Hourly Forecasting of Long Term Electric Energy Demand Using Novel Mathematical Models and Neural Networks

Abstract

In this work, hourly forecasting of long term electric energy demand is achieved using mathematical models and Artificial Neural Network (ANN) approaches. Previous works regarding energy demand forecasting either treated the problem of long term prediction over yearly averages, or considered hourly prediction using a very short term time lag, such as a few hours. The methods proposed in this work produce predictions with hourly accuracy despite the time lag of "years", making the model suitable for long term prediction. Several functions for mathematical modeling and different ANN structures are applied and tested for achieving small forecasting errors. The proposed mathematical models of the load are compared with different ANN model outputs in the sense of Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE). The mathematical models are observed to provide a simple, intuitive and more generalized form, whereas the ANN models provided specified models that are better fine-tuned for the available data. The suitability of these methods is illustrated and verified using 4-year-long real-life hourly load data taken from the Turkish Electric Power Company.