基于模型平均的青岛市空气质量分析

Air Quality Analysis of Qingdao Based on Model Averaging

摘要: 近年来，人民生活水平提高，人民对美好生活的向往从过去的有没有转变成了好不好，在物质文明和精神文明相协调的过程中，空气质量问题成为众多学者关注的焦点。针对当前研究中存在着空气质量影响因素复杂并且传统单一的模型预测不稳定的情况，本文以青岛市为研究对象，基于模型平均方法对空气质量指数进行分析和预测，主要工作如下：(1) 首先，利用python爬虫获取2018~2023年青岛市空气质量指数AQI以及大气污染物浓度(PM2.5、PM10、CO、NO₂、O₃)数据，并对原始数据进行标准化处理和描述性分析，了解当前空气质量变化情况，揭示青岛市空气质量的年际变化趋势和季节性特征。(2) 基于变量相关性分析的基础上，构建多种回归模型，并且利用AIC准则筛选代表性子模型，并且引入四种模型平均方法，如平滑AIC (S-AIC)、平滑BIC (S-BIC)、Mallows模型平均(MMA)和Jackknife模型平均(JMA)，通过赋予不同权重确定策略对多个回归模型进行融合，并利用真实数据进行预测和比较，采取绝对误差、均方误差等指标对模型性能进行评估。并且和传统机器学习模型进行比较，以验证本次研究中所提出方法预测的有效性。本文研究结果表明，近年来，随着政府推出一系列绿色政策，积极响应生态文明建设，青岛市空气质量整体呈改善趋势。在预测性能方面，S-AIC和S-BIC模型平均方法在短期预测中具有较高精度和稳定性，而MMA和JMA模型平均方法在长期预测中表现出更好的稳定性和适应性。综合研究表明，模型平均方法能够有效降低单一模型预测过程中存在的不确定性，同时，本文的研究结果也为后续城市环境管理和空气质量污染防控提供了科学依据。

Abstract: In recent years, with the improvement of people’s living standards, the pursuit of a better life has shifted from “whether it exists” to “whether it is of high quality”. In the process of coordinating material and spiritual civilization, air quality has become a major concern among researchers. Considering the complexity of influencing factors and the instability of traditional single-model predictions, this study takes Qingdao as the research object and analyzes and predicts the Air Quality Index (AQI) based on model averaging methods. The main contributions are as follows: (1) First, Python-based web crawling techniques are employed to collect AQI data and atmospheric pollutant concentrations (PM2.5, PM10, CO, NO₂, and O₃) in Qingdao from 2018 to 2023. The raw data are standardized and subjected to descriptive statistical analysis to explore the overall air quality conditions, as well as the interannual trends and seasonal characteristics. (2) Based on correlation analysis of variables, multiple regression models are constructed. Representative sub-models are selected using the Akaike Information Criterion (AIC). Furthermore, four model averaging approaches, including smoothed AIC (S-AIC), smoothed BIC (S-BIC), Mallows Model Averaging (MMA), and Jackknife Model Averaging (JMA), are introduced. Different weighting strategies are applied to combine multiple regression models. The predictive performance is evaluated using real data with metrics such as mean absolute error (MAE) and mean squared error (MSE). And compare it with traditional machine learning models to verify the effectiveness of the method proposed in this study. The results show that, in recent years, with the implementation of a series of green policies and the promotion of ecological civilization, the air quality in Qingdao has shown an overall improving trend. In terms of predictive performance, S-AIC and S-BIC exhibit higher accuracy and stability in short-term forecasting, while MMA and JMA demonstrate better stability and adaptability in long-term forecasting. Overall, model averaging methods effectively reduce the uncertainty associated with single-model predictions. The findings of this study provide a scientific basis for urban environmental management and air pollution control in the future.