独立多灾种年度财产损失综合风险的信息扩散合成范式研究

The simple sum of risks is not equal to the integrated risk caused by multi-hazard. From the weighted integration to Copula model, from classification matrix to fuzzy information granulation, it is difficult to form a universal method to assess integrated risks. In this project we consider an integrated risk as the result of combining effects of multivariate random variables, and that the combination mechanism is too complex to construct the theory distribution of the variables. Due to that risk analysis is usually done with incomplete information, i.e., involving both of random uncertainty and fuzzy uncertainty, in this project we will develop the information diffusion theory and methods to deal specifically with such issues. By using the discrete probability distributions estimated by the information diffusion techniques, we would solve the problem of compositing independent multiple hazards in terms of risk, and have a universal method to assess integrated risks. To increase the information available on geographical units, we will study the diffusion of information technology for incomplete spatial information. To increase the information available about hazard-affected bodies, we will use the Internet of intelligences to collect risk information perceived by the people living in the studied region, and to deal with the information by the diffusion method on a factor space. The assessed integrated risk will have period information. Such risk is easier to be changed into an annual risk. Then, we can establish an information diffusion paradigm to assess integrated risk of annual property loss. It would provide a scientific and technological support for government and enterprises to manage integrated risks of natural disasters. To demonstrate the paradigm, we will employ it to assess the integrated risk of annual property loss caused by earthquake and flood in the townships of the Fujiang river basin, Santai County, Sichuan Province.

单灾种风险简单相加，不等于多灾种综合风险。从加权综合到Copula模型，从等级矩阵到模糊信息粒化，均难形成普适性综合风险评估方法。综合风险是多随机变量综合作用的产物，形成机理十分复杂，很难构造有价值的理论分布。风险分析通常是在信息不完备条件下进行，涉及随机和模糊不确定性，本项目将发展处理这类问题的信息扩散理论和方法，从随机变量离散化概率分布角度，解决独立多灾种叠加问题，建立普适性综合风险评估方法。为弥补地理单元上的缺失信息，将研究空间不完备信息的扩散技术；为增加承灾体的可用信息，将用智联网获取研究区民众在风险感受方面的信息，并在因素空间中进行扩散处理。评估出有时段信息能年化处理的综合风险，从而建立年度财产损失综合风险的信息扩散合成范式，为政府和企业进行自然灾害综合风险管理，提供科技支撑。本项目将评估四川省三台县涪江流域中各乡镇的地震和洪涝年度财产损失综合风险，以检验相关的理论和方法。

人们常用加权综合法或Copula模型等方法，试图加入主观经验，避开多灾种综合风险评估时的小样本问题。但是，评估结果的可靠性没有保障，可用性较低。本项目发展了能优化处理小样本的信息扩散方法，用于估计多灾种联合概率分布和风险承受体关于多灾种的脆弱性，提供了客观评估多灾种综合风险的多维信息扩散理论和方法。为弥补地理单元上的缺失信息，本项目建立了以背景数据为媒介，通过自学习离散回归方式，将一些地理单元上获得数据信息，扩散到空白单元上去的地理空间信息扩散模型，解决了数据不连续数学插值法失效等问题，精度高于BP-人工神经网络和地理加权回归模型。通过直观地展示信息扩散技术可消除台风统计资料中出现的平缓和跳空现象，以及计算机仿真实验，证明了本项目建立的模型可靠性较高。用容量为20的单灾种样本进行风险估计时，信息扩散模型相比直方图模型，能减少大约12%的误差。为在线获取研究区民众的风险感受信息，得到微观灾情数据，本项目研制出了将地图制作成网页前台脚本JavaScript文件嵌入智联网的技术，解决了由应用程序接口(API)接入商业化在线地图，所产生的不可维护和不安全性等问题。将地理空间信息扩散模型嵌入，研制出对灾情具有“采点外推”功能的智联网，解决了根据历史数据统计公式“隔空判灾”的精度低，且不可在线更新的问题。32人次历时三年，在四川省三台县对历史和项目期间的地震和洪涝财产损失情况的调研，获得了必要的数据，进行了地震和洪涝年度财产损失综合风险评估。研究结果表明，在智联网的支持下，用信息扩散模型，可进行乡镇级地理单元上的多灾种综合风险评估。本项目还探讨了用形态发生算法更新自然灾害风险，研究了自然灾害对中国整体实际GDP的影响，尝试让智联网在重大自然灾害精准救助中发挥作用，以及在信息不完备条件下估算新型冠状病毒病(COVID-19)死亡风险。