Forecast And Application Of Schedulable Capacity Of TCLs Based On Big Data Analysis Method

Due to the shortage of traditional fossil energy and environmental protection,in recent years,our country has vigorously developed distributed energy generation.The increasing penetration of distributed energy in the power system,which has brought great pressure to the safe and stable operation of the power system.The extensive demand response resources from the demand side,especially the thermostatically controlled loads(TCLs),can provide various ancillary services for the power system.Because of its fast response capability to the power grid dispatching,it will help maintain the stability of the power system and relieve the operation pressure of the grid.The ability of TCLs to participate in various services is evaluated by the schedulable capacity of the TCLs,which is the basis for the TCLs to participate in the various ancillary services of the power system.This paper focuses on the prediction of schedulable capacity of TCLs and its application in distribution network.The main works and innovations are as follows:1.Considering the comfort requirement of users with TCLs,a real-time schedulable capacity calculation model of TCLs is put forward based on real-time data of TCLs operation states.2.According to the big data scenarios that may be encountered in the prediction of the TCLs,we use big data analysis methods to predict the ultra-short time-day schedulable capacity of the TCLs.It mainly includes using distributed file system to store data,and using the parallel big data algorithm to build the intra-day schedulable capacity prediction model with TCLs.The scheduling algorithms used in this paper include the parallel decision tree algorithm,the parallel random forest algorithm and the parallel gradient boosting decision tree algorithm.Finally,the simulation is carried out by the big data platform of Hadoop and Spark.We use the rolling forecast method to train the model with a step length of 15 min.Then the experimental results are analyzed and results show that the parallel prediction algorithm we proposed can effectively improve the prediction accuracy and speed.3.Applying the above TCLs rolling forecasting results and considering the characteristics of the large number of nodes and the distributed distribution of nodes in the distribution network,the centralized clearing of the power transaction meet high cost.Therefore,a peer to peer transaction model of power distribution in a regional distribution network is proposed.The TCLs with fast response capability can sell its own schedulable capacity according to the prediction results of the schedulable capacity and participate in the power transaction in the regional distribution network to realize the energy balance of distribution network.