【作者】 关龙；

【导师】 刘志刚；

【作者基本信息】 西南交通大学 ， 电力系统及其自动化， 2012， 硕士

【摘要】 在当今社会,广大用户不仅要求保证供电的连续性,而且要求保证供电的可靠性,然而,配电网规模的不断扩大,极大增加了故障发生的概率。为快速准确定位故障原因,一些人工智能的理论方法被用于配电网故障诊断,如基于经验的专家系统。但是传统专家诊断系统难以诊断经验之外的故障,系统的可移植性与可维护性较差。基于模型诊断方法(Model-based diagnosis, MBD)则直接利用电压电流等量测量来判断故障元件,可以在保护装置和断路器动作前进行故障定位,具有一定的故障预警功能,在产生更加严重后果之前及时排除故障并采取对应保护措施,已广泛应用于工程领域,航天领域,电路诊断等领域。首先,本文分析了课题的背景与意义,配电网故障诊断的研究现状。简述了MBD方法的基本思想、诊断步骤、常见存在的问题等。其次,MBD方法中一个关键的问题就是如何求解冲突集的最小碰集,传统碰集搜索算法搜索效率和正确率无法满足复杂系统故障诊断实时性的要求。由于离散二进制粒子群算法(BPSO, Binary Particle Swarm Optimization)算法有较快的收敛速度和较高的计算效率,引入改进的离散二进制粒子群算法用于求解冲突集的最小碰集,将计算最小碰集问题转化为0、1表示的二值空间问题。通过实际建模、编程和实验表明,二进制粒子群算法比HS-Tree、Boolean Algebra方法、遗传算法等算法搜索效率更高,可节约1/2-1/3的搜索时间,当问题规模较大时并未出现内存溢出问题,取得了较好的效果。接着,MBD方法存在着一定的不确定性问题,其诊断结论为一组或者多组引发故障的元件集合,导致实际应用效果不佳。本文引入贝叶斯概率方法进行诊断识别,利用最小诊断候选来区分可能状态集合并且降低计算的复杂度,由后验概率方法对诊断候选进行故障概率排序,从而以后验概率形式量化了诊断识别的衡量标准,并通过诊断实例验证方法的可行性与有效性。最后,通过完善MBD配电网故障诊断过程中存在的问题,结合配电网本身结构特点,给出一个基于MBD方法的配电网故障诊断方案。以两个配电网诊断实例,通过PSCAD软件来建立系统元件模型,通过不同的故障模式,采集配电网不同故障部位和不同故障类型时故障数据信息,实验结果表明本文方法可以快速准确的得到诊断结论,并且与故障假设相一致,验证了诊断方案的可行性与有效性。更多还原

【Abstract】 In the modern society, how to guarantee the continuity and reliability of power supply is very important, but with continuous enlargement of the scale of power network, the probability of its failure is increasesing. In recent years, many scholars use the theory and method of artificial intelligence in fault diagnosis of the electric distribution network in order to find out the cause of the malfunction accurate and fast, such as the expert systems based on experiences. But the expert systems could not diagnose the faults beyond experiences and are very difficult for system transplanting and maintaining. Model-based diagnosis (MBD) diagnoses faulty components using measurement value of voltage and current, and it can locate the fault before protection device and breaker active. This method has a function of failure warning, which can get rid of fault before it turns worse and take measures. It is widely used in engineering, aerospace fields, circuit diagnostic and so on.Firstly, this paper analyzes the current studies on the fault diagnosis of the electric distribution network, the background and the significance of the subject. The basic ideas, the basic processes, the main problems existing in MBD were introduced.Secondly, for MBD (Model-Based Diagnosis), a key step is to compute the minimal hitting sets from the minimal conflict sets, however the search efficiency of commonly used minimal hitting set algorithms could not satisfy the real-time requirement of complex systems. The fast convergent and high efficiency characteristic of BPSO (Binary Particle Swarm Optimization) makes it used extensively. The minimal hitting set (that is, candidate diagnosis) could be calculated from the obtained minimal conflict set based on improved BPSO, so that the algorithm maps the minimal hitting sets problem to0/1integer programming problem. The results of modeling, programming and testing show that the search efficiency of the improved BPSO is much better than HS-Tree, Boolean Algebra, genetic algorithm and other commonly used minimal hitting set algorithms and could save the time of1/3-1/2. The improved BPSO algorithm avoid the problem of memory overflow in the case of large-scale problem, which showed excellent performances in fault diagnosis of distribution network based on MBD.Thirdly, the diagnosis results based on model-based diagnosis (MBD) were a group or several groups of fault components set involved uncertainty, so the practical application effect is not good. The Bayesian probability method is adopted to identify the diagnosis. We could get the possible system state according to the minimal diagnosis candidates, and the posteriori fault probabilities of each diagnosis candidates could be calculated by Bayesian probability method. So, we could quantize the measure standard of the diagnosis identification in the form of the posteriori fault probabilities. Finally, an example of actual electric distribution network was presented to verify the feasibility and validity of this method through modeling, programming and testing.In the end, considering the structure of the electric distribution network and the demands in the fault diagnosis of the electric distribution network, this paper presents a scheme in which the theories of MBD is applied to the distribution system for fault diagnosis. Taking two10kV distribution networks as diagnosis example, model building by the software of PSCAD, when fault occurs, the fault information of the distribution network could be accurately measured. The results of modeling, programming and testing show that the MBD method had excellent performances in fault diagnosis of distribution network, and verified the feasibility and effectiveness of the method.更多还原