A GCN-based availability analysis method for mobile network with spatial–temporal effect awareness

The spatial–temporal effect induces complex dynamic changes in the states of network, increasing the risk of mobile network unavailability. Current availability analysis methods are difficult to establish the network spatial–temporal dynamic variation pattern. Machine learning methods avoid the weakness of explicitly modeling effect patterns. However, these methods usually neglect spatial topology changes and abrupt temporal dynamics. To address the above problem, we propose a GCN-Based Availability Analysis Method for Networks with Spatial–Temporal Effect Awareness (Gstea). In the spatial domain, the spatial variation is captured by graph adaptation and graph convolution. In the temporal domain, the Ek (ERP-based time-aligned with k-shape integration to preserve shape similarity) fully aligns the temporal sequence to identify abrupt network changes and generates enhanced graph signal propagation. Experimental results show that Gstea outperforms the best-performing baseline methods in various complex network scenarios (average improvement range from 7% to 26%). This demonstrates that Gstea can provide a more accurate evaluation in networks with spatial–temporal effect.