| 引用本文: | 谢 敏,闫圆圆,诸言涵,吴亚雄,刘明波.基于向量序优化的多目标机组组合[J].电力自动化设备,2015,35(7): |
| XIE Min,YAN Yuanyuan,ZHU Yanhan,WU Yaxiong,LIU Mingbo.Multi-objective unit commitment based on vector ordinal optimization[J].Electric Power Automation Equipment,2015,35(7): |
|
| 摘要: |
| 机组组合在数学上可建模为含连续、离散变量的动态优化问题，对于大规模电力系统，其最优解的求取不可避免地存在维数灾的弊端。以发电机组的煤耗量和购电费用为优化目标，引入向量序优化理论对大规模多目标机组组合问题进行求解。采用BP神经网络对表征集合进行快速评估，确定选定集合，在保证足够好解个数的前提下大幅降低计算量，缩短求解时间。以某省级实际电力系统为例，考虑水电、核电、生物质能、气电、火电等多种类型的复杂电源结构，选取典型日96点负荷曲线形成该日发电机组日启停计划和出力安排优化方案，将向量序优化求解结果与基于GAMS-BARON解法器的混合整数非线性规划（MINLP）法的计算结果进行对比分析，结果表明采用所提方法求解到满足工程实际需要的足够好解，其计算速度是传统MINLP法的7.608倍，说明所提方法是可行且有效的。 |
| 关键词: 机组组合 多目标 优化 向量序优化 足够好解 BP神经网络 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金青年基金资助项目（50907023）；国家重点基础研究发展计划（973计划）项目（2013CB228205） |
|
| Multi-objective unit commitment based on vector ordinal optimization |
|
XIE Min, YAN Yuanyuan, ZHU Yanhan, WU Yaxiong, LIU Mingbo
|
|
School of Electric Power，South China University of Technology，Guangzhou 510640，China
|
| Abstract: |
| Unit commitment can be mathematically modelled as a dynamic optimization problem containing both continuous and discrete variables. For large-scale power system，a dimensionality curse is inevitably encountered during the optimal solution search. The vector ordinal optimization with the generator coal consumption and power purchase cost as its optimization objectives is introduced to solve this problem. The BP neural network is adopted to quickly evaluate the characterization set for determining the candidate set，which significantly decreases the computational load and time while the quantity of good-enough solution is ensured. With a structurally-complex provincial power grid as an example，which has different types of power source：hydropower，nuclear power，biomass energy，thermal power，etc.，the optimal scheme of daily unit on／off plan and output arrangement is based on the 96-point load curve of a selected typical day，the calculative results obtained by the proposed method and the MINLP（Mixed Integer NonLinear Programming） method based on GAMS-BARON solver are compared，which show that，to obtain a good-enough solution meeting the need of practical engineering，the calculation speed of the proposed method is 7.608 times of that of MINLP method，verifying its feasibility and effectiveness. |
| Key words: unit commitment multi-objective optimization vector ordinal optimization good-enough solutions BP neural network |
