Predicting the production in shale gas wells based on CNN-LSTM-ATT model
As a kind of considerable energy resources,shale gas is one of principal forces in production growth of natural gas in China.To predict its production accurately is crucial to the rational planning of shale gas utilization.Thus,correla-tion analysis on production dimension was implemented in an effort to solve a difficulty in this prediction,such as complex influential factors with dynamic changes,and further to improve forecast accuracy.The tubing pressure,casing pressure and water production as independent variables together with the gas production as an dependent variable were put into one prediction model.Then,a combined CNN-LSTM-ATT(convolutional neural network-long short-term memory-attention mechanism)model was constructed for multi-variable prediction.In this model,CNN is used to extract characteristics from production data,ATT enhances the importance of characteristics to input effect,and LSTM is good at learning how to deal with time series data.Results show that(ⅰ)the correlation analysis can screen out the dimension affecting intensely the pro-duction prediction,which is also of great significance to subsequent prediction;(ⅱ)the multi-variable prediction through the combined neural network model can better forecast the next production trend for shale gas wells;and(ⅲ)better predic-tion effect has been obtained from this combined model than that from single neural network.In conclusion,with good ap-plicability,the constructed model can increase the prediction accuracy on shale gas production and offer more reasonable prediction,which is meaningful to guide shale gas development.
NETL
NSTL
万方数据
