Very Short-Term Electricity Demand Forecasting using Adaptive Exponential Smoothing Methods

Forecasting of future electricity demand is crucial for the efficient management of power systems. In latest years, due to privatization and deregulation of the power industry, accurate electricity forecast for the next several minutes has become an important research area for the real-time scheduling of electricity generation. For this lead time, univariate methods such as exponential smoothing can be useful on account of its computational efficiency and its reasonable accuracy. This paper presents the development of three new electricity demand forecasting models, based on the use of the adaptive Holt's exponential smoothing technique in a parallel implementation. Holt's methods are adaptive in the way that their smoothing parameters are fitted every time when a new observation is recorded. Real-world case study data based on the French Half-hourly electricity demand are presented; in order to illustrate the proficiency of the proposed approaches. With an average MAPE of 0.491%, the effectiveness of the third proposed model is clearly revealed.