| Popis: |
Intermittent generation, such as wind and solar, are popularizing stochastic simulation of power grids. Stochastic forecasts of these uncertainties are typically characterized by their marginal distributions. However, dependencies between forecasts also influence the grid significantly, but their specification is not straightforward. We propose Correlated Sampling (CS), a novel, purely data-driven technique that deduces spatial and temporal correlations from historical measurements, and samples stochastic forecasts from provided marginal distributions while imposing these correlations. By incorporating correlations between (potentially many) intermittent injections, CS greatly improves peak demand predictions, congestion analysis, spinning reserve optimization, power quality, etcetera, which will save utilities huge costs. In an experiment, CS computes confidence intervals (CIs) of demand aggregated over multiple substations. These CIs are either 3 times as narrow or 2 times as accurate as the CIs of two ‘naive’ techniques that omit estimating correlations. This shows that CS enables much more realistic power system analyses. |
