PSO聚类和梯度算法结合的RBF神经网络优化

Optimization of RBF Neural Network Based on Combination of PSO Clustering and Gradient Algorithms

针对制约径向基函数RBF神经网络发展及其应用的瓶颈问题,提出一种基于粒子群优化算法PSO的改进K-means聚类思想,以确定其隐节点的数目。结合梯度算法,通过最小化目标函数调节隐节点的数据中心、宽度和输出权值,最终达到优化RBF神经网络的目的;同时,将优化后的网络应用于滚动轴承故障模式识别。试验结果表明,该方法能自适应地确定RBF神经网络隐节点的数目并调整其结构参数,使网络具有较快的收敛速度和较高的收敛精度,从而准确地识别滚动轴承的故障模式。

Aiming at the bottleneck problem that restricts the development and application of radial basis function （RBF） neural network,the improved K-means clustering concept based on particle swarm optimization （ PSO） has been proposed to determine the number of hidden nodes. In addition,by combining the gradient algorithm,via minimizing the objective function to adjust the data center and width of the hidden nodes and weights of output, the optimization of RBF neural network is achieved. The optimized network is applied to the failure pattern recognition of roiling beatings. The result of test indicates that the method can adaptively determine the number of hidden nodes of RBF neural network and adjust the structural parameters to make the network featuring faster convergence speed and higher convergence precision ,thus it can accurately identify the failure pattern of rolling bearings.