【摘要】 生活小区污水排放量数据是采用模型法进行污水管网检漏的重要基础条件，但现有的直接监测法、定额法和给水量法存在监测量大、精度低和数据难获取等问题。考虑到生活小区的建筑特征信息（如建筑面积、住户数量等）与排水行为密切相关，通过收集上述易获取的建筑特征信息，结合极端梯度提升（XGBoost）机器学习算法的优异非线性拟合能力，具有实现生活小区污水量高精度预测的潜力。为此，对多个典型生活小区的污水排放量进行监测，基于小区建筑特征信息并结合XGBoost算法，建立生活小区污水量预测模型。结果表明，预测模型的主要参数为住户数、建筑面积、占地面积、竣工时间和排水时间，经预处理和超参数优化后，模型在一天中08:00—24:00的污水量预测误差较小，测试集的平均相对误差为11%，优于定额法和给水量法的42%和17%。更多还原

【Abstract】 The volume of domestic sewage discharge in residential communities serves as a fundamental input for leakage detection in sewer networks using model?based approaches. However, conventional methods such as direct monitoring, quota?based estimation, and water consumption?based estimation suffer from limitations including extensive monitoring requirements, low accuracy, and poor data accessibility. Given the close correlation between sewage discharge behavior and building characteristics(such as total floor area and number of households),this study explored the potential of using easily accessible building characteristic information, combined with the strong nonlinear fitting capabilities of the XGBoost machine learning algorithm, to achieve high?accuracy predictions of sewage discharge volumes. To this end, sewage discharge data were collected from multiple representative residential communities. A predictive model for community?scale sewage discharge was developed by integrating building characteristic variables with the XGBoost algorithm. The results indicated that key predictive characteristics included the number of households, total floor area, land area, completion year, and typical drainage hours. After data preprocessing and hyperparameter optimization, the model achieved high prediction accuracy during 08:00 and 24:00, with an average relative error of 11% on the test set, which outperformed the quota?based(42%) and water consumption?based(17%) methods.更多还原