一种FIR神经网络的在线学习算法在系统辨识中的应用

An On-line Learning Algorithm of FIR Neural Network in System Identification

通过在指标函数中增加遗忘因子,提出一种FIR神经网络的在线学习算法.该算法具有随训练的样本进度即时更新权值的在线训练能力,与同类方法相比,因没有复杂的矩阵计算,使计算时间大大减少.编制了算法程序,对真实的六机架热连轧机试验数据进行了辨识,所耗用的计算时间远远小于实际数据的采样时间,验证了该算法的在线辨识能力.

Through adding oblivious factors to cost function, an on-line learning algorithm of FIR neural network is proposed. It has the ability of temporal update weights with training sample process. Compared with the other algorithms of the same kind, it is less time-consuming because of no complex matrix mathmatical calculations. Program is done and system identification has been completed successfully on actual datasets of 6 stands tandem hot mill. When simulation is performed, less time is required than the sampling time of actual datasets. It verifies the ability of the on-line identification of this learning algorithm.