| 455 | 0 | 32 |
| Downloads | Citas | Reads |
Accurate prediction of the Air Quality Index(AQI) is crucial for environmental governance, pollution early warning, and the development of effective prevention and control policies. To address the limitations of existing Transformer-BiLSTM-based prediction models—such as challenges in processing time-frequency data, insufficient capture of local temporal variation dependencies, and inadequate extraction of salient features—this study proposes an enhanced air quality prediction model. This model integrates a feature fusion embedding layer, a residual attention encoder, and a Temporal Convolutional Network(TCN). The proposed model decomposes the input time series into the frequency domain using Fast Fourier Transform, and the filtered frequency-domain features are fused with time-domain features after high-pass filtering to construct a more comprehensive feature representation. The residual attention encoder consists of one multi-head self-attention block and two residual blocks, where the multi-head self-attention mechanism works in conjunction with the Squeeze-and-Excitation attention mechanism to enhance the model's focus on critical features associated with air quality changes. The stacked residual blocks help mitigate issues of gradient vanishing and explosion during model training, thereby improving the model's representation capability and stability. Dilated convolutions within the TCN are employed to expand the receptive field and effectively capture local temporal dependencies that are often overlooked by global attention mechanisms. Finally, the bidirectional structure of BiLSTM enhances the model's sequential understanding and fully leverages contextual information in both forward and backward directions. Experiments conducted using a dataset containing pollutant concentrations and meteorological data from 12 districts in Wuhan City in 2024 demonstrate that, compared to baseline models, the proposed model reduces the root mean square error, mean absolute error, and mean absolute percentage error in AQI prediction by 53.1%, 57.5%, and 14.79 percentage points, respectively, achieving superior predictive performance and offering a novel and effective approach for air quality forecasting.
[1] 唐颖潇，秦明月，蔡子颖，等.基于松弛逼近方法的天津PM2.5延伸期数值预报研究[J].中国环境科学，2023,43(9):4503-4511.Tang Yingxiao,Qin Mingyue,Cai Ziying,et al.Study on extended-range numerical prediction of PM2.5 in Tianjin based on relaxation approximation method[J].China Environmental Science,2023,43(9):4503-4511.
[2] 屈坤，王雪松，张远航.基于人工神经网络算法的大气污染统计预测模型研究进展[J].环境污染与防治，2020,42(3):369-375.Qu Kun,Wang Xuesong,Zhang Yuanhang.Research progress on statistical prediction models of atmospheric pollution based on artificial neural network algorithms[J].Environmental Pollution and Control,2020,42(3):369-375.
[3] Dawar I,Singal M,Singh V,et al.Predicting air quality index using machine learning:a case study of the Himalayan city of Dehradun[J].Natural Hazards,2024,121(5):1-27.
[4] 苏静，娄英斌，刘语薇，等.基于机器学习的大气NO2浓度预测模型[J].生态毒理学报，2024,19(3):61-69.Su Jing,Lou Yingbin,Liu Yuwei,et al.Prediction model of atmospheric NO2 concentration based on machine learning[J].Asian Journal of Ecotoxicology,2024,19(3):61-69.
[5] 陈镇，刘润，罗征，等.基于机器学习的珠三角秋季臭氧浓度预测[J].环境科学，2024,45(1):1-7.Chen Zhen,Liu Run,Luo Zheng,et al.Ozone concentration prediction in the Pearl River Delta in autumn based on machine learning[J].Environmental Science,2024,45(1):1-7.
[6] Zhan Haolin,Zhu Xin,Hu Jianming.A probabilistic forecasting approach for air quality spatio-temporal data based on kernel learning method[J].Applied Soft Computing,2023,132:109858-1-109858-17.
[7] Che Jinxing,Hu Kun,Xia Wenxin,et al.Short-term air quality prediction using point and interval deep learning systems coupled with multi-factor decomposition and data-driven tree compression[J].Applied Soft Computing,2024,166:112191.
[8] Chen Shuang,He Li,Shen Shinan,et al.Improving air quality prediction via self-supervision masked air modeling[J].Atmosphere,2024,15(7):856-856.
[9] Feng Yang,Kim J S,Yu J W,et al.Spatiotemporal informer:a new approach based on spatiotemporal embedding and attention for air quality forecasting[J].Environmental Pollution,2023,336:122402.
[10] 丁子昂，乐曹伟，吴玲玲，等.基于CEEMD Pearson和深度LSTM混合模型的PM2.5浓度预测方法[J].计算机科学，2020,47(增刊1):444-449.Ding Ziang,Le Caowei,Wu Lingling,et al.PM2.5concentration prediction method based on CEEMD Pearson and deep LSTM hybrid model[J].Computer Science,2020,47(S1):444-449.
[11] Wu Haixu,Xu Jiehui,Wang Jianmin,et al.Autoformer:decomposition transformers with auto-correlation for long-term series forecasting[J].Advances in Neural Information Processing Systems,2021,34(5):2241922430.
[12] Zhou Tian,Ma Ziqing,Wen Qingsong,et al.FEDformer:frequency enhanced decomposed transformer for long-term series forecasting[C]//Chaudhuri K,Szepesvári C,Jegelka S,et al.Proceedings of the 39th International Conference on Machine Learning.Baltimore,Maryland,USA:PMLR,2022:27268-27286.
