神经网络技术在试验精馏塔控制中的应用

Application of Neural Networks Technique in Tentative Distillation Tower Control

为了预测精馏塔底部产品的成分 ,建立了 4层前馈神经网络结构 ,作为动态系统的正向模型 ,并采用BP学习算法对神经网络进行了训练 ;建立了动态系统的神经网络逆模型 ,作为系统的控制器 ;采用神经网络内模控制结构 ,根据精馏塔第 2级的温度 ,对底部产品成分进行控制 .试验表明 ,神经网络法与气相层析法相比 ,能够以任意精度逼近任意非线性映射 ,更快地提供成品估算值 ,使控制系统更及时地采取措施 。

In order to forecast the product composition in bottom of the distillation tower, a four layers multi layer perceptron (MLP) neural network is set up. The neural network is the positive orientation model of dynamic system predicting the product composition. Back propagation algorithm is introduced to neural network train.The contrary model of neural network of the dynamic system is established as a controller. Applying the neural network inner model control construction, the product composition in bottom is controlled according to the second degree temperature of the distillation tower. Experiment shows that the neural network can approach discretional nonlinear mapping with freewill precision such that sooner provide the estimation of product composition than the meteorology analysis, and can make the control system react in time so as to improve the control effect.