A DEEP LEARNING-BASED DEMAND FORECASTING SYSTEM FOR PLANNING ELECTRICITY GENERATION

Abstract

In today's world, where economic and industrial development continues, the importance of electrical energy is constantly increasing. Energy demand should be forecast as precisely as possible to reduce lost energy costs in the system, to plan generation expenditures appropriately, to ensure that market players are not economically harmed, and to deliver quality and uninterrupted energy to system consumers. Balancing the electric energy supply and demand of the system is possible with a forecasting plan. Our research aims to generate hourly electricity consumption load forecasts for the period 2018-2021 using Turkish Electricity Consumption Data and meteorological data, with the addition of time and public holiday features. The forecasting performance of the models is evaluated by training multiple machine learning models and deep neural network-based time series models with the data. When the prediction results of our load demand forecasting problem were evaluated, it was seen that deep learning methods gave higher results in prediction success compared to machine learning models. It has been observed that the prediction success of the LSTM model, one of the deep learning methods we use, is higher than the RNN and GRU models. The analysis envisages the elimination of mismatches between energy supply and demand.