Abstract:
Understanding the impact of rainfall characteristics on urban stormwater quality is important for stormwater management. Even though significant attempts have been undertaken to study the relationship between rainfall and urban stormwater quality, the knowledge developed may be difficult to apply in commercial stormwater management models. A data mining framework was proposed to study the impacts of rainfall characteristics on stormwater quality. A rainfall type-based calibration approach was developed to improve water quality model performance. Specifically, the relationship between rainfall characteristics and stormwater quality was studied using principal component analysis and correlation analysis. Rainfall events were classified using a K-means clustering method based on the selected rainfall characteristics. A rainfall type-based (RTB) model was independently calibrated for each rainfall type to obtain optimal parameter sets of stormwater quality models. The results revealed that antecedent dry days, average rainfall intensity, and rainfall duration were the most critical rainfall characteristics affecting the event mean concentrations (EMCs) of total suspended solids, total nitrogen, and total phosphorus, while total rainfall was found to be of negligible importance. The K-means method effectively clustered the rainfall events into four types that could represent the rainfall characteristics in the study areas. The rainfall type-based calibration approach can considerably improve water quality model accuracy. Compared to the traditional continuous simulation model, the relative error of the RTB model was reduced by 11.4 % to 16.4 % over the calibration period. The calibrated stormwater quality parameters can be transferred to adjacent catchments with similar characteristics.
摘要:
了解降雨特征对城市雨水水质的影响对雨水管理具有重要意义。尽管已经进行了大量尝试来研究降雨与城市雨水质量之间的关系,开发的知识可能难以应用于商业雨水管理模型。提出了一种数据挖掘框架来研究降雨特征对雨水水质的影响。开发了一种基于降雨类型的校准方法,以提高水质模型的性能。具体来说,利用主成分分析和相关分析研究了降雨特征与雨水水质的关系。根据选定的降雨特征,使用K均值聚类方法对降雨事件进行分类。针对每种降雨类型独立校准了基于降雨类型(RTB)的模型,以获得雨水水质模型的最佳参数集。结果表明,之前的干旱天数,平均降雨强度,降雨持续时间是影响总悬浮固体事件平均浓度(EMC)的最关键降雨特征,总氮,和总磷,而总降雨量被发现的重要性微不足道。K-means方法有效地将降雨事件分为四种类型,可以代表研究区域的降雨特征。基于降雨类型的校准方法可以大大提高水质模型的准确性。与传统的连续仿真模型相比,在校准期间,RTB模型的相对误差降低了11.4%至16.4%。校准后的雨水水质参数可以传输到具有类似特征的相邻集水区。