Probabilistic load forecasting in Europe: Capturing meteorological, socio-economic and political risks
提出一个跨24个欧洲国家的概率性中期电力需求预测模型，分解需求为日历、温度、社会经济和政治等成分，并联合建模不确定性，在超过9年的每小时数据上优于标准基准。
Electricity in Europe is delivered through an interconnected grid that requires continuous balancing of supply and demand while large-scale storage remains limited. Trading, contracting and generation decisions therefore rely on high-resolution mid-term load forecasts capturing cross-country dependencies and uncertainty in meteorological, socio-economic and political conditions. Yet fine-resolution models at this horizon remain scarce—and probabilistic multivariate frameworks across countries rarer still. We propose a novel probabilistic mid-term forecasting model for hourly electricity demand that is multivariate across 24 European countries. Demand is decomposed within an interpretable Generalized Additive Model (GAM) into calendar and temperature effects, including a climate trend, an endogenously retrieved unit-root socio-economic and political component, and short-term autoregressive deviations. Uncertainty in these components is modeled jointly across countries and propagated through forecasted trajectories. In a forecasting study based on more than nine years of hourly data (2015–2024), the model outperforms standard benchmarks in terms of Continuous Ranked Probability Scores. The latent socio-economic component is shown to align with external macroeconomic, energy-market and uncertainty indicators. Beyond probabilistic forecasting, the trajectory-based design enables gigawatt-level attribution of individual drivers under risk scenarios. We demonstrate this by showing how extreme weather events translate into country-specific demand deviations, revealing elevated cold-weather vulnerability in countries with high shares of electric heating.