American
International
University-
Bangladesh

Students & Supervisors

Abstract

Nowadays, understanding the demand for electricity is an very important component of contemporary energy systems, mostly as nations transition to smart grids and incorporate renewable energy sources. Constantly growing urbanization, industrialization, and climate change are all putting a lot of strain on Bangladesh's national grid. As a result, supply and demand can’t match up with each other. Popular statistical models have mostly been used but they are ineffective due to the complexity, nonlinearity, and weather-dependent nature of electricity use. In this recent situation, artificial intelligence (AI) offers new opportunities for creating forecasting systems that are more accurate and flexible. To solve this problem, we have used open-access datasets relevant to Bangladesh that includes historical energy consumption data (1975–2021), regional demand-weather records of the Sylhet division, and national daily power demand and generation reports (2019–2024). Preprocessing steps added filling missing values, normalizing the data, and generating new features from weather variables. We developed a comparative modeling framework that applied AI techniques such as Random Forest, XGBoost, and Long Short-Term Memory (LSTM) networks alongside baseline statistical models, including ARIMA and linear regression. Model performance was evaluated using standard error metrics such as RMSE, MAE, and MAPE. Our study reveals that AI-based models always do better than old-fashioned ones. LSTM models, in instance, cut predicting errors by up to 30% when it came to capturing daily and seasonal changes compared to ARIMA. The temperature turned out to be the best predictor of peak demand, and adding weather data made the model even more accurate. Regional case studies from Sylhet showed how AI models can change based on changes in demand in specific areas. Long-term estimates at the national level predicted a constant increase in electricity demand, which was closely linked to forecasted population and industrial expansion. The study represents that using AI for forecasting greatly enhances Bangladesh's ability to predict energy demand over the long and short terms. These findings imply that forecasting systems must incorporate meteorological and regional data because standard models aren't always reliable which is a matter to look over. More precise forecasts help Bangladesh's future smart grid use more renewable energy while reducing the risk of load shedding and grid instability that improves the efficiency.

Keywords

Publication Details

• Type of Publication:
• Conference Name: 1st International Conference on Science and Humanities for Sustainable Development
• Date of Conference: 23/10/2025 - 23/10/2025
• Venue: Dhaka University of Engineering & Technology (DUET), Gazipur
• Organizer: Dhaka University of Engineering & Technology (DUET)