Research And Application Of Data-Driven Abnormal Conditions Detection Technology For Petrolchemical Equipment

Petrochemical industry is a high-risk field where danger or disaster occur at any time.Any carelessness may lead to accidents,equipment damage and other serious consequences.In order to ensure the safety and reliability of petrochemical production process,the research on abnormal working condition detection technology of petrochemical equipment has important practical significance and engineering value.Typical equipment such as cracking furnace and heating furnace are taken as the object.A data-driven approach is adopted to study the abnormal detection method of petrochemical equipment.The data acquisition,pre-processing,abnormal detection method and technology of DCS system based on OPC technology are mainly researched.The software for abnormal change detection of petrochemical equipment working condition data is designed and developed.The specific research work is as follows.(1)Aiming at the problem that it is difficult to obtain DCS data in industrial field,a DCS data acquisition system is developed based on OPC UA communication standard to realize data acquisition.The OPC UA server module and client module included in the working condition data acquisition system are tested and verified based on the mature OPC communication test tool.The results show that a reliable communication link based on OPC protocol can be established by the developed OPC UA server and client to meet the requirements of data acquisition tasks.(2)Aiming at the problem of data loss in working condition data acquisition task,a classification and processing method of working condition data loss value is proposed.For the case of single point missing data,a repair method based on Lagrange three-point interpolation method is proposed.Aiming at the continuous multi-point data loss,a repair method based on BSVR prediction is proposed.Based on the measured data,the above method is simulated to verify the effectiveness and feasibility of the classification and processing method of working condition data loss value proposed in this paper.(3)Aiming at the problem of abnormal change detection of one-dimensional petrochemical units working condition data,two one-dimensional working condition data abnormal change detection algorithms based on NLOF and CHQ are proposed from the perspective of density and filtering respectively.Based on the one-dimensional working condition data,the above algorithms are simulated and calculated to verify that the two proposed algorithms are sensitive to the abnormal temperature data at the outlet of cracking furnace and can accurately identify the abnormal data contained in the experimental data.(4)Aiming at the problem of abnormal change detection of multi-dimensional petrochemical units working condition data,based on the deep learning method,a hybrid neural network prediction model DLA integrating DCC,LSTM and Attention for multi-dimensional working condition data is proposed.Modeling and calculation are carried out to verify that the DLA prediction model can predict the trend of multi-dimensional data with high accuracy.The abnormal change detection of multi-dimensional data is realized by comparing the residual data.(5)Based on the algorithm researched in this paper,the abnormal condition detection software system driven by the working condition data of petrochemical equipment is designed and developed.According to the functional requirements,the working condition data acquisition module,data processing module,abnormal change detection module and visualization module are developed,and the usability of the software is verified based on the measured data.Aiming at the research on the key technology of data-driven abnormal condition detection of petrochemical equipment,the main innovative work of this paper is as follows.A DCS condition data acquisition system based on OPC technology is developed.A condition data missing supplement method based on BSVR is proposed.An improved LOF algorithm and a onedimensional condition data variation detection algorithm based on filtering method are proposed.A multi-dimensional working condition data variation detection method based on new hybrid neural network is established.The effectiveness and feasibility of the proposed method are verified by numerical simulation and calculation analysis.The research results can provide technical support for the detection of abnormal working conditions of petrochemical equipment.