Study On Mid-term Prediction Of Wind Power Considering Meteorological Features

With the depletion of fossil energy,the demand and exploration of clean energy around the world are becoming more and more urgent.Due to the easy availability and renewability of wind resources,utilizing wind power generation is of great significance for implementing national energy policies and saving fossil energy consumption.However,wind power generation has strong randomness,a certain degree of volatility and intermittent characteristics,which can easily deteriorate the safety and stability of the power grid.Therefore,in order to ensure the safety and economy of the power system,we need to make further research on wind power forecasting,improve the existing forecasting work,and optimize the accuracy and reliability of the forecasting model.To address the problem of mid-term forecasting of wind power,this thesis proposes a mid-term forecast model of wind power by using the improved Support Vector Machines based on the Prophet algorithm to improve forecast accuracy and reliability.We then propose a hybrid quantile regression model based on the stacking integration strategy,which integrates the P-SVM and random forest,as well as blends quantile regression with kernel density estimation for higher prediction accuracy.The key contributions of this work are organized as follows:First of all,the collected wind power and climate data of a wind power plant in the coastal area are preprocessed.The characteristics of future wind energy resources are analyzed.The random forest is adopted for feature sorting to achieve effectiveness evaluation and screening of multi-dimensional features.The results provide data support for the subsequent research work..Secondly,considering the facts that future climate change characteristics and data mining are not effectively involved in most previous studies,this thesis proposes an improved P-SVM wind power mid-term forecast method based on the Prophet algorithm.We compare this improved algorithm with other traditional models in different scenarios.The prediction results of this method and classical algorithms are shown by the data analysis with relevant indicators to show the superiorities of the P-SVM model.Finally,to further reduce the prediction error caused by the influence of extreme meteorological factors of a single algorithm and avoid uncertain information in the operation of the power system,this thesis proposes a combined mid-term probability prediction method based on the stacking strategy.The proposed method is compared with other probability prediction models in the subsequent experimental analysis of numerical examples,demonstrating its superiority via data analysis with relevant indicators.