Research On BP Neural Network Optimization By Cuckoo Search Algorithm

Artificial neural network (ANN) has been successfully used in various fields, such as pattern recognition, machine learning, signal processing and information infusion. However, improper selection of learning algorithms and topology may lead to low forecasting precision. Therefore, many researchers have merged modern metaheuristic algorithms into traditional neural networks with the aim of improving its efficiency.As a new kind of heuristic search technique, CS has only one parameter and its optimization process is simple. Especially, CS method can make a good trade-off between exploration and exploitation. Thus, CS is incorporated into the neural networks and utilized to find the best weights and bias. A new kind of neural network is constructed in terms of the obtained optimal thresholds and weights. The main contributions can be given as follows:Based on the description of the CS algorithm and BP network, a new kind of neural network, called CSBP, is proposed. In this method, the training error is considered as fitness function. CSBP can optimize the weights and bias of the neural networks, and has been successfully used to solve the Italian wine classification problem, and the results show that the prediction error of the CSBP is significantly less than traditional neural networks (such as BP and GRNN). In order to make CSBP implement in the best way and improve efficiency, three parameters:population size (NP), maximum number of iterations (Maxgen) and discovery rate (pa), were studied through an array of experiments. CSBP is insensitive to three parameters, which further show the high efficiency and strong robustness of the proposed method.