基于统计算法的城市犯罪情报分析

发布时间：2018-04-11 11:08

本文选题：犯罪情报 + 统计算法　；参考：《武汉大学》2009年博士论文

【摘要】：自上个世纪80年代以来,随着计算机技术的不断发展,针对不同领域的犯罪情报分析算法和模型层出不穷。特别是在国外,针对犯罪的行为特性、心理特性、地域特性等开展了大量有价值的定量研究,其研究成果已经广泛的应用于实际工作。但是在国内,由于各种原因针对犯罪分析的定性研究远远多于定量研究；同时,针对社会治安综合状态的犯罪分析仍属空白领域；而如何有效的针对犯罪情报进行科学分析,找出其变化内蕴含的信息,准确分析和把握社会治安形势变化趋势,是公安机关各项工作顺利开展、保持社会稳定有序的重要基础。在公安部科研项目：“基于数据仓库的警情预测分析系统”(项目编号：2005hbstyycx115)和“警务信息综合平台和警用GIS的数据交换模型研究”(项目编号：2007hbstyycx065)的资助下,本论文围绕城市犯罪情报信息分析模型和算法开展了深入研究。本论文主要做了如下工作： 1.针对社会现实中社会治安形势的“规律性波动”和“警示性波动”现象,在客观分析犯罪情报研究领域相关成果基础上,指出了犯罪情报分析在趋势分析预警领域的薄弱之处,探讨了在研究犯罪情报分析中使用基于统计的入侵检测思想的可行性,指出了需要重点解决的几个方面问题。 2.提出了基于指数衰减的城市犯罪分布特征向量,用于描述社会治安综合状态。该向量由若干重叠时间窗内分布特征按一定算法衰减、叠加后用以刻画城市社会治安综合态势。由于多次考虑了历史因素的影响,该向量能较好的反应城市犯罪统计特征的时间变化特性,既能表示当前统计数据的特性,又能够蕴含历史数据特性和现实表征。由于采用递推的方法进行构造,该向量使用较为方便。提出了基于卡方检验的原始数据预处理算法(Chi-square testing based Distribution Classifying Algorithm, CDCA),该算法通过对各类犯罪数据进行分类、合并和剔除,获取对犯罪数据进行Poisson逼近的最优区间。 3.提出了基于复合统计特征向量的假设检验算法(Composite Statistic Profiling-Vector Hypothesis-test Algorithm, CSPHA)。CSPHA算法设计了分别体现数据收集、数据提取和融合、数据分析时间特征的三种时长,在不同的时间范围内收集业务系统中的分离数据、提取其统计特征值并构造检验向量；设计了有效的系统异常检验式和参照异常检验式以对当前统计量进行检验。实验系统和实验室检测结果验证了CSPHA算法对常态城市社会治安综合态势的准确把握。 4.提出了蕴含统计分量关联关系的ECSPHA算法(Enhanced Composite StatisticProfiling-Vector Hypothesis-test Algorithm)。ECSPHA算法选取替代协方差矩阵以放大反应不同类型犯罪形式之间关联特性,优化了检验量以提高对复杂系统中不同参量关联的敏感性。实验结果表明ECSPHA对单一统计数据不敏感,对关联统计数据较敏感,较好的实现了对特殊条件下的社会治安状态把握。 5.提出了一个基于历史分布数据的犯罪情报分析算法(Long-term Historical Profile based Information Processing Algorithm, LHPIPA)。该算法分析各个统计特征量在较长历史环境中的分布特性,通过一个标准正态分布随机变量将不同特征量测量值进行归一化。算法充分考虑实际工作中具有丰富经验的专家意见,在计算异常指标时引入了人工的决策干预。实验结果表明了LHPIPA相对CSPHA有较好的人工科学决策和时空拓展特性。
[Abstract]:Since the last century since 80s, with the continuous development of computer technology, according to the different areas of the criminal intelligence analysis algorithms and models emerge in an endless stream. Especially in foreign countries, the behavior of crime, psychological characteristics, regional characteristics and carry out a quantitative and a lot of valuable research, its research results have been widely used in practical work. But in China, due to various reasons for the qualitative study of crime analysis is far more than the quantitative research; at the same time, according to the comprehensive analysis of the social security status crime is still a blank field; and how to effectively carry out scientific analysis for criminal information, to find out the change in the information contained, accurately analyze and grasp the changing trend of the social security situation, public security organization of the work carried out smoothly, the important foundation to maintain the stability and order of the society. In the research project of the Ministry of Public Security: "pre warning based on data warehouse Measurement and analysis system "(project number: 2005hbstyycx115) and" research model of police information integrated platform and police GIS data exchange "(project number: 2007hbstyycx065), this paper focuses on City criminal information analysis model and algorithm to carry out in-depth research. This paper mainly do the following work:
1. according to the law of social reality in the social security situation of "volatility" and "warning wave" phenomenon in the research field of objective analysis of criminal intelligence related on the basis of the results of the analysis of criminal intelligence analysis pointed out that the weakness of early warning in the field of trend, discussed in the study of crime in the information analysis the feasibility of using statistical intrusion detection theory based on the pointed out several problems need to be solved.
2. the city crime distribution feature vector based on exponential attenuation is used to describe the comprehensive social order. The state vector is composed of several overlapping time window distribution algorithm according to certain attenuation, superimposed to characterize the comprehensive situation of the city social security. Because of the multiple influence factors of history, time variation characteristics of the vector can reflect the city crime the statistical characteristics of good, not only can represent the characteristics of the current statistical data, but also contains the history and the characteristic of the data representation of reality. By using recursive method to construct the vector more convenient to use. The chi square test of the original data preprocessing algorithm based on (Chi-square testing based Distribution Classifying Algorithm, CDCA), the the algorithm through the classification of various types of crime data, merger and acquisition was removed, optimal Poisson approximation of crime data.
3. proposed composite statistical hypothesis testing algorithm based on feature vector (Composite Statistic Profiling-Vector Hypothesis-test Algorithm, CSPHA.CSPHA) algorithm is designed which reflects the data collection, data extraction and data fusion, time characteristic analysis of three kinds of long, isolated data service system in different time range, extract the statistical characteristic value and construct the test vector; design of effective system of inspection and inspection according to abnormal abnormal type of the current statistics to test. Test results verify the laboratory experimental system and CSPHA algorithm to accurately grasp the situation of social security in normal city.
4. proposed ECSPHA algorithm contains statistical correlation component (Enhanced Composite StatisticProfiling-Vector Hypothesis-test Algorithm.ECSPHA) algorithm to select alternative covariance matrix to magnify the correlation characteristics between the reactions of different forms of crime, to optimize the test in order to improve the sensitivity of different parameters related in complex systems. The experimental results show that the ECSPHA of single statistical data is not sensitive, sensitive the related statistical data, to achieve a better grasp of the special conditions of social security.
5. propose a criminal intelligence analysis algorithm based on historical data distribution (Long-term Historical Profile based Information Processing Algorithm, LHPIPA). The algorithm of distribution analysis of characteristics of all the statistics in the long historical environment, normalized by a standard normal distribution of random variables with different characteristics of measurement values. The algorithm takes full account of the expert opinions have rich experience in the practical work, making artificial intervention applied in calculation of abnormal indicators. The experimental results show that the LHPIPA CSPHA had better artificial decision-making and development of time and space characteristics.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2009
【分类号】：D917

【参考文献】