电网技术
電網技術
전망기술
POWER SYSTEM TECHNOLOGY
2015年
1期
48-54
,共7页
电动汽车%多代理系统%分布式优化%峰谷差
電動汽車%多代理繫統%分佈式優化%峰穀差
전동기차%다대리계통%분포식우화%봉곡차
electric vehicles%multi-agent system%distributed optimization%peak-valley difference
为最大化电力公司利益，设计了一种用于协调电动汽车充电的多代理系统，并在满足电动汽车车主充电需求及变压器容量限制的前提下，提出一种以负荷峰谷差最小为目标的分布式优化算法。利用分时电价算法初步优化后得到理想的充电时间区间，在充电区间内应用优化算法避免新的负荷尖峰，引入训练学习机制以使负荷曲线达到削峰填谷的效果。根据用户的驾驶习惯，采用蒙特卡洛方法模拟用户的充电需求，对电动汽车在无序充电、单次优化充电以及引入训练学习机制充电3种情况下的电网负荷进行了仿真分析。研究结果表明：单次优化可以避免负荷尖峰，但不能优化峰谷差；而引入训练学习机制后在减小峰谷差方面有显著作用，而且该分布式优化有更高的计算效率，适于实际应用。
為最大化電力公司利益，設計瞭一種用于協調電動汽車充電的多代理繫統，併在滿足電動汽車車主充電需求及變壓器容量限製的前提下，提齣一種以負荷峰穀差最小為目標的分佈式優化算法。利用分時電價算法初步優化後得到理想的充電時間區間，在充電區間內應用優化算法避免新的負荷尖峰，引入訓練學習機製以使負荷麯線達到削峰填穀的效果。根據用戶的駕駛習慣，採用矇特卡洛方法模擬用戶的充電需求，對電動汽車在無序充電、單次優化充電以及引入訓練學習機製充電3種情況下的電網負荷進行瞭倣真分析。研究結果錶明：單次優化可以避免負荷尖峰，但不能優化峰穀差；而引入訓練學習機製後在減小峰穀差方麵有顯著作用，而且該分佈式優化有更高的計算效率，適于實際應用。
위최대화전력공사이익，설계료일충용우협조전동기차충전적다대리계통，병재만족전동기차차주충전수구급변압기용량한제적전제하，제출일충이부하봉곡차최소위목표적분포식우화산법。이용분시전개산법초보우화후득도이상적충전시간구간，재충전구간내응용우화산법피면신적부하첨봉，인입훈련학습궤제이사부하곡선체도삭봉전곡적효과。근거용호적가사습관，채용몽특잡락방법모의용호적충전수구，대전동기차재무서충전、단차우화충전이급인입훈련학습궤제충전3충정황하적전망부하진행료방진분석。연구결과표명：단차우화가이피면부하첨봉，단불능우화봉곡차；이인입훈련학습궤제후재감소봉곡차방면유현저작용，이차해분포식우화유경고적계산효솔，괄우실제응용。
To maximize the benefit of power company, a multi-agent system (MAS) to coordinate the charging of electric vehicles (EV) is designed. Under the premise of meeting the charging demand of EV owners and according to the restriction of transformer capacity a distributed optimization control algorithm, in which the minimized peak-valley load difference is taken as objective, is proposed. Firstly, the preliminarily optimized time-of-use (TOU) price algorithm is used to obtain ideal charging time interval; secondly, in the obtained time interval the optimization algorithm is utilized to avoid the occurrence of new peak load; finally, the training-learning mechanism is led into to achieve the effect of peak load shifting. According to the driving habits of EV owners the Monte Carlo method is used to simulate the charging demand of EV owners, and the power grid load conditions under three charging situations, namely the out-of-order charging, single-time optimization charging and the charging with leading in training-learning mechanism, are simulated. Results of this research show that adopting single-time optimization charging the new peak load can be avoided, however the peak-valley difference cannot be optimized; leading in training-learning mechanism plays obvious role in decreasing peak-valley difference and the distributed optimization control algorithm possesses higher computing efficiency, so it is suitable for actual application.
