Presentations

Abstract

Wind energy trading markets are moving to 15-minute time scale. Wind traders are exposed to higher financial risk as imbalance price volatility fluctuates with more frequent extreme values. New Numerical Weather Prediction (NWP) models have improved wind energy forecasting in recent years. However, wind ramps, characterized as sudden changes in wind speed causing large energy production swings in a matter of minutes, are difficult to predict in time and space. These minute-scale wind ramps increase energy trading imbalance penalties for wind farms, and sometimes even disrupt TSO reserve energy scheduling and planning. At large wind farm clusters such as in the North Sea, a single, large wind ramp can cause a local energy production swing from 500 MW up to a few GWs in a matter of minutes. This can lead to expensive mitigation measures or direct damage to the TSOs infrastructure and operations. Ramps can also cause severe wind turbine damage, and lead to increased insurance premiums.  Scanning wind lidars can continuously measure a 2D map of the wind field over a radius of more than 10 km around the instrument. This enables us to capture the space–time coherence of the wind field. In particular, the scanning lidar can resolve spatial structures such as wind ramps and track their evolution in time. This presentation demonstrates a new, real-time algorithm that automatically detects the boundary of a wind ramp event at the edge of the lidar scan and tracks its evolution in space and time. This allows for early alerts to wind farms when wind ramps of interest are approaching. By following ramp evolution across successive scans, the algorithm can predict its trajectory and calculate the expected time of impact on the wind farm. Multiple studies focus on general wind energy production forecast KPIs such as RMSE improvement compared to persistence. This work focuses specifically on wind ramps. A quantitative validation specific to wind ramp forecasting is presented. The lidar-based ramp forecasts are compared to actual wind ramps detected by reference SCADA and onsite anemometer data. Wind ramps are split into different categories associated with shifts in wind regimes, transient wind ramps or turbulent wind ramps. The corresponding wind ramp forecast performance KPI’s are computed and presented as a confusion matrix. This is performed for multiple forecast horizons. The results are derived from a two-year campaign conducted at a coastal, onshore wind farm. In the demonstration campaigns, the algorithm successfully forecasts more than 90% of important wind ramp events. The ability to forecast wind ramps minutes ahead and to quantify forecast performance allows wind energy owners, operators, traders and TSOs to evaluate the financial benefit for their wind asset operations and safety.