Research On Short Term Wind Power Forecasting Based On Data Driven

With the vigorous development of wind power technology in recent years and the continuous expansion of the grid connected scale of wind power,the strong randomness of natural wind leads to irresistible uncertainty in wind power output,which brings great pressure on the stable operation of wind farms and power systems,system peaking and consumption.Accurate wind power forecasting is important to improve the quality of wind energy and promote the steady development of wind power.The quality and size of sample data and the selection of input parameters of the prediction model have a direct impact on the accuracy and robustness of data driven based wind power prediction model.The purpose of this paper is to address the abnormality,redundancy,high dimensionality,and finiteness of the raw SCADA data.A wind farm 2MW wind turbine was used as the research object to carry out deterministic and probabilistic wind power forecasting research in terms of anomaly data cleaning,input feature parameter selection,feature fusion,and fusion of aerodynamic mechanism data and actual measurement data.The main research contents of this dissertation are as follows:1.Based on the pauta criterion,the quartile method,the sliding standard deviation method and the Density Based Spatial Clustering of Applications with Noise(DBSCAN),a combination algorithm based anomaly data cleaning method was proposed considering the effects of different characteristic partitions of wind speed and power;data retention rate,pearson coefficient,standard deviation and coefficient of variation were used to quantify and evaluate the cleaning effect of anomaly data and improve the validity of sample data.2.The single factor analysis method was used to analyze the influence law of different meteorological parameters on the output power of wind turbines;Based on the entropy value method and evaluation indexes such as Pearson coefficient,Spearman coefficient,Kendall coefficient,mutual information coefficient and gray correlation coefficient,an entropyweighted comprehensive correlation evaluation method was proposed to quantify the meteorological parameters The correlation degree of meteorological parameters,unit control parameters and state parameters on the output wind power;The dynamic correlation input feature selection method was used to construct deterministic and probabilistic wind power prediction models based on BP,PSO-LSSVM and CNN,and the impact of different input feature selection methods on the accuracy of wind power prediction was comparatively studied.3.An input feature fusion method based on entropy-weighted integrated correlation evaluation was proposed to comparatively study the influence of with and without feature fusion on the accuracy of wind power prediction based on prediction accuracy evaluation indexes such as normalized maximum error,normalized mean error,normalized mean rootmean-square error and coefficient of determination.The results show that the method can fully utilize all raw data containing weak correlation without increasing the model dimensionality,and effectively improve the accuracy of deterministic and probabilistic wind power prediction.4.The wind turbine power and load calculation model was established based on the aerodynamic principle,and the correctness of this mechanism calculation model was verified by combining the wind field measured data;A wind power prediction method based on the fusion of aerodynamic mechanism data and measured wind field data was proposed,and IEC normal and turbulent conditions case studies were carried out;Simulation data of IEC-DLC1.1conditions for wind power and waving loads,combined with measured data,were used to investigate the effect of different scales of mechanistic and measured data on prediction accuracy.;Simulation data of IEC-B class high turbulence conditions for power and waving load,combined with measured low turbulence conditions data,were used to study the effect of having or not having high turbulence conditions data on the prediction accuracy and to provide a reference for improving the applicability of the prediction model.