Application Of Improved TCN-Attention Network Based On Logging In Reservoir Porosity And Permeability Prediction

Reservoir porosity and permeability are key parameters for oil exploration and production,which determine oil and gas storage and oil and gas production capacity respectively.The success of many oil and gas exploration work mainly depends on finding a reservoir with sufficient porosity and penetration to support feasible commercial development.Considering how the porosity and permeability of the sand body change,determining the petrophysical properties of the reservoir,and what distribution model to use are all important for reservoir prediction.Now,many reservoir prediction models such as porosity-depth or porosity-temperature curves,and burial history reconstruction have been proposed.However,due to these models cannot take into account the impact of all control factors,most prediction models still have some limitations.At present,some conventional machine learning algorithms have been applied to the prediction of porosity and permeability.Among them,shallow machine learning methods often require experts with professional knowledge to select and extract feature parameters.Due to the limited learning ability of shallow machine learning in the case of fewer samples,its generalization ability for complex nonlinear problems will also be restricted.In order to solve the problem of feature extraction and insufficient learning,deep learning arises to solve the problem of feature extraction and insufficient learning,although deep learning can rely on deeper training to extract more useful information and enhance memory ability compared with shallow learning.However,for massive logging data,there are still problems such as insufficient information access ability,and excessive stacking of layers leading to gradient explosion and dispersion.In order to improve the above defects,a TCN-Attention network based on logging data is improved and applied to the prediction of reservoir porosity and permeability parameters.Firstly,in order to solve the problem that the neural network is difficult to train due to the uneven distribution of lithology in strata.The multi-layer GMM model based on EM algorithm is used to preprocess the input logging data.Even in the same block,although there is little difference in lithology distribution,the distribution of central and marginal wells in the same depth block is still quite different.This is because it is necessary to find the data points of the same formation depth without being disturbed by the data points of horizontal depth.The multi-layer GMM model filters the data points that are not in the same stratum depth through the continuous average operation of the initialization information,so as to achieve the effect of accurate prediction of stratum lithology distribution.Secondly,the clustered logging data are sent to the TCN-Attention network to prepare for the subsequent parallel training.The role of TCN is to compress information so that the network has a wider field of vision for logging data.Attention is to assign importance to the compressed information.Important information is given high weight,and less important information is given low weight.In addition,an adjustable ’length’ parameter is proposed in Attention,which can more flexibly adjust the length of information needed for importance assignment.It is conducive to exploring the relationship between part and whole and reducing the number of parameters required for training.Thirdly,the TCN-Attention network based on logging data is applied to public logging data set.In comparison with other traditional algorithms,LSTM and TCN network,the visual prediction effect is compared with the mean square error(MSE),root mean square error(RMSE)and mean absolute error(MAE)values,which verifies that the network has good prediction effect for reservoir porosity and permeability.In the prediction process of long-sequence logging data,TCN-Attention can not only retain information for a long time,but also extract the importance of all information,and achieve the purpose of accurate prediction.Finally,the TCN-Attention network verified by public data sets is applied to predict the porosity and permeability of low permeability oilfield.The stability and accuracy of the network are verified by the actual prediction performance.