【作者】 闫晓强；

【导师】 张宏伟；

【作者基本信息】 天津大学 ， 环境工程， 2004， 硕士

【摘要】 大气是人类不可缺少的生存元素，大气质量的好坏直接影响着人类的生活质量和健康水平。大气污染作为环境问题的主要分支受到世界各国的普遍重视，污染控制工作势在必行。而大气质量预测工作是环境管理部门从宏观角度出发制定决策、提出污染防治措施的前提，当前，在大气污染的短期预测预报领域，应用较多的是结合具体地理环境、气象因素的高斯扩散模式的修正，对于中长期的数值预测，这类模型由于其大量参数的介入而使预测工作变得复杂，且参数确定本身就包含了很多的模糊因素，妨碍着预测工作的精度。因此，本文以华北某市1986～2000年各季度的降尘月均值监测数据为例，从统计学角度出发，对大气质量的中长期数值预测方法进行了系统性地研究，旨在寻找那些既可以达到一定的精度要求又可以避免应用大量参数的简化模型和组合预测模型。本文中主要的大气质量中长期预测模型有：时间序列分析法灰色系统理论法人工神经网络法最优化组合预测法灰色系统理论与时间序列相结合的GM-AR法灰色系统理论与神经网络相结合的GM-ANN法时间序列与神经网络相结合的AR-ANN法小波神经网络组合预测法通过编制相应的程序，除了对这些模型进行纵向比较之外，还在每种模型内部进行了横向比较和结果分析。实例考核证明，在环境信息不够充足的情况下，单纯利用一定历史时期的污染监测资料，应用统计学方法，对于大气质量的中长期预测，只要模型选择适当，完全可以满足实际工作中对预测的期望和要求。更多还原

【Abstract】 Air is the essential element for human being and air quality influents directly the life condition and health level of people. As the main branch of the environmental problems, air pollution is being paid more and more attention by countries of the world. There is no time to delay for air pollution control. Air quality forecasting is the precondition for the environmental administration to make decisions and bring forward pollution prevention and cure measures effectively. Currently, in the field of the air pollution short term forecasting, the correction models of Gaussian distribution are widely used, which vary according to the specific geographic environment and weather factor. Such kinds of models are complicated because there are too many parameters being adopted. However, estimation of the parameters includes certain fuzzy factors itself, which inevitably results in an unsatisfactory prediction. Therefore, based on the principle of statistics and taking the fall dust monitoring data of each season monthly average from year 1986 to 2000 in some city of North China as a study case, medium- term and long-term forecasting methods to air quality were studied systematically, which aimed at finding simplified models and combined forecasting models. That is, finding the models which can both reach the required precision and avoid adopting too many parameters. The main air quality forecasting models discussed in this paper were as follows:Time series analysis modelGrey modelArtificial neural network modelOptimal combination modelGM-AR model (combination of Grey model and Auto-regression model in time series analysis)GM-ANN model (combination of Grey system model and Artificial Neural Network model) AR-ANN model (combination of Auto-regression model and Artificial Neural Network model)Wavelet neural network modelThrough corresponding computer programs, longitudinal as well as transverse comparisons to the forecasting effect between the models were made. Then, with a study case, this paper concluded that, under the situation of lacking environmental information, as long as suitable model was chosen by applying statistics to analyze historical monitoring data, it could satisfy the engineering requirement to medium-term and long-term air quality forecasting.更多还原