Real-time monitoring and optimization in geothermal operations requires a dependable predictive model that incorporates the thermodynamic and thermo-economic behavior that takes place in the power plant. The effect of ambient temperature on an air-cooled binary cycle powerplant is too complex and costly to model using a physics-based predictive model. Fit-for-purpose data-driven predictive models offer a cost-effective and efficient alternative to physics-based models for energy production prediction and online optimization. Dynamic neural networks offer powerful and flexible models that can be readily constructed, trained, and updated using feedback from the real process and relevant historical monitoring and performance data. We present a deep learning-based approach for online prediction and optimization of geothermal operations while accounting for disturbances of ambient temperature and brine supply. The model predicts power output and thermal efficiency by propagating the measurements variable and by including the influence of control and disturbance variables in the form of a feed-forward neural network. Once trained, the deep learning model is used to predict and maximize the predicted power production by automatically adjusting the pump speed. We present the workflow in detail and demonstrate its performance using quasi-dynamic model in the form of thermodynamic flowsheet simulation as well as the field data from a binary cycle geothermal power plant.

Press the Back button in your browser, or search again.

Copyright 2022, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.