The dynamic relief supplies estimation for urban flood disaster still remains an important and challenging topic in emergency response. Traditional relief supplies estimation mainly depends on static census data rather than dynamic spatio-temporal population distribution, which may easily lead to a serious imbalance between supply and demand for relief resources. The emergence of big data originating from web mapping service, social media, crowdsourcing system and other methods provides alternative data sources to understand the dynamic distribution of urban population, which can help accurately estimating the relief supplies demand when urban flood occurs. This paper proposes a big data-driven dynamic estimation model of relief supplies demand, which merges the dynamic population distribution from Baidu big data and Multilayer Perceptron (MLP) neural network so as to improve the accuracy and timeliness. Taking Wuhan as an example, a specific day in summer flooding period was chosen to dynamically predict three kinds of relief materials. The results indicate that the proposed model is more feasible to estimate the relief supplies because it is integrated with MLP neural network trained by national-wide historical flood disaster cases and Baidu big data. The utility of Baidu big data for flood-affected population statistics has been shown to be more comprehensive than that of the traditional demographic methods. Thorough analysis exhibits the superiority of our proposed model with respect to accuracy and dynamics. This allows our proposed model to be more widely applied in pre- and post-disaster operations activities carried out by government and humanitarian aid organizations.

对城市洪水灾害的动态救灾物资估算仍然是应急响应中一个重要且具有挑战性的话题。传统的救济物资估算主要依靠静态人口普查数据，而不是动态的时空人口分布，这很容易导致救济物资供需之间的严重失衡。来自网络制图服务，社交媒体，众包系统和其他方法的大数据的出现为了解城市人口的动态分布提供了可替代的数据源，可以帮助准确估计城市洪水发生时的救济物资需求。本文提出了一个大数据驱动的救济物资需求动态估计模型，该算法将百度大数据的动态人口分布与多层感知器（MLP）神经网络相融合，以提高准确性和及时性。以武汉为例，选择夏季汛期的某一天来动态预测三种救济物资。结果表明，该模型与受全国历史洪涝灾害案例和百度大数据训练的MLP神经网络相集成，是估计feasible灾物资更可行的模型。事实证明，百度大数据在受洪灾人口统计中的应用比传统的人口统计方法更为全面。全面的分析显示了我们提出的模型在准确性和动力学方面的优越性。