Operation of Power Systems Against High Dimensional Weather Forecasts

Accommodating high-dimensional weather uncertainty is an unavoidable challenge in renewable-heavy power systems. We describe a means of incorporating raw forecast data into a multistage optimization framework able to exploit spatial and temporal correlation. We derive an optimal low-order representation of the uncertainty that maximizes information content while retaining the ability to balance the grid. The result is a so-called reserve policy, a correlated generalization of existing grid mechanisms, described by a relatively small number of parameters. We demonstrate this with a model of the GB grid in 2030, driven by raw weather data from the UK Met Office.