[Zurück]

M. Spiegel, T. Strasser:
"Experiences with Meteorological Models for Asset Scheduling in Local Energy Communities and Microgrids";
Vortrag: 3rd CIGRE SEERC Conference Vienna 2021 (CIGRE SEERC Vienna 2021), Vienna, Austria (online); 30.11.2021; in: "Proceedings of the 3rd CIGRE SEERC Conference Vienna 2021 (CIGRE SEERC Vienna 2021)", CIGRE, (2021), S. 1 - 10.

Local energy community networks and microgrids are built to meet current challeng-es in the power system such as reducing CO2 emissions by increasing the share of renewables, increasing economic benefits, and enhancing power supply resiliency. One lever that can be pulled is to deploy controllers that presciently schedule the operation of all controllable assets in advance. Control algorithms may range from simple rules that include predicted load and infeed to complex optimization problems that minimize operation costs such that a resilient operation is guaranteed. Often several modelling assumptions are made to quantify volatile generation within the scheduling horizon and to optimally operate the assets. In case these assumptions are violated, the economic performance as well as the resiliency of the power sys-tems can be seriously affected.

This work addresses the impact of common assumptions regarding meteorological inputs. It uses an exemplary meteorological dataset to assess the impact the as-sumptions have on the forecasted renewable energy generation. These forecasts are particularly needed to optimally schedule local energy community/microgrid as-sets and are thus vital for the system performance. Additionally, the paper showcas-es the impact of plant models on the eligibility of studied input assumptions and as-sesses the value of scheduling-time information such as numerical weather fore-casts. Therefore, several models of wind speed and global horizontal irradiation are trained and evaluated on two independent sets of measurements. The accumulated daily wind speed, irradiation, as well as the daily capacity factor are calculated to estimate the quality of the input assumptions in predicting renewable infeed within a typical scheduling horizon.

One common modelling assumption is that a meteorological observable is inde-pendent from previous observations. It is demonstrated that the probability of days with high and low solar irradiation and wind speed is considerably underestimated on using the independence assumption. Consequently, probabilities of days with high and low renewable generation are systematically underestimated. It is shown that discrete Markov models that additionally consider one previous observation can ameliorate the goodness of fit and accurately estimate the distribution of infeed.

To quantify the effects of scheduling-time information on the accuracy, a re-forecasting dataset is used to estimate the distribution of meteorological observables given a forecasted value. Although it is demonstrated that scheduling-time forecasts can reduce underestimation of extreme values, their main benefit is found in reduc-ing the uncertainty when predicting the most likely observation. Such a reduction may directly result in an improved economic performance in case reserves can be safely reduced.
By applying two exemplary plant models that estimate the infeed, the effects of the meteorological assumptions on the infeed distribution, and the prediction errors are demonstrated. It is shown that the plant model can influence target metrics such as the prediction error significantly. Hence, the plant characteristics need to be consid-ered when assessing the eligibility of assumptions. By assessing common input models, their implications, and alternative formulations, it is believed that this work aids the design and validation of scheduling algorithms in the context of local ener-gy communities and can help to improve the overall system performance.

Local Energy Community, Microgrid, Asset Scheduling, Meteorological Input, Dis-crete Markov Model, Forecasting Deviation

https://publik.tuwien.ac.at/files/publik_298972.pdf