Research On Load Forecasting Method Based On Incremental Updated Algorithm For Association Rules

In recent years,with the rapid expansion of China's power industry,both enterprises and users pay more attention to the research of load forecasting,which is an important index of the power system.According to different forecast targets,load forecasting can be divided into short,medium and long three categories.Among them,short-term forecasting is based on daily forecasting and its value is mainly affected by meteorological and time factors.The keys to improve the accuracy of load forecasting are the factors which influence the load change and the load forecasting model.Fast and accurate load forecasting allows forecasters to keep track of the load changes and provide reliable support for business decision-making.Combined with the association rules mining technology,this thsis carries on a thorough research to the load forecasting.The research mainly focuses on the following aspects:(1)An improved incremental updated algorithm for parallel association rule is proposed based on parallel frequent pattern-tree algorithm and the Spark distributed processing framework.This algorithm optimizes the structure of the frequent pattern tree as well as the parallel grouping strategy and reduces the time and space complexity.The experiment results demonstrate that the algorithm is feasible and has high mining efficiency under dynamic big data environment.(2)A short-term load forecasting method is designed based on the improved parallel association rules algorithm.The method can quickly and accurately catch the factors which influences load changes from the mass load data and output the predicted values.The experiment results demonstrate that the predicted load data generated by this method has high accuracy and reliability.(3)An urban short-term load forecasting system is set up,which includes a pretreatment module,an association rules mining module and a forecasting module.The association rules mining module performs mining on the preprocessed data to screen out the attribute factors closely related to the load and output the result to the forecasting module;the forecasting module calculates the future load via a multiple regression model.