基于大数据分析的输变电设备状态评估基础理论与方法

The traditional condition assessment of power equipment is mainly based on the mechanism and causality model built through theoretical analysis, simulation and experimental testing. In this project, large amount of information, including equipment condition information, grid operation information and meteorological information, which are collected through the gradually improved power information platform, are utilized to analyze the relations and trends of equipment state evolution from the perspective of inherent law of data itself. A big data mining analysis method is proposed for the condition assessment of equipments based on power operation and maintenance data. A data-driven personalized equipment condition assessment and fault prediction model is built, thus providing a new idea and technique for the sophisticated assessment and predication of the equipment state. The following aspects will be researched in the project:1)State assessment oriented multi-source (grid, equipment and environment) heterogeneous information feature analysis and data normalization modeling method;2)Methods of multi-scale anomaly detection on real-time data flow of power equipment state;3)Principle and implementation strategy of integrated learning on the pattern of power transmission equipment character, and dynamic assessment model of key performance of equipment based on big data samples;4)Multi-dimensional relation recognition methods applicable to the evolution analysis of power transmission equipment to build a data-driven relation analysis and fault prediction model.

传统的输变电设备状态评价一般是基于理论分析、仿真和试验测试等手段建立的机理和因果关系模型。本项目拟利用日渐完善的电力信息化平台收集的大量设备状态信息、电网运行信息和环境气象信息，从数据本身内在规律分析的角度研究设备状态演变的关联关系和发展趋势，提出适用于设备状态评估的电力运行检修大数据挖掘分析方法，建立基于数据驱动的设备状态个性化评估和故障预测模型，为设备状态的精细评价和预测提供全新的解决思路和技术手段。主要研究内容包括：1）面向设备状态评估的电网、设备、环境等多源异构信息特征数据表达模型和规范化建模方法；2）研究输变电设备状态实时数据流多尺度异常检测的方法；3）研究输变电设备个性规律集成学习的原理和实现策略，建立基于大数据样本的设备关键性能动态评价模型；4）研究适用于输变电设备状态演变分析的多维度关联关系识别方法，构建数据驱动的设备状态关联分析和故障预测模型

本项目提出了一系列适用于设备状态评估的电力运行检修大数据挖掘分析方法，建立了基于数据驱动的设备状态异常检测、关键性能评估、故障诊断和预测模型，为设备状态的精细评价和预测提供全新的解决思路和技术手段，有效提高了电力设备状态评估的准确性和效率。主要进展和成果包括：1）提出了基于关联规则和主成分分析的输变电设备关键参数体系构建方法；2）建立了基于滑动时间窗时空联合聚类分析、多元时间序列分析以及转移概率序列无监督学习的输变电设备状态异常检测模型；3）提出了基于高维随机矩阵分析实现输变电设备关键性能评估的方法和算法模型；4）研究了适用于输变电设备状态演变分析的复杂多层多维关联关系识别方法，提出了基于概率图的关联关系高效挖掘方法，构建了基于大数据样本深度学习的设备故障诊断和预测模型。