Deep-Neural-Network-based Process Data Simulation Model for Production Well of a Geothermal Power Plant

Some production wells of geothermal power plants sometimes shut down unintentionally. Before the unintentional shutdown, unstable fluctuations in pressure and flow rate at the wellhead are observed. Because of the difficulties in observing the ground, it is constrained to predict the performance of the steam at the wellhead. If the abrupt shutdown symptoms can be detected and the shutdown prevented, then the power generation continuation can be achieved. The authors have proposed a method for detecting anomalies using 1-Dimensional convolutional neural networks (1D-CNN) to predict this phenomenon. There is an empirical rule that opening the bypass valve can prevent a sudden pressure drop. It is necessary to conduct a comparative experiment between cases relevant to valve operation. This comparative experiment is difficult to conduct because the condition of the well differs from case to case, and therefore the authors could not measure it. Thus, the authors proposed another 1D-CNN model to predict the pressure at the wellhead. It was verified that this model steadily predicts the pressure performance several hours in advance. Consequently, the authors can virtually compare the past measured data with the predicted data to investigate the validity of the valve operation.