PREDICTING WELL-STIMULATION RESULTS IN A GAS-STORAGE FIELD IN THE ABSENCE OF RESERVOIR DATA WITH NEURAL NETWORKS

Selection or candidate wells for stimulation treatment to increase the ir productivity is a challenging task, A systematic approach that uses a three-layer backpropagation neural network, introduced in this pape r, assists engineers in predicting post-stimulation well performance t o select candidate wells for stimulation treatment. This approach can also be used to optimize the stimulation design parameters. Unlike con ventional simulators that are based on mathematical modeling of the fr acturing process, the process introduced in this paper uses no specifi c mathematical model, As a result, access to explicit reservoir data, such as porosity, permeability-thickness, and stress profile, is not e ssential. This is a major advantage over conventional hydraulic fractu ring simulators, which can translate to considerable savings because i t eliminates the need for expensive data collection. The application o f this methodology to a gas-storage field is presented in this paper. The developed neural network can predict the postfracture well deliver ability with approximately 95% accuracy. These results were achieved i n the absence of reservoir data (permeability, porosity, thickness, an d stress profiles) that makes conventional fracture simulation impossi ble. This process is currently being used to select candidate wells fo r future stimulation treatment in the aforementioned field.