Research On Fracturing Productivity Potential Evaluation And Parameter Optimization Of Shale Gas Based On Data Driven Methods

The fracturing mechanism of shale gas is complex,and the influencing factors of productivity are characterized by multiple parameters,mutual influence and complex relationship,which are difficult to be quantitatively evaluated.The conventional single-factor analysis method has been unable to meet the needs of productivity evaluation.Shale gas development process,because of the lack of reasonable evaluation and understanding of shale gas well productivity potential,fracturing operation parameter design is not according to difference of shale gas well production capacity potential and manifest the differences,so there will be some geological condition is good,but due to the fracturing operation parameter design is not reasonable and low production wells,production capacity of this kind of wells has not been fully digging,because of the high nonlinear and complexity of underground reservoir dynamics,it is difficult to apply traditional optimization methods to this problem.To solve these problems,a data-driven shale gas classification method is proposed in this paper.Through XGBoost prediction,feature construction and K-mean++ clustering,sample Wells are successfully divided into four categories according to geological,drilling and fracturing parameters of shale gas wells: category 1 is the wells with good geological conditions,high productivity potential and highest actual productivity;category 2: wells with medium to upper geological conditions,high productivity potential and low actual productivity;category 3: wells with moderate geological conditions and high actual productivity;category 4 refers to wells with poor geological conditions,low productivity potential,and low actual productivity.In order to improve the fracturing productivity of the wells with medium geological condition,high production potential and low actual fracturing productivity,a parameter optimization model of fracturing scheme for shale gas wells based on deep Q network was established in this paper.Experimental results show that the surrogate model constructed through data driven methods has the advantages of strong adaptability and high flexibility for complex engineering scene.The fracturing productivity of the two experimental wells increased by 15.8% and 10.9% respectively after optimization,which proves that DQN can realize the self-optimization of fracturing operation parameters of shale gas wells in the way of variable step size search and synchronous optimization of multiple variables,and verifies the advantages of deep reinforcement learning algorithm in fracturing scheme parameter optimization of shale gas wells.In this paper,a proppant screenout early warning model based on LSTM is established to predict the change trend of construction pressure of shale gas wells in time,so as to avoid sand screenout accidents in the fracturing process and realize the early warning of fracturing sand plugging accidents.At the same time,an intelligent fracturing model based on Actor-Critic is established.Real-time optimization of the flow rate and sand ratio during the operation cycle enables the shale gas fracturing operation to proceed normally.This paper,by using data driven methods to fully tap the implied information between shale gas well productivity factors,and for the first application of depth of reinforcement learning algorithm on shale gas fracturing parameters optimization problem,changed the traditional optimization algorithm search step length decision way,can in a relatively short time can get the optimal fracturing operation parameters.The optimization framework based on DQN and ActorCritic can also be applied to many other similar scenarios involving computationally expensive simulation models,and has broad application prospects in the oil and gas industry.