Research On Power System Short-term Load Forecasting Based On Gradient Boosting Decision Tree

Electric power industry is the basic pillar industry to ensure all-round coordinated development in China.The stable operation of power system affects the prosperity and development of the country and the people's good life.With the reform of China's power industry system and the continuous optimization of power supply structure,short-term load forecasting,as an indispensable part of power market analysis and forecasting,is becoming more and more important.Accurate short-term load forecasting technology provides technical support for the power sector to formulate dispatching plans,and guarantees the safe and stable operation of the power grid.At the same time,in the gradually improved power market,the high-precision short-term load forecasting technology provides the basis for economic dispatch for generation side,transmission and distribution side and power purchase side,so that they can pursue higher economic benefits under meeting the basic needs of users.Therefore,short-term load forecasting has long been one of the focuses of scholars and experts at home and abroad.Beginning with the influencing factors of short-term load forecasting accuracy,this paper studies load forecasting based on decision tree theory,ensemble learning,fuzzy theory and other related knowledge.Through research,some progress has been made.Inspired by the idea of ensemble learning,this paper integrates CART decision tree model with Gradient Boosting algorithm.Through theoretical analysis,it is proved that Gradient Boosting algorithm can improve the accuracy of load forecasting,and a short-term load forecasting model based on gradient lifting decision tree is proposed.In order to increase the learning space of a single CART model and improve the generalization performance of the forecasting model,Shrinkage idea is introduced into the model.The simulation analysis of actual power load data in a certain area proves that the model has higher forecasting accuracy,but still lacks stability.This paper innovatively applies GBDT algorithm to the field of short-term load forecasting,and achieves the desired results,which provides a certain basis for the following research.Benefiting from the basic principle of random forest algorithm,this paper introduces Bagging algorithm to the GBDT model.Bagging algorithm randomly sample the training samples according to a certain proportion,which can reduce the training samples of a single GBDT model.And Bagging algorithm realize parallel training and prediction of multiple models,which is conducive to saving prediction time and improving the generalization performance of the model.At the same time,the membership function in the fuzzy theory is used to process the sample data,and the uncertain factors are summarized by the way of human thinking,which can find the potential information and rules between input vectors and target variables.On this foundation,a short-term load forecasting model based on fuzzy Bagging-GBDT is proposed.Analysis results of an example show that the model has more stable and accurate forecasting performance than GBDT model.