一种基于ICA-FNN的多模型高压网络保护设备异常状态风险预警方法

An ICA-FNN-based multi-model early warning approach for the abnormal state risks in high-voltage network protection devices

继电保护自动设备是确保高压网络安全稳定运行的主要防线之一。但基于当前主网拓扑结构复杂、线路架构繁多、分布跨度较大的应用场景环境下,保护设备的潜在运行异常甚至故障难以完全避免;同时,保护设备种类、功能、分布的多样化也对设备的缺陷管理与检修反措提出了挑战。故亟待研究兼顾时效性与全面性的设备异常状态风险自动预警技术。为此,针对继电保护自动设备,提出一种基于数据挖掘的异常状态风险实时检测模型。其中,首先采用独立成分分析方法,生成独立分量的线性组合以面向海量异构监测数据实施降噪,能够有效提升高维数据条件下的运算效率;其次,构建深度学习前馈神经网络,使用端到端的训练方法以实现时间序列的异常检测,能够有效缓解多类别时序条件下的运算复杂度。最后,以某地区主网保护系统设备异常数据作为仿真实例,实验结果验证了所设计模型的异常检测性能,能够助推设备安全风险的自动识别和及时应对。

Protection relay system is one of the main defense lines to ensure the stable operation of high-voltage networks.However,within the scenarios with the more complex network topology,the line architecture and the distribution,it is difficult to eliminate the potential operating anomalies or even failures.Also,the diversification of the protection equipment types,functions and locations poses new challenges to the defect management and equipment maintenance.Therefore,the automatic early warning technology of equipment abnormal state risk which considers both the timeliness and comprehensiveness should be studied.To this end,a real-time detection model of abnormal state risk based on data mining is proposed in this paper.Firstly,the independent component analysis is used for mass heterogeneous monitoring data to implement noise reduction.This can effectively improve the computational efficiency under high-dimensional data conditions.Secondly,the feed-forward neural network deep learning method which deploys the end-to-end training process to achieve time series anomaly detections is utilized.This can effectively alleviate the multi-category timing conditions of computational complexity.Finally,the protection system equipment in one area is exploited as empirical study,the results verify the abnormal detection performance of the designed model,which can promote the automatic identification and timely response of the protection relay system.