Fault Diagnosis And Product Quality Prediction In Petroleum Production

Petroleum is one of the important economic pillars of our country.The production process of petroleum products is complex,and there are strict requirements for the control of production equipment and the management of the production environment.In recent years,with the application of Internet of Things technology in industrial production and the rapid development of big data analysis technology,industrial fault diagnosis and product quality prediction based on machine learning has become one of the research hotspots.According to the actual industrial scene,selecting the appropriate intelligent algorithm and model is one of the key issues of research.Industrial production collects large amount of data,the data has high feature dimensions,and contains a large amount of noise.Therefore,data preprocessing and feature engineering are required.Efficient feature engineering can help the model achieve higher performance and more accurate results.At present,machine learning has achieved good research results in many fields,but deep learning has achieved better performance in some scenarios.This paper analyzes and studies the fault diagnosis and product quality prediction of the production process for the petroleum production process.(1)Troubleshooting of the production processThe petroleum production process is complex,and the number of production equipment is huge.Once equipment fault occurs,it not only affects the production of petroleum,but also leads to safety accidents.Therefore,it is crucial to find faults in time.This paper analyzes the historical data of petroleum production and finds that there is a problem of data imbalance in existing data sets.The types and quantities of fault cases are limited,and it is impossible to provide sufficient samples for the training of classification models,thus affecting the accuracy of the model..Intelligent algorithms such as random forest and support vector machine are currently used in fault diagnosis of industrial production.In order to solve the problem of fault diagnosis under high feature dimension and data imbalance,this paper proposes a modal based on the one class support vector machine classification.The fault diagnosis model of random forest and single-class support vector machine,the model can obtain high fault diagnosis accuracy under the condition of data imbalance.(2)Petroleum product quality predictionPetroleum quality testing has high requirements for testing equipment and testing technology,and the testing cycle is long,and the quality results cannot be fed back in time to adjust equipment parameters and production processes.Therefore,combining the production equipment and environmental parameters to predict the quality of petroleum products can save a lot of testing costs and feedback results more quickly.According to the quality feedback,the production status can be judged so that the equipment parameters and production process can be adjusted in time.At present,the partial least squares method is applied to the prediction of petroleum product quality,and the partial least squares method is a combination of principal component analysis,correlation maximization and multiple linear regression.In this paper,the partial least squares method is improved,and the petroleum product quality prediction regression model based on KPCA and RNN is proposed,which can better fit the nonlinear relationship between input and output of petroleum product quality prediction.The PCA selects features according to the information contained in the feature,thereby reducing the dimension of the feature space,which can save the model training time and space consumption,and improve the accuracy of the model understanding of the feature information.KPCA can be better used for complex nonlinear feature selection.The neural network can realize complex nonlinear mapping and has strong fitting ability in the case of high feature dimension.Through the comparative analysis and verification between the models,it is proved that the proposed KPCA-RNN model has higher accuracy and stability.