Recently Artificial Neural Networks (ANNs) have been gaining an important role in the analysis of complex power cycles, since they have the potential to reduce the computational effort in designing and control of power plants operating conditions compared to rigorous thermodynamic models.

This paper presents a novel methodology for the prediction and optimization of the performance of thermoelectric power plants at design conditions using ANNs. The methodology involves a preliminary study to randomly generate the dataset of input variables (i.e., power plant operating conditions) and evaluate the dataset of output variables (i.e., energy and economic performance indicators) via thermodynamic simulation. Using these datasets, ANNs are trained and validated. Finally, the ability of ANN algorithms to replicate thermodynamic models is assessed in terms of absolute relative errors and coefficient of determination. The proposed methodology is flexible with regard to the type of power plants to be replicated and the extent of the investigation, that can be easily adapted by properly selecting the set of input and output variables.

To prove its feasibility, the methodology is applied to a coal-fired power plant and a triple-pressure reheat combined cycle. In both case studies, the methodology provided a very good accuracy in predicting the power plant behavior and optimizing their energy or economic performance.

This content is only available via PDF.
You do not currently have access to this content.