DAS光纤应变和应变率演化分析及状态识别OA

To improve the hydraulic fracturing model and optimize completion design parameters,this study presents a relational model between fracture-induced fiber-optic strain and strain rate based on the displacement discontinuity algorithm,clarifies their evolution characteristics during fracturing and pump shut-in,and develops a fracture parameter interpretation method based on low-frequency distributed acoustic sensing(DAS)data.The study results show that during the propagation of a single injection fracture,the strain evolution of the fiber in the monitoring well presents four stages:weak stretching,shrinkage convergence,banded convergence,and compressive strain shrinkage after pump shut-in;correspondingly,the strain rate evolution presents four stages:strain rate enhancement,heart-shaped convergence,banded convergence,and strain rate reversal after pump shut-in.There are differences in the variation laws of strain and strain rate between fibers at the same position in monitoring wells at different distances.Before the fracture reaches the monitoring well,both the fracturing impact points and non-fracturing impact points generate tensile strain,and the strain rate of the former increases continuously.After the fracture reaches the monitoring well,the tensile strain of the fracturing impact points increases,while the tensile strain of the non-fracturing impact points decreases and turns into compressive strain.The strain rate of the former first increases sharply and then decreases sharply,whereas the strain rate of the latter drops sharply from positive to negative and then rises again.After pump shut-in,the strain of both types of impact points gradually decreases,and the strain rate undergoes a reversal characteristic opposite to that during fracture impact.To clarify the sensitivity of fiber-optic strain and strain rate to fracturing and fiber parameters,parametric sensitivity analysis is conducted,and a fracturing impact identification method based on convolutional neural network(CNN)is proposed.A total of 840 and 420 samples are used to train the classification recognition and time recognition models,respectively.The F1-score of the classification recognition model on the test set is 1,and the determination coefficient R2 of the time recognition model on the test set is 0.997,which indicates that the CNN model exhibits good performance in both classification recognition and time recognition,and verifies the feasibility of using CNN for real-time event monitoring with low-frequency DAS strain rate data.This proposed method provides an efficient and economical solution for adjacent-well fracturing monitoring,which can accurately reflect the fracturing state,provide early warning for fracturing impacts,and reduce monitoring costs and operational risks.