一种基于对数区间隔离森林的电力调度数据异常检测集成算法

An Anomaly Detection Ensemble Algorithm for Power Dispatching Data Based on Log-interval Isolation

准确的电力调度自动化系统异常检测对电力系统安全稳定运行有重要意义。该系统具有业务种类繁多、业务逻辑交互复杂等特点,带来了调度监测数据维度多、空间分布多样的特性;现有基于机器学习的离线数据异常检测方法,存在对局部异常等特殊异常检测精度与检测效率难以有效兼顾等问题。提出了一种基于对数区间隔离的电力调度数据异常检测方法。针对数据维度之间的分布差异特性,运用马氏距离度量方法,基于每个样本点到数据分布中心的马氏距离,设计了对数区间隔离策略,构建多个子树,并将其整合成对数区间隔离森林异常检测器,筛选出数据集中的异常样本,兼顾检测精度和检测效率。公开数据集和某省级电网调度中心业务数据集作为训练与测试样本,验证了所提方法在异常检测AUC值等综合性能上的先进性及其在实际系统应用中的可行性。

Accurate abnormal detection of the power dispatching automation system is of great significance to the safe and stable operation of the power system.The system has the characteristics of having various kinds of services and complex interaction of business logics,which results in the multiple dimensions and diverse spatial distribution of the dispatching monitoring data.The existing off-line data anomaly detection methods based on machine learning are unable to effectively balance the detection accuracy and detection efficiency of the local anomalies and other special anomalies.Therefore,an electric power dispatching data anomaly detection based on the logarithm interval isolation method is put forward.For the distribution differences between the data dimensions,the Mahalanobis distance measure method is used.On the basis of the Mahalanobis distance between each sample point and the center of the data distribution,a logarithmic interval isolation strategy is designed.According to the strategy,many tall trees are constructed and integrated into a logarithmic interval isolation forest anomaly detector,which selects a data set of abnormal samples with both detection accuracy and efficiency.With a service data set from the server and a provincial power grid dispatch center as the training and testing samples,the advanced performance of the proposed method in the AUC value in the anomaly detection and its feasibility in the actual system application are verified.