Research On Medium-term Wind Power Prediction Based On Meteorological Data

With the continuous expansion of wind power installation and the further improvement of the power grid system,the state attaches more and more importance to the consumption of new energy power,and the requirements of wind farms and Internet electricity are becoming more and more strict.The latest power prediction assessment standards increase the requirements of medium-term power prediction,and the prediction time covers the power prediction of 0-10 days in the future,requiring the accuracy of no less than 70% on the 10 th day.Through the in-depth study of the mid-wind power forecast of wind farms,constantly improving the accuracy of the mid-wind forecast can minimize the assessment loss of wind farms,and it is of great significance for the consumption of new energy.Considering the long forecasting time of the mid-wind power,it is difficult to predict.Based on meteorological data,this paper makes an in-depth study on the mid-wind power forecast of wind farms.In order to assist wind farm operation and maintenance personnel to develop more accurate and efficient operation and maintenance strategies,a method of feature discretization processing of ultra-long period meteorological data is proposed.A classification filter model based on rule screening and Bin algorithm was proposed to realize efficient and rapid mass cleaning and filtering of the wind power curve.Combining these two methods,the characteristic system required for effective wind power prediction is established.In order to improve the accuracy of prediction,the method of linear fitting is proposed to solve the optimal point position of wind tower based on the wind speed of the unit,and Pearson correlation coefficient method is proposed to determine the unit parameters of power prediction.Finally,by comparing the prediction effects of ARIMA model,Prophet model,SARIMAX model and ML model,XGBoost model is used to predict the wind power of the wind farm in the next 10 days.The prediction results meet the national standard for the accuracy of wind power prediction in the next 10 days.The following is the main content of this article:(1)In view of the short operation time of some existing units,the operation data of the units is small,which is not enough to provide a valuable reference for the operation and maintenance of wind farms,especially the planned maintenance.Under such circumstances,through the characteristic discretization method to process ultra-long-term meteorological data and apply it to the process of lean operation and maintenance of wind farms,the potential meteorological change laws of wind farms can be obtained,and these results or laws can be used as a basic tone for long-term operation and maintenance of wind farms,which can provide help for power generation enterprises to find suitable operating windows,and can also help power generation enterprises formulate macro operation inspection plans,seasonal maintenance plans,etc.,thereby reducing unplanned and blind maintenance work.Loss of power generation can be minimized.(2)Study on wind speed-power curve cleaning based on SCADA data.By filtering the wind power curve by mass classification,it can facilitate the daily maintenance of wind farm and unit fault handling.In this paper,the isolated forest algorithm and the classification filtering model based on rule screening and Bin algorithm were respectively adopted to conduct mass classification filtering on the operating data of 24 units of a wind farm for one year.Finally,the classification filtering model based on rule screening and Bin algorithm was evaluated from multiple dimensions of cleaning degree and cleaning time.The algorithm can meet the requirements of mass cleaning air power curve.(3)Considering the urgent demand of wind power industry for medium-term(next 10 days)wind power prediction,this paper studies wind power prediction for the next 10 days based on wind farm meteorological data.In order to improve the accuracy of the prediction model and obtain a good prediction result,the meteorological data of 5 points of the wind farm combined with the output power of the whole wind farm was selected as the data set of the study.In the aspect of data processing,linear fitting method is proposed to find the best point position of wind tower,and Pearson correlation coefficient method is proposed to determine the unit parameters of power prediction.In terms of the construction of feature engineering,the feature discretization method mentioned in the previous chapter is used to process the data set.Meanwhile,the classification filtering model based on rule screening and Bin algorithm in Chapter 4 is used to filter the abnormal power value.In terms of the selection of prediction model,by verifying mixed models such as ARIMA,Prophet,SARIMAX and ML(Machine Learning),Boosting algorithm was selected to predict the wind power of the next 10 days.In order to verify the effectiveness of the above method and the generalization ability of the algorithm,Wind farm data from another area were selected for verification.In this paper,different data processing methods and prediction models are adopted to further improve the accuracy of prediction model and achieve the purpose of reducing the assessment cost for power generation enterprises.