本文针对北京、上海、广州、武汉四个典型城市的PM2.5浓度数据，构建了基于RNN循环神经网络、LSTM长短期记忆网络和GRU门控循环单元三种深度学习模型的空气质量预测系统。通过对比分析不同模型在各城市数据集上的预测性能，探究了深度学习技术在PM2.5浓度预测中的应用效果。研究发现，引入门控机制的LSTM和GRU模型能够有效捕捉PM2.5时间序列中的长期依赖关系，在预测精度和稳定性方面优于传统的RNN模型。其中，GRU模型在多个数据集的性能最佳，在预测精度、训练效率和模型稳定性方面达到了较好的平衡。消融实验揭示了隐藏层神经元数量和学习率等关键参数对模型性能的影响规律。另外，不同城市PM2.5数据的特性差异,导致模型在不同数据集上的表现有所不同,需要针对性地调整模型设计。本研究探索了不同深度学习模型在我国典型城市上的预测效果，为未来进一步完善预测模型和拓展应用场景和环境管理和决策提供有力支持。

This study introduces an advanced air quality forecasting system powered by three cutting-edge deep learning architectures: Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs). The system is rigorously trained and validated using PM2.5 concentration data collected from four prominent Chinese cities: Beijing, Shanghai, Guangzhou, and Wuhan. By conducting an in-depth comparative analysis of the predictive performance of these models across diverse urban datasets, this research endeavors to shed light on the efficacy of deep learning techniques in anticipating PM2.5 concentrations. The findings underscore the superior capabilities of LSTM and GRU models, which leverage gating mechanisms to effectively capture long-term dependencies in PM2.5 time series, resulting in enhanced prediction accuracy and robustness compared to conventional RNN models. Among the three architectures, GRU emerges as the frontrunner, exhibiting the most favorable balance between prediction precision, training efficiency, and model stability across multiple datasets. Ablation experiments further elucidate the influence of critical parameters, such as the number of hidden layer neurons and learning rate, on model performance. Notably, the study underscores the inherent heterogeneity in PM2.5 data characteristics across cities, necessitating city-specific fine-tuning of model architectures. This research offers valuable insights into the application of deep learning models for air quality forecasting in major Chinese cities, providing a solid foundation for the refinement of prediction models and the expansion of their application domains in environmental management and policy-making.