PM-PCNet: A Deep Learning Model for PM2.5 Prediction Integrating Numerical Forecast
Yiqun Yin, Bowen Zhou, Dan Niu, Zengliang Zang, Huixian Zhang, Hua Gao
Southeast University National University of Defense Technology
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
PM2.5 pollution is a major environmental issue in China in recent years.${}^{2}$Accurate prediction of PM2.5 concentration is crucial. Traditional methods rely on numerical models based on aerodynamics. In recent years, deep learning has been widely used in PM2.5 prediction. Previous deep learning methods mainly used observational data from monitoring stations. This paper proposes a deep learning PM2.5 prediction model that integrates numerical model forecast data, called PM-PCNet. The model consists of three parts: periodic feature extraction, trend feature extraction, and covariate relationship extraction. First, the input time series is decomposed using discrete wavelet transform (DWT). The decomposed wavelet coefficients are then fed into the wavelet feature extraction module. By utilizing a discrete wavelet attention module, the model captures both local and global dependencies within the time series. These features are subsequently reconstructed back into the time domain during the reconstruction stage, completing the extraction of periodic features. Trend feature extraction is accomplished through multiple layers of decomposition modules. A Transformer encoder module is used to compute attention mechanisms in the feature dimension. This effectively learns the relationships between covariates. The model is validated using PM2.5 prediction data from Changsha in 2022 and 2023. Results show that PM-PCNet achieves lower errors than numerical models and other deep learning methods. The model demonstrates good predictive performance and practical application potential.
逐年被引趋势
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
物理Air Quality Monitoring and Forecasting
Air Quality and Health Impacts · Traffic Prediction and Management Techniques