滑动数据窗口驱动动的的贝叶斯-高斯网络及其在非线性系统辨识中的应用

Sliding-data-window-driven Bayesian-Gaussian neural network and its application to modeling of nonlinear system

工业控制场合中,需要获取非线性被控对象的结构特性,而系统动态响应的数据直接从外部特征上反映了非线性系统结构关系.为了充分利用非线性动态系统响应过程中的数据,本文提出了一种基于滑动数据窗口(sliding data window)的贝叶斯-高斯神经网络(SW-BGNN)模型.该模型将数据融合于网络模型结构中,借助于贝叶斯推理和高斯假设,利用滑动窗口数据,实现非线性动态系统的辨识和预测.整个SW-BGNN本身需要确定的参数很少,因此运算的时间很短,适合于非线性动态系统的在线辨识.将SW-BGNN应用于几个非线性动态系统的辨识和预测,仿真试验结果表明了SW--BGNN模型的有效性.

In industrial control, the structure of the nonlinear dynamic system is determined by using the dynamic data of the controlled object. In order to make full use of the data obtained from the dynamic response process of the nonlinear dynamic system, a novel Bayesian-Gaussian neural network based on sliding-window（SW-BGNN） is proposed which combines the Bayesian reasoning formula with the Gaussian assumption. Based on the data in the sliding window, the operation process of SW-BGNN reasonably predicts the output of the nonlinear dynamic system in terms of a small number of parameters of the SW-BGNN. The SW-BGNN has limited computation time which makes it suitable to onlinear identification applications. Examples of identification and prediction of nonlinear dynamic system are presented.Simulation results show the effectiveness of the SW-BGNN method.