Electricity Demand Prediction: Case of Turkey

It is critical to generate accurate power demand forecasts for balancing supply and demand, as well as for the efficient operation and management of electricity generation. Accurate forecasts can also help to prevent energy wastage, reduce operating costs, and minimize environmental impacts. Several approaches for this task have been proposed in the related literature, including machine learning, statistical modeling, stochastic/fuzzy/gray modeling, and metaheuristics. The present study initially discusses the importance of accurately predicting electricity demand, providing an extensive overview of linear functions, neural network architectures, and regression models concerning their relevant concepts and applications, and then furnishes a case study of monthly electricity demand prediction in Turkey using data from 1998 to 2017. Considering several independent variables (IVs), namely average temperature, average precipitation, consumer price index, total industrial production index, total import, and total export, the performance of an Artificial Neural Network (ANN) model was compared to that of a Multiple Linear Regression (MLR) model and an Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) model. The former has been shown to outperform the other two in terms of MAPE and R2, indicating its capability to better explain the nonlinear relationship between the IVs and energy consumption. As a potential avenue for future research, to develop more comprehensive models that can generate reliable forecasts, the impacts of weather and climate change on energy demand should be evaluated in different countries and regions adopting hybrid models that combine different methodologies such as metaheuristics and machine learning.