Indoor air quality prediction and purification system optimisation based on DSTGCN model

Indoor air quality in modern buildings increasingly affects residents’ health, yet existing prediction models suffer from limited accuracy, weak spatial feature extraction, and inefficient optimisation. To address these issues, this study proposes an indoor air quality prediction model based on a dynamic spatiotemporal graph convolutional network integrating generative adversarial networks, lightweight gradient boosters, and pruning algorithms. Virtual samples generated by the adversarial network alleviate sample scarcity, lightweight gradient boosters optimise feature selection, and pruning algorithms reduce computational complexity. Based on accurate prediction results, a dynamic optimisation scheme for air purification systems is developed to achieve active equipment control. Experimental results show that the proposed model achieves a PM2.5 prediction mean absolute error of 2.25 μg/m3 with a processing time of 130.7 ms. After optimisation, the indoor air quality compliance rate reaches 99.04%, outperforming comparison models. The proposed method provides an effective solution for intelligent indoor air quality control and energy-efficient purification system optimisation in building environments.