K-GCRNN: A Kolmogorov-Arnold Approach for Multi-Band Spectral Prediction

As the global proliferation of wireless devices intensifies, the availability of spectrum resources is increasingly constrained. Consequently, developing predictive methods based on spectrum resource management is crucial. In this paper, we propose a new model to predict the spectrum. This model processes spectral data by utilizing Graph Convolutional Networks (GCNs) and Gated Recurrent Units (GRUs). In addition, the model is processed with a Kolmogorov-Arnold network(KAN) after the output of each time step, which enhances the adaptability to environments with complex spatio-temporal dynamics. Experiments show that this comprehensive approach not only enhances prediction accuracy but also expands the model's capacity to address complex data issues.