基于长短时记忆神经网络和支持向量数据描述的间歇过程监测方法

A batch process monitoring method based on long short term memory network and support vector data description

间歇过程操作是化工过程中的一种重要生产方式。与连续过程不同,间歇生产不是在一个稳定的工作状态运行,而是根据设定的原料比例、操作条件所对应的操作轨迹运行。因此间歇过程数据具有多阶段性、动态时变性和非线性等特性,传统的监测方法难以应用于对间歇过程生产运行状态的监测。为了解决这个问题,提出了一种新的间歇过程监测策略。首先基于长短时记忆神经网络提取变量间的动态、非线性关系来建立回归模型。然后以回归模型得到的残差作为监测对象,由于残差已经不再有多阶段、动态特征,可以直接应用连续过程监测的方法对残差建立监测模型。最后选择支持向量数据描述对正常数据集的残差进行超球体描述,以测试集样本点到超球体中心的距离作为监测统计量来实现故障点的检测。本方法被应用于青霉素发酵仿真过程,结果表明本方法成功将复杂的间歇过程监测转化成较为简单的连续过程监测问题来处理,且能够及时对过程故障进行报警。

Batch process is an important operating mode in chemical process.Different from continuous processes,batch process production is not operated in a stable operating point,but a preset trajectory according to raw material ratio and operating conditions.Therefore,process data in batch process display complex characteristics including multi-stage,dynamic time-varying and nonlinear,which means traditional monitoring methods cannot be directly applied to monitor the operating status in batch process.To solve this problem,a novel process monitoring strategy is proposed in this paper.Long short-term memory network is first applied to establish a regression model by extracting the dynamic and nonlinear feature among process variables.Then residuals obtained from the regression model is considered as the monitoring object.Since the multi-stage and dynamic characteristics no longer exist in the obtained residuals,the continuous process monitoring methods can be directly applied to establish a monitoring model for residuals.Finally,the support vector data description is selected to describe the residuals of the normal data set in a hypersphere,and the distance from the test set sample point to the center of the hypersphere is used as a monitoring statistic to detect the faults.The method is applied to a simulated penicillin fermentation process,the results show that complex batch process monitoring can be transformed into a relatively simple continuous process monitoring problem to deal with,and the process faults can be detected in time.