洪涝灾害给城市安全发展带来新的挑战，“逢暴雨必涝”成为众多城市的真实写照，灾害治理急需从被动应急管理向常态化防控转变。应急物资是防范化解重大风险的“稳定器”和“压舱石”，也是打赢抢险救援战的有力保障。当前在着力解决“储什么、储多少、如何储”的问题上，鲜有以应急物资需求等级为基准预测需求量。因此，急需开展以洪涝灾害为切入点，以需判储探究各类应急物资等级及需求量的研究。

      本文以洪涝灾害应急物资为研究对象，根据洪涝灾害及物资需求特征，从种类、响应、数量三个维度开展应急物资需求分析。在种类维度，结合生命周期理论将应急管理全过程划分为“事前静态防-事中动态控-事后综合补”三个阶段，以此研判有效处置所需各类应急物资的时间及数量需求；将应急物资分为生活保障、临时住宿、公共卫生及抢险救援四大类，进一步细化为9小类27项明确物资。在响应维度，在应急物资影响因素分析的基础上，建立洪涝灾害应急物资需求等级指标体系；以主客观结合为主导思维，提出熵权-模糊层次分析（EM-FAHP）组合赋权确定指标权重；对各指标进行模糊化处理并建立隶属度函数，以评估各项应急物资需求等级。在数量维度，第一步基于灰狼算法（GWO）优化极限学习机（ELM）构建应急物资需求预测模型，确定受灾人口、季节系数等10项影响因素，收集40组历史数据作为样本集，应用主成分分析法对其进行降维预处理后，将其作为输入数据代入构建的GWO-ELM模型，以预测紧急转移安置人口及受灾面积，进一步计算集中安置人口数；第二步结合安全库存理论，依据预测结果与各类物资间的配比关系，确定不同等级下各类应急物资估算关系式，最终以蓝田“8.19”洪涝灾害为例，估算所需应急物资的数量，并提出相关管控建议措施。

       研究结果表明建立的应急物资需求指标体系中，各层级指标一致性程度检验均小于0.4，可进行EM-FAHP组合赋权；系统聚类分析结果与划分结果一致，可应用该指标体系评估应急物资需求等级；GWO-ELM模型精度为98.24%，其预测值与实际值的拟合效果显著优于BP神经网络，表明使用灰狼算法优化极限学习机可提高模型的预测精度。

       由此可得出：以应急物资需求等级为奠基估算各类物资需求量，并将具有特定功能的物资组进行模块化储备，可实现应急物资的高效预储。本文将在一定程度上加强了灾前、灾中、灾后阶段的衔接，助推应急物资从供需“失衡”到有效满足需求的转变，为应急物资配备提供有益借鉴。

        Flood disaster have brought new challenges to the development of urban safety, with "every rainstorm must be waterlogging" becoming a true reflection of many cities. Disaster management urgently needs to change from passive emergency management to normalized prevention and control. Emergency materials are the ' stabilizer ' and ' ballast ' to prevent and resolve major risks, and also a powerful guarantee to win the rescue war. At present, in the problem of "what to store, how much to store and how to store", there is little demand forecast based on the demand level of emergency materials. Therefore, there is urgent to carry out  research on the level and demand of various emergency materials based on flood disaster.

       This paper takes flood disasters emergency materials as the research object, and analyses the demand for emergency materials in three dimensions: category, response and quantity, according to the characteristics of flood disasters and the demand for materials. In the category dimension, the whole process of emergency management is divided into three stages: "static prevention beforehand - dynamic control during the event - comprehensive replenishment afterwards" by combining the life cycle theory, so as to determine the response and quantity demand of various types of emergency materials required for effective disposal; the emergency materials are divided into four categories: life support, temporary accommodation, public health and rescue. It is further refined into 9 small categories and 27 clear items of materials. In the response dimension, on the basis of analysis of the influencing factors of emergency materials, an index system for the demand level of emergency materials in flood disasters is established; with the combination of subjective and objective as the dominant thinking, a combination of EM-FAHP is proposed to assign weights to determine the index; each index is fuzzified and an affiliation function is established to assess the demand level of each emergency material. In the quantity dimension, the first step is to construct an emergency material demand forecasting model based on the Grey Wolf algorithm (GWO) optimized Extreme Learning Machine (ELM), determine 10 influencing factors such as the affected population and seasonal coefficient, collect 40 sets of historical data as a sample set, apply the principal component analysis method to pre-process them by dimensionality reduction, and then substitute them into the GWO-ELM model constructed as input data to predict the emergency transfer resettlement population and the damage area, and further calculate the number of centralized resettlement population. The second step is to combine the safety stock theory with the prediction results and the proportional relationship between various types of materials to determine the equation for the estimation of various types of emergency materials under different levels, and finally take the "8.19" flood disaster in Lantian as an example to estimate the quantity of emergency materials demand and propose relevant control measures.

         The results show that in the established emergency materials demand index system, the consistency degree test of index at each level is less than 0.4, and the EM-FAHP combination can be assigned; the results of the system clustering analysis are consistent with the division results. Thus the index system can be applied to assess the emergency materials demand level; the accuracy of the GWO-ELM model is 98.24%, and its predicted values fitted the actual values significantly better than the BP neural network. Using the GWO to optimise the ELM could improve the prediction accuracy. This leads to the conclusion that estimating the demand for each type of materials based on the level of demand for emergency materials, and modularizing the reserve of material groups with specific functions, efficient pre-storage of emergency materials can be achieved. This paper will, to a certain extent, strengthen the link between the pre-disaster, disaster and post-disaster phases, and help to transform the supply of emergency materials from an "imbalance" to effectively meet demand, providing a useful reference for the provision of emergency materials.