A Two-Stage Ultra-Short-Term Wind Power Forecasting Method Based on Transitional Weather Identification and Meteorological Prediction Error Propagation

Accurate and reliable wind power forecasting is crucial for the safe and economical operation of power systems. However, under transitional weather conditions, the forecasting errors of meteorological variables such as wind speed tend to increase, introducing significant input noise into wind power prediction models and reducing their reliability. To address this, this paper focuses on analyzing the error propagation mechanism of meteorological input variables and proposes a strategy to improve short-term wind power forecasting accuracy under transitional weather conditions. First, transitional weather periods are identified by analyzing the fluctuation characteristics of multi-dimensional meteorological variables. Then, a two-stage model combining Sparse Variational Gaussian Process (SVGP) and Noisy Input Gaussian Process (NIGP) is proposed. In this model, noisy input data (wind speed predictions) are decomposed into true data (actual wind speed) and noise data (forecast errors), which are modeled independently. By considering the propagation process of input noise in wind power forecasting and making necessary corrections, the SVGP-NIGP model provides more accurate deterministic forecasts and higher-quality interval predictions. Experimental results show that the proposed method significantly enhances wind power forecasting accuracy under transitional weather conditions, offering an effective solution to address the uncertainties arising from meteorological forecasting.