Well Control Optimization of Waterflooding Oilfield Based on Deep Neural Network

A well control optimization method is a key technology to adjust the flow direction of waterflooding and improve effect oilfield development. The existing mainly based on algorithms numerical simulators. In face larger models, longer periods, or reservoir models with large number optimized wells, there are many variables, which will cause algorithm convergence difficulties costs. application not good because problems time length, few comparison schemes, only fixed frequency. This paper proposes new multi-input deep neural network. takes production history data as main input saturation field auxiliary establishes network for learning, forming dynamic prediction model instead conventional Based model, series generation, prediction, comparison, carried out find best plan reservoir. calculation results examples show that (1) compared single-input multiple inputs has better accuracy, close simulator; (2) multiple-input fast speed, schemes effect.