Air pollution in industrial zones poses significant threats to environmental and public health, especially in developing regions, highlighting the necessity for accurate forecasting to guide pollution management. This study presents an Optuna-enhanced hybrid Transformer-TimesNet model aimed at improving time series forecasting of six critical pollutants (SO2, NO2, CO, O3, PM10, and PM2.5) in Xinyang Industrial Zone, Xiamen, China. Utilizing air quality data from 2019 to 2023, the model combines the Transformer’s strength in capturing long-range dependencies with TimesNet’s expertise in handling complex temporal patterns. Advanced preprocessing techniques were employed to address both linear and non-linear data components, and Optuna was used for hyperparameter tuning, enhancing model stability and predictive accuracy. Comparative experiments demonstrated the hybrid model’s superior performance against traditional statistical methods, machine learning models, and deep learning approaches, evaluated through metrics such as MAE, RMSE, SMAPE, and R2. The model’s capability to accurately capture long-term pollutant trends underscores its reliability and validity as a predictive tool for policymakers and environmental managers. These results contribute to theoretical advancements in environmental monitoring and offer practical solutions for public health protection and pollution mitigation, demonstrating the potential of hybrid deep learning models in addressing complex forecasting challenges.