The Application Of Wind Speed Index In Field And Numerical Prediction Revision For Wind Power Forecasting

The wind power resource is renewable and clean energy. Because of the volatility and uncertainty of wind power output, it has constraints for combination wind power to the grid. The key means to solve this problem is to accurately predict wind power output of wind farm. However, wind power forecasting involves complex factors. It makes the formulation and forecasting of wind speed has been the key technology of wind power prediction. In this paper, according to the measured data of wind speed and wind power output in a large wind farm in the North, also in the light of the technology requirements of "Function Specification of Wind Power Forecasting System", the relevant technology and method of wind power prediction and statistical revision are discussed from different aspects by using numerical forecast products.1. According to the characteristics of the large wind farm, also based on analyzing the factors of influence on wind power, Three kinds of wind speed index are put forward. a. average wind velocity in field/Vb. total sum of wind speed cubed in wind fieldc. effective total sum of wind speed cubed in wind field among them, Vi is the wind speed of fan in wind farm and k is the number of fan that are actual open in wind farm.Using three kinds of index, the statistical relationship between index and wind power output in farm is analyzed respectively. Results show that the index χ which is combined with the features of fan has the best fitting effect. It has a significantly difference of the fitting precision among the three index by analyzing the error of fitting output power in wind farm which estimated by measured wind speed index in field. The index χ has the best fitting effect and the index V fit better. On this basis, putting the foresting wind speed from the BJ-RUC numerical product into the three kinds of index to get the numerical forecasting value, the prediction error is analyzed. The error shows that the numerical prediction and the true value have a large gap of these three indexes. Therefore, the index V and the index χwill be regard as the factor in wind power prediction by considering the simple and feasible in calculation and the feasibility of revision.2. Exploration of statistical revised method of wind speed index numerical prediction and the improvement of accuracy of wind power prediction.1) Ultra-short-term wind power forecasting (0-4h). Dynamic correction of delay errors from velocity numerical forecast was made by using neural network to revise wind speed. So the wind power was predicted by the revised wind speed. After revised, mean absolute error of wind speed decreased by 48.7% and mean absolute error of wind power down 58.2%.2) Short-term wind power forecasting (0-24h). Due to the extension of the forecast time, autocorrelation of the numerical prediction error weakened. The ultra-short-term (0-4h) wind power forecasting methods are no longer applicable, so for the wind power short-term forecast (0-24h) using other methods.a. Research show that effective total sum of wind speed cubed in wind field which defined according to power curve of wind turbine can fit the total output power of wind farm. The statistical revised method is explored for 0-24h predicted value of effective total sum of wind speed cubed in field based on BJ-RUC model output. Using regression analysis method to revised the error, the mean absolute error of wind speed decreased by approximately 33.9% and the root mean absolute error reduced about 25.5%. The correlation between and the revised wind speed and measured wind speed is improved, and the error index of total power is also improved after revised.b. Correction by using the diurnal variation of numerical prediction error. The error between the 0-24h numerical forecast wind speed and measured mean wind speed in field of BJ-RUC meteorological model is analyzed. Pointing at the characteristics of diurnal variation of error, daily error values are harmonic analyzed and harmonic parameters C0、C1、C2、φ1、φ2 are obtained. Neural network (BP) model is established to predict the harmonic parameters according to actual daily error values of 08-19h. Then 24 hours numerical forecasted wind speed is revised according to the estimated errors of 24 hours by BP model. Through independent sample test, absolute error of predicted average wind speed decreased from 2.77 m/s to 2.01 m/s and absolute error of wind power prediction was reduced from 0.134 to 0.092 after the revision.3. Direct Application of numerical forecasting products.1) The method is to establish the neural network model by using the relationship between numerical forecast wind speed and wind power output. The wind power prediction of the 0-4h test was carried out according to the measured wind power data and BJ-RUC model output data. Results show that the method make the error indicators of predicting wind power improve obviously. The mean absolute error of wind power reduced of 60.4%.2) The output data of wind and other meteorological elements by the BJ-RUC numerical forecast model is collected. And the corresponding relation is established using different measured wind speed and its numerical forecast value with the total output power respectively. Using numerical prediction values directly fitting the total wind power relations, errors of different wind speed index fitting formula do not vary much. Although the fitting effect is increased (the root mean square error and mean absolute error improved about 10%), the correlation coefficient had no significant difference. It is visible that the effect of this method is not obvious.