【作者】 章政；

【导师】 肖登明；

【作者基本信息】 上海交通大学 ， 电力系统及其自动化， 2007， 博士

【摘要】 电力变压器是电力系统中分布广泛、结构复杂、造价昂贵的重要电气设备之一,担负着电压转换和电能传送的重任,它们的安全运行直接关系到整个电力系统的稳定性和安全性。对电力变压器进行绝缘状态检测和潜伏性绝缘故障诊断一直是电力部门面临的重要问题之一。因此,研究电力变压器的故障诊断技术,提高电力变压器的维护水平,具有重要的现实意义。油中溶解气体分析(Dissolved Gas Analysis,简称DGA)是目前分析电力变压器等充油电气设备绝缘状态最重要的检测方法,为诊断电力变压器内部故障提供了重要依据。DGA技术通过分析绝缘故障机理和大量的实际故障数据来发现油中溶解气体的组分含量与故障的类型和严重程度之间的关系,进而归纳整理出诊断不同故障的规则。由于这些常规的故障诊断规则主要来源于统计方法和经验积累,在实际应用中仍然存在许多不足之处。近年来,神经网络、模糊理论、灰色系统、粗糙集理论以及专家系统等人工智能技术的发展为变压器绝缘故障诊断提供了新的研究途径,逐步成为变压器绝缘故障诊断技术的主要研究方法。但是,采用这些传统的人工智能方法建立故障诊断模型时,往往需要依靠专家的诊断知识和经验事先确定诊断模型的结构等重要的模型要素,或者确定模型结构的方法是局部寻优的,这在一定程度上阻碍了电力变压器绝缘故障诊断系统的发展和推广。随着计算机技术和人工智能技术的发展,让计算机自动发现系统内在的规律并自动建立故障诊断系统模型已成为计算智能的研究热点。遗传编程(Genetic Programming,简称GP)算法是计算智能理论中进化算法的重要分支之一,其灵活的动态模型结构表示和借鉴生物界自然选择与遗传机制的全局搜索能力使得它在数据挖掘、控制理论、电子工程、模式识别等研究领域取得了广泛的成功。鉴于GP算法的特点,本文将GP算法首次引入电力变压器绝缘故障诊断,以DGA数据为特征参数,提出了四种以GP算法为基础的变压器绝缘故障诊断模型。论文主要包括以下内容:①基于GP和判别函数的变压器绝缘故障诊断模型。判别函数法是模式分类中常用的方法,虽然常规的线性判别函数法用于故障诊断具有简单、方便的特点,但是它是以更多还原

【Abstract】 Power transformer is one of widely distributed, complex and expensive equipment in the power system. It undertakes the heavy task of voltage conversion and power transmission. And its safety state plays a great effect on the stability and security level of power system. It is one of the most important issues for electricity sector to monitor and detect the potential insulation faults of the power transformers. Therefore, it is of great realistic significance to study the fault diagnosis technology and raise the level of maintenance of power transformers. Currently, Dissolved Gases Analysis (DGA) is the most important means to analyze the incipient insulation fault statement of oil-filled power equipments. The principle of DGA is based on the insulation failure mechanism and a lot of actual DGA data to discover the relationship between dissolved gases with the fault reasons and severity of power transformers and conclude the diagnosis rules from it. However, when applying these rules in the practical application, there are still many deficiencies because these rules are mainly generated from the accumulated experience and statistical methods.In recently years, the great development of artificial intelligence technology, such as neural network, fuzzy theory, grey system, rough set and expert system, has provided new research ways to diagnose the faults of power transformers. And it has gradually become the main research method. However, when constructing the diagnosis model based on these traditional artificial intelligence technologies, it usually need to preset the important issues of diagnosis model, such as the model structure, by the experts’knowledge and experience. And the model’s structure is usually locally optimized. In a certain extent, it hinders the development and promotion of insulation fault diagnosis system for power transformers. With the development of computer technology and artificial intelligence, allowing computers to discover the inner relationships of system and build the fault diagnosis models automatically has become a hot research issue in the domain of computational intelligence. Genetic programming (GP) algorithm is one of the important branches of evolutionary algorithms in the theory of computational intelligence. GP has achieved wide success in the domain of data mining, control theory, electrical engineering and pattern recognition because of its flexible expression of model structure and global searching ability.In this dissertation, with the characteristics of GP and DGA, four GP-based insulation fault diagnosis models for power transformers have been proposed. The main contents are as更多还原