Ex.3 Translate from English into Russian:

Data source

The data in this paper are obtained from four real wind farms located in the northern part of Hebei Province, China. The geographical positions of the wind farms (including wind farms A, B, C, and D) are shown in Fig. 1. In this paper, wind farm A is selected as energy storage sizing target, which named TWF. Other wind farms named RWFs, are used to analyse the effects of spatial-temporal correlation on energy storage sizing for TWF A. The data of each wind farm contain wind power data from 1 January 2014, to 31 December 2014. The wind power data have been normalised according to the installed capacities of each wind farm. The time interval between adjacent sampling points is 1 h.

In order to obtain the wind power forecast and error data, the forecast method in is used to predict data of TWF and RWFs. The combination forecast method proposed in is effective to correct the short-term wind power forecast by building an error prediction model using the regression learning algorithm (including Support Vector Machine and Extreme Learning Machine), to obtain better prediction results. Based on the method mentioned above, the wind power forecast data of 8760 sampling points, i.e. one year, are generated. The forecast is generated once per day by blocks of 24 h, and the forecasted series start at midnight (0 o'clock). Correspondingly, the forecast error can be obtained as the difference between the forecast value and actual value. In order to further illustrate the forecast error characteristics, the MAE and RMSE results of forecast error are shown in Fig. 2.

Fig. 2 shows the MAE and RMSE of wind power forecast errors of TWF A. As observed, with the prediction horizon increasing, MAE and RMSE values gradually increase, implying the prediction accuracy decreases with the increase of the prediction horizon. Generally, the RMSE and MAE are different for the different forecast methods. However, the tendency above is usually similar to other forecast methods.