Smart Operation of Gas Turbine Combined Cycle Plants: Prediction and Improvement of Thermal Efficiency at Part Load

This thesis investigates various operational aspects of Gas Turbine Combined Cycle Power Plants (GTCC). GTCC power plants are expected to play an increasingly important role in the balancing of supply and demand in the electricity grid. Although originally meant for predominantly base load operation with high efficiencies, market circumstances, namely the increasing supply of unpredictable wind and solar power, force these units to be operated frequently across a wide range of load settings. The required flexibility opens a need for models, that can predict the plant performance accurately at design point as well as off-design conditions. The models and performance data, made available by equipment manufacturers, are usually too general to be applied for accurate prediction and optimization purposes. Adding to this, the electricity producing companies usually do not possess detailed design information, so that creating accurate process models presents an extra challenge. The chapters of this thesis are dedicated to the proposal of several methods for overcoming challenges related to the operation of existing gas turbine combined cycle plants in current and future energy markets. All models and procedures developed in the framework of this thesis, are applied to the Alstom KA-26-1 GTCC as a case study. Two of these units were installed at the Maxima Power Station in Lelystad for GDF SUEZ Energy The Netherlands. The current study is first placed in the broader context of the developments in the electricity market, and of research fields related to this subject. The importance of accurate simulation tools is motivated, and the potential role of uncertainty management and optimization is put forward. The implications of the market developments are synthesized into a concise problem statement. The productive core of GTCC plants is the gas turbine, especially when there is no external firing in the steam cycle. A step wise method for accurately modeling the design and off-design steady state performance of gas turbines is presented. Tuning performance models to measured data typically available to an engine user is an important task. Therefore, a method for achieving this is proposed and applied to a case study: the GT26, an industrial gas turbine (part of the KA-26 plant) with two sequential combustor components. The results of this modeling effort indicate that the accuracy decreases towards part load. Thermodynamic modeling of the steam cycle, although a widely practiced discipline, still presents some challenges in case of industrial-size units. Second-law analysis is often added to thermodynamic flow sheet calculations; this can be enhanced by analyzing the interaction between plant components with the help of a novel procedure presented in the thesis. For this purpose, the plant model is calculated over a randomly and uniformly distributed set of input conditions, calculating the (second law) thermodynamic losses of major components for every case. (The term numerical experiment is used for this procedure.) After this, the resulting data is processed and visualized to reveal expected as well as unexpected mutual relations between the losses of individual plant components. When gas turbine and steam cycle models, and computer models in general, are applied to make predictions, and economical decisions are based on these models, there is always an amount of uncertainty present with respect to the validity of the predictions. Quantification and reduction of this uncertainty can be of significant value for stakeholders. In this context, an existing method for statistical analysis and calibration of computer models, the Kennedy & O'Hagan framework, is applied to the the previously presented gas turbine and steam cycle models. The purpose is to enhance the accuracy of (especially) part load efficiency prediction by calibrating the models with the available (industrial) measurements. The mathematical tools applied in this framework are explained, along with the manner in which it is applied to the gas turbine and steam cycle models respectively. For plant performance prediction, it is necessary to integrate the models, so that uncertainties in one model are propagated through the next. Two methods are described for achieving this: integration of the models can be done either before or after calibration. The two stochastic integration methods are applied to predict the efficiency of the case study plant. While both methods produce accurate results, there is an indication that integration after calibration is slightly more accurate. The most important objective for the current study, besides accurate performance prediction, is the proposal of efficiency optimization methods. The final part of the thesis illustrates methods for analyzing efficiency improvement possibilities of existing (gas turbine combined cycle) power plants, and optimizing part load efficiency with steady state plant models. Firstly, the data from the numerical experiment mentioned earlier are processed. By comparing how strong the exergy losses in major components are correlated to overall thermal efficiency of the plant, the low pressure steam turbine is shown to be the component whose thermodynamic losses have the largest effect on the variations in overall plant efficiency. However, it is also known that gas turbine losses represent the largest exergy loss. This seeming contradiction is thoroughly explained in the thesis. By using a clustering algorithm, operational regimes are revealed with respect to the losses in the low pressure steam turbine and gas turbine. Efficiency optimization is performed at ambient conditions corresponding to these distinct operational regimes. The results of optimization indicate that the optimum set of operational settings is different for each of the identified regimes, thereby confirming that they are distinct regimes. After using deterministic models for the efficiency maximization, model uncertainty is incorporated in the calculations, and the stochastic models presented earlier are applied. The difference with the previous optimizations is that in this case, the applied model has been proven to give more accurate results, and it provides the statistical distribution and expected value of the plant efficiency, not just a deterministic value. The results of optimization under uncertainty are compared to results of deterministic optimization under equal ambient conditions: the resulting optimal operational settings for both cases are shown to be similar in many aspects; differences are analyzed and put into perspective. The final part of the thesis synthesizes the main conclusions and recommendations from the previous sub-studies and places them in the general context of the research field. Suggestions are proposed for possible applications of the proposed methodologies to problems which are outside the scope of the thesis.Process & EnergyMechanical, Maritime and Materials Engineerin