论文部分内容阅读
针对生产过程中参数容易受外界影响而改变 ,传统的系统辩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神经网络的神经元权值的不断调整 ,实现离线辩识和在线辩识 ,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 ,神经网络的自我学习能力应用到动态建模中能以较高的精度逼近实际系统 ;其“在线更新”特点将能进一步应用到预测控制、自适应控制和随机控制等领域。
In view of the fact that the parameters of the manufacturing process are easily influenced by the outside world, the traditional method of system identification is hard to get the actual situation of the mathematical model. The dynamic online modeling based on the self-learning and self-adaptive function of neural network is proposed. This article has carried on the simulation research to this kind of method. Generated by the computer simulation input signal: random signal or M-sequence pseudo-random signal, input to the production process ubiquitous first-order lag object. Through the continuous adjustment of the neuron weights of the three-layer BP neural network, offline recognition and on-line identification are realized until the step response curve of the neural network almost overlaps the actual system step response. The simulation results show that the self-learning ability of neural network can be applied to the dynamic modeling to approximate the actual system with high accuracy. The “online update” feature will be further applied to the fields of predictive control, adaptive control and stochastic control.