Using Engineering Enhanced AI to Forecast Combined Cycle Power Plant Performance in the Presence of Uncertain Weather Conditions

Abstract

Abstract Characterizing the performance of gas fired combined cycle power plants is critical to economic dispatch decisions. Dispatch models rely on an accurate prediction of Gas Turbine and combined cycle power output and heat rate. Approaches for generating performance characteristics range from correction curves to detailed thermodynamic performance models. Unfortunately, most techniques are either too simplified, require significant expertise, or are manually labor intensive. Furthermore, these performance estimation techniques do not intrinsically capture uncertainty due to the inherent variability of the weather and state of the asset. This paper proposes a physics-enhanced Artificial Intelligence technique for automatically characterizing power plant performance including uncertainty due to weather effects. The model uses a layered sub-model approach to rapidly learn power plant performance without the need for extensive data preparation. The proposed technique is used to evaluate for accuracy, ease of use, and level of automation. The new technique is accurate to within 1% and provides a power and efficiency forecast one week out. The technique is also applicable to other power generation assets and scaling techniques, and challenges will be discussed. An overview of the automation framework is provided including discussion on modeling approaches, AI approaches used, modeling techniques, and use cases.