Optimizing Power Plant Operations Through Machine Learning Based Electricity Demand Forecasting

• Main Authors: De Silva, Chamoda, Jayawardana, Vihan, Gunasekara, Pasindu, Medawatta, Tharani, Jayakody, Anuradha, Lokuliyana, Shashika
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Electricity; Generation; Machine Learning; Machine learning algorithms; Predictive models; Production; Regression tree analysis; Sociology; Stacking; Tuning
• ISSN: 2837-5424
• DOI: 10.1109/ICAC60630.2023.10417372

Balancing the supply and demand of power has become one of the greatest challenges with the rapid growth of population and industrial and commercial development. This study delves into electricity generation forecasting, which plays a pivotal role in optimizing power systems' growth and operation. It not only aids energy suppliers in measuring electricity consumption and preparing for future demands but also empowers power distributors to align production with future electricity requirements. Focusing on Sri Lanka's power supply, primarily sourced from thermal and hydroelectric power plants, the past two decades have witnessed a steady annual increase of 3.4% in peak electricity demand. To tackle this surging demand, the study employs a machine learning approach. The parameters used for this prediction are number of consumer accounts, monthly selling price, monthly electricity consumption, population, gross domestic product, and total generation. The predictions are made using a few machine learning algorithms, and the stacking ensemble method is also applied to those algorithms to retrieve a higher accuracy level. The machine learning models used are Lasso Regression, Ridge Regression, Random Forest, Linear Regression, Decision Tree, and XGBoost. Moreover, to increase the accuracy of the models, methods like feature engineering and hyperparameter tuning using GridSearchCV and Randomized-SearchCV were used, and as mentioned earlier, model stacking was also used.