Research On Evaluation Method Of Fracturing Effect In Tight Reservoirs Based On Deep Learning

Hydraulic fracturing technology is a necessary technique for developing tight oil and gas reservoirs,and identifying fracturing effects is crucial for subsequent oil and gas production and wellbore optimization.Currently,fracturing effect evaluation methods based on sonic logging are widely used,such as reflection wave imaging and orthorhombic dipole anisotropy,but these data processing methods are greatly influenced by the selection of processing parameters and are easily affected by human factors,requiring significant manpower and time costs to manually process data.To address this issue,this article proposes a deep learning-based evaluation method for fracturing effects in tight reservoirs.The input dimension of array acoustic logging signal is small,the shape is complex and contains noise.The conventional deep learning model has a shallow number of layers,and it is difficult to correctly learn such data features.A deeper network can learn more advanced features,and Res Net is a commonly used model in deep learning.,the network can be designed deeper in the way of residual connection;Dense Net makes the network have a strong feature reuse ability in the way of dense connection;the SE module can enhance the network’s attention to important features;Based on the above characteristics,this paper research on the evaluation of fracturing effect by SE＿Res Net,Res Net and Dense Net.At the same time,Using the improved WGAN-GP to generate adversarial samples to enhance the network’s generalization ability,based on the original dataset,the network’s anti-disturbance ability is trained to improve the robustness of the model to small disturbances;Determine the optimal structure of the model by increasing or decreasing the number of residual blocks and dense blocks in the model.The main work of the paper is as follows:(1)Due to the small amount of collected array acoustic logging data,it is difficult to meet the requirements for training deep learning models.Therefore,this article uses an unsupervised data enhancement method to expand the dataset,and demonstrates the effectiveness of dataset expansion through comparative experiments.In order to better learn the characteristics of the data,this article has improved the WGAN-GP generator and discriminator models.After dataset expansion,1000 generative adversarial samples were obtained,including 500 positive samples and 500 negative samples.The experimental results show that the WGAN-GP generator model can effectively learn the distribution of real signals,and dataset expansion can significantly improve the performance of the model.(2)Construct SE＿Res Net,Res Net,and Dense Net models suitable for array acoustic logging data,and use the method of controlling variables to fine-tune the model parameters,selecting parameters that simultaneously maximize network stability and generalization ability to improve model performance.On the basis of support from the original dataset,test the sensitivity of the models to different gradient descent algorithms,learning rates,and loss functions,analyze the sensitivity of model parameters,and provide references for model parameter selection and hyperparameter optimization.(3)Train the three deep learning models with the expanded dataset and analyze the performance of different deep learning models based on experimental results to determine the optimal model structure.The experimental results show that the deep learning-based evaluation method for tight reservoir fracturing is feasible,and the residual connection method of Res Net is more suitable for the dataset in this paper,achieving higher accuracy.The SE module further improves the accuracy of the model by adaptively adjusting the weights of feature map channels.(4)By establishing a hydraulic fracturing model to simulate dipole acoustic logging data with different radial extension lengths of the fracturing crack,we further studied the method of using deep learning models to evaluate the radial extension length of tight reservoir fracturing cracks.SE＿Res Net and Res Net models were constructed to evaluate the radial extension length of fracturing cracks.The experimental results show that on the simulated data,SE＿Res Net and Res Net can accurately evaluate the radial extension length of fracturing cracks,and the evaluation results of SE＿Res Net are more accurate.This paper studies the evaluation method of tight reservoir fracturing effect based on deep learning models.By collecting array acoustic logging data and constructing deep learning models,the research on the evaluation method of tight reservoir fracturing effect based on deep learning has been completed.