Short-Term Electricity Demand Forecasting Using a Functional State Space Model

In the past several years, the liberalization of the electricity supply, the increase in variability of electric appliances and their use, and the need to respond to the electricity demand in real time has made electricity demand forecasting a challenge. To this challenge, many solutions are being proposed. The electricity demand involves many sources such as economic activities, household need and weather sources. All of these sources make electricity demand forecasting difficult. To forecast the electricity demand, some proposed parametric methods that integrate main variables that are sources of electricity demand. Others proposed a non parametric method such as pattern recognition methods. In this paper, we propose to take only the past electricity consumption information embedded in a functional vector autoregressive state space model to forecast the future electricity demand. The model we proposed aims to be applied at some aggregation level between regional and nation-wide grids. To estimate the parameters of this model, we use likelihood maximization, spline smoothing, functional principal components analysis and Kalman filtering. Through numerical experiments on real datasets, both from supplier Enercoop and from the Transmission System Operator of the French nation-wide grid, we show the appropriateness of the approach.