Study Of Data Imputation And Rapid Loss Assessment Of High Casualty Fires Of China

Fire plays an important role in the development of human being. It promotes the society development and human progress, but creates disasters as well. Since the reform and open, the economy of China develops rapidly, resulting in the rise of activity frequency of production and living, the accumulation of social wealth, and the increase of fire numbers and losses. Thereinto, the high casualty fires which cause huge number of deaths and injuries draw the public attention especially. The process of high casualty fires requires the cooperation of multiple departments, which is beyond the scope of general risk management. Focusing on the high casualty fires, the data imputation and rapid loss assessment are studied, in order to solve the essential problems of emergency management of high casualty fires and provide theoretical foundation and assessment system.To tackle the data missing problem of high casualty fires, a principle component regression based numerical data imputation method and a multiple correspondence analysis based categorical data imputation method were proposed. Experiments of different missing rates and accident levels were designed and root mean squared error and correct rate were used to evaluate these2methods, respectively. The former method was compared with mean, mode, k-means cluster, fuzzy c-means cluster, and k-nearest neighbor imputation methods, and the latter method was compared with mode, multinomial logistic regression, and radial basis function network imputation methods. The results show that the former method performs better under low missing rate and high accident level, the latter method performs better for the attributes of place and cause under different missing rates and accident levels, and these two proposed methods perform the best when the missing rate and accident level are both under consideration.To realize the rapid loss assessment for high casualty fires, a consequence index from the view of fire consequence and a situation index from the view of fire situation were constructed basing on the analytic network process. The fomer index includes the attributes of death, injury, burned area, and property loss, and the latter index includes18attributes of3clusters of time, place, and cause. Two indexes were calculated and the consequence index was used for loss assessment from the aspect of fire consequence, and the situation index was used for loss assessment from the aspect of fire situation basing on the relationship between the2indexes. The correct rates of classification for different accident levels were compared between the proposed method and multinomial logistic regression, multi-layer perceptron network and radial basis function network. It is indicated that the results of the proposed method has a higher correct rate and is more conservative. Basing on the consequence index and situation index, the rapid loss assessment is realized with easily acquired input and conservative acceptable output.Considering the aspects of exposure and resistance&recovery,38basic indexes of status of socioeconomic, demographic, and infrastructure&lifeline were collected. Data were standardized with z-score function and adjusted. Seven principle components were extracted by principle component analysis, and the social vulnerability index was synthesized by weighted addictive model. By analyzing the social vulnerability index, the time and space distributions of social vulnerability were revealed. The results show that the social vulnerability is high for the east and west, and low for the north, middle, and south by space, and the formation model of the high social vulnaerabilty of east and west are different:it is resulted from the unbalanced development for the east (e.g. Shandong, Anhui, Jiangsu, Henan, and Jiangxi) and the slow development for the west (e.g. Xizang, Sichuan, Yunnan, Guizhou, Gansu, and Qinghai); is decreasing and approaching each other by time, and the good linear trend can be used for forcasting to tackle the data collection problem and simplify the calculation process. The sensitivity analysis shows that the social vulnerability index is sensitive to the change in space scale, and stable to the change in time scale.The summation of consequence index and the standardization of social vulnerability index were calculated basing on the consequence index and social vulnerability index, and a simple area rapid loss assessment method was proposed. With the results of area loss index, the comprehensive loss status was assessed and graded for all provincial administrative region of China from2004to2010. Furthermore, a data envelopment analysis based area rapid loss assessment method was proposed basing on the4inputs of summation of consequence index and7outputs of7principle component scores. The assessment and grading were conducted as well. These two methods were compared and the results show that the former one is mainly affected by the results of summation of consequence index, and is believed to be suitable for assessments where the losses should be emphasized; the latter one avoid the assumption of the relationship between loss consequence and social vulnerability, is capable of handle multi inputs and outputs, assess from a more comprehensive aspect, give more sonservative results, and is more suitable and applicabel for assessments where both the losses and the social vulnerability should be considered evenly.