二级RBF神经网络用于化工过程的动态监控

文章提出用二级ＲＢＦ神经网络来实现化工过程的动态监控。第一级ＲＢＦ神经网络称作预测网络，用于预测未来一段时间内的有关状态量。第二级ＲＢＦ神经网络称作诊断网络，它根据预测量判断它们是否相应于某种事故状态，如果是，则提示操作人员采取相应的措施。这样可起到防患于未然的作用。ＲＢＦ神经网络的基础是ＲＢＦ插值，文章提出了改进的ＲＢＦ插值方法。文章又提出了提前一个时间步进行预测以获得预测结果修正项的方法。另外，文章还提出对诊断网络输出结果进行变换使能合理反映事故可能性的方法。文章所述方法被用于加氢裂化脱丁烷塔开工过程的动态监控，取得了满意的结果。

A Two-stage RBF Neural Network Used for the Dynamic Monitoring of Chemical Processes

In this paper, we proposed the monitoring of chemical processes with a two-stage RBF neural network The first stage is called the prediction network, which predicts the future state The second stage is called the fault diagnosis network, which checks the prediction result to identify whether there will be any faults and diagnoses the category of the faults identified In this way the monitoring of the process can be completed To improve the performance of a RBF neural network, we gave out a theorem and a revisionary method about RBF interpreation Then we proposed the other two methods in this paper They are used to improve the output of the two RBF neural networks respectively, by which the sample numbers of the two networks can be grately reduced We used the two-stage RBF neural network to the monitoring the simulation of start process of the debutanizing column in the hydrocracking fracnation section, which showed a great success