Eventos Anais de eventos
ENCIT 2022
19th Brazilian Congress of Thermal Sciences and Engineering
Artificial Neural Networks to predict a coal-fired power plant efficiency: comparative study
Submission Author:
Renata Rech de Souza , RS
Co-Authors:
Renata Rech de Souza, Lara Werncke Vieira, Paulo Smith Schneider, Paola Mendes Albino
Presenter: Renata Rech de Souza
doi://10.26678/ABCM.ENCIT2022.CIT22-0548
Abstract
The energy sector is responsible for almost three-quarters of the world’s CO2 emissions and consequently for a great part of climate change, which makes it necessary for the sector to be at the center of climate change solutions. In this respect, coal-fired power plants have an important role since they provide about 40% of electricity worldwide and coal is the largest source of carbon emissions into the atmosphere. That makes the efficiency of these plants a critical parameter when it comes to environmental control and CO2 emissions mitigation. In this regard, machine learning models can capture unique characteristics of the system without any previous knowledge of the process, which makes them an important tool to represent systems that are too complex or contain many unmeasurable disturbances. Artificial Neural Networks (ANN) are a powerful machine learning method and there are many different structures of neural networks that have been developed to address a wide range of prediction problems. The present work proposes the evaluation of three different structures to predict a coal-fired power plant’s efficiency in order to enhance its performance. The neural networks considered are a traditional Feedforward Neural Network (FNN), a simple Recurrent Neural Network (RNN), and a Long Short-Term Memory network (LSTM). The objective is to identify the differences between the structures since the RNN and the LSTM are models specially adapted to learning timeseries data and FNN models problems where input data has a timeless impact on the output data. Therefore, it is possible to compare the structures and identify the variation of the parameters in time and determine how important this is for the improvement of the plant's performance. For this purpose, the system under analysis is the steam generator of the PECEM power plant, a coal-fired power plant located in Ceará, Brazil, with data from 2018 to 2020. Thus, as the results of the application of the neural networks, there are efficiency predictions and variations over time, positive and negative aspects of each approach used as well as the network that best represents the system.
Keywords
Machine learning for thermal systems, Recurrent Neural Networks, coal-fired power plant, Efficiency prediction
