An improved spatial upscaling method for producing day‐ahead power forecasts for wind farm clusters

Abstract

Large‐scale day‐ahead wind power forecasting (WPF) for wind farm clusters (WFCs) can enable dispatching agencies to formulate scientifically sound power generation plans and enhance the robustness of power grids. Most available WPF methods for WFCs only involve mathematical models and rarely consider spatial correlation factors. This necessitates further improvements to forecasting systems. In this study, to increase the day‐ahead WPF accuracy for WFCs, fractal transform theory is introduced to optimize the process of WPF for WFCs through spatial upscaling and establish a day‐ahead WPF model for WFCs based on an improved spatial upscaling method. First, a WFC is partitioned into subclusters. Then, using fractal transform theory, an affine relation is established between the local output of each subcluster and the overall output of the WFC. Finally, a day‐ahead power forecast is produced for the WFC through spatial upscaling of the forecast for the subcluster with the highest grey relational grade with the output of the WFC. The applicability of the proposed forecasting model is examined using historical measured data for a large WFC in north‐eastern China. The case study results show that the proposed forecasting model outperforms the summation method and the statistical upscaling method in terms of forecasting accuracy.

Steady-state deduction methods of a power system based on the prediction of large-scale wind power clusters

“…The statistical upscaling method only needs to linearly upscale the predicted output of the reference wind power farm to obtain the cluster prediction result. This method can offset potential correlation factors between different wind power farms' data and has good dynamic adaptability, but the selection criteria for the reference wind power farms are difficult to determine (Yang et al, 2022). The cluster division method divides the wind power farms in the region into several wind sub-clusters according to the fluctuation patterns of power and meteorological data and establishes a separate prediction model for each sub-cluster.…”