海绵城市建设进入系统化全域推进阶段，结合观测数据和汇水区特征因地制宜进行低影响开发设施布局是重要的研究课题。本研究以泸州市中心城区沿江雨水汇水区为案例，探讨了基于汇水区空间特征和实测数据清洗的低影响开发（LID）设施布设多目标优化问题。研究收集了降雨和排口流量监测数据，识别典型监测异常类型，制定城市雨水管网排口异常流量数据自动检测方法清洗数据。构建EA-LSTM模型识别影响径流量的关键空间特征，以此为依据提出一种基于空间约束条件的LID设施空间配置综合策略，并采用NSGA-II多目标优化算法，确定了各汇水区内最优的LID设施布设方式。本研究的主要结论如下：（1）将城市雨水汇水区雨水管网末端排口流量监测数据的异常类型分为噪声、随机尖峰和响应模式异常三类，制定城市雨水管网排口异常流量数据自动检测方法。经过异常值检测后，10个汇水区的LSTM模型的R2值由检测前的0.1 ~ 0.5提升至0.5 ~ 0.9，NSE值从检测前仅有1个汇水区大于0提升至9个汇水区的NSE值大于0.4。（2）选取15个景观格局指数表征汇水区空间特征，结合汇水区不透水率、平均坡度和区域面积构建基于汇水区空间特征的EA-LSTM模型。通过Sobol全局灵敏性分析方法和Pearson相关系数识别出影响径流量的三个关键汇水区空间特征，分别为分维数、平均形状指数、欧几里得最邻近距离。其中分维数和平均形状指数与径流量呈正相关，欧几里得最邻近距离与径流量呈负相关。（3）融合多源数据将不透水层划分为建筑、道路和以广场为主的其他不透水面，为LID设施布设设定空间约束条件。基于影响径流量的关键汇水区空间特征，提出LID设施空间配置综合策略，该策略与随机布设方法相比在径流控制效果上具有较大优势。（4）采用NSGA-II多目标优化算法，实现了径流控制能力、投资成本和经济效益的最优平衡。优化后各汇水区的径流控制能力提升幅度在0.84 ~ 1.67之间，15年回收期内LID设施的总投资成本在0.40 ~ 1.66千万元之间，年经济效益在0.90 ~ 29.52万元之间。
The construction of sponge city has entered a stage of systematic and comprehensive advancement. It is an important research topic to plan the layout of low-impact development facilities according to locol conditions, combining observational data and the characteristics of the catchment areas. This study takes the central urban area along the river in Luzhou City as a case study to explore the multi-objective optimization problem of LID (Low Impact Development) facility layout based on the spatial characteristics and the cleaning of measured data of rainwater catchment areas. The research initially collects rainfall runoff monitoring data, identifies typical monitoring anomalies, and establishes an automatic detection method for abnormal flow data at urban rainwater pipe outlets to clean the data. An EA-LSTM model is constructed to identify key spatial features affecting runoff, and a comprehensive strategy for LID facility spatial configuration based on spatial constraints is proposed. The NSGA-II multi-objective optimization algorithm is employed to determine the optimal LID facility layout within each rainwater catchment area. The main conclusions of this study are as follows:(1) The abnormal types of flow monitoring data of the urban stormwater management outlets are categorized into three types: noise, random spikes, and response pattern anomalies, and an automatic detection method for abnormal flow data is established. After anomaly detection, the R2 values of the LSTM models for the 10 rainwater catchment areas increased from 0.1 to 0.5 before detection to 0.5 to 0.9, and the NSE values increased from only one area being greater than 0 before detection to nine areas having NSE values greater than 0.4.(2) Fifteen landscape pattern indices are selected to characterize the spatial features of the rainwater catchment areas, and combined with the impervious rate, average slope, and regional area to construct an EA-LSTM model based on the spatial characteristics of the rainwater catchment areas. Through Sobol global sensitivity analysis and Pearson correlation coefficient, three key spatial features of the rainwater catchment areas affecting runoff are identified, namely fractal dimension, mean shape index, and Euclidean nearest neighbor distance. The fractal dimension and mean shape index were positively correlated with runoff, while the Euclidean nearest neighbor distance was negatively correlated.(3) Multi-source data are integrated to divide the impervious layer into buildings, roads, and other impervious surfaces mainly consisting of squares, setting spatial constraints for LID facility layout. A comprehensive strategy for LID facility spatial configuration based on key rainwater catchment area spatial features is proposed, which shows significant advantages in runoff control effects compared to random layout methods.(4) The NSGA-II multi-objective optimization algorithm is used to achieve the optimal balance of runoff control capacity, investment cost, and economic benefits. After optimization, the increase in runoff control capacity in each rainwater catchment area ranged from 0.84 to 1.67, the total investment cost of LID facilities within a 15-year recovery period ranged from 4.0 to 16.6 million yuan, and the annual rainfall-related economic benefits ranged from 0.10 to 2.95 million yuan.
