Improving the Next 24-Hour Wind Power Forecast using Daily Total Wind Energy Forecast
The rates of renewable power plants in newly added power plants are constantly increasing. Wind energy is one of the most preferred renewable energy sources due to its high profitability. Since wind power generation depends on nature, it is not clear how much electricity will be produced at which hour and this is not desirable situation for the management of the electricity grid. Therefore, wind power forecasts are made every day by wind power plants (WPPs). Wind power forecasts are also important for WPPs. The penalties arising from the estimation errors affect the profitability of WPPs. In this study, a wind power forecast model for the next 24-hours is introduced. Data of a WWP located in Turkey is utilized. The model consists of three stages named as clustering, forecasting and correction. In the first stage, 12-hours data frames are clustered according to the wind direction changes. Then in the second stage, wind power forecast for the next-24 hours is implemented using Artificial Neural Network (ANN). Forecast inputs are the wind speed forecast, wind direction forecast, air temperature forecast, air pressure forecast, wind speed forecast at T-24h and measured wind speed at T-24h. In the third stage, a correction process is constructed to improve the forecast values found at the end of Stage-2. A daily total energy forecast is made to improve the forecast values. Forecasted next 24-hours values,obtained at the end of Stage-2, are updated according to the daily energy forecast values. Min. wind speed, max. wind speed, mean wind speed, generated total energy of the previous three days are the inputs. The final forecasts are compared with the forecasts found by the WPP. While the forecast error made by WPP is 12.34% NMAE, the error of the proposed forecast is found as 11.65% NMAE. Updating the next 24-hours forecast values with daily energy forecasts has increased the accuracy of the forecast. It has been concluded that the proposed approach is successful and can be integrated into the currently used wind power forecasting systems.
Keywords - Artificial Neural Networks, Short-term Wind Power Forecast, Wind Power Forecast.