Research On Ultra-short-term Combined Forecasting Method Of Wind Farm Power Based On Wind Process Pattern Recognition

Due to the large-scale development of non-renewable energy,the global climate problem is increasingly serious,and wind power is widely concerned and studied at home and abroad because of its clean and pollution-free excellent characteristics.In recent years,the installed capacity of global wind power is increasing year by year,but the randomness and fluctuation of wind power will lead to voltage flicker and frequency fluctuation of power grid,threatening the safe and stable operation of power system.Among the measures to promote wind power consumption,forecasting the output power of wind farm in the future is an effective means.Therefore,this paper studies the wind farm power prediction method of ultra-short-term.Wind power prediction can be divided into direct power prediction and wind speed prediction,and then get the power at the future moment through the wind-speed-power mapping model.This paper chooses the second method for research.At present,in the wind speed prediction research,the original wind speed sequence is generally divided into different wind processes first,and the classification method is mostly mathematical method or fixed time period,without considering the combination with the operation characteristics of the fan.In addition,the traditional research rarely considers the future change trend of wind speed.In view of these shortcomings.In this paper,a wind process classification method combined with fan operation characteristics and ultra-short-term combined prediction method based on pattern recognition of future wind process are proposed.The method mainly includes the following five stages.Firstly,the operating characteristics of wind turbines are analyzed,and the thresholds of cut wind speed and rated wind speed are defined.Then the original wind speed sequence is divided into a series of wind processes.Secondly,according to the variation trend of fan output power in different wind processes,the wind processes are divided into six types of wind modes,and the feasibility of the classification method is verified.Thirdly,the Adaboost classification model was used to predict the wind patterns at the future time to understand the change trend of wind speed at the future time.Fourthly wind speed forecasting model is established,this paper select the BP neural network,LSTM neural network,support vector machine(SVM)and LightGBM as a child model,through training the model and get the best combination of six kinds of mode,according to the prediction of wind model,dynamic choice corresponding prediction mode of the combination model,get the final forecast wind speed.Finally,the wind speed-power mapping model is established to obtain the predicted power at the corresponding time.In order to prove the effectiveness of the proposed method,the wind speed power data of a wind farm in Inner Mongolia,China are used for simulation verification,and the results show that the proposed method has higher prediction accuracy compared with other prediction models.