Posters

Abstract

The ramp event is characterized as a rapid increase or decrease in wind speed within a short time interval, which consequently induces substantial fluctuations in the power output of wind farm. Such abrupt variations may compromise the secure and stable operation of the power grid and, under severe circumstances, may even lead to local frequency deviations or voltage collapse. Nevertheless, the inherently turbulent nature of atmospheric wind fields results in wind speed variations that are highly nonlinear and strongly stochastic, posing significant challenges to the detection and characterization of ramp events. Therefore, the accurate identification of wind-speed and power-output ramp events, as well as a comprehensive understanding of their underlying physical mechanisms and statistical properties, remains an urgent research topic with considerable engineering relevance. The Swinging Door Algorithm (SDA) is a trend‑oriented data compression method that offers advantages such as the simple algorithm structure and high computational efficiency, making it widely used in industrial and engineering applications. In this study, the SDA is implemented in three steps: (1) a fixed‑width time window is selected to segment the time series; (2) data within each segment that satisfy the algorithm’s threshold criteria are retained; (3) the window is continuously advanced to compress the entire sequence. By suppressing minor fluctuations and retaining only trend‑representative points, the algorithm reliably extracts the key temporal patterns embedded in all data. To enhance the accuracy of ramp event identification, this study systematically evaluates the influence of different threshold parameters on recognition performance. The results show that a time threshold of 4 hours, a wind‑speed threshold of 6 m/s, and a power threshold of 1000 kW comparatively optimal detection accuracy. Considering the operational characteristics of different wind turbine types, this study recommends the following parameters to enhance the algorithm’s adaptability: (1) setting the wind‑speed threshold to two‑thirds of the difference between the cut‑in wind speed and the rated wind speed; (2) setting the power threshold to two‑thirds of the rated power. This study further conducts the analysis of duration, magnitude, and occurrence probability of ramp events, and compares the distribution characteristics across complex terrain conditions. The main findings are as follows. (1) Ramping characteristics of an individual turbine: Statistical results for a turbine located near the Meteorological Mast show 177 upward and 188 downward ramp events over one year. The durations predominantly concentrate around 4 hours, with hourly wind‑speed variations averaging approximately 2 m/s. (2) Wind‑farm‑level characteristics: Annual statistics for all turbines indicate that the wind farm experiences approximately 150~250 ramp events per year, corresponding to a cumulative duration of about 24~35 days. (3) Terrain influence: Complex mountainous terrain exerts a notable influence on the occurrence of ramp events. The frequency is closely correlated with the elevation and averaging wind speed at turbine position. The proportion of time in one year of the ramp events ranges from 6.5% to 9.8% for each wind turbine, with the overall average of 7.8%.