摘要
针对单来源污染物数据的空气质量指数(AQI)预测问题,构建了基于CNN-LSTM-AdaBoost混合框架的AQI预测模型。通过卷积神经网络(CNN)提取污染物间的局部交互特征,利用长短期记忆网络(LSTM)构建长期时序趋势,采用自适应提升(AdaBoost)集成学习模块动态加权不同时间尺度的预测结果,实现AQI高精度预测。同时引入随机森林特征归因,识别关键污染物,采用沙普利加性解释(shap)分析不同时间尺度特征对预测值的贡献度。通过对北京市2022-2025年监测站点数据进行实证研究表明,该混合模型的决定系数R^(2)达到0.9866,较传统LSTM模型有显著提升,且在可解释性方面表现出色。
Aiming at the problem of air quality index(AQI)prediction based on single-source pollutant data,an AQI prediction model based on the hybrid framework of CNN-LSTM-AdaBoost was constructed.Local interaction features among pollutants were extracted through the convolutional neural network(CNN),while long-term temporal trends were constructed by using the long short-term memory network(LSTM),and the Adaptive Boosting(AdaBoost)ensemble learning module was adopted to dynamically weight the prediction results of different time scales to achieve high-precision AQI prediction.Simultaneously,random forest feature attribution was introduced to identify key pollutants,and Shapley additive explanations(shap)analysis was used to analyze the contribution of features at different time scales to the predicted values.Empirical studies utilizing monitoring stations’data from Beijing(2022-2025)demonstrate that the determination coefficient R2 of this hybrid model reaches 0.9866,significantly outperforming traditional LSTM models while exhibiting superior interpretability.
作者
周子渊
张梓萱
郭晓梅
ZHOU Zi-yuan;ZHANG Zi-xuan;GUO Xiao-mei(School of Computer and Cyber Sciences,Communication University of China,Beijing 100024,China)
出处
《计算机工程与设计》
北大核心
2025年第8期2365-2372,共8页
Computer Engineering and Design
关键词
空气质量指数
卷积神经网络
长短期记忆网络
自适应提升算法
集成学习
特征归因
深度学习
air quality index
convolutional neural network
long short-term memory
adaptive boosting algorithm
ensemble learning
feature attribution
deep learning
