时间序列挖掘算法在生产安全事故中的应用研究

发布时间：2018-11-13 13:48

【摘要】：我国每年由安全问题造成的经济损失约占GDP总量的6%,给国家和人民带来了极大的损失,因此对生产安全事故的预测就显得十分重要。传统的生产安全事故分析主要包括统计分析、回归模型、灰色模型等,不利于采取措施预防事故发生。本文在针对生产安全事故数据研究时结合了一元时间序列理论,多元时间序列理论,在理论应用方面有一定的创新,将时间序列预测模型应用到生产安全的预测中,特别是将多元时间序列的二元时间序列的向量移动平均自回归方法运用到生产安全事故中,以近10年的数据作为实证,从多方面分析事故的发生的趋势和影响因素,对其提出指导性意见和建议,及时对生产安全事故采取措施。本文主要展开了一下几个方面的工作:1.从国家安监局官网获取数据,通过正则表达式获取生产安全事故及其简介信息。通过对较大事故进行数据预处理,通过可视化展示其数据特点。2.将一元时间序列ARIMA模型与三次指数平滑模型以15年生产安全较大事故序列数据拟合及预测结果相比较,从事故死亡人数、起数及发展趋势相对比,得到一元时间序列预测模型残差为-21. 92,相对误差0. 266,比三次指数平滑有更高的准确性和可靠性。3.研究多元时间序列理论基础,将多元时间序列建模并应用于生产安全事故的预测,与之前的几个模型相比较,得出相对误差0.2452,更加准确的结果。将死亡人数和事故起数二元时间序列做相关分析,并以其增长率为时间序列分析建模,预测未12个月的数值。4.本文将定性分析与定量预测相结合,用于对生产安全事故的预判。定性分析:通过较大事故简介信息将事故分为不同类型,其中交通事故类型事故占比最多;对于较大事故发生的年份分析,其中以2005年为转折点,2005年以前较大事故呈上升趋势,2005年后呈逐年下降趋势;针对较大事故发生的地域方面得出较大事故主要其中在西南山区较多,地质不稳定区域。定量分析:通过对生产安全事故序列的建立指数平滑、一元时间序列、多元时间序列的模型,调整模型参数,选择最佳模型对较大事故序列进行预测。得到未来一年生产安全较大事故的走势。通过走势及时提出预防和减少事故发生的对策措施,为国家宏观决策提供支持。
[Abstract]:The annual economic loss caused by safety problems in China accounts for about 6% of the total amount of GDP, which brings great losses to the country and the people, so it is very important to predict the accidents of production safety. The traditional analysis of production safety accidents mainly includes statistical analysis, regression model, grey model and so on, which is unfavorable to take measures to prevent accidents. This paper combines the theory of monadic time series and the theory of multivariate time series in the research of production safety accident data. It has some innovation in the application of the theory, and applies the time series prediction model to the prediction of production safety. In particular, the vector moving average autoregressive method of binary time series of multivariate time series is applied to production safety accidents. Taking the data of nearly 10 years as an example, the trend and influencing factors of accidents are analyzed from many aspects. To provide guidance and suggestions, timely production safety accidents to take measures. This paper mainly carried out a few aspects of the work: 1. The data are obtained from the official website of the State Safety Supervision Bureau, and the production safety accidents and their brief information are obtained by regular expression. Through the data preprocessing to the big accident, the data characteristic is displayed through the visualization. 2. 2. The ARIMA model of monadic time series and the cubic exponential smoothing model are compared with the data fitting and forecasting results of 15 years of production safety larger accident series, and the number of deaths in accidents, the number of deaths and the development trend are compared in terms of the number of accidents, the number of deaths, and the trend of development. The residual error of the unary time series prediction model is -21. 92, relative error 0. 266a higher accuracy and reliability than cubic exponential smoothing. This paper studies the theoretical basis of multivariate time series and applies it to the prediction of production safety accidents. Compared with the previous models, the relative error is 0.2452, which is more accurate. The binary time series of death and accident are analyzed, and its growth rate is used as time series analysis model to predict the value of less than 12 months. This paper combines qualitative analysis with quantitative prediction to predict production safety accidents. Qualitative analysis: the accident is divided into different types through the large accident brief information, among which the traffic accident type is the most; For the year analysis of major accidents, the turning point is 2005, before 2005, the major accidents are on the rise, and after 2005, the trend is decreasing year by year. In view of the region of large accidents, it is found that the major accidents are mainly in the southwest mountainous area and the geological unstable area. Quantitative analysis: through establishing exponential smoothing, univariate time series, multivariate time series model, adjusting model parameters, selecting the best model to predict the larger accident sequence. Get the trend of production safety accidents in the coming year. The countermeasures to prevent and reduce accidents are put forward in time through the trend to provide support for national macro-decision-making.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X915.4;O211.61

【参考文献】