Abstract: The great success of the North American shale gas revolution has posed a profound impact on the global energy market, attracting great interest from both industry and academia. Accurate prediction of sweet spots is essential to determine the well location and enable a high after-stimulation productivity for the fractured wells. However, there exist in the traditional approach to the shale gas sweet spot prediction. In this study, Fox Creek, a commercially developed shale gas producing area in Canada, is taken as an example to investigate the geological and engineering factors which can be utilized to identify the sweet spot area of shale gas reservoirs. A modified residual neural network approach is proposed to determine the main controlling factors and establish a prediction model for the sweet spot area. Results show that the main controlling factors affecting the sweet spot area of shale gas formations are porosity, permeability, shale content, burial depth, formation pore pressure, shale brittleness index and fracturing parameters (horizontal length, number of fracturing stages, total placed proppant and total fluid injection). The gas production reached 0.94 and 0.85 in the modified residual neural network algorithm in the test and training respectively. The distribution of shale sweet spots has been contoured based on the prediction model of the modified residual neural network. It is shown that the shale sweet spots locate along the Duvernay boundary in the west and south and deteriorated toward the northeast. This prediction model of shale gas sweet spot area provides a reliable foundation for the subsequent efficient development of shale gas.

Key words: shale gas, sweet spot, controlling factors, improved residual neural networks, Western Canada Basin