Forecast of Renewable Curtailment in Distribution Grids Considering Uncertainties

Abstract

Renewable energies curtailment induced by grid congestions increase due to grown renewable energies integration and the resulting mismatch of grid expansion. Short-term predictions for curtailment can help to increase the efficiency of its management. This paper proposes a novel, holistic approach of a short-term curtailment prediction for distribution grids. The load flow calculations for congestion detection are realized by taking different operational security criteria into account, whereas the models for the node-injections are adjusted to the characteristic of each grid node specifically. The determination of required curtailment based on the resulting congestions considers uncertainties of component loading and its corresponding probability. The forecast model is validated using an actual 110 kV distribution grid located in Germany. In order to meet the requirements of a forecast model designed for operational business, prediction accuracy, and its greatest source of error are analyzed. Furthermore, a suitable length of training data is investigated. Results indicate that a six month time period for maintenance gains the highest accuracy. Curtailment prediction accuracy is better for transmission system operator components than for distribution system operator components, but the Sørensen Dice factor for the aggregated grid shows a high match of historic and predicted curtailment with a value of 0.84 and a low error for curtailed energy, which makes 2.23% of the historic curtailed energy. The model is a promising approach, which can contribute to improvement of curtailment strategies and enable valuable insight into distribution grids.

Predicting Renewable Curtailment in Distribution Grids Using Neural Networks

“…Fig. 1 shows the interrelation between the models and the components (each grid component type is assigned a number) of the 110 kV distribution grid (the grid topology was first presented in [14]). To identify congestions, injections into the grid must be known.…”

“…As for these wind farms the installed capacity is the only information available, an adjusted wind turbine power curve is used to calculate generated wind power from wind speed model data (for further information regarding the data see Section III-A). A more detailed description of the calculation of WP can be found in Section II-B.1 of our previous paper [14]. The resulting power is used as target feature for the following ANN.…”

“…A description of the used distribution grid can be found in Fig. 7 of our previous publication [14]. In the contingency analysis the load flow for every outage is calculated and in the end the maximal loading for each line is selected.…”

“…Since grid operation require a high level of security, the presented assessment of the model's accuracy is a first step towards evaluating the integration of ANNs into operative grid management. As parametric models already exist that provide the required predictions [13], [14], a comparison of the ANNs with the parametric approaches is also conducted. Because measured transformer power are recorded in SCADA (supervisory control and data acquisition) systems, past grid injections are available and former grid congestions can be reconstructed (as shown in [14]).…”

“…As parametric models already exist that provide the required predictions [13], [14], a comparison of the ANNs with the parametric approaches is also conducted. Because measured transformer power are recorded in SCADA (supervisory control and data acquisition) systems, past grid injections are available and former grid congestions can be reconstructed (as shown in [14]). This data can then be utilized to predict upcoming curtailments in distribution grids using ANN.…”