基于数据挖掘技术的大气环境预测研究

发布时间：2018-04-08 09:47

本文选题：大气环境　切入点：污染物浓度　出处：《北京交通大学》2017年硕士论文

【摘要】：随着我国环境监测网络覆盖面的深入,各环境监测站点产生并积累了大量的监测数据。这些数据目前只是用来数据查询,数据的潜在价值还没有挖掘出来,因此利用这些历史数据找到大气污染物浓度变化的趋势和规律,并且设计开发大气环境预测系统是十分有必要的。本文对大气环境预测模型展开细致分析和探讨,提出两种大气环境预测模型,分别针对短期(1小时到4天)和中长期(4天到21天)情况下的预测,并在两种预测模型基础上搭建大气环境预测系统,方便用户了解未来大气污染物浓度。本文的主要研究工作有:(1)数据获取与数据预处理。编写网络爬虫脚本从北京市环境保护监测中心获取原始数据,网络爬虫脚本使用HttpClient技术实现模拟浏览器发送请求,使用Jsoup技术完成对网页源码信息的解析。对得到的原始数据使用数据清洗的方法,删除超出正常范围和相互矛盾的数据,使用数据变换的方法,完成不同量级、量纲的归一化。(2)基于多元线性回归的短期预测模型研究。通过优化建模方法、增加输入因子,提出多元线性回归优化模型。通过实验对比,建模方法为逐步线性回归,增加季节因素和其他污染物浓度两个输入因子,能较为准确的预测未来大气污染物浓度并且适用于大气污染物浓度的短期预测。(3)基于遗传神经网络的中长期预测模型研究。针对传统BP神经网络全局搜索能力不足、容易陷入局部最优、训练速度慢等问题,提出BP神经网络与遗传算法相结合并且将遗传算法的交叉概率和变异概率随适应度的变化而变化。经过改进的神经网络局部寻优能力强、擅长全局搜索、训练时间短并且适用于大气污染物浓度的中长期预测。(4)大气环境预测系统的设计与实现。该系统使用jQuery技术和Ajax技术将监测点在地图上展示,并在对应的监测点位置显示未来大气污染物浓度,使用Spring MVC框架实现后台架构设计,将算法模块与系统功能模块结合,完成大气环境预测模块快速计算和直观显示预测结果的功能。
[Abstract]:With the development of environmental monitoring network in China, a large number of monitoring data have been generated and accumulated in environmental monitoring stations.These data are currently used only for querying data, and the potential value of the data has not yet been mined, so use these historical data to find trends and patterns of concentration changes in atmospheric pollutants.And it is necessary to design and develop the atmospheric environment prediction system.In this paper, the atmospheric environment prediction model is analyzed and discussed in detail, and two kinds of atmospheric environment prediction models are put forward, which are respectively for short term (1 hour to 4 days) and medium and long term (4 to 21 days).Based on the two prediction models, an atmospheric environment prediction system is built to facilitate users to understand the future concentration of atmospheric pollutants.The main research work in this paper is: 1) data acquisition and data preprocessing.The web crawler script is written to obtain the original data from Beijing Environmental Protection Monitoring Center. The web crawler script uses HttpClient technology to simulate the browser to send requests, and Jsoup technology is used to complete the analysis of the web page source code information.The method of data cleaning is used to remove the data beyond the normal range and the contradictory data, and the method of data transformation is used to complete the research on the short-term prediction model based on multivariate linear regression.By optimizing modeling method and adding input factor, a multivariate linear regression optimization model is proposed.Through experimental comparison, the modeling method is stepwise linear regression, adding two input factors of seasonal factor and other pollutant concentration.It can accurately predict the future concentration of atmospheric pollutants and is suitable for short-term prediction of the concentration of atmospheric pollutants. (3) based on genetic neural network, the study of medium and long term prediction model is carried out.The traditional BP neural network has insufficient global search ability, easy to fall into local optimum and slow training speed, etc.Combining BP neural network with genetic algorithm, the crossover probability and mutation probability of genetic algorithm are changed with fitness.The improved neural network has strong local optimization ability, is good at global search, has short training time and is suitable for medium and long term prediction of atmospheric pollutant concentration.The system uses jQuery technology and Ajax technology to display the monitoring points on the map, and shows the future concentration of air pollutants in the corresponding monitoring points. The system uses the Spring MVC framework to realize the background architecture design, and combines the algorithm module with the system function module.The function of fast calculation and visual display of prediction results is completed in the atmospheric environment prediction module.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X831;TP311.13

【参考文献】