Study On The Similarity Of Wind Speed And Its Application In Short Term Wind Speed Prediction

The volatility of wind energy affects the security and stability of power system. To the reliable operation of power system, curate wind speed forecasting is an important guarantee. This paper start with the diurnal variation of wind speed and the surface temperature of the underlying surface which is closely related to the characteristics of the 24 hour weak periodic. By finding the similarity sequence of wind speed data, this paper combines forecasting with Dynamic Time Warping (DTW) algorithm, Spatial Nearest Neighbor (SNN) Method and Pearson Correlation Coefficients (PCC) to predict short-term wind speed.Because the defects of original wind speed data, it cannot be directly used for wind speed prediction, first we need to make quality control.So the research work mainly includes the quality control of the original wind speed data and the forecasting of the short-term wind speed. And the most important innovations are: This paper proposed an integrated fill model of the defect wind velocity based on entropy, and a kind of improved generalized regression neural network is used to forecast the short term wind speed. In the quality control of the original wind speed data, the most important problem is the filling of the defect data. An ensemble interpolation method to study on the similarity of wind speed is proposed Firstly, in the wind farm, on the basis of SNN, depend on the equal-likelihood of the sub-models, DTW and PCC are introduced to measure the similarity of wind speed series between the missing wind speed turbine and other turbines, and then 3 alternative sub-models are constructed. Secondly, the Particle Swarm Optimization (PSO) is used to train the composition of the Generalized Regression Neural Network (GRNN). Lastly, an ensemble interpolation model based on entropy weight is constructed using the superior sub-models. In the wind speed forecasting, dynamic time warping method are used respectively. Extract and evolve most similar to the wind Speed date of several sub sequences. Based on the dynamic time warping method and correlation coefficients Method generalized regression neural network sub Model Prediction Unit are established. The specific parameters of each sub model use PSO in order to Global optimization. The mean values of the two sub models forecasting results are used as the final results of the combined forecasting method. Finally, the forecast results were compared with Back Propagation (BP) Neural Network、GRNN and Autoregressive Integrated Moving Average Model (ARIMA). And based on similarity principle, it is feasible to carry out wind speed forecasting. The effect is better than that based on extrapolation method.