Prediction Method And System Application Of Key Corrosion State Parameters In Oil Refining Process

Corrosion poses a serious threat to the safety and long-cycle stable operation of oil refining units.Through the means of monitoring and detection of refinery equipment corrosion environment and state identification is an important part of the refinery enterprise corrosion management.The current corrosion management is still difficult to identify the corrosion status,long period,high cost,and data-driven analysis and prediction is an effective way to solve the problem of ideas and methods.In this paper,for the refining process corrosion occurrence law and the existing corrosion-related historical data,the establishment of data mining-based corrosion key parameters prediction method process and key parts of the prediction model.First,according to the water quality analysis and laboratory data collected from the circulating water plant and the corrosion results measured by the corrosion weight loss experiment,a BP neural network-based prediction model for the corrosion rate of circulating water in oil refineries is established,and PCA is applied to reduce the dimensionality of the original high-dimensional data,compare the impact of the reconstructed data dimensionality on the prediction results,and optimize the BP neural network model parameters in combination with PSO.Second,for the characteristics of refinery process production and safety management,a prediction method including data source analysis,data conversion,data evaluation and pre-processing,model construction,etc.oriented to data mining and application requirements is proposed.Based on the data of eight process operating parameters and media analysis and assay of corrosion multi-influence variables of the low temperature atmospheric tower overhead system,the corrosion rate is taken as the key parameter to reflect the corrosion status of the equipment,and the regression prediction model of key parameters at atmospheric tower overhead based on random forest is established.By using Isolation Forest to detect outliers on the original field data and applying Symbiotic Organisms Search algorithm to optimize the model parameters,the prediction model obtained a better prediction performance after 5-fold crossvalidation.Meanwhile,the prediction performance of multiple machine learning algorithms was compared.Then,for the problem of data noise in corrosion monitoring data,the EMD algorithm was applied to noise reduction of the time-series data,and the timeseries data were transformed into standard data set in the form of sliding window,and the time-series corrosion rate prediction model and the time-series corrosion depth prediction model were established based on LSTM,respectively.The effects of different data input and prediction length on the prediction results were compared.Finally,with the refinery unit site corrosion management as the guide,the refinery unit corrosion intelligent prediction and early warning system is established.The system integrates site corrosion-related data,data processing algorithms and corrosion prediction models to realize the prediction and early warning of key corrosion parameters of the site equipment.This paper proposes a corrosion prediction model with certain generalization performance for the key parameters of the refinery process,and establishes a corrosion management system applicable to the refinery site,which provides guidance for refinery corrosion management and identification of safety hazards,and has certain application value.