带动量批处理梯度下降法对模型的动态辨识

针对生产过程中参数容易受外界影响而改变，传统的系统辨识方法难以得到精确的数学模型的实际情况，介绍一种用改进的BP神经网络辨识对象模型的方法。采用串一并联型的辨识结构；针对BP算法收敛速度慢，容易陷入局部最小的缺点，提出了带动量的批处理梯度下降的方法；为了更有效地辨识对象的动态过程模型，其输入／输出加上按拍延迟线。用MATLAB对该改进的BP神经网络辨识方法进行了设计、仿真和性能分析，结果表明：该方法具有良好的辨识能力。

Dynamic System Identification Based on the Conjugate Gradient Method with Momentum and Batch Techniques

This paper aims to introduce an improved BP neural network for system identification, because it is difficult to get accurate mathematic model using the traditional system identification. It puts forward a conjugate gradient method with momentum and batch techniques in order to resolve the slow astringency of BP algorithm and avoid falling into the local minimum. It also adds the delay -time beat lines to the input and output to effectively identify the dynamic processing model. The results of MATLAB program show that this method can make a better identification.