过程控制中被控对象神经网络模型的设计

NEURAL NETWORK MODELLING FOR A DYNAMIC PROCESS PLANT IN PROCESS CONTROL

根据控制过程的特点提出了一种新的动态系统辨识方法 将输入、输出信号、滤波后的数据用于训练静态前向神经网络 ,辨识出控制系统工作点处一线性动态模型的参数 ,从而预测过程的输出 文中详细论述了设计方法及算法原理 ,并通过试验对比表明 ,这种模型精度高、计算量少 ,对噪声不敏感 ,特别适用于运行过程复杂、干扰因素多。

Presents a new dynamic system identification method based on the operating points of the process plant. A feedforward neural network is trained using filtered inputs and outputs to identify the plant parameters at the operating points. The parameters are then used by a linear dynamic model to predict the future process output. The proposed identification algorithm was analyzed in detail. Experiments were performed and the results were compared with existing dynamic neural network model, showing that the new method is fast and accurate, and is particularly useful in noisy nonlinear identification.