| The rapid development of electric vehicles and the large-scale construction of charging posts lead to a huge charging load connected to the power grid,which will pose a huge challenge to the safe operation of the local power grid.Therefore,there is an urgent need for a unified management of large-scale electric vehicles charging to ensure the stable operation of the power grid under the premise of meeting the users' requirements.In order to manage the charging behavior of users uniformly,a method was proposed to guide the users to orderly charge with a reasonable time-of-use price based on the prediction of future load data.By analyzing and sorting daily load data of charging station in a region in the past two years,fuzzy clustering and improved gray relational analysis were used to obtain similar day sets of the days to be measured.Then the load data in the sets was used to train daily load of the days to be measured by using GA-BP neural network,predicting the load.Hadoop cloud platform was built,the storage and read-write of historical data in HDFS file system were completed,and GA-BP training and prediction in Map/Reduce were carried out.The results show that the Hadoop cloud platform performs similar day selection and GA-BP prediction,improving the efficiency and prediction accuracy.According to the prediction results,the charging behavior was controlled in an orderly manner,and a multi-objective optimization model was established.The multiscale mutation particle swarm optimization method was used to optimize the user cost and load fluctuation.Particle swarm optimization could reduce the user cost and the load fluctuation.Multi-scale mutation avoided particles group getting into the local optimum in the process of decreasing mutation operator.The results show that the use of forecasted results for time-of-use price adjustment can reduce the user cost to a certain extent and improve the power quality so as to achieve the purpose of "peak load shifting". |
PDF Full Text Request