Quantifying Inter-Well Connectivity and Sweet-Spot Identification through Wavelet Analysis and Machine Learning Techniques

Abstract This study leverages wavelet analysis and machine learning (ML) techniques, including a 1D Convolutional Neural Network (1D CNN), to analyze inter-well connectivity and pinpoint an optimal new drilling location (sweet spot) based on datasets from five wells. The dataset utilized in this work includes well logging data of porosity, permeability, and water saturation at different depths of the wells. A 1D CNN was used to extract important features from the dataset. Wavelet analysis and correlation techniques were applied to the feature space extracted by the 1D CNN, revealing inter-well connectivity. Well-pairs with the highest correlation scores indicated enhanced inter-well communication. For identifying a sweet spot, machine learning regression models, including Gaussian Process (GPR), K-Nearest Neighbours (KNN), Gradient Boosting (GB), and Extreme Randomized Trees (ERT), were trained and tested to predict properties across the field. Locations with high porosity, high permeability, and low water saturation were assessed to identify sweet spots. The Wavelet analysis was then used to detect and analyze inter-well communication between identified locations and existing wells, aiding in identifying a new optimal drilling location relative to the five wells.