<p>Wireless Sensor Networks (WSNs) face challenges such as dynamic environmental conditions, energy constraints, and fault occurrences, necessitating adaptive and resilient routing strategies. This paper introduces Fault-tolerant Dynamic Graph Attention Network (FDGAN), a comprehensive methodology designed to address these challenges. In the network modeling phase of the FDGAN, the sensor nodes are initialized with unique identifiers, and fundamental characteristics are defined, forming the basis for subsequent analyses. Graph construction employs different matrices, providing a robust representation of the WSN’s topology. The FDGAN consists of the embedding layer, attention mechanism, neighbourhood aggregation layer, multi-head attention layer, and output layer. The embedding layer captures crucial node attributes, such as residual energy, node location, and connectivity information, forming the foundation for subsequent computations. The attention mechanism dynamically assigns attention coefficients, allowing the network to focus on relevant nodes during computations, contributing to dynamic adaptation, fault tolerance, and energy efficiency. The neighbourhood aggregation layer computes weighted feature aggregation from neighboring nodes, incorporating adaptive dynamic graph updates and position modules to enhance fault tolerance. The multi-head attention layer captures diverse information through independent attention computations, optimizing routing decisions and sleep scheduling. The output layer aggregates outputs for predicting fault-tolerant, energy-efficient dynamic routing paths. FDGAN’s adaptability is further exemplified during fault handling processes, where the attention layer adjusts coefficients based on fault-related features, prioritizing resilient pathways. Overall, FDGAN presents a promising solution for addressing challenges in WSNs, offering adaptability, fault tolerance, and energy efficiency to enhance network performance and reliability.</p>