Electricity consumption forecasting with deep learning models

Current trends in population growth, industrialization, and technological advancements are driving a significant increase in global electricity consumption. While renewable energy sources are making significant strides, fossil fuels still remain the primary source of electricity generation, posing challenges due to resource limitations and environmental concerns. To address these challenges and optimize energy use, accurate prediction of electricity demand is crucial. Therefore, in this thesis dissertation, deep learning models based on long short-term memory (LSTM) network, convolutional neural network (CNN), and ensemble learning combining both architectures are proposed for short-term (next 24 hours) electricity consumption forecasting. The models utilize not only electricity consumption data but also additional features including timestamps and relevant meteorological parameters such as temperature, relative humidity, and wind speed. Two geographically diverse datasets encompassing approximately 2.5 years of hourly electricity consumption data as well as meteorology and timestamp data were utilized for training and evaluating the models. Extensive experimental studies demonstrated that the proposed models utilizing appropriate feature sets can achieve normalized root mean square error (N-RMSE) values as low as "0.16", normalized mean absolute error (N-MAE) values as low as "0.13", and mean absolute percentage error (MAPE) values as low as "4.05%". In conclusion, this dissertation presents not only effective models for short-term electricity consumption forecasting but also valuable insights into the impact of meteorological features on forecasting performance. These contributions can guide future research efforts in developing even more accurate and robust forecasting methods.