neural networks, multi-objective optimization, distributed network reconfiguration and others optimization issues. Because the algorithm has the characteristics of a simple structure, less parameters, robustness and it’s easy to understand and program. In recent years the bat algorithm attracts more and more attention to the scholars, and becomes the important research in the field of computational intelligence. However, because of the bat algorithm has just been proposed shortly. There are still many shortcomings in itself, such as slow convergence in late, lack of search activity and it’s easy to fall into local optimum. Bat algorithm has greatly limited the application of itself. This thesis will use the integration strategy improving the performance of the basic algorithm for preventing this shortcoming. This paper will discuss to improve the study of the BA algorithm theory and to expand the application of BA algorithm.The research results obtained are as following:(1) The basic bat optimization algorithm accuracy is not high and it has the precocious phenomenon. So a simplified adaptive bat algorithm based on frequency (FSABA) which is inspired by the frequency of bat echolocation is proposed. We use bat population diversity for the search of the global, and at the same time once out of range value, the value will be instead of a value randomly selected. With the experiments showed, the new algorithm is proposed to effectively prevent premature and improve the convergence speed and accuracy.(2) Based on the basic bat algorithm, we use the characteristic equation method to analyze the convergence of the bat algorithm. Accordingly we propose a new improved algorithm and use the same method to prove the convergence of the new bat algorithm for converging to the global optimum.(3) Based on the study of weights and bias input of extreme learning machine which is random to basic complex network structure. The new bat algorithm will be applied to extreme learning machine to train and to obtain a more reasonable input weights and bias. So as to improve the unknown data accuracy of the ultimate learning machine for prediction and obtains a more reasonable network structure. The proposed method is applied to classification applications in order to obtain a satisfactory precision.