Pressure Prediction of Oilfield Water Injection Pipeline Network Based on Bayesian-Optimized CNN-LSTM
Predicting node pressure in oilfield water injection systems is a critical method for energy conservation,consumption reduction,and optimized scheduling in actual oilfield production.Traditional pressure prediction methods typically use pipe network balance calculations.However,due to years of operation,issues such as scaling,perforation,and corrosion inside the pipeline network cause significant changes in the pipeline's internal di-ameter and friction coefficient.This results in substantial discrepancies between balance calculation results and measured values.Therefore,this paper proposes an oilfield water injection network pressure prediction model based on a combination of Bayesian-optimized Long Short-Term Memory(LSTM)networks and Convolutional Neural Networks(CNN).Firstly,the self-attention mechanism captures the correlations between nodes in the input sequence,and CNN is used to obtain spatial features between nodes.Then,LSTM performs sequence modeling,and the output features of CNN and LSTM are concatenated and fed into Deep Neural Networks(DNN)to predict the network node pressure.Finally,the Bayesian optimization method is used to optimize the hyperparameters of each sub-sequence predic-tion model.The proposed model is validated using production data from a domestic oilfield water injection network system and compared with six other models.Experimental results show that the Bayesian-optimized CNN-LSTM prediction model achieves RMSE,MAE,and R2 values of 0.045,0.035,and 0.967 at monitoring node FWA,respectively,significantly out-performing the comparison models.The model demonstrates good generalization ability and can effectively improve the prediction accuracy of node pressure in oilfield water injection networks.
oilfield water injectionpressure predictionlong short-term memory networksCNN-LSTMBayesian optimization
NETL
NSTL
万方数据
