Wilfred M. Walsh

Spatial Load Forecasting With Communication Failure Using Time-Forward Kriging

Short-term and very short-term load forecasting are essential for power grid management operations such as automatic generation control, unit commitment scheduling, and transmission loss estimation. Most existing forecasting techniques require that the load data be available up to the current time step. In recent years, cyber attacks increasingly threaten the secure operation of power grids. One potential cyber threat is communication failure which will affect load forecasting. When communication failure happens and actual load data are not available, forecasting accuracy suffers. To overcome this problem, we propose a time-forward kriging based approach to forecast load with and without communication failure. This technique has the ability to forecast the load by utilizing load information from neighboring regions. The proposed method has been tested using NYISO and PJM load data with 5-min and 1-h intervals, respectively. Our results show that the proposed method is capable of forecasting load under communication failure with acceptable accuracy and improved accuracy when compared with other forecasting techniques.

Probabilistic accumulated irradiance forecast for Singapore using ensemble techniques

The performances of Bayesian model averaging (BMA) and ensemble model output statistics (EMOS) in producing intra-day accumulated solar irradiance forecast in tropical Singapore by utilizing global model numerical weather prediction (NWP) outputs are compared. The effect of the predictive probability density function (PDF) choices for the BMA and EMOS methods is investigated as well. The BMA and EMOS methods are shown to be better than climatology and simple bias-corrected ensemble methods. There is, however, no significantly best methods among various variants of the BMA and EMOS, although employing skew-normal conditional predictive PDF for BMA seems to improve the probabilistic forecast calibration. The skew-normal PDF is chosen based on the PDF of the observation data.