Elevated cognitive workload (CW) in air traffic controllers can markedly increase operational errors and safety risks, making rapid and accurate CW identification essential for ensuring the safe and efficient operation of the aviation system. Electroencephalogram (EEG) has played an important role in cognitive workload recognition due to its high temporal resolution and objectivity, yet existing methods still suffer from strong signal nonstationarity and relatively limited representational dimensions, which undermines recognition stability and interpretability. To address these issues, this study developed a high fidelity airport air traffic control simulation platform and collected EEG data under low workload and high workload tasks. A feature representation framework was proposed by integrating power spectral density (PSD) features with microstates dynamics. A binary classifier was then built using a support vector machine (SVM). A sliding window analysis identified a 10 s window as an appropriate monitoring scale that balances information content and real time responsiveness, achieving an accuracy of 83.7 percent and a high workload recall of 82.0 percent. Ablation comparisons based on the receiver operating characteristic (ROC) curve and the area under the curve (AUC) showed that the fused feature model achieved an AUC of 0.92, significantly outperforming models using PSD features alone or microstates features alone. These findings demonstrate that the proposed multi domain feature fusion and interpretable modeling approach can reliably capture neurophysiological evidence associated with cognitive workload and provides an effective method for accurate workload assessment in air traffic controllers.