Research And Application Of Data Driven Fault Prediction Algorithm For Reciprocating Compressor

Reciprocating compressor as an important dynamic equipment in petrochemical industry,its safe operation can not only provide a guarantee for the stable output of petrochemical products,but also has an important significance for protecting the safety of field workers.With the continuous development of information technology in the petrochemical industry,the data of reciprocating compressor has appeared the characteristics of multi-source,heterogeneous and massive.Through big data analysis technology,it is possible to realize the fault early warning of reciprocating compressor equipment.Taking the operation data of p301 a reciprocating compressor in a petrochemical plant in Qingdao as the research object,this paper summarizes the algorithms of data noise reduction and feature selection,the time series data prediction model based on neural network and the state recognition model,and applies them to the fault prediction of p301 a.The specific research contents of this paper are as follows:Firstly,the basic structure and working principle of reciprocating compressor are studied,the common faults of reciprocating compressor are analyzed,and the data source and characteristics of P301 A are introduced in detail.Secondly,the common data denoising and feature extraction algorithms are studied,and the preprocessing work is done for P301 A parameter data,including using EEMD algorithm to denoise the data,and selecting the key parameter features through Relief-F algorithm.Thirdly,three time series prediction algorithms based on neural network are studied and applied to P301 A.By comparing the prediction accuracy of BP,RNN and LSTM,the LSTM algorithm model with the highest prediction accuracy is selected as the parameter trend prediction algorithm in p301 a fault prediction.Finally,a combined model based on lstm-cnn is proposed to predict the failure of reciprocating compressor.In the simulation experiment,the state data of reciprocating compressor is input into the state recognition model based on CNN for training,and then the prediction output of LSTM model is input into the state recognition model,and finally the output state type is used to predict the fault.Through the simulation experiment,the difference between the actual fault and the predicted fault is compared,and the feasibility of the method is verified.