基于SMOTE-ENN-GWO-SVM模型的低成本直饮水管道漏损预警

Low-cost early warning for direct drinking water pipeline leakage based on SMOTE-ENN-GWO-SVM model

摘要

【目的】直饮水管道漏损是我国供水行业长期面临的难题，不仅影响居民正常用水，还造成大量饮用水资源浪费，给供水企业带来显著经济损失。当前漏损检测技术存在设备成本高、智能化水平不足以及数据类别不平衡等挑战。【方法】针对这些问题，提出一种基于恒压供水原理的低成本管道漏损预警模型。基于恒压供水原理，利用供水变频器的频率、电压、电流、转速等运行参数的动态变化，结合智能算法与机器学习技术构建SMOTE-ENN-GWO-SVM漏损预警模型，采用SMOTE算法对样本数据进行过采样以平衡类别分布，利用ENN算法清洗噪声样本，最后通过灰狼优化算法调整SVM的关键超参数，以提升模型性能。【结果】结果表明：SMOTE-ENN-GWO-SVM模型准确率为98.16%,F1为0.952 3,皆优于对比模型。【结论】该方法显著提高了漏损识别的准确性和鲁棒性。其低成本、高精度的特点可满足直饮水系统在实际应用中对漏损检测的敏捷响应与可靠性要求，为提升城市水资源管理效率提供了技术支撑。

Abstract

[Objective] Leakage in direct drinking water pipelines has long been a persistent challenge in China's water supply industry. It not only affects residents' normal water usage but also leads to substantial drinking water waste and brings significant economic losses to water supply enterprises. Current leakage detection technologies face challenges such as high equipment costs, insufficient intelligence levels, and imbalanced data categories. [Methods] To address these issues, a low-cost pipeline leakage early warning model based on the principle of constant pressure water supply was proposed. Building upon the principle of constant pressure water supply, the dynamic changes in operational parameters of water supply frequency converter, such as frequency, voltage, current, and rotational speed, were used to establish the SMOTE-ENN-GWO-SVM leakage early warning model by integrating intelligent algorithms and machine learning technologies. The SMOTE algorithm was used to perform oversampling on the sample data to balance the class distribution, the Edited Nearest Neighbors(ENN) algorithm was applied to clean noisy samples, and the grey wolf optimizer was used to adjust the key hyperparameters of the support vector machine(SVM) to enhance model performance. [Results] The result showed that the SMOTE-ENN-GWO-SVM model achieved an accuracy of 98.16% and an F1 score of 0.952 3, both outperforming the comparative models. [Conclusion] This method significantly improves the accuracy and robustness of leakage detection. Its characteristics of low cost and high precision meet the practical application requirements for responsive and reliable leakage detection in direct drinking water systems, providing technical support for improving the efficiency of urban water resource management.

关键词

直饮水供水管道 / 支持向量机 / 灰狼优化算法 / 不平衡数据集 / 恒压供水 / 供水变频器 / 漏损预警 / 影响因素

Key words

direct drinking water pipeline / support vector machine / grey wolf optimizer / imbalanced dataset / constant pressure water supply / water supply frequency converter / leakage warning / influencing factors

基金资助

国家重点研发计划(2016YFB0502305)

省级科技成果转化引导重点项目(CGZH2018042)