| 仇翔*,蒋文泽*,吴麒*,张宝康*,葛其运**.基于专家知识与监测数据联合驱动的高压开关柜状态评估[J].高技术通讯(中文),2024,34(7):776~786 |
| 基于专家知识与监测数据联合驱动的高压开关柜状态评估 |
| State evaluation for high-voltage switchgears by combined domain-knowledge-driven and monitored-data-driven methodology |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2024. 07. 011 |
| 中文关键词: 高压开关柜(HVS)； 状态评估； 参数学习； 知识与数据联合驱动； 贝叶斯网络(BN) |
| 英文关键词: high-voltage switchgear (HVS), state evaluation, parameter learning, knowledge-driven and data-driven, Bayesian network (BN) |
| 基金项目: |
| 作者 | 单位 | | 仇翔* | (*浙江工业大学信息工程学院杭州 310014）
(**科润智能控制股份有限公司衢州 324100) | | 蒋文泽* | | | 吴麒* | | | 张宝康* | | | 葛其运** | |
|
| 摘要点击次数: 423 |
| 全文下载次数: 374 |
| 中文摘要: |
| 高压开关柜(HVS)作为电力系统的关键设备，对其工作状况进行有效评估可以保障电力系统的安全稳定运行。在工程实践中，由于高压开关柜长期服役于潮湿、高温等恶劣环境下，不可避免的传感器失效或人为因素会导致其设备状态数据存在随机缺失的现象，从而破坏了数据的完整性和可用性，使得对数据质量要求较高的数据驱动方法难以直接用于解决高压开关柜状态评估的问题。为了解决上述问题，研究了一种基于专家知识和监测数据联合驱动的高压开关柜状态评估方法。首先，对高压开关柜系统的内部构成进行了深入分析，并根据区域中设备的功能不同将其分为电缆室、母线室和断路器室三大区域。其次，进一步分析了系统状态、各区域状态及其关键部件状态两两之间的因果关系，从而建立了适用于高压开关柜状态评估的三层贝叶斯网络(BN)拓扑结构。然后，引入专家领域知识设计了适用于高压开关柜系统的3种约束罚函数，并通过求解带有约束的优化问题，改善了不完整数据集下的贝叶斯网络参数估计性能，进而实现了对高压开关柜系统状态的精确评估。最后，在自主设计的10kV高压开关柜样机上开展了对比验证实验，结果表明，相比于支持向量机(SVM)方法和反向传播(BP)神经网络方法，本文所提方法在状态评估精度上更具优势。 |
| 英文摘要: |
| High-voltage switchgear (HVS) is an essential component of the power system, and its state evaluation is of great significance for maintaining stability and safety of the system. In fact, HVS is usually served in such harsh environments with high temperatures and humidity. It is common knowledge that the state of HVS is lost randomly due to unavoidable sensor failure and uncontrollable human factors, which has limitations on the integrity and availability of data. As a result, data driven methods which require high-quality data are hard to be directly applied for state evaluation of HVS. To solve the above problem, an integrated state evaluation approach is proposed by combining domain-knowledge-driven methodology with monitored-data-driven methodology. First, the internal structure of the HVS is in-depth analyzed and it is categorized into three regions pursuant to the terms of the device utility, i.e., the cable-room region, busbar-room region, and circuit breaker-room region. Then, a three-layer Bayesian network (BN) topology for state evaluation of HVS is conducted by the causality analysis of the relationship between each pair of system states, regional states, and basic device states. While the expert domain knowledge is adopted and three kinds of constraint penalty functions matching the BN model are developed. Since the constrained optimization problem is solved, the parameter estimation performance of BN model under incomplete data sets is improved and the accurate state evaluation of the HVS is accomplished. Finally, comparative experiments are carried out on the self-designed 10kV HVS prototype. The results show that the proposed approach is able to achieve the goal of accurate state evaluation and has superior performance in items of both accuracy and convergence compared with support vector machines (SVM) and back propagation (BP) neural network. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
