Load Forecasting for Trakya University

Abstract

Electricity demand prediction involves analyzing historical and current electricity usage, identifying factors influencing load estimation, and employing various methods and algorithms to project future demand. The primary objective of demand forecasting is to deliver cost-effective, reliable, and highquality energy to consumers. To achieve this, electricity demand forecasting is segmented into different time periods, encompassing production, transmission, and distribution levels. Time-based classifications categorize load forecasting into very short-term, short-term, mid-term, and long-term forecasts, each serving specific planning purposes. Forecasts extending beyond a year are considered long-term, those spanning one week to one year are categorized as mid-term, forecasts covering one hour to one week are short-term, and estimates less than an hour fall under very short-term forecasts. The factors utilized to estimate electrical load vary based on the forecasting duration. Long-term load forecasting primarily relies on parameters like gross national product and population, while short-term load forecasting predominantly utilizes meteorological data. Electricity consumption fluctuates on an hourly basis throughout the day. Analyzing these consumption variations and forecasting power plant activity initiates with short-term load forecasting. An additional role of short-term load forecasting is optimizing energy usage economically. Addressing escalating energy requirements and ensuring sustainable energy supply commences with accurate load forecasting. Consequently, load forecasting applications are progressively gaining significance to meet the evolving energy demands. In this study, a short-term load forecast for Trakya University have been made. During the estimation and analysis processes, error rates have been revealed by using decision trees (DT), artificial neural networks (ANN) and least squares method (LSM). It is seen that the ANN algorithm comes to the forefront in predicting behavioral characteristics. The performance of the DT algorithm in behavior analysis is not unsuccessful. However, there is a higher margin of error in the prediction process using the DT algorithm compared to the ANN application. It has been revealed that LSM cannot be used as a load forecasting algorithm.