我所殷辰堃老师课题组完成的论文《Outlier Detection of Traction Energy Consumption Based on Local Density and Cluster for Time Series Data》于2023-05在《IEEE Data Driven Control and Learning Systems Conference (DDCLS)》正式发表。论文摘要如下：
Abstract: Traction energy consumption accounts for 40%-50% of the total energy consumption of urban rail transit. Outlier detection of traction energy consumption data is the key technology of traction energy consumption fluctuation analysis. To accurately detect abnormal traction energy consumption, this paper first proposes a calculation method of the typical value of traction energy consumption indicator based on the combination of the ARIMA model and XGBoost algorithm. Its core idea is to extract residual item information based on the XGBoost algorithm, and then integrate all information for modeling; Then, based on the calculated typical values, this paper uses an incremental local density and cluster-based outlier factor (iLDCBOF) method to detect outliers. The experimental results show that the prediction effect of the ARIMA+XGBoost hybrid model is better than that of a single model. The proposed method can effectively detect abnormal energy consumption values in data streams.
引用信息：Chengxi Zhang, Jing Xun, Zhihui Ji, Chenkun Yin, Jiang Cao. Outlier Detection of Traction Energy Consumption Based on Local Density and Cluster for Time Series Data. IEEE Data Driven Control and Learning Systems Conference (DDCLS), 2023-05. DOI: 10.1109/ddcls58216.2023.10166787