Optimization Of Compressor Operation And Maintenance Strategy Based On Data-Driven

Natural gas is a clean energy widely used in China.Due to its flammable and explosive characteristics,most of the transportation methods of natural gas are pipeline transportation.Compressors are key equipment in long-distance natural gas pipelines,and the stability,safety,reliability,and economy of compressor units during operation have an important impact on the economic benefits and safety production of natural gas pipeline transportation enterprises.This article has optimized the compressor operation and maintenance strategy from the perspective of data,designed matching algorithms and software,verified using real compressor data,and carried out pilot applications.The main content of the paper is as follows:(1)Collect dynamic data during compressor operation from on-site real-time monitoring systems and database history records,and collate static data of the compressor from maintenance records,account data,spare parts delivery records,shutdown reports,and other documents to form a comprehensive compressor database with multiple models,types,and sites,laying a high-quality information foundation for research topics.Based on this,adaptability analysis is conducted for the characteristics of data and compressor operation and maintenance process,and the overall architecture of the new compressor operation and maintenance strategy is introduced.Three main lines of the model are determined: optimizing the maintenance decision-making model using historical compressor failure data such as fault database and maintenance records,screening potential risks using historical compressor operation data such as alarm records and loading time,and using exhaust pressure The real-time monitoring data of the compressor such as lubricating oil temperature assists in fault diagnosis.(2)Analyze the characteristics of historical compressor fault data,design a maintenance decision model for a specific type of compressor,and optimize the compressor maintenance outline.Based on the relevant data of "AXX5000 inverter and BXX452 electric drive compressor unit",a specific case analysis was conducted.According to the specific functions,the compressor was divided into 11 systems,and the components of each system were analyzed.On the basis of the failure mode and impact analysis(FMEA)analysis method,the "entropy weight method" was introduced to evaluate the degree of component failure risk,and a logic decision diagram was designed based on the reliability centered maintenance(RCM)idea,The items in the compressor maintenance plan that are under maintenance or over maintenance have been corrected and adjusted,forming a set of compressor maintenance strategy optimization methods based on historical fault data,and finally integrating the functions into the software.(3)Sensors collect a large amount of status information during the operation of the compressor,and the information is saved to the server to form historical operation data of the compressor.Experts mine the information in these data to assess the deterioration of the compressor,and ultimately identify specific equipment with potential risks.Taking the data of71 screw air compressors as an example,we searched for equipment with fault risks from three aspects: operation data overrun alarm,key indicator analysis,and power consumption analysis.Conduct key troubleshooting for equipment alarms,and analyze whether the compressor has faults based on its functions and environment;Conduct horizontal and longitudinal comparative analysis of the pressure loading time of the air compressor to screen for machines with abnormal performance;Analyze whether the energy consumption of the compressor conforms to the factory design from the perspective of power transmission efficiency,and timely identify abnormal machines.(4)In order to quickly form the preliminary diagnosis results of the compressor,two fault diagnosis algorithm models are designed for real-time data collection of the compressor,respectively realizing two functions: fault type detection and operation status prediction.The wolf swarm algorithm is introduced into the LM-BP neural network model.The wolf swarm algorithm is used to search for the optimal input weight ratio of the neural network.The Levenberg-Marquardt algorithm is a variation of the Newton method and is used to minimize the square sum of nonlinear functions.Both of these methods can improve the accuracy and convergence speed of the BP neural network classification algorithm.The new algorithm can diagnose real-time data and determine whether the compressor has failed,And speculate on the type of fault;In order to predict the degree of compressor degradation in advance,a CWD-LSTM(composite variable,wavelet transform,depth automatic encoder,and short-term memory neural network)prediction algorithm was designed.The algorithm separates the main variables and auxiliary variables,and then performs a third order wavelet decomposition to counter the interference of background noise on the prediction.The auxiliary variable is dimensionally reduced by depth self coding(DAE)to obtain a two-dimensional feature vector,which is input into the LSTM neural network as an information enhancement.The main variables and eigenvectors predict a set of wavelet coefficients,and then reconstruct the wavelet coefficients to obtain the predicted values.This CWD-LSTM algorithm can predict the key indicators of the compressor and send an alarm for abnormal data in advance.This paper proposes a set of data driven optimization methods for compressor operation and maintenance strategies.The core idea is to deeply mine compressor data to obtain effective information to help enterprises improve compressor operation and maintenance efficiency,reduce compressor operation risks and costs,and reduce the maintenance workload of frontline employees.