论文部分内容阅读
提出了基于双层机器学习的动态精馏过程故障检测和分离的方法,检测的阈值为正常工况训练的网络输出值与样本的残差.通过对比网络预测值和实测值的偏差检测故障,检测到故障时,启动另一网络对动态过程自适应拟合异常工况数据.网络的预测值与实测值的偏差小于阈值时,拟合成功.通过对两个网络进行结构解析找到造成输出变量异常波动的输入变量.将该方法运用到脱丙烷精馏塔中,检测出过程中的故障,并分离出与故障源相关的变量,表明该方法准确、有效.
The method of fault detection and separation in dynamic distillation process based on two-layer machine learning is proposed.The detection threshold is the network output value and the sample residual in normal working conditions.By comparing the deviation of network forecast value and measured value, When a fault is detected, another network is started to fit the dynamic process adaptively to the abnormal working condition data. When the deviation between the predicted value and the measured value of the network is less than the threshold value, the fitting succeeds. By structural analysis of the two networks, Abnormal fluctuations of the input variables.This method is applied to the de-propane distillation column to detect the process of failure and isolate the source of the variables associated with the failure, indicating that the method is accurate and effective.