Application of Machine Learning Method of Data-Driven Deep Learning Model to Predict Well Production Rate in the Shale Gas Reservoirs

Reservoir modeling to predict shale reservoir productivity is considerably uncertain and time consuming. Since we need simulate the physical phenomenon of multi-stage hydraulic fracturing. To overcome these limitations, this paper presents an alternative proxy model based on data-driven deep learning model. Furthermore, study not only proposes development process a model, but also verifies using field data for 1239 horizontal wells from Montney formation in Alberta, Canada. A neural network (DNN) multi-layer perceptron was applied cumulative gas production as dependent variable. The independent variable largely divided into four types: well information, completion fracturing data. It found that prediction performance better when principal component with contribution 85% analysis extracts important information multivariate data, predicting DNN 6 variables calculated through importance analysis. Hence, develop reliable sensitivity hyperparameters performed determine one-hot encoding, dropout, activation function, rate, hidden layer number neuron number. As result, best mean absolute percentage error improved at least 0.2% up 9.1%. novel approach can be other formations. useful guide economic future plans nearby reservoirs.