Session: 08-01: Gas Turbine Research and Developments

Paper Number: 154036

Taking the Guesswork Out of Power Plant Dispatch 

Power plant dispatch requires accurate forecasts of plant capabilities, especially for gas turbines operating in variable weather conditions. Traditional methods—such as correction curves based on historical performance or complex thermodynamic models—are often either too labor-intensive or overly simplified. This paper builds on previous work by using a refined digital twin modeling framework to enhance predictions for both simple and combined cycle performance under changing weather conditions. With a set of training data, the resulting model requires only three ambient conditions—pressure, temperature, and humidity—to predict plant heat rate and power output. Paired with a weather forecast model, the system can run turnkey. This approach is both accurate and user-friendly, addressing the challenges associated with traditional techniques.

 

To further improve the prediction capabilities of the digital twin model and ease dispatch decisions, this work incorporates extrapolation techniques, augmentation modeling, and dispatch optimization. Enhanced extrapolation expands predictions beyond the training data range, overcoming a limitation of neural networks, which are used in the digital twin framework. Augmentation models improve predictions by estimating the impact of gas turbine performance upgrades, such as evaporative coolers, inlet fogging, chillers, and steam injection, on inlet conditions. Dispatch optimization forecasts individual plant capabilities seven days ahead across a fleet, factoring in weather conditions, maintenance schedules, fuel costs, and startup/shutdown expenses. This integrated approach is designed to simplify dispatch decision-making, reduce fuel consumption, lower operational costs, and minimize excessive plant run-times.

 

The paper details the modeling methodologies and calibration techniques that ensure high fidelity between the models and the physical power plants. Case studies from gas turbine plants are presented to demonstrate the effectiveness of the technology in predicting performance and optimizing dispatch decisions, considering competing factors such as maintenance, performance, and operational costs. The result of this work is a method for developing data-driven digital twin models that simplify and improve power plant performance predictions and dispatch decisions. The benefits associated with these models include improved asset availability, increased efficiency, and reduced fleet-wide emissions.

Presenting Author: Christopher Perullo Turbine Logic

Presenting Author Biography: Chris Perullo is the Director of Engineering at Turbine Logic. Mr. Perullo has 15 years of experience in gas turbine and combined cycle design, modeling and simulation, and analysis. He leads digital twin development which encompasses health monitoring, instrumentation fault detection, anomaly detection, and performance prediction. Before joining Turbine Logic, he was a Senior Research Engineer at Georgia Tech where he focused on a wide variety of advanced gas turbine simulation and monitoring methods for entities including NASA, the FAA, and major large frame and aircraft engine OEMs.

Authors:

Corson Teasley Turbine Logic
Andrew Gerlings Chevron
Christopher Perullo Turbine Logic
Lea Boche EPRI
David Noble EPRI