Data and code supporting: Unsupervised Learning Workflow for Clustering Pressure Transient Responses in Fractured Reservoirs

This repository provides a complete and reproducible unsupervised learning workflow for clustering pressure transient responses in fractured reservoirs. The dataset and code support systematic analysis of pressure transient behaviour simulated for a large ensemble of geologically constrained discrete fracture networks under multiple matrix and fracture permeability configurations. The zip file includes:<br><strong>DFN dataset:</strong> 4,850 geologically consistent DFN geometries generated with GeoDFN from a literature-informed design of experiments and Latin Hypercube Sampling. The detailed DoE and the code used to generate the DFNs are provided.<strong>Pressure transient dataset:</strong> Synthetic well-test pressure responses for the same DFNs under three matrix–fracture permeability configurations (Datasets A–C), together with the MRST (EDFM) scripts used to generate the synthetic pressure transients from the DFN geometries.<strong>Machine-learning workflow:</strong> Code for preprocessing pressure and derivative signals, computing DTW distances, performing K-medoids clustering to group responses by transient shape, and training a Random Forest classifier to quantify which fracture network properties control flow-behaviour separation.<strong>Cross-dataset comparison:</strong> Scripts for cluster alignment between datasets, Sankey visualisation of transitions, and comparison of diagnostic signatures across permeability configurations.<br>This release is intended as a reusable research resource. The DFN dataset can serve as a benchmark for DFN modelling, upscaling, and flow simulation in fractured media. The pressure transient dataset supports development and evaluation of pressure-transient interpretation methods and machine-learning workflows. The analysis code provides an extensible template for clustering time series data, and can be adapted to other transient data types, such as thermal transients in geothermal systems.

搜集汇总

背景与挑战

背景概述

该数据集提供了用于裂缝性储层压力瞬态响应聚类的无监督学习工作流的完整可重复数据与代码，包含4850个地质一致的离散裂缝网络（DFN）几何及其在三种基质-裂缝渗透率配置下的合成试井压力响应，并附有预处理、DTW距离计算、K-medoids聚类和随机森林分类的代码，支持DFN建模、压力瞬态解释方法开发和机器学习工作流评估。