Air Quality Index Prediction Based on Informer Model Combined with Multi-Scale Wavelet Transform

随着全球城市化和工业化的快速发展，空气污染问题越来越严重，能准确预测空气质量变换对社会意义重大。本文用小波变换与Informer结合的模型预测武汉市东湖生态旅游风景区的空气质量指数(AQI)。研究选取2023年1月1日至2024年12月31日每小时该区域的气象数据，针对数据的非平稳性，用小波变换对数据去噪，提高其平稳性，例如对不同数据类型选择适配的小波基：AQI、PM2.5、PM10数据选择db6小波基；对于SO2、NO2、O3、CO数据采用sym6小波基等。将处理后的数据输入Informer模型中进行长短期预测。并构建LSTM、Informer、Autoformer和小波变换与Informer结合模型，在不同时间段对比预测，通过MAE和RMSE等指标评估。发现结合模型在各时间段预测中，指标值均优于其他模型，表明该模型在AQI长短期预测中有更高的精度和稳定性。With the rapid development of global urbanization and industrialization, the problem of air pollution has become increasingly serious. Accurately predicting the changes in air quality is of great significance to society. This paper uses a model combining wavelet transform and Informer to predict the Air Quality Index (AQI) of the East Lake Ecological Tourist Scenic Area in Wuhan. The study selects hourly meteorological data of the area from January 1, 2023 to December 31, 2024. To address the non-stationarity of the data, wavelet transform is used to denoise the data and improve its stationarity. For example, the db6 wavelet basis is selected for AQI, PM2.5, and PM10 data, while the sym6 wavelet basis is used for SO2, NO2, O3, and CO data. The processed data is then input into the Informer model for long-term and short-term predictions. LSTM, Informer, Autoformer, and the combined model of wavelet transform and Informer are constructed, and predictions are compared in different time periods. The MAE and RMSE indicators are used for evaluation. It is found that the combined model has better indicator values in all prediction periods, indicating that this model has higher accuracy and stability in long-term and short-term AQI predictions.