Study On Productivity Prediction Of Coal Seam Gas Wells Based On Depth Belief Network

As a new type of environmental energy,coalbed methane plays a very important role in the energy structure of our country.In the actual production process of coal seam gas well,in order to improve the production efficiency as much as possible,it is necessary to adjust the production system according to the change of productivity,and if we can accurately predict the short-term future of coal seam gas well,The change of medium-term or long-term production capacity is bound to provide a reliable scientific basis for the work of drainage,thus improving the efficiency of drainage to a certain extentThis paper firstly expounds the prediction method based on BP neural network,and uses BP neural network’s powerful nonlinear prediction ability to predict the productivity of CBM wells,but such prediction models are easy to fall into local minimum and increase the number of hidden layers.The problem of causing gradient dispersion.Then,the BP neural network is optimized by the search optimization ability of the genetic algorithm.Although the above defects are avoided to a certain extent,there is still a problem that the shallow model has low precision in the face of multidimensional data.In view of the above problems,this paper applies the model combining deep belief network and conjugate gradient method to the production capacity prediction of coalbed methane wells,fully considering the geological factors that have a great influence on the production capacity of coalbed methane wells in the actual production of coalbed methane,and using standardized methods to calculate the data concentration.The average of each attribute and the standard deviation of the data set complete the normalization of the data,and the experimental method is used to determine the optimal structure of the DBN model and the way to update the weightBased on the characteristics of coalbed methane drainage data,this paper constructs a prediction model based on deep belief network and applies it to the prediction of coalbed methane production capacity.Compared with the traditional BP neural network,this model avoids the traditional BP neural network model.The value and the problem of gradient dispersion with the number of hidden layers,and the prediction accuracy is improved based on the GA-optimized BP neural network.Compared with shallow models,it shows better predictive ability when faced with multi-dimensional geological factors.