*With increasing regulation and the push for clean energy, power plants are becoming increasingly complex. This complexity, combined with the need to optimize performance, means that power plant modeling is more important than ever. However, traditional modeling approaches such as physics-based models do not capture true performance of a power plant component. The complexity of factors such as coupling and noise makes them difficult to model and computationally expensive, so they are generally not considered in real-time applications. In a complex system such as a gas turbine, this creates significant disparities between models and actual performance. This study explores the use of machine learning tools to predict gas turbine performance. A long short-term memory (LSTM) model, a form of recurrent neural network, was trained using operational datasets from a 100 kW recuperated gas turbine power system designed for hybrid configuration. The LSTM turbine model was trained to predict shaft speed, outlet pressure, and outlet temperature. First, the performance of both the machine learning model and physics-based model were compared against experimental data of the gas turbine system. Next, the machine learning turbine model was incorporated into a system model of the test facility which was comprised of physics-based component models such as a compressor, air plenum, combustor, and piping. The performance of the system model using the LSTM turbine was then compared with a system model using the physics-based turbine model. Results show that the machine learning model predicted turbine performance better than a traditional physics-based model when fed facility data as an input. Furthermore, the system model using the LSTM turbine performed closer to real plant operation than the system model using the physics-based turbine model.*


Using Machine Learning to Increase Model Performance for a Gas Turbine System

Division / Sponsors: 
Power Division & Nuclear Engineering Division (NED)

Power Conference
Paper Proceedings: https://asmedigitalcollection.asme.org/POWER/POWER2020/volume/83747

International Conference on Nuclear Engineering
Paper Proceedings: https://asmedigitalcollection.asme.org/ICONE/ICONE2020/volume/83761

POWER-ICONE - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.