基于神经网络的过程系统动态建模

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,提出利用神经网络的自学习、自适应功能实现动态在线建模。本文对这种方法进行了仿真研究。由计算机产生仿真输入信号：随机信号或Ｍ序列伪随机信号,输入到生产过程中普遍存在的一阶纯滞后对象。通过三层ＢＰ神经元权值的不断调整,实现离线辨识和在线辨识,直到神经网络的阶跃响应曲线几乎和实际系统的阶跃响应重叠。仿真结果表明,

System dynamic modeling based on neural network

Usually, the parameters of the real time system will be changed from time to time, and the traditional system identification methods are not so precise to model the system. Therefore, considering the "self-adaptive and self-learning "characteristics of a neural network, this paper brings out a new method to use neural network to realize dynamic modelling and do some useful simulation research. The computer generated M series as input signals to excite a normally existing model (with pure delay and first order). Then, the BP neural network modified its neural values to minimize the errors.After off-line identification and on-line identification, the neural network reaches almost the same step response as the real time system's step response. The simulation research shows that the neural network can model the procedure system with higher accuracy and easier flexibility than the traditional identification method. It also implies that it can further be used in the forecasting control, adaptive control or the random system control fields.