Medium-range probabilistic forecast of wind power ramps based on the hybrid multi-model-analog ensemble using self-organizing maps

Masamichi Ohba, Daisuke Nohara, Shinji Kadokura
Central Research Institute of Electric Power Industry, Abiko, Chiba, Japan


In this study, self-organizing maps (SOM) is applied to analyze and establish the relationship between atmospheric synoptic patterns over Japan and wind power generation. SOM is employed on sea level pressure data derived from atmospheric reanalysis over the Tohoku region in Japan, whereby a two-dimensional lattice of classified weather patterns is obtained. Our analysis extracts seven typical patterns that are linked to frequent occurrences of ramp events. By using TIGGE medium-range grand ensemble forecast data, possibility of application to probabilistic weekly forecasts of power generation and ramps based on the obtained SOM is discussed.


We use operational medium-range ensemble forecasts from five of the leading global NWP centers: ECMWF, JMA, NCEP, CMC and UKMO. The used forecast length is 216-h, and the total ensemble size is 168. Only the ensemble forecasts initialized at 1200 UTC are used here to create products for the multi-center grand ensemble based on the data from the five leading centers.
Each SOM node defines wind power generation corresponding to each weather pattern. Based on this link between the SOM-derived weather patterns and related local wind power generation, we can obtain a forecast PDF from atmospheric variables of the GSM. This can be regarded as an alternative to analogue ensemble techniques presented in the previous studies while we use multiple (SOM-based) analogs under PP approach.


The wind power generation and ramp probability are derived from the multiple SOM lattices based on the matching of output from a global forecast model to the weather patterns on the lattices. Since this method effectively takes care of the empirical uncertainties from the historical data, wind power generation and ramp is probabilistically forecasted from the global model forecast. The predictability skill of the forecasts for the wind power generation and ramp show the relatively good skill score under the grand-ensemble compared with the single center approaches. Further analysis and improvement of the method will be done in the future study.


Understanding of the impact of WP evolution on integrated wind power production and variability is important for ensuring energy security as well as the operational management of the grid. In addition to evaluating the wind power potential, forecasting the variability of the wind power output is one of significant challenges of wind power management. The meteorological forcing of wind ramp events is complex and still remains largely unexplained. In this study, the impacts of synoptic-scale WP evolutions on wind ramp events are examined by classifying the SLP field using SOM and investigate its applicability of the probabilistic forecasts based on the multi-center grand-ensemble.


A better understanding of large-scale wind ramps and their links to WP variability might ultimately lead to improved forecasting of wind power production. In the future, we will further explore the weather pattern-wind power relation based on SOM to the development of a stochastic prediction of wind power and ramps.