[Research Data] AI-Driven Assessment of Urban Water Meter Errors: A Sustainable Framework for Optimizing Resource Efficiency and AI-Driven Assessment of Urban Water Meter Errors: A Sustainable Framework for Optimizing Resource Efficiency and Reducing Carbon Footprint

Accurate water metering is critical for addressing non-revenue water (NRW) losses and advancing sustainable urban water management. This study proposes an AI-enabled framework to optimize meter replacement cycles, integrating advanced metering infrastructure data with semi-supervised learning to mitigate metrological errors and reduce lifecycle greenhouse gas (GHG) emissions. Through case studies of two Chinese cities (Y and W), we developed an Apparent Metering Error (AME) metric that quantifies water loss by synthesizing flow-dependent error curves and usage patterns. A neural network model, trained on 144 mechanical meters (MMs) and validated against 140,000 residential meters, demonstrated that aging MMs exhibit systematic under-registration (AME: −6% to −2%), driven by low-flow measurement failures and prolonged service durations. Results revealed that optimized replacement strategies could reduce GHG emissions and lower costs compared to conventional practices. The framework offers a scalable solution for global cities grappling with NRW and climate resilience challenges.

精准的水表计量对于解决无收益水（non-revenue water, NRW）损失、推进可持续城市水管理而言至关重要。本研究提出一种用于优化水表更换周期的人工智能赋能框架，将先进计量基础设施数据与半监督学习（semi-supervised learning）相结合，以缓解计量误差并降低全生命周期温室气体（greenhouse gas, GHG）排放。通过对中国两座城市（Y市与W市）开展案例研究，我们构建了表观计量误差（Apparent Metering Error, AME）指标，该指标通过综合流量依赖型误差曲线与用水模式来量化水损失。我们基于144台机械水表（mechanical meters, MMs）训练了神经网络模型，并以14万台住宅水表进行验证，结果显示老化机械水表存在系统性计量欠报现象（表观计量误差范围为-6%至-2%），其诱因包括小流量计量失效与服役时长过长。研究结果表明，相较于传统更换方案，优化后的水表更换策略可减少温室气体排放并降低成本。该框架可为全球面临无收益水与气候韧性挑战的城市提供可规模化推广的解决方案。