Due to the uncertainty and fluctuation of wind power generation, probabilistic prediction for regional wind power generation is critical to accurately quantify the uncertainty of meaningful information to the dispatching departments of power grid. This paper proposes an approach of very short-term probabilistic prediction for regional wind power generation based on optimal performance-based nonparametric prediction intervals (OPNPIs). First, the deterministic prediction for regional wind power generation considering the division of wind farms based on the detrending-based partial cross-correlation analysis (DPCCA) is studied. Based on the deterministic prediction and its prediction errors, the OPNPIs are proposed considering the reliability and overall performance for the uncertainty analysis. Furthermore, a regulating coefficient is studied to further enhance the performance of PIs. Effectiveness of the proposed method is verified through multistep PIs of 15-minute based on the real wind power generation data.

With the development of the smart grid, the distribution system operation conditions become more complex and changeable. Furthermore, due to the influence of observation outliers and uncertain noise statistics, it is more difficult to grasp the dynamic operation characteristics of distribution system. In order to address these problems, by using projection statistics and the noise covariance updating technology based on the SageHusa noise estimator, for distribution power system with outliers and uncertain noise statistics, a robust adaptive cubature Kalman filter forecasting-aided state estimation method is proposed based on generalized-maximum likelihood type estimator. Furthermore, an adaptive strategy, which can enhance the filtering accuracy under normal conditions, is presented. In the simulation part, the branch parameters and node load parameters of the test system are appropriately modified to simulate the asymmetry of the three-phase branch parameters and the asymmetry of the three-phase loads. Finally, through simulation experiments on the improved test system, it is verified that the robust forecastingaided state estimation method, presented in this paper, can effectively perceive the actual operating state of the distribution network in different simulation scenarios.

To enhance the performance of the prediction intervals (PIs), a novel very short-term probabilistic prediction method for wind speed via nonlinear quantile regression (NQR) based on adaptive least absolute shrinkage and selection operator (ALASSO) and integrated criterion (IC) is proposed. The ALASSO method is studied for shrinkage of output weights and selection of variables. Furthermore, for the better performance of PIs, composite weighted linear programming (CWLP) is proposed to modify the conventional linear programming cost function of quantile regression (QR), by combining it with Bayesian information criterion (BIC) as an IC to optimize the coefficients of PIs. Then, the multiple fold cross model (MFCM) is utilized to improve the PIs performance. Multistep probabilistic prediction of 15-minute wind speed is performed based on the real wind farm data from the northeast of China. The effectiveness of the proposed approach is validated through the performances’ comparisons with conventional methods.