安全风险的贝叶斯估计分布特征与预测应用研究

发布时间：2018-03-08 03:32

本文选题：安全风险　切入点：贝叶斯统计　出处：《华南理工大学》2014年硕士论文　论文类型：学位论文

【摘要】：风险研究是当前研究的热点问题，，如何定义安全风险度量，使安全风险度量能反映相关的差异信息，并可以通过相应的分布特征预测和控制安全事故出现的可能性，是实际应用中亟待解决的问题。本文提出了适合高校资源配置安全风险的新定义，通过对引起风险差异信息的分布进行研究，建立风险评价与预测模型。该模型应用在高校资源配置风险问题上，依据投入产出绩效、偏离期望绩效差异、差异随机出现的可能性等综合评价资源配置过程可能导致的风险程度，取得的预测效果良好。本文的主要研究工作如下： (1)将安全风险定义为各种影响因素的差异信息所引起的不确定因素，认为安全风险是受差异值与差异可能性的作用所引起的。在高校资源配置风险问题中，基于绩效评价模型，获取2007-2011年72所高校的绩效评价结果，以绩效与期望绩效（平均绩效与最大绩效）的偏离程度作为引起风险的差异信息。 (2)利用贝叶斯估计差异信息的分布参数，动态研究绩效偏离期望差异的分布变化特征，使能获取差异分布参数估计量的变化规律，提高风险程度预测的可靠性，使差异随机出现可能性的预测偏差尽可能减少；在配置风险问题中，高校绩效与期望的偏离情况是服从正态分布的，由此建立多层贝叶斯模型，动态获取了每年高校绩效与期望绩效偏离的分布，结果表示高校历年绩效与期望的偏离分布参数波动较小。 (3)利用K-Means聚类算法综合差异信息和差异可能性等信息，寻求能代表不同风险等级的中心样本，采用高斯隶属度函数以模糊隶属度的形式描述7级风险程度。此风险评价模型以最大隶属度对应的风险等级与综合风险等级作为高校的配置风险评价结果，与高校现状较吻合。 (4)在风险评价模型基础上，建立基于BP神经网络的综合风险等级区间预测模型，用于风险的动态预测。对高校的绩效差异信息以及风险评价结果建立预测模型，预测的准确率达到了94.72%。
[Abstract]:Risk research is a hot issue in the current research. How to define the safety risk measurement so that the safety risk measurement can reflect the relevant difference information, and can predict and control the possibility of the safety accident through the corresponding distribution characteristics. It is an urgent problem to be solved in practical application. In this paper, a new definition of security risk suitable for university resource allocation is put forward. By studying the distribution of risk difference information, a risk evaluation and prediction model is established. The model is applied to the risk problem of resource allocation in colleges and universities. According to the input-output performance, deviating from the expected performance difference and the possibility of random occurrence of the difference, the risk degree that the resource allocation process may lead to is evaluated, and the prediction effect is good. The main work of this paper is as follows:. 1) the security risk is defined as the uncertain factor caused by the difference information of various influencing factors, and it is considered that the security risk is caused by the effect of the difference value and the difference possibility. In the problem of university resource allocation risk, it is based on the performance evaluation model. The results of performance evaluation of 72 colleges and universities from 2007 to 2011 were obtained, and the deviation between performance and expected performance (average performance versus maximum performance) was taken as the difference information of risk. 2) using Bayes to estimate the distribution parameters of the difference information, dynamic research on the distribution and variation characteristics of the performance deviating from the expected difference, so as to obtain the variation rule of the difference distribution parameter estimator, and improve the reliability of the risk degree prediction. In the problem of allocation risk, the deviation between performance and expectation in colleges and universities is normally distributed, and a multilayer Bayesian model is established. The dynamic distribution of the deviation between college performance and expected performance is obtained dynamically, and the results show that the deviation distribution parameters of performance and expectation in the past years of colleges and universities fluctuate little. (3) using K-Means clustering algorithm to synthesize the information of difference and possibility of difference, and to seek a central sample which can represent different risk levels. Gao Si membership function is used to describe the risk degree of grade 7 in the form of fuzzy membership degree, and the risk evaluation model takes the risk grade corresponding to the maximum membership degree and the comprehensive risk grade as the evaluation result of collocation risk in colleges and universities, which is consistent with the present situation of colleges and universities. 4) on the basis of risk evaluation model, a comprehensive risk grade interval forecasting model based on BP neural network is established, which is used for dynamic risk prediction. The performance difference information and risk evaluation results of colleges and universities are forecasted. The accuracy of prediction reached 94.72.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：G647

【参考文献】