机械与电子2026，Vol.44Issue(5)：120-126,7.
基于集成隔离森林与业务感知动态选择的电力调度数据异常检测方法
Power Dispatch Data Anomaly Detection Based on Integrated Isolation Forest and Business-aware Dynamic Detector Selection
摘要
Abstract
To address the challenges in the power dispatch automation systems,such as diverse busi-ness types,multi-dimensional monitoring,and complex anomaly patterns,this paper proposes an anomaly detection method based on business-aware hybrid selection integration.Building upon the existing Loga-rithmic Interval Isolation Forest(LIIF)framework,the method incorporates Mahalanobis distance metrics and dimensional correlation analysis to enhance local anomaly identification within multi-dimensional fea-ture spaces.A business-aware hybrid selection ensemble strategy is introduced:first,base detectors with performance biases are pre-filtered through an isolation forest-based static screening to improve true/false value reliability.Subsequently,a dynamic ensemble selection is performed for each test sample based on the consistency,business relevance,and measurement point correlation indicators,thereby balancing de-tection accuracy with computational efficiency.Experimental results on public datasets and business data-sets from a provincial power grid dispatch center demonstrate that the proposed method outperforms exist-ing approaches in comprehensive metrics such as AUC.It effectively identifies typical power dispatch a-nomalies,including data jumps,application disconnections,and remote telemetry table failures.关键词
电力调度数据异常检测/局部异常/马氏距离/集成学习/多维度关联隔离Key words
power dispatch data anomaly detection/local anomalies/Mahalanobis distance/ensemble learning/multi-dimensional correlation isolation分类
信息技术与安全科学引用本文复制引用
吕勃翰,郭岩,李慧勇,孟沛彧,孙焜,高鑫..基于集成隔离森林与业务感知动态选择的电力调度数据异常检测方法[J].机械与电子,2026,44(5):120-126,7.基金项目
中国南方电网有限责任公司技改项目(000500SZ25010011) （000500SZ25010011）
