A Comparative Study of Data and Physically Based Gas Turbine Modeling for Long-Term Monitoring Scenarios: Part I — Thermodynamic Performance Prediction Without Design Information

Abstract

Within the ongoing global transition process towards renewable energies, gas turbines can play a significant role due to their ability to provide flexible and dispatchable power that compensates for the inherent volatility of renewable power generation. While being important for the stability of the electricity grid, this flexible mode of operation may result in a significant increase in thermo-mechanical stress for the gas turbine components. The demand for a constantly high level of performance and availability despite these challenges requires the employment of comprehensive monitoring tools. While different monitoring tools may vary in detail, a common core element is the inherent model capability to predict the ideal operational characteristics of the gas turbine for varying boundary conditions. In literature, various approaches are suggested to generate these gas turbine models. Within the present study, the authors apply a data-driven and a physically based modeling approach to two real long-term monitoring scenarios and compare different evaluation metrics. The overall goal of the study is the identification of advantages and disadvantages of the investigated modeling approaches depending on the monitoring scenario. The first part of the study takes the perspective of a gas turbine operator mainly focusing on the monitoring of the thermodynamic performance parameters. This perspective is characterized by the availability of a comprehensive set of long-term operational data on the one hand but the lack of detailed design information regarding the component characteristics of the operated gas turbine on the other hand. In the present study, the operational data set is provided by an E-class gas turbine that is operated in a Chinese combined cycle power plant. The physically based modelling approach used for this scenario is mainly based on a combination of heat- and mass balances representing a simplified thermodynamic gas turbine process. In addition, publicly available component maps are modified and subsequently integrated into the model. The corresponding data based modelling approach utilizes the set of long-term operational data as input parameters for the development of an artificial multi-layer perceptron neural network model with one hidden layer. The development steps conducted within the present study include the selection of adequate input and output parameters, the pre-processing of the data set for training and a sensitivity analysis regarding the number of neurons in the hidden layer. In summary, the results show that the data based model approach outperforms the physically based model approach based on an evaluation of the RMSE and the nRMSE. However, both the data based model approach and the physically based model approach are able to capture the main operational characteristics of the investigated gas turbine within the complete load range making both approaches suitable approach for long-term monitoring scenarios.