武汉市PM2.5浓度预测的统计分析及仿真

发布时间：2018-01-06 12:27

本文关键词：武汉市PM2.5浓度预测的统计分析及仿真　出处：《华中科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：细颗粒物PM2.5是雾霾产生的重要因素,其成因复杂,危害严重。为实现对PM2.5浓度的预测并加强对其污染的控制,开展相关的研究及基于研究上的污染治理工作意义极其重大。然而传统方法难以充分利用历史数据中的有用数据进行精确预报。本文将BP神经网络用于武汉市PM2.5浓度的预测,借助MATLAB软件实现了此模型的建立、训练和仿真。首先,利用时间序列模型对PM2.5的污染进行了预测,进而通过对API的六个指标绘制散点图及相关关系的分析构建主成分分析模型和逐步回归模型,通过对比找出相对最优模型,并根据此模型预测了下一个月的PM2.5浓度值。由于传统建模法建立的模型大多是线性的,预测精度往往难以保证,而BP神经网络能提供较为理想的处理非线性问题的模型,因此考虑将BP神经网络用于PM2.5的预测,在365个测试样本中,预报结果与实际完全吻合的为240天,占65.8%;相差不超过一级的为337天,占92.3%。预报结果与实际情况基本一致,从实践上证明了BP神经网络运用于PM2.5预报的可行性,为城市PM2.5的污染预报提供了新的思路和方法。
[Abstract]:PM2.5 fine particles is an important factor in haze, the causes are complex, serious harm. In order to realize the prediction of PM2.5 concentration and strengthen the control of the pollution, to carry out related research and Research on the pollution control work is of great significance. Based on the traditional method to make full use of useful data in historical data for accurate prediction in this paper BP. Neural network is used to predict the concentration of PM2.5 in Wuhan City, the implementation of the model established by MATLAB software, training and simulation. Firstly, using time series model to predict the pollution of PM2.5, and then through the six indicators of API drawing scatter plot analysis and correlation of the construction of the model of principal component analysis and stepwise regression model, through comparison to find out the relative optimal model, and predict the concentration of PM2.5 a month depending on the value of this model. Since the establishment of the traditional modeling method of the model are mostly The linear prediction accuracy is often difficult to guarantee, and the BP neural network can provide the ideal model for dealing with nonlinear problems, so consider using BP neural network for the prediction of PM2.5, in the 365 test samples, the prediction results are in agreement with the actual 240 days, accounting for 65.8%; but no more than a class is 337 day, accounting for 92.3%. prediction results consistent with the actual situation, from the practice proved the feasibility of using BP neural network to forecast PM2.5, provide new ideas and methods for pollution prediction of city PM2.5.

【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X513

【相似文献】