基于改进RBF神经网络的动态测量误差预测方法

A dynamic measurement error predicting method based on improved RBF network

该文针对传统的RBF神经网络预测方法的局限性,引入次胜者受惩(RPCL)算法和递归正交最小二乘算法(ROLS),进行了动态测量误差实时预测算法的研究。理论分析和预测实例表明,该方法预测精度明显高于传统的方法,具有很强的学习与泛化能力,在处理动态测量误差序列的预报问题和提高动态测量精度方面具有很高的应用价值。

In view of the limitation of a predicting method using the conventional radical basis function （RBF） neural network ,this paper puts forward the algorithm of rival penalized competitive learning （RPCL） and recursive orthogonal least square（ROLS） and makes a study of dynamic measurement errors. The theoretic analysis and predictive examples show that the method has powerful learning and universalized capability. Its predietve accuracy is higher than that of a conventional method.