The increasing integration of renewable energies into modern electricity grids creates new challenges for grid operators and energy traders. Wind and solar energy are highly dependent on the weather and are therefore particularly characterized by spatial and temporal fluctuations. Therefore, precise and reliable power forecasts for hours and days into the future are essential for grid stability, energy trading, and cost reduction. Moreover, the increasing availability and quality of power measurements enables a variety of methods to adapt weather and power forecasts to these measurements, with combining these diverse approaches often leading to improved accuracy. However, in an environment with changing conditions, such as additional construction of wind turbines or different regularization of the grid operators, past behavior can become outdated quickly. Adaptive combination offers the potential for further enhanced forecasting accuracy and reliability. This paper presents presents an adaptive combination method that produces a combined forecast based on power forecasts from different weather models. Next to the usual predictors, we use additionally spatio and temporal features. This approach is applied to different baseline and Machine Learning models. We compare online machine learning methods such as Online Sequential Extreme Learning Machines (OsELMs) that adaptively combine predictions with other benchmark approaches and individual forecasting models. One particular focus is on the systematic analysis of defined update intervals (daily, weekly, monthly). During each interval, the adaptive learning process accumulates new measurement and forecast data and adapts its weights automatically in order to optimize the prediction quality. Both the combination and the adaptive training of the ML models show that the updates lead to improved forecasting accuracy.
Lipskij et al. (Sun,) studied this question.