Air pollution has become a serious public health and environmental issue, driven by quick urbanization, industrialization and a growing number of cars on the road, which necessitates smart city air quality monitoring systems with real-time intelligent capabilities. The current air quality forecasting techniques both rely on spatial learning or temporal analysis and do not provide an adaptive environmental control over distributed IoT sensor networks. Considering these limitations, this research aims at proposing a novel Graph-Aware Transformer Neural Control (GAT-NC) framework for IoT-based environmental monitoring and air quality assessment. The proposed GAT-NC algorithm combines the spatial dependency learning capability of Graph Attention Networks (GAT), long-range pollution pattern analysis using Transformer-based temporal modeling, spatiotemporal feature fusion, adaptive nonlinear control generation and attention-based learning mechanisms for intelligent environmental regulation. The framework is based on the air quality data received from the IoT sensor network alongside the environmental parameters of temperature, relative humidity and absolute humidity from the UCI Air Quality dataset with 9,358 hourly observations of pollutants like CO, NOx, NO₂, Benzene and NMHC. The experimental results show that the proposed GAT-NC model can outperform the baseline models in terms of accuracy (98.1%) and minimum RMSE (0.072). The framework is robust, stable and reliable through statistical validation using t-test and ANOVA. Based on the reserach, the proposed GAT-NC framework is efficient, scalable, and intelligent solution for real-time air quality prediction and adaptive air quality control in IoT-based smart city infrastructures.