基于数据分析的给水管网爆管辨识与定位研究

Water loss is a great concern of water utilities. Smart pipe burst warning as a pivotal task of water loss control will became the foundation of Smart Water Utility in water distribution networks (WDNs). The intelligent burst decision support based on the real-time pressure data analytics is prone to be affected by the uncertainties, and also the theoretical basis of the dynamic partition of networks is unexplored, which adds more challenges to the pipe burst detection and pinpoint problems. Therefore, the project is planned to set up the pressure sensor placement upgrade model based on burst detection. The linearization theory of the nonlinear model in water distribution networks is investigated to optimize the sensor placement model. Moreover, the time series pressure data are forecasted by the multi-variable Kalman Filter. The results of deviation, boundary, and trend analysis and times are coupled to implement a four dimensions Bayesian Network framework. The probability of the burst event is computed based on the data analysis and synthesis. Finally, the targeted zone of the burst is dynamically identified by the hydraulics mechanism that the monitoring pressure reflects the burst flow. The probability of the burst for each pipe is computed in the target zone using the spatial-geostatistics method. This study will give some insights into the machine learning theory and automatic partition theory in WDNs, which can provide the scientific evidence for pipe burst detection and location in WDNs and the theoretical basis for resolving the water loss problem.

给水管网漏损是供水企业极为关注的问题，智能爆管预警作为管网漏损控制的重要工作，未来将成为智慧水务建设的基石。基于实时压力数据分析的智能化爆管决策支持易受到数据不确定性的影响，且给水管网动态分区理论基础尝未明析，均给爆管的辨识与定位问题增添了更多挑战。由此，本研究拟建立面向爆管检测的压力监测点优化升级模型，探索给水管网理论计算中非线性化模型的线性化求解理论；然后，基于多变量卡尔曼滤波预测压力时间序列数据，构建偏差、边界和趋势结果与时间的四维度贝叶斯网络框架，基于此数据解析与综合方法辨识爆管事故的发生概率；最后，通过探讨压力监测点对爆管流量的水力响应机制，动态划分爆管目标区域，在管网爆管虚拟分区内逐条定位分析管道爆管概率。研究成果将拓展机器学习理论和管网自动分区理论，也为给水管网爆管智能预警提供充分的科学依据，为解决管网漏损问题奠定理论基础。

供水管网漏损是水务公司重点关注的问题，智能爆管预警作为管网漏损控制的重要工作，未来将成为智慧水务建设的基石。本项目围绕智能爆管预警中的难题，按照爆管“监测—辨识—定位”的研究思路，首先构建基于爆管检测为目标的既有压力监测点优化升级模型，建立时间序列监测数据的多维度解析框架，辨识管网爆管事故的发生,最后识别爆管分区，计算各管道的爆管发生概率。创新成果包括：1）基于成本-效益分析理论，建立了面向爆管检测的两阶段压力监测点优化布设方案决策框架，量化了传感器精度、漏点数量和爆管规模等工程参数对监测点布设方案的影响，解决了最优监测点数量确定的问题；2）从监测数据序列的时序相关性维度，建立了基于自适应卡尔曼滤波的流量预测-异常诊断方法，从用水序列变化模式的维度，基于聚类分析算法，解析了正常用水模式特征，构建了模式库，进一步地建立了实时流量序列重构-动态检测阈值的爆管诊断方法，解决了爆管事故快速诊断问题；3）挖掘了监测点压力对爆管流量变化的水力反馈机制，建立了爆管影响的管网动态虚拟分区方法，进一步地利用地理空间统计方法，确定了分区内的各管道的爆管发生概率值，解决了爆管定位不准的问题。研究成果为供水管网爆管智能预警和管网漏损问题提供了充分的科学依据，已经在深圳供水管网和大连供水管网进行了初步应用。在本项目的资助下，共发表论文9篇，其中在《Water Research》、《Journal of Water Resources Planning and Management》、《Urban Water Journal》和《Journal of Hydroinformatics》等期刊上发表SCI论文6篇，申请发明专利3项。