A Machine Learning Based Modeling and Analysis of Solar-Assisted Thermal Power Plant in Bangladesh

Abstract

Global climate change, driven by carbon emissions from fossil fuels, has accelerated the global transition toward renewable energy sources. A major challenge with renewable energy, particularly solar energy, is its intermittency, which makes accurate forecasting crucial for effective energy management. This thesis addresses two critical aspects of solar energy utilization forecasting Direct Normal Irradiance (DNI) to enhance the reliability of solar energy production and optimize the energy management of a solar-assisted regenerative Rankine cycle to maximize power generation using available solar resources. These two studies complement each other by focusing on both the prediction of solar energy availability and the efficient utilization of that energy in a thermal power plant. The first study explores advanced statistical, ensemble, and deep learning models for short-term DNI forecasting in Bangladesh. By analyzing geographical data and identifying optimal solar energy locations, the study applies models such as Facebook Prophet, SARIMAX, XGBoost, Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Artificial Neural Networks (ANN). The performance of each model was evaluated using error metrics like R^2, MAE, MSE, and RMSE. Among machine learning models, XGBoost performed the best (MAE: 2.70, R^2: 0.93), while CNN was the top-performing deep learning model (MAE: 2.33, R^2: 0.991), demonstrating the effectiveness of these approaches in forecasting solar irradiance. Building on these predictions, the second study focuses on optimizing power generation in a solar-assisted regenerative Rankine cycle. The study examines various repowering configurations by closing one or more of the six feedwater heater (FWH) extractions and integrating solar energy from a Concentrated Solar Power (CSP) plant. Depending on DNI availability, the heat from the CSP plant is either used directly or stored in a Thermal Energy Storage (TES) system to be utilized during peak electricity demand. By simulating different DNI conditions, the study found that repowering could enhance the original cycle's 200 MW output to a maximum of 241.3 MW, depending on the closed extractions and thermal input from the solar system. However, this increase in power output was accompanied by a decrease in thermal efficiency from 43.63% to 39.45%, which is justified as additional power input is provided by solar energy. The study simulated energy management during the operation of the power plant, exploring various repowered cycle configurations to ensure the efficient utilization of solar energy. This energy, whether received directly or from the Thermal Energy Storage (TES) system, was optimized to meet the varying electricity demands. Together, these studies form a comprehensive approach to addressing the challenges of intermittent solar energy. Accurate DNI forecasting ensures reliable energy availability, while efficient management of solar energy within the Rankine cycle ensures optimal power generation. This combined approach not only supports Bangladesh’s commitment to Sustainable Development Goal 7 (SDG 7) but also offers broader insights into the integration of renewable energy in thermal power plants globally.