Font Size: a A A

Schedulable Capacity Forecasting For EVs Based On Machine Learning And Its Application In Energy Management System

Posted on:2021-01-29Degree:MasterType:Thesis
Country:ChinaCandidate:S L ZhangFull Text:PDF
GTID:2392330614959849Subject:Power system and its automation
Abstract/Summary:PDF Full Text Request
The rapid growth of electric vehicles brings great challenges to the safe and stable operation of power grid.However,the technology of Vehicle-to-grid & Grid-to-vehicle(V2G & G2V)can provide distributed energy storage service for power grid,which is an important part of smart grid in the future.The key to the realization of V2 G technology is to quickly achieve accurate electric vehicle schedulable capacity forecast(EVSCF).However,with the increasing application scale of electric vehicles,the data scale of electric vehicles is growing explosively,which puts forward higher requirements for data mining analysis and the modeling of EVSCF model.At the same time,under the background of big data driving,traditional machine learning algorithm will no longer suitable for EVSCF model.This paper focuses on the application of big data analysis and machine learning methods in EVSCF and the application of EVSCF in microgrid energy management.The contents of work and innovations include:1.Against the shortcomings of traditional machine learning algorithm,based on the big data platform of Hadoop and Spark parallel processing framework,traditional random forest algorithm and gradient boosting decision tree algorithm are parallelized.And the parallel random forest algorithm and parallel gradient boosting decision tree algorithm are constructed,as well as long short-term memory network algorithm in deep learning and applies them into EVSCF to solve the time-consuming and prediction accuracy problems of processing massive data under the condition of single machine.2.Considering the multi-time scale demand of power grid for electric vehicle scheduling,three-time-scale EVSCF models,real-time with one minute,ultra-short-term with one hour and one-day-ahead with 24 hours,are established.In order to verify the proposed method,different from traditional probabilistic modeling method,this paper takes the actual operation data of electric vehicle buses as data source.And it compares the prediction results of three algorithms of parallel random forest algorithm,parallel gradient boosting decision tree algorithm and long short-term memory network algorithm based on the built big data platform and deep learning platform.The simulation results show that prediction accuracy of the proposed parallel gradient boosting decision tree algorithm is better than the other two algorithms,in which the MAPE error value and the RMSE error value of ultra-short-term EVSCF model can reach 3.37% and 3.96%,which effectively reduces prediction error and calculation time.3.The Energy Management System for Microgrids(MG-EMS)model,including photovoltaic,wind power generation,energy storage and electric vehicle,is constructed.And the rolling prediction method of electric vehicle schedulable capacity is put forward innovatively.According to the prediction results of photovoltaic,wind power generation and loads at different times,as well as the rolling prediction results of electric vehicle schedulable capacity,participate in EMS regulation dynamically.After the comparison of no V2 G and V2 G scenarios,the results of example analysis show that the model proposed in this paper can not only ensure economic benefits of new energy generation and daily use needs of electric vehicles,but also achieve the goal of economic scheduling.
Keywords/Search Tags:Electric Vehicle, Machine Learning, Schedulable Capacity Forecast, Microgrid, Energy Management Strategy
PDF Full Text Request
Related items