中文摘要：

提出一种基于差分进化算法（DE）的径向基函数神经网络（RBFNN）模型,用于预测直线伺服系统的定位误差。该模型用差分进化算法训练径向基函数（RBF）网络隐层中心位置、宽度和输出层连接权重。为了评价优化后RBF网络预测的精度,运用部分误差样本进行训练和仿真。构建了以数字信号处理器（DSP）为核心的直线电动机定位误差实验平台,根据误差校正值进行误差实时补偿实验。仿真和实验结果表明：经过DE算法训练的神经网络模型对工作台的误差具有良好的学习能力和泛化能力,与单纯RBF网络、基于遗传优化的RBF神经网络相比,该建模方法具有更高的定位精度。

英文摘要：

A novel Radial Basis Function Neural Network（RBFNN）model based on learning algorithm Differential Evolution（DE）is proposed to predict the positioning error of linear servo system.The DE algorithm automatically adjusts the width and positions of hidden layer RBF centers as well as the weights of output layer.In order to evaluate the accuracy of optimized RBF network prediction method,part of the error samples are used to train and simulate.A DSP-core linear motor positioning error experimental platform was built,the error compensation experiments are conducted.The simulation and experimental results indicate that RBFNN error model trained by the DE algorithm has a good learning ability and a generalization ability,the DE-RBFNN possesses superior positioning accuracy than RBFNN,GA-RBFNN.