[1]黄海东,刘世通,吴美琼.基于BiTCN和QRF的DMA用水量区间预测[J].中国给水排水,2026,42(5):55-62.
　HUANGHaidong,LIUShitong,WUMeiqiong.Interval Prediction for Water Demand in DMA Based on BiTCN and QRF[J].China Water & Wastewater,2026,42(5):55-62.
点击复制

基于BiTCN和QRF的DMA用水量区间预测

中国给水排水[ISSN:1000-4062/CN:12-1073/TU] 卷: 第42卷 期数: 2026年第5期 页码: 55-62 栏目: 出版日期: 2026-03-01
Title:
Interval Prediction for Water Demand in DMA Based on BiTCN and QRF
作者:
黄海东1， 刘世通2， 吴美琼1
（1.广西水利电力职业技术学院 土木建筑工程学院，广西 南宁 530023；2.北部湾大学 建筑工程学院，广西 钦州 535011）
Author(s):
HUANG　Haidong1， LIU　Shitong2， WU　Meiqiong1
（1. School of Civil Engineering and Architecture， Guangxi Vocational College of Water Resources and Electric Power， Nanning 530023， China； 2. College of Civil Engineering and Architecture， Beibu Gulf University， Qinzhou 535011， China）
关键词:
用水量区间预测;  双向时间卷积网络;  分位数回归森林;  开普勒优化算法;  独立计量区
Keywords:
interval prediction for water demand;  bidirectional time convolutional network;  quantile regression forests;  Kepler optimization algorithm;  district metered area
摘要:
独立计量区（DMA）用水量区间预测有助于供水管网管理者做出合理的漏损诊断。然而，如何保证较高预测区间覆盖率的同时具有较窄的平均宽度，是用水量区间预测需要解决的一个技术难点。为此，提出了一种融合双向时间卷积网络（BiTCN）和分位数回归森林（QRF）的区间预测模型。首先，基于用水量数据的特点，通过数据重构生成合适的样本集；然后，利用BiTCN模型提取用水量数据深层次的时序特征；接着，将所提取的特征输入QRF模型，初步构建BiTCN-QRF模型；最后，利用开普勒优化算法（KOA）优化BiTCN-QRF模型，并将优化后的模型用于区间预测。以广西都安县某DMA为例，对所提模型的有效性进行验证。结果表明，在95%、90%、85%、80%和70%置信水平下，所提模型均能实现高质量的区间预测，并且整体性能优于其他参与对比的模型。因此，该模型可作为基于流量的DMA实时漏损诊断的有效辅助工具。
Abstract:
Interval prediction for water demand in district metered area (DMA) is conducive to making appropriate leakage diagnosis for managers of water supply networks. However, ensuring higher coverage probability of the prediction interval while maintaining a lower average width remains a technical challenge in water demand interval prediction. To this end, an interval prediction model combining the bidirectional time convolutional network (BiTCN) with quantile regression forests (QRF) was proposed. First, based on the characteristics of water demand data, a suitable sample set was generated through data reconstruction. Then, the BiTCN model was used to extract deep-level temporal features of the data. Next, the extracted features were input into the QRF model to preliminarily construct the BiTCN-QRF model. Finally, the model was optimized using the Kepler optimization algorithm (KOA) and applied for interval prediction. The effectiveness of the proposed model was validated using data from a DMA located in Du’an County, Guangxi. The results showed that the proposed model could achieve high-quality interval prediction at 95%, 90%, 85%, 80%, and 70% confidence levels. Moreover, its overall performance was better than that of other comparative models. Based on these findings, the proposed model can serve as an effective auxiliary tool for flow-based real-time leakage diagnosis in DMA.
更新日期/Last Update: 2026-03-01