Back to the programme printer.gif Print

Delegates are invited to meet and discuss with the poster presenters in this topic directly after the session 'Forecasting: Maximizing grid deliverability and leading your business processes to profitability' taking place on Thursday, 13 March 2014 at 11:15-12:45. The meet-the-authors will take place in the poster area.

Danka Todorovic Faculty of Electrical engineering, University of Belgrade, Serbia

Printer friendly version: printer.gif Print


Partially linear model for wind farm power forecasting: case study of Banat Region, Serbia


Previous research of wind energy’s potential in Serbia present Banat region as a promising regarding the use of this renewable energy source. Banat region is situated in the north-eastern part of Serbia. Topography of the region shows a very flat terrain. The basic characteristics that favour the region regarding the possibility of using wind energy are: good wind energy potential, accessible facilities and low cost of building wind farms, cheap wind turbine transportation costs and good climatic conditions. This region is currently developing several projects of wind farms with a total power of over 1000 MW.


Micro-locations of perspective wind farms are very similar in terms of wind climatology, which was confirmed by a special wind speed measurements. From the aspect of an electric power system, this region can be observed as a single aggregated wind turbine, which injects 0-1000MW in one node of a power system, whereby injections are directly related to the characteristics of the wind. The major issue of wind energy development in Serbia is the concentration of wind turbine projects in one region, which could greatly and negatively affect the error in the prediction of wind energy production at the system level.

Main body of abstract

Due to tight connection between forecasting system and integration in power system, the main goal of this paper is to present a model for perspective wind farm power forecasting in this region.
This paper analyzes errors in the prediction of wind speed and wind-farm production for several micro-locations in the target region. The analysis used data on hourly wind speed forecasts for 12 to 36 hours in advance for the various models of prediction (ETA, WRF and GFS). The measurements of wind speed were conducted at two locations in the target region and performed a linear correlation analysis between predicted and measured values. It was concluded that deviations from the linear model were not equal throughout the range of speeds, and that some nonlinear functions would fit better. Correlation analysis was performed for different segments of the speed and thus, a linear function depending on the segment was obtained. Using linear regression model for each segment the errors in the prediction of wind speed were determined for all three forecasting models. Errors that this model makes in the wind-farm forecasting have been determined and presented graphically.
Comparative analysis of the standard linear model, which is implemented in the whole range of wind speeds, and the proposed partially linear model shows that the error in the prediction of wind farm production can be significantly reduced if the linear model is applied to each segment in the range of wind speeds.


The idea of filtering input wind speed and directing it to the specific segment has contributed to the significant reduction of error in wind-farm production forecasting. Small deviations between forecasted and the real power make this model good basis for the development of professional software for wind-farm production forecasting in the Banat. Since the Banat region is characterized by very similar climatology of the wind, the conclusions arising from the analysis conducted in this paper practically can refer to the entire region.

Learning objectives
This model enables the predictability of production on an hourly level for the day ahead with accuracy of over 90%, which allows the planning of production and balancing capacity in the power system, and the optimum implementation of thermo-hydro-wind coordination.
The proposed model is tested on the example of the Banat region in Serbia, but it has general importance and can be a good approach for wind farm production forecasting in any region.