考虑人工裂缝展布及井间干扰的多井图卷积与门控循环单元耦合页岩气井产量预测方法

打开文本图片集

中图分类号：TE312 文献标志码：A

Abstract：Inordertopredict shalegaswellproductionmore accuratelyand effciently，considering 3Dwellnetworksand multi-stage hydraulicfracturinghorizontalwelsinadvancedshalegasreservoirdevelopment，anovelIMWs-GCN-GRUcoupledmachine learning method was proposedconsidering hydraulic fracturedistributionandinter-wellinterference.Firstly，a method forconstructing theadjacency matrix was proposed based on a comprehensive consideration of various factors such as thespatial positionsofwells，fracture distributio，ndreservoir permeability，whichcanaddressthe limitationofcurrntapproaches thatconsidersinter-wellinterferenceonlywithinverticalwellinjectionand productionnetworks.Agraphconvolutional neural network（GCN）technique was employed toexplore spatial featuresamong shale gas wels.Secondly，agaterecurrent unit（GRU）method was utilized toextracttemporal featuresover diferenttime periods，thusa novel shalegas wel machine learningapproach to predictproductionwas formulated thatconsidersboth spatialand temporal characteristics.The accuracy ofthenew modelwas validated incomparisons withconventional numericalsimulationresults.Finally，basedonthe actual productionwelleadofshalegaswels，differentadjacencymatrixconstructionmethods werecomparedandanalyzed. Theresults show that incorporating fracturedistribution along with reservoir permeabilityonthe basis of adjacency matrixconstructioncan further enhancetheprediction accuracyof thenew model.Acomparison with traditional machinelearning methods such as LSTM，GRU，and RNN demonstrates that this model can achieve higher accuracy，surpasing 90 % ，for shale gas wells with inter-well interference.

Keywords：production prediction；shale gas well;inter-well interference；graph convolution network；gate recurrentunit

页岩气作为重要的清洁能源，对保障国家能源安全和“碳达峰、碳中和”目标实现具有重要意义[]。（剩余14857字）