The Application Of Seismic Attributes In The Reservoir Prediction

Predicting spatial changes of reservoir parameters accurately is of realistic and far-reaching significance in oil-gas exploration and development.Seismic attributes,obtained from seismic data using mathematical transformation,can find out useful information hidden underground effectively and convert it into parameters closely related to reservoir lithology,physical properties as well as characteristic parameters so as to serve for geological interpretation and reservoir description directly.The cores of seismic attribute technology are attribute optimization and reservoir prediction.There are more than 100 kinds of seismic attributes.Although high-dimensional data give extremely abundant and detailed information about reservoir,they increase the workload,consume limited resources and even cause dimension disaster.In terms of seismic optimization,this paper mainly adopts principal component analysis(PCA)and locally linear embedding(LLE)to transform seismic attribute data from high-dimensional space to low-dimensional space in order to remove the redundant information from the original data and select a few attributes that are more effective and representative.In practical application,LLE acts better in dimension reduction due to the non-linearity of attribute data.Attribute optimization not only simplifies relevant calculation process,but also improves reservoir prediction accuracy.The error back propagation neural network(BPNN)is widely used in reservoir prediction.However,it still has some disadvantages,such as converging slowly and falling into local extremum easily.Therefore,this paper first introduces wavelet analysis tool,which uses its zooming feature and time-frequency local property to speed up the convergence rate.Secondly,genetic algorithm(GA),a kind of global algorithm,is adopted to achieve the optimal network configuration,which not only avoids the local extremum but also improves the operation efficiency and stability of the network.Prediction results of three methods show that genetic algorithm optimized wavelet neural network(GA-WNN)has the fastest convergence speed and the highest prediction accuracy.