[13] 潘理虎，张梦麟，樊光瑞，等.基于小波分解的增强时间延迟感知交通流量预测[J].计算机应用，2025,45(11):3649-3657.Pan Lihu,Zhang Menglin,Fan Guangrui,et al.Enhanced time-delay aware traffic flow prediction based on wavelet decomposition[J].Journal of Computer Applications,2025,45(11):3649-3657.
[14] 赵玉帛，丁晓格，刘露毅，等.基于机器学习Mod-EMD-BiLSTM组合模型的碳价格预测方法研究[J].工业技术经济，2025,44(7):124-134.Zhao Yubo,Ding Xiaoge,Liu Luyi,et al.Research on carbon price prediction method based on machine learning Mod-EMD-BiLSTM hybrid model[J].Journal of Industrial Technology Economics,2025,44(7):124-134.
[15] 黄铝文，刘宇航，屈昆仪，等.LSTM与Informer融合预测冠层区域温度[J].农业工程学报，2025,41(8):222-232.Huang Lüwen,Liu Yuhang,Qu Kunyi,et al.Temperature prediction in canopy region by fusing LSTM and Informer[J].Transactions of The Chinese Society of Agricultural Engineering,2025,41(8):222-232.
[16] Jia Guangjie,Zhang Xu,Shen Yijun,et al.Intermittent multivariate time series spindle thermal error prediction under wide environmental temperature ranges and diverse scenario conditions[J].The International Journal of Advanced Manufacturing Technology,2024,132(9/10):4625-4643.
[17] 赵彩敏，刘国红.基于改进的LeNet-5卷积神经网络的人脸表情识别[J].许昌学院学报，2021,40(2):113-116.Zhao Caimin,Liu Guohong.Facial expression recognition based on improved LeNet-5 convolutional neural network[J].Journal of Xuchang University,2021,40(2):113-116.
[18] 闫鹏程，张孝飞，尚松行，等.LIF结合LSTM神经网络的矿井水源识别[J].光谱学与光谱分析，2022,42(10):3091-3096.Yan Pengcheng,Zhang Xiaofei,Shang Songxing,et al.Mine water source identification using LIF combined with LSTM neural network[J].Spectroscopy and Spectral Analysis,2022,42(10):3091-3096.
[19] Yang Xingyao,Xiong Xinyu,Yu Jiong,et al.Cross-domain sequential recommendation based on Fourier Transform and contrastive variational augmentation[J].Computers and Electrical Engineering,2024,120:109681.
[20] Song Weiwei,Xu Ming,Li Haixing,et al.Spatial-frequency fusion for retinal vessel segmentation[J].Biomedical Signal Processing and Control,2025,110:108054.
[21] 田源，高树国，邢超，等.基于多头注意力机制和TCN-BiLSTM的IGBT剩余寿命预测方法[J].电气工程学报，2025,20(3):69-77.Tian Yuan,Gao Shuguo,Xing Chao,et al.Remaining life prediction method of IGBT based on multi-head attention mechanism and TCN-BiLSTM[J].Journal of Electrical Engineering,2025,20(3):69-77.
[22] Wang Yanan,Deng Hongxing,Wang Yunfei,et al.CenterNet-LW-SE net:integrating lightweight CenterNet and channel attention mechanism for the detection of Camellia oleifera fruits[J].Multimedia Tools and Applications,2024,83(26):68585-68603.
[23] Shao Yiqin,Tang Jiale,Liu Jun,et al.Multivariable system prediction based on TCN-LSTM networks with self-attention mechanism and LASSO variable selection[J].ACS Omega,2023,8(50):47798-47811.
[24] 胡衍坤，王宁，刘枢，等.时间序列模型和LSTM模型在水质预测中的应用研究[J].小型微型计算机系统，2021,42(8):1569-1573.Hu Yankun,Wang Ning,Liu Shu,et al.Research on the application of time series models and LSTM models in water quality prediction[J].Journal of Chinese Computer Systems,2021,42(8):1569-1573.
[25] 张彦虎，范敬勇.用GRU循环神经网络优化CMAQ预测结果[J].中国环境监测，2024,40(6):70-79.Zhang Yanhu,Fan Jingyong.Optimizing CMAQ prediction results with GRU recurrent neural network[J].Environmental Monitoring in China,2024,40(6):70-79.
[26] Feng Xiaowei,Fu Jiayu,Zhang Zhili,et al.A study on long-term industrial electricity consumption and carbon emission forecasting methods based on TimesNet models[J].Journal of Physics:Conference Series,2024,2814(1):012019.
[27] Tang Weijie,Chen Jiayan,Chen Dongjiao.Predicting EV battery state of health using long short term degradation feature extraction and FEA TimeMixer[J].Scientific Reports,2025,15(1):2200.
Basic Information:
DOI：10.13637/j.issn.1009-6094.2025.1505
China Classification Code:TP18;X51
Citation Information:
[1]Cui Yanrong,Wang Chunzhou.Air quality prediction model based on improved Transformer-BiLSTM[J].Journal of Safety and Environment,2026,26(08):3300-3309.DOI:10.13637/j.issn.1009-6094.2025.1505.
2026-04-28
2026-04-28
2026-04-28