Study Of Feeder Load Forecasting System For Smart Distribution Grid

The forecasting feeder load can make the operation decisions, stabilize the loadfluctuations, and guide users consuming energy, which is significant for ensuring thesafe and economical operation of distribution grid. As the small distributedgenerations (DGs) such as wind and photovoltaic developing rapidly, their energyhave obvious random intermittent since the influence of intensity uncertainties ofwind and sunlight. The intermittent exacerbates the load fluctuations on distributionfeeder, and interferes the safe operation of distribution grid. It takes new challengesfor the feeder load forecasting of smart distribution grid.In this paper, according to the different characteristics between the feeder loadof smart distribution grid and the traditional one, feeder load of smart distributiongrid are divided into three parts: the user’s load, the negative load of wind DG andthe negative load of PV DG. Considering its intermittent impact on feeder loadforecasting, the power produced by DG is defined as a negative load according tothe direction of the power flow, and the concept of net load is defined.On the basis of divided loads, the features for feeder load of smart grad feederload are analyzed so that deciding prediction methods and the required data. Theuser’s and the PV load patterns are established using the methods of C-fuzzyclustering and K-means clustering.The different methods of forecasting feeder loads are presented separatelyaccording to the characteristics of feeder loads. The load forecasting models basedon robust regression and improved Elman neural network are established. The windpower forecasting model based on ARMA time series that de-noised by wavelet iscreated. The PV load forecasting model is established by using GRNN neuralnetwork. The forecasting method of net load is determined based on thereorganization. The average relative errors and the VAR values are used to evaluatethe prediction accuracy of methods, and analysis and risk of forecasting error.Data warehouse is designed according to the required forecasting data, and thefeeder load forecasting system of distribution grid is established based on datamining. The system includes the following main functions: load analysis,establishment of various load patterns, forecasting feeder load, analysis ofprediction error and testing prediction methods. Finally, a feeder which connectedwith DGs in a distribution grid is taken as an example to verify the effectiveness ofthe designed feeder load forecasting system of smart distribution grid.This research is supported by the National Grid Corporation’s project "Theresearch of key simulation technology for controlling to running the smart distribution grid"(DZB17201200260).