Spatiotemporal sparse autoregressive distributed lag model with extended Regressors for regional wind power forecasting

Large-scale wind power forecasting is critical for the secure and economic dispatch of power systems, and the spatiotemporal correlation is suggested to improve forecasting accuracy. In this paper, a novel L1-regularized and Correlation-Constrained Autoregressive Distributed Lag Model with Extended Regressors (LRCC-ARDLX) is proposed for sparse regional wind power prediction, which sufficiently considers spatiotemporal correlations of regional wind farms and effectively coordinates the heterogeneities in different farms. This method first designs a spatiotemporal correlation quantification model, including dynamic time warping, spatially representative wind farms, and their derived information screening, to provide a unified framework for the spatial interdependence of regional wind farms evolving. An autoregressive distributed lag (ARDL) model applicable to multiple wind farm inputs is developed based on this. By introducing spatial-temporally correlated multi-cluster prior information, an extended regressor is proposed to enhance the representation capability, and a two-stage spatial-temporal sparsity criterion is proposed to narrow the optimization scope, thus achieving rapid prediction of large-scale regional wind power while ensuring accuracy. Experiments involving 80 operating wind farms from 4 regions distributed over a wide spatial extent demonstrate the generalization and interpretation of LRCC-ARDLX compared to other commonly considered benchmarks.