Patch-Transformer\nModeling for Production Forecasting\nin Tight Gas Reservoirs: A Case Study of the Yan’an Field

Accurate prediction of production rates in fractured\nhorizontal\ngas wells is crucial for optimizing development strategies in tight\ngas reservoirs. Traditional prediction methods, such as decline curve\nanalysis, often rely on simplifying assumptions, making it challenging\nto capture the complex nonlinear dynamics of tight gas production.\nConsequently, prediction accuracy remains limited even with extensive\nhistorical data. This study introduces a novel production forecasting\nmodel based on the Patch-Transformer architecture, leveraging five\nreadily available production parameters: the daily production time,\nwellhead pressure, wellhead temperature, daily water production, and\ndaily gas production. The model captures both local and global temporal\ndependencies by segmenting the data into patches, thereby enhancing\nthe predictive accuracy. The model was trained and tested using production\ndata from 37 fractured horizontal wells in the Yan’an gas field.\nThe model achieved an average <i>R</i><sup>2</sup> of 0.9984\non the training set and 0.9932 on the test set, demonstrating exceptional\npredictive accuracy. The Patch-Transformer model effectively captures\nthe complex temporal dependencies and nonlinear relationships in production\ndata, making it a powerful tool for forecasting gas well performance.\nBy providing accurate predictions, this model has the potential to\nsignificantly enhance production optimization, decision-making, and\nresource management in tight gas reservoir development, ultimately\nleading to improved productivity.