Research On Diagnosis Method Of Abnormal Condition In Transfer Station Based On Deep Learning

Oil and gas gathering station is the core part of oilfield ground engineering construction.The diagnosis of the operation condition for oil and gas station is crucial to oil and gas production system.In the context of AI oilfield construction,it’s possible to diagnose and analyze the operation of station process through the data collected by station monitoring facilities.In this paper,SCADA data from an oil field transfer station is taken as the research object,and deep network is taken as the diagnostic tool,to study the data processing method and construction method of abnormal condition diagnosis algorithm.In view of the multi-source heterogeneity of oil and gas station monitoring data,regularization is determined as a data reconstruction method.Aim at the problem of noise caused by unstable performance of station signal collector or fluctuation of working conditions,wavelet method is used to separate the noise.Proposed data preprocessing flow which is suit for transfer station.In view of the severe imbalance of data for all conditions in transfer station,analyze the improvement effect of under-sampling,stochastic over-sampling,SMOTE oversampling,or other data balance methods on the data imbalance of transfer station.Improved Time-GAN is used to generate the few class samples from transfer station database.The effects of input type,network structure and model parameters on the quality of generated data is discussed.PCA and T-SNE is used to visualize the distribution of generated data and original data.Aiming at the characteristics of much working conditions in the transfer station,the adaptability of machine learning models such as CNN,SVM,MLP and ResNet to abnormal condition diagnosis task of transfer station is analyzed.The effects of input type,network structure and model parameters on the diagnostic accuracy of the transfer station anomaly diagnosis model is discussed.The reliability of the diagnosis results is verified by the correlation analysis method based on mutual information.The structure design and parameter setting of deep networks are difficult,relying on expert experience and repetition test.To solve this problem,an adaptive construction method of deep network is proposed.Genetic algorithm is applied to optimize the network structure and parameter range of CNN,constructed the optimal CNN model for abnormal condition diagnosis task of transfer station.