基于PSO-BP神经网络的车险客户风险等级评估模型研究

发布时间：2018-05-07 02:42

本文选题：客户风险 + 层次分析法　；参考：《宁夏大学》2017年硕士论文

【摘要】：机动车辆的迅猛发展,使人们出行方式发生巨大变化,人们对车辆的依赖性加重。其中,车险作为一种保护类风险投资,以70%的业务量占财产险公司的首位,更是受到保险公司的关注。但是,在竞争日益激烈的保险公司中,整个车险业务系统管理仍不完善,导致保险公司对客户的赔付过多,整个行业利润明显下降。所以,在不断增长的车险客户历史数据中,有效利用这些数据进行车险客户风险等级评估模型的构建,不仅可以及时识别投保客户的潜在风险、提供客户投保建议;还可以为金融研究、社会管理提供有力的依据,具有巨大的社会和商业价值。为此,本文展开如下研究工作:1.车险客户风险评估相关研究概述。概述风险管控的理论基础,强调风险管控的重要性,详细分析车险客户风险评估指标因素,包括车辆风险因素、驾驶人风险因素、环境因素和投保特征因素,并结合定性和定量分析的层次分析法,对量化过的风险指标数据进行加权处理,搭建整个车险客户风险评估指标体系,最后对车险客户风险评估的统计模型方法和人工智能方法进行介绍。2.BP神经网络方法和粒子群算法分析研究。详细介绍BP神经网络方法和粒子群算法(PSO)的原理基础,概括BP神经网络算法、粒子群优化算法的实现思想和优缺点,分析粒子群算法的历史发展进程,提出一种可以克服PSO算法易陷入局部搜索、过早收敛的问题的改进PSO算法,并通过标准函数测试,验证该改进的PSO算法的有效性。3.基于改进PSO-BP神经网络车险客户风险等级评估模型构建。针对BP神经网络模型参数设置不足,分析具有全局寻优功能的粒子群优化算法对BP神经网络的优化原理,并利用改进的PSO算法确定BP神经网络的权值和阈值,构建基于改进PSO-BP神经网络算法的车客户风险等级评估模型。基于以上研究,对某车险行业客户风险评估数据进行了实例仿真,并把实验结果与BP模型和PSO-BP模型相比,分析得出PSO-BP车险客户风险等级评估模型在收敛速度和准确度均有很大提升。
[Abstract]:With the rapid development of motor vehicles, great changes have taken place in people's travel mode and people's dependence on vehicles is becoming more and more serious. As a kind of protection risk investment, auto insurance accounts for the first place in property insurance companies with 70% of business volume, and is paid more attention by insurance companies. However, in the increasingly competitive insurance companies, the management of the whole auto insurance business system is still imperfect, which leads to the insurance companies pay too much to customers, and the profits of the whole industry decline obviously. Therefore, in the growing historical data of auto insurance customers, the effective use of these data to build a risk rating evaluation model of vehicle insurance customers, not only can identify the potential risks of insured customers in time, and provide customer insurance advice; It can also provide a powerful basis for financial research and social management, with great social and commercial value. Therefore, the following research work is carried out in this paper: 1. Summary of relevant research on risk assessment of auto insurance customers. This paper summarizes the theoretical basis of risk management, emphasizes the importance of risk control, and analyzes in detail the risk assessment index factors of vehicle insurance clients, including vehicle risk factors, driver risk factors, environmental factors and insurance characteristics. Combined with the analytic hierarchy process of qualitative and quantitative analysis, the quantitative risk index data are weighted to set up the whole risk evaluation index system of automobile insurance customers. Finally, the statistical model method and artificial intelligence method of vehicle insurance customer risk assessment are introduced. 2. BP neural network method and particle swarm optimization algorithm are analyzed and studied. The principle of BP neural network method and particle swarm optimization (PSO) is introduced in detail. The realization ideas, advantages and disadvantages of BP neural network algorithm and particle swarm optimization algorithm are summarized, and the historical development process of PSO algorithm is analyzed. An improved PSO algorithm is proposed to overcome the problem of local search and premature convergence of PSO algorithm, and the validity of the improved PSO algorithm is verified by the standard function test. Based on the improved PSO-BP neural network, the risk rating model of vehicle insurance customer is constructed. Aiming at the deficiency of BP neural network model parameter setting, this paper analyzes the optimization principle of BP neural network based on particle swarm optimization algorithm with global optimization function, and uses the improved PSO algorithm to determine the weight and threshold of BP neural network. Based on the improved PSO-BP neural network algorithm, the evaluation model of vehicle customer risk level is constructed. Based on the above research, the customer risk assessment data of a vehicle insurance industry are simulated, and the experimental results are compared with BP model and PSO-BP model. It is concluded that the convergent speed and accuracy of PSO-BP vehicle insurance customer risk rating model are greatly improved.
【学位授予单位】：宁夏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F842.634;TP183

【参考文献】