Research On Efficiency Prediction Method Of Pumping Well System Based On Deep Learning

The system efficiency of pumping wells is a key evaluation criterion for the oil extraction capacity and technical application effects of the pumping equipment,and it is also an important working condition indicator of the entire pumping system.Therefore,it is of great significance to study the system efficiency of the pumping wells.There are many factors affecting the system efficiency of the pumping wells,including more than 30 kinds of factors,such as oil well natural factors,man-made adjustable factors and natural factors,which maintain a complex relationship with each other.Relying only on expert experience for manual selection,the workload is large and it is easy to miss the hidden influencing factors.The system efficiency prediction of pumping wells is a time series forecasting problem that uses historical pumping well operation data to predict the future system efficiency,which takes the influencing factor series as the input and the target system efficiency as the output.Traditional methods for predicting the system efficiency of pumping wells adopt mechanism models and fail to fully consider the complexity,uncertainty and timeliness of the problem,and the prediction effect is not ideal.In view of the poor performance of traditional pumping well system efficiency prediction methods,this paper first improves the long-term memory(LSTM)network,which is a common algorithm for time series prediction,and proposes a self-attention-mechanism-based DSALSTM time series model.The proposed model improves the accuracy of model prediction and training speed by introducing a dual self-attention mechanism.Secondly,on the optimization of influencing factors of pumping well system efficiency,a single evaluation criterion is difficult to comprehensively evaluate the quality of feature subsets,so a multi criteria integration based optimization method of pumping well system efficiency attribute is proposed.Different evaluation criteria are fused by multi criteria integration strategy to optimize feature subsets more comprehensively and reasonably.Finally,on the basis of the previous two steps of work,this paper constructs a DGR-DSA-LSTM-based system efficiency prediction algorithm framework for pumping wells.The first step is to use the system efficiency attribute optimization method of pumping wells based on multi-criteria integration to select a subset of the system efficiency characteristics of the pumped well.The second step is to construct a DSALSTM-based system efficiency prediction model for pumping wells,and conduct model training and prediction.Experiments prove the effectiveness of the proposed method,and it provides a new idea for the research of the system efficiency prediction of pumping well.