An Ultra-Short-Term Wind Power Forecasting Approach Based On Component Divide And Conquer

As a new type of energy,wind energy has received extensive attention.However,the wind energy is greatly affected by external factors,so that wind energy has large volatility and uncertainty,which causes the accuracy of ultra-short-term wind power prediction is not high.The existing ultra-short-term wind power prediction methods do not fully consider the influence of wind speed components of different frequency and wind direction on the prediction accuracy.In this paper,considering the role of wind direction,a hybrid forecasting method of ultra-short-term wind power based on component divide and conquer is proposed.The main work of this paper is as follows:Firstly,an ultra-short-term wind power forecasting framework based on component divide and conquer is proposed.Through mechanism analysis,it is found that the following two factors comprehensively affect the power of the unit: on the one hand,due to the influence of the inertia and dynamic performance of the unit,the wind speed with different frequencies has different influence on the power of the unit,and the uncertainty degree of the wind speed component with different frequencies is different;On the other hand,the wind turbine set is placed in a wind farm.Under the influence of barrier and other factors,when the wind speed is the same but the wind direction is different,the output power of the wind turbine set is also different.Based on this,this paper firstly decomposes the wind speed and power,obtains the components of different frequencies,synthesizes the effect of wind direction,using the neural network to establish the forecasting model of wind speed component to power component,realizes component divide and conquer,and fuses the forecasting wind power component into final forecasting wind power.Secondly,a new hybrid decomposition method combining wavelet packet decomposition and ensemble empirical mode decomposition is proposed.In the proposed decomposition method,the original wind turbine measured data are firstly decomposed by wavelet packet decomposition,and the original data are divided into high-frequency components and low-frequency components.Then the high-frequency components are decomposed into a set of intrinsic mode function components by ensemble empirical mode decomposition.Finally,combined with the measured data of wind farm,the applicability of the proposed combined decomposition method for wind power prediction is verified.Thirdly,The Elman neural network wind power forecasting method considering wind direction is proposed.In the proposed forecasting method,Elman neural network forecasting models are established based on the low-frequency components and intrinsic mode function components respectively.Considering the influence of wind direction on wind power prediction,the wind direction data was taken as an input vector of Elman neural network to forecast the wind power component,and the forecasting wind power components were fused to obtain the final forecasting ultra-short-term wind power through neural network.Finally,combined the monitoring data of wind farm,the accuracy of Elman neural network power prediction method considering wind direction is verified.Fourthly,testing and improvement are carried out for the proposed ultra-short-term wind power forecasting approach based on component divide and conquer.The results show that the decomposition method combining wavelet packet decomposition and empirical mode decomposition can decompose the fluctuating original wind speed and wind power data into stable components step by step,reduce the influence of wind turbine data volatility on the prediction results,and improve the accuracy of wind power prediction.In addition,considering the influence of wind direction,the components of wind speed with different frequencies obtained by decomposition are respectively used for the prediction modeling of wind power components based on Elman neural network,and the method of integrating the forecasting wind power components through the neural network can further improve the forecasting accuracy of wind power and more accurately reflect the trend of wind power changes.