摘要
容错是应用分布式并行计算系统时必须解决的一个关键难点。在基于广域网络的新型计算环境下实现基于进化算法的电力系统优化应用时,需要对大量个体进行频繁的迭代评估,现有的各类容错技术难以实现对此类应用的高效容错。文中结合进化算法概率性搜索、个别个体失效不会影响系统整体性能的特点,提出以父代个体取代未按时返回的子代个体的方式实现容错,并结合基于差异进化算法的无功优化问题对所提出的方法进行了仿真分析。IEEE118节点系统测试表明该方法能以优化性能的降低为代价实现高效容错。在该容错手段支持下,可通过采用更大范围网络计算资源基础上更大的群体规模,取得一致性更好、更接近全局最优的解。
Fault tolerance plays a key role in the success of distributed computing systems. Under widely distributed computation circumstance, the process of evolutionary algorithm based optimization of power systems involves substantial evaluation. Current fault tolerance schemes that do not consider the Characteristic of specific applications do not fit the evolutionary algorithms based applications. In evolutionary algorithm, the optimal resolve is searched according to certain probability. Hence failure of separate individuals won＇ t be fatal to the operation of the whole system. A simple and effective approach using differential evolution （DE） is proposed to realize the fault tolerance in parallel optimization of reactive power flow. In the proposed method the filial generation not being able to return is substituted by the corresponding parent generation. Simulation in the IEEE 118 system shows that at the cost of performance deterioration, the system can handle individual failure and realize fault tolerance conveniently and effectively. However, based on the approach proposed, we can use evolutionary algorithm with larger population size togain better performance even with a very high fault probability.
出处
《电力系统自动化》
EI
CSCD
北大核心
2007年第21期15-19,共5页
Automation of Electric Power Systems
基金
国家自然科学基金重大项目(50595414)
高等学校优秀青年教师教学科研奖励计划资助项目(104123)~~
关键词
无功优化
容错
分布式计算
差异进化
optimization of reactive power flow
fault tolerance
distributed computing
differential evolution
