应用BP神经网络预测石脑油热裂解产物收率

采用BP神经网络模型建立了石脑油裂解产物收率的预测方法。BP神经网络模型的输入层设12个结点,输出层设22结点,设一层隐含层。在保证学习训练数据具有代表性的情况下,BP神经网络模型的预测结果与实验数据相比,误差约为5%。BP神经网络模型的预测结果比非线性回归方法的预测结果要好。BP神经网络模型的外延性不强,外延的部分数据预测结果偏差较大。在能够保证基础学习训练数据的准确性和合理选取的条件下,BP神经网络模型能够应对乙烯装置原料变化频繁的情况。

Applying BP Neural Networks to Predict Product-Yields of Naphtha Steam Cracking

A method for predicting product-yields of naphtha steam cracking was established by means of BP neural network model.The model consists of three neuron layers: input layer with 12 nodes,output layer with 22 nodes and hidden layer.If training data are of representative,the results obtained by neural network model can be well in accordance with experimental results and its errors are less than 5%.The results obtained by neural network are more accurate than those obtained by non-linear regression.Extension of BP neural network is unsatisfactory and its errors are beyond control.Under conditions of accurate and reasonable training data,BP neural network model is a powerful tool for predicting product-yields of naphtha steam cracking especially when composition of the feedstock fluctuates.