| 贾翔宇,刘涛,翟勇龙.基于群体用户画像的燃气用气异常检测方法研究[J].唐山学院学报,2025,38(6):19-22 |
| 基于群体用户画像的燃气用气异常检测方法研究 |
| Research on Gas Consumption Anomaly Detection Method Based on Group User Profiling |
| 投稿时间：2025-02-26 |
| DOI：10.16160/j.cnki.tsxyxb.2025.06.004 |
| 中文关键词: 群体用户画像 主成分分析 聚类 随机森林 异常阈值分析 |
| 英文关键词: group user profiling principal component analysis clustering random forest anomaly threshold analysis |
| 基金项目:河北省教育厅科学研究项目(ZC2024187);河北省高校人工智能赋能教改专项课题(2025RGZN071);河北省高校创新创业教育教学改革研究与实践项目(2023CXCY290) |
|
| 摘要点击次数: 425 |
| 全文下载次数: 227 |
| 中文摘要: |
| 针对燃气用户异常检测中个体画像方法泛化能力不足的问题,文章提出一种基于群体用户画像的燃气用气异常检测模型。首先,基于燃气计量时序数据,构建包含统计特征、分布特征和行为特征的多维特征体系;其次,通过KMO检验与Bartlett球形检验验证特征相关性后,采用主成分分析实现特征空间降维;最后,融合K-Means聚类与随机森林算法构建基于群体用户画像的燃气用气异常检测模型,并设计具有时序自适应特性的动态阈值决策机制。实证分析结果表明,经5折交叉验证后模型在存量用户检测中的准确率达86.67%,新用户误判率控制在9.8%以内,鲁棒性较传统个体检测方法有较大提升。 |
| 英文摘要: |
| To address the insufficient generalization capability of individual profiling methods in gas user anomaly detection,this paper proposes a gas consumption anomaly detection model based on group user profiling. Firstly, a multi-dimensional feature system comprising statistical, distributional and behavioral features is constructed from gas metering time-series data. Then, after validating feature correlations through KMO test and Bartlett spherical examinations, principal component analysis is applied for feature space dimensionality reduction. Finally, a gas consumption anomaly detection model based on group user portraits is constructed by K-Means integrating clustering and random forest algorithms, with a dynamic threshold decision mechanism possessing time-series adaptive characteristics. Empirical analysis shows that after five-fold cross-validation, the model has achieved an accuracy rate of 86.67% for existing users, with a misjudgment rate for new users kept below 9.8%. Compared with traditional individual detection methods, robustness has been significantly improved. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|