基于STARMA模型的城市暴雨积水点积水短时预测

近年来城市暴雨出现突发和多发态势,导致城市内涝灾害频繁发生,威胁着城市居民的生命和财产安全。随着城市降雨积水监测网的建立,获得分钟尺度的降雨和积水时序监测数据成为可能,实现了城市内涝的实时监控。但目前对监测数据的利用仍显不足,缺乏对其深度分析挖掘,造成监测系统“只监不控”的局面。本文基于城市降雨积水监测网的监测数据,根据积水时间相关性、降雨空间相关性以及降雨积水序列相关性,构建降雨积水的时空自相关移动平均模型(STARMA),对城市暴雨积水点积水过程进行短时预测。STARMA模型已被广泛应用于交通预测、环境变量预测以及社会经济领域,特别是在时空过程机理不清楚、多因素时空变量影响的情况下效果较好。本文首次将该模型应用到降水积水过程拟合和积水短时预测上,同时在方法上改进了传统单变量的STARMA模型,建立降雨和积水双变量的STARMA模型模拟降雨积水过程。并以北京市2012年“7.21”事件降雨积水过程为研究对象,以丰北桥、花乡桥、马家楼桥和六里桥4个积水监测点为例,建立降雨积水的STARMA模型,以5 min为步长作积水5、10、15 min三步预测。验证结果表明,该模型在降雨积水过程中拟合效果较好,模型短时预测精度较高。该项研究能够有效地利用监测数据,提高信息预警和应急指挥能力,为市政防汛或交通等部门提供决策支持。

Short-term forecasting of waterlogging at urban storm-waterlogging monitoring sites based on STARMA model

Storm struck cities frequently and often suddenly in recent years, leading to urban inundation and threatening life and property in these cities. With the establishment of urban storm-waterlogging monitoring network, real-time and time series rainfall and waterlogging data at the temporal resolution of a few minutes can be easily acquired. Real-time monitoring of inundation thus can be achieved and this provides new ways for research of inundation in cities. At present, however, there is a general lack of in-depth data mining and analysis of the observed data, which leads to the fact that urban inundation monitoring systems are used only for monitoring purposes and waterlogging control is not an integrated part of the system. Based on the monitoring data of urban inundation monitoring system, according to the temporal autocorrelation of waterlogging, spatial correlation of rainfall and the correlation of rainfall and waterlogging, a spatial and temporal auto regressive and moving average model (STARMA) has been built for short-term forecasting of waterlogging in this study. Auto regressive and moving average model (ARMA) is one of the correlation analysis methods of time series data. ARMA combined with spatial analysis methods generates the STARMA model. STARMA model is an effective means for modeling the spatiotemporal processes of geographic phenomena, especially when the mechanism of the spatiotemporal processes is unclear or multiple spatial and temporal variables are involved. STARMA model has been applied in traffic prediction, environment variable prediction and the social and economic fields. In this study, the model is applied in rainfall and waterlogging process simulation and short-term forecasting for the first time. In order to simulate rainfall and waterlogging processes, the traditional STARMA univariate model is modified to create a bivariate model of rainfall and waterlogging. Based on urban inundation monitoring data on 12 July 2012 in Beijing, using Fengbei Bridge, Huaxiang Bridge, Majialou Bridge and Liuli bridge as examples, the STARMA models were built respectively to predict water depth with a 5 minute step at 5, 10, and 15 minutes. The modeling process included creating rainfall stations' spatial weight matrix, model identification, parameter estimation and model verification. The STARMA model form was determined by autocorrelation function and partial autocorrelation, in addition to A-information criterion or Bayesian information criterion. Spatial weight values were calculated by inverse distance weighting (IDW). Parameter estimates were derived by the least square method. The simulation results show that the STARMA model predictions fit well with observed data and accuracy of short-term forecasting is high. The root mean square error (RMSE) is about 0.03, the relative square error (RSE) is about 0.01 and the average error rate is about 5%. When the prediction time increased from 5 to 15 minutes, prediction accuracy slightly decreased. This method improves prediction accuracy and reliability as compared to traditional hydro model simulation and prediction. The research uses urban inundation monitoring network data to predict short-term waterlogging. On the one hand it takes full advantage of the monitoring data, and on the other hand it improves the ability of disaster early warning and emergency command, thus provide decision support for related government departments.