Equivalent-gate identification in And-Inverter Graphs (AIGs) aims to identify logic gates with identical or highly similar Boolean functionality, which can support logic optimization, redundancy detection, technology mapping, and functional verification. However, structurally heterogeneous AIGs contain complemented edges with explicit phase-transition semantics, and functionally equivalent gates may be distributed across different logic-cone structures and circuit scales. These characteristics pose challenges for learning-based equivalentgate identification methods, as they usually do not explicitly model complemented-edge-induced phase transitions during representation propagation and may have limited ability to align functionally equivalent gate representations under diverse logic-cone structures and circuit-scale variations. To address these challenges, this paper proposes PhaseConeGNN, a phaseaware logic-cone graph neural network for equivalent-gate identification in AIGs. Specifically, PhaseConeGNN employs phase-aware propagation, stable representation propagation, and a mixture-of-experts fusion mechanism to capture inversion-sensitive signal semantics, alleviate representation drift in heterogeneous AIG structures, and integrate logiccone information at different depths, thereby improving the identification of functionally equivalent gates under diverse circuit structures. Comprehensive