Abnormal Conditions Warning Method For Refining Process Based On Data Cross-verified

The occurrence of abnormal conditions in refining process is often suddenly and accidentally,and will often spread quickly and be difficult to control once it happens,Abnormal conditions early warning method monitors the trend factors and abnormal characteristics inside data at early stage of accidents by data mining,so that can take timely containment measures to maintain the system security before the accident occurs.At the same time,because of the abnormal conditions early warning method requires high data quality,once the data monitoring system break,it will generate abnormal data that can affect the functions of petrochemical devicse and safety early warning systems.So that the diagnosis and recovery of abnormal data are equally important.Therefore,diagnosing and restoring abnormal data,ensuring the accuracy of data,mining abnormal characteristics inside data,then predict the occurrence of accidents and take measures to prevent it,is of great significance for the safe operation of refining process.(1)Refining process have complex parameters,large fluctuation of data and obvious noise.Traditional monitoring and diagnosis methods for abnormal data are not stable enough to meet the requirements of accurate diagnosis of abnormal data in refining process.Therefore,a diagnostic model for abnormal data of refining process was proposed.Firstly,established a parameter soft-sensing estimation model to fit parameters,then abnomalous data were discriminated according to residual values.At the same time,the outlier detection algorithm based on Gauss kernel density and the joint mutual information algorithm were used to judge the data and the correlation between data.Finally,using the form of joint index monitoring,synthesizing the three methods mentioned above,if two or more of them exceed the alarm limit,the data will be judged to be abnormal.The method was applied to the abnormal diagnosis of temperature parameters of polymerizer.The results showed that compared with traditional methods,this method effectively reduced the false alarm rate and was suitable for the abnormal data diagnosis of refining process.(2)In order to solve the problems of traditional methods that haved short prediction period and low prediction accuracy in the of data recovery for refining devices,a data soft-sensing recovery method based on optimized RNN-LSTM was proposed.Used PCA to reduce the dimension of data,a RNN neural network model based on LSTM architecture was constructed.Elastic network regularization Adam algorithm and genetic evolution algorithm were used to optimize the model.The results showed that compared with the traditional BPNN and PSO-LSSVM models,the optimized RNN-LSTM reduced the error index by more than 50%,reduced the oscillation amplitude and frequency of the fitting data,and achieved good performance in the recovery of alarm data.(3)Aimed at the problems that traditional abnormal conditions early warning method have narrow scope of application,single forecasting means and lack of in-depth data mining,a parameter risk early warning method for refining process based on loss function was proposed.The loss function was used to evaluate the expected loss caused by parameter deviation quantitatively,and the residual time model was used to calculate the probability of abnormal operating conditions,then obtained the risk value caused by parameter deviation.Achieve the risk early warning.Applied the methed in early warning of different parameters of different devices in petrochemical enterprise.Case study showed that the method can identify the early deviation of state parameters and detect the abnormal parameters about 10 minutes in advance.It provided an early warning method to identify the risk of abnormal conditions in the long-term operation of refining process.