Research On Reservoir Prediction Method Based On Improved Long And Short-term Memory Neural Network

With the gradual transformation of oil and gas exploration targets from structural oil and gas reservoirs to lithological oil and gas reservoirs,the petroleum industry has put forward higher and higher requirements for the accuracy of reservoir prediction.Machine learning algorithms have the advantages of high efficiency and accuracy,so various machine learning algorithms have been introduced into the research of reservoir prediction.Representative algorithms used in reservoir prediction include neural networks,support vector machines,random forests,etc.These algorithms have achieved good results.However,these algorithms generally have a flaw: they cannot effectively use the time sequence information in the data.The seismic data contains information about the upper and lower strata of the interface,which has typical time sequence characteristics.The long and short-term memory neural network can use time sequence information in the input data to improve the prediction effect of the neural network by introducing the gate control loop structure.Through the improvement of long and short-term memory neural network reservoir prediction technology,time sequence information can be used more effectively,thereby improving the accuracy of reservoir prediction.The paper systematically analyzed the principles of seismic reservoir prediction.Research on methods of selecting seismic attributes based on seismic attribute extraction;Based on the analysis of the long and short-term memory neural network algorithm,the loop structure of the long and short-term memory neural network is improved,thereby improving the utilization effect of the long and short-term memory neural network on time sequence information.Use long and short-term memory neural algorithms to build the mapping relationship model between seismic attributes and reservoir characteristics;The method in this paper is applied to the North Sea Industrial Area in the Netherlands to predict the natural gamma and porosity,and compare it with the prediction effect of the conventional deep fully connected neural network algorithm.The test results show that the long and short-term memory neural network can better describe the reservoir characteristics and has high accuracy.The research results showed that the long and short-term memory neural network can construct a nonlinear mapping model between seismic attributes and reservoir parameters,and realize the conversion from seismic attributes to reservoir parameters.