Urban Water Demand Regression Modeling for California Water Suppliers

Urban water demand modeling with regression identifies explanatory factors of water use in cities. A generalized demand modeling approach was developed for over 400 urban water supply agencies in California. Using standardized data from self-reported sources for agencies across the state, a batch-processing approach was used to create standardized urban water demand models. The models were developed to test the validity of a simplified and generalized demand modeling approach using monthly available data. Semilog, multivariate regression models were developed for each urban water supply agency. Consumption from residential (single- and multi-family), commercial, industrial, and institutional water use were considered as outcome variables. Explanatory variables include indicator variables for months in a calendar year, periods of water conservation requirements during a 2011-16 severe drought, population, and water rates. The models were of reasonable fit, with adjusted R-squared values ranging from 0.6-0.99. Visual inspection revealed that the monthly models captured trends with reasonable accuracy. The time frame for models was 2013-18, a period with standardized available data through statewide reporting. The modeling approach has been subsequently further extended to incorporate additional climate variables (precipitation and evapotranspiration) for sector-specific models. The models are intended to understand explanatory factors of demand through a generalized modeling approach and not intended to be used for water supply operations without further refinement and testing. The approach can be adapted to many types of cities.基于回归分析的城市需水量建模，可识别城市用水的解释性影响因子。研究团队为加利福尼亚州超400个城市供水机构开发了一套广义需水量建模方法：借助全州范围内各机构自主上报的标准化数据，采用批处理流程构建标准化城市需水量模型，旨在基于可获取的月度数据验证简化广义需水量建模方法的有效性。针对每个城市供水机构，均构建了半对数多元回归模型（semilog multivariate regression models），将住宅（单户与多户住宅）、商业、工业及公共机构的用水量作为因变量。解释变量涵盖：日历年内各月份的指示变量、2011至2016年严重干旱期间的节水管制时段、人口规模以及水价。模型拟合效果良好，调整后决定系数（adjusted R-squared）取值范围为0.6至0.99；目视检验结果表明，月度模型可较为精准地捕捉用水趋势。本模型的时间跨度为2013至2018年，该时期内全州统一上报的数据均已完成标准化处理。后续该建模方法已进一步拓展，可为分部门模型纳入额外气候变量（降水与蒸散量）。本模型仅旨在通过广义建模方法解析需水量的影响因子，未经进一步优化与测试，不得用于供水运营决策。该建模方法可适配多种类型的城市。