高准确度电网孤岛效应协同检测模型研究

Along with the high penetration of distributed sustainable energy such as wind and solar in China’s power system, Islanding detection has become one of the most key issues to protect the distribution network and the entire grid. Due to the drawbacks of the traditional islanding detection methods, such as the unavoidable disturbances to the power quality and the inevitable non detection zone, this proposal introduce a totally new approach to set up a principal component model for passive islanding detection, by introducing process control methods and artificial intelligent methods, key issues such as the computational efficiency and accuracy of detection model can be solved. In this proposal, key technologies including principal component analysis, dynamic and non-linear probability principal component analysis will be studied theoretically to set up islanding detection model. In order to reduce the false alarms caused by slow changing but normal process, a recursive update approach will be also studied to calculate the model as well as the control limits. For the non-linear and non-Gaussian distributed data, artificial intelligent method and optimization method are cooperated to set up the monitoring model, eliminating the non-linearity existed in the power data so that the accuracy of the islanding detection model can be guaranteed. Moreover, multimode heterogeneous data integration technology is studied, and the high efficiency computational engine is designed, thus the possible relationships between the grid environment, weather condition, the geographic factor and the happening of power gird islanding event can be revealed through deep learning strategy. Therefore, this proposal not only can provide a solid technology background to support the smart distribution network to realize the automated islanding detection, but also maintain the safety of power equipment and personnel.

随着以太阳能、风能为代表的分布式发电在我国电网的大量接入和高度渗透，孤岛效应检测已成为配电网保护的一个重要方面。传统的孤岛效应检测存在干扰电能质量、难以消除检测盲区等缺点，本项目提出一种新型被动式孤岛效应检测模型，引入过程统计与人工智能方法，解决数据量大时检测模型计算效率和准确度低的技术难题。从理论上全面研究线性主元分析、动态和非线性概率主元分析方法，建立孤岛效应检测模型；深入研究迭代更新方法实时计算模型及控制限，解决缓慢时变过程引起的误报；针对电网非线性与非高斯分布特征数据，提出人工智能与优化算法相结合的协同模型，消除电力数据的非线性关系，实现高准确度电网孤岛效应检测；研究多模异构电力大数据融合技术，设计高效计算引擎，深度挖掘孤岛效应与电网的环境、气候、地理等因素的关联关系。该研究可为智能配电网实现自动化孤岛效应检测运行提供关键技术基础，维护和保障了电网设备和人员生命安全。

本课题围绕新型被动式孤岛检测模型建立方法，引入过程统计与人工智能策略，解决数据量大时检测模型计算效率和准确度低的技术难题。提出了基于数据驱动的主元分析系统方法，充分利用高精度的、全网时间同步的广域测量信息资源，成功建立孤岛效应检测模型，高效、快速的实现孤岛效应检测，并根据提取主元的贡献率能够简便快速识别孤岛发生地点；当电网状态异常时，出现丢失数据、误报数据等情况，导致过程数据不符合正态分布，此时线性模型无法正确检测孤岛效应，本课题继续研究了概率主元分析、核主元分析模型建立的方法，通过概率的方法解决了丢失数据的实际问题，提高检测准确率；针对缓慢变化过程，设计了迭代更新主元算法，动态更新模型参数，实时动态监测电网输电线路的暂态扰动状况。研究了人工智能方法与传统方法的结合应用，设计时空关联分析模型，针对输电线路跳闸、电压暂变与天气、环境等因素进行关联分析，挖掘可靠关联规则，同时研究了深度学习方法在大规模电动汽车充电负荷数据分析与预测中的应用；最后，针对电网多模异构数据特征，将大数据技术引入电力系统应用分析，搭建了基于内存分布式计算框架Spark的电网大数据分析管理平台，对城市电网的三个主要应用进行研究，包括基于CIM/E文件的电网拓扑图构建、电网负荷削减并行计算、以及电压暂降关联规则挖掘与三维展现案例。本课题将主元分析系统方法应用到基于同步检测向量数据的孤岛效应模型中，能够保持现有电网结构、充分利用高精度的，全网时间同步的广域测量信息资源，解决传统方法可能引入的干扰对电能质量的损坏，基于多个对象同步分析处理，减少孤岛效应检测时间，提高检测准确度，消除检测盲区，维护分布式发电并网系统以及大电网的正常运行；搭建了大数据管理分析平台，结合人工智能和大数据技术，消除信息孤岛、实现数据贯通。在研究团队的共同努力之下，本项目取得的成果已全面超额完成指标，共完成期刊论文13篇，会议论文8篇，申请相关专利10项。