Hourly electricity demand forecasting using Fourier analysis with feedback

Abstract

Whether it be long-term, like year-ahead, or short-term, such as hour-ahead or day-ahead, forecasting of electricity demand is crucial for the success of deregulated electricity markets. The stochastic nature of the demand for electricity, along with parameters such as temperature, humidity, and work habits, eventually causes deviations from expected demand. In this paper, we propose a feedback-based forecasting methodology in which the hourly prediction by a Fourier series expansion is updated by using the error at the current hour for the forecast at the next hour. The proposed methodology is applied to the Turkish power market for the period 2012–2017 and provides a powerful tool to forecasts the demand in hourly, daily and yearly horizons using only the past demand data. The hourly forecasting errors in the demand, in the Mean Absolute Percentage Error (MAPE) norm, are 0.87% in hour-ahead, 2.90% in day-ahead, and 3.54% in year-ahead horizons, respectively. An autoregressive (AR) model is also applied to the predictions by the Fourier series expansion to obtain slightly better results. As predictions are updated on an hourly basis using the already realized data for the current hour, the model can be considered as reliable and practical in circumstances needed to make bidding and dispatching decisions.

Keywords

• Time series analysis
• Prediction
• Forecast
• Fourier series
• Modulation
• Feedback