Day Ahead Load Forecasting of Educational Institutions Using Multiple Linear Regression and Artificial Neural Network

Load forecasting is pivotal for efficient energy management and planning within institutional power supply systems. Especially in case of the institutions having their own renewable or alternative energy sources, the generation and load scheduling plays a vital role. This study utilizes daily energy consumption data from on-site energy meters and corresponding weather data from forecasting websites such as AccuWeather, The Weather Channel and Weather Underground to predict the energy demand for the institution one day ahead. After selecting the weather data source based on accuracy and data availability, five weather variables with the highest correlation to daily energy consumption were chosen as input features. Datasets were created for autumn, winter, and spring, with season-specific variables with respect to the place of study in the state of Assam, India. Multiple Linear Regression (MLR) and Artificial Neural Network (ANN) models were developed for each dataset, and their accuracies were tested using forecast weather data. The results illustrate the effectiveness of both approaches, providing insights into their performance, accuracy, and applicability for short-term load forecasting in educational institutions, having a predictable pattern of use of loads, but with complex dependency on multiple weather variables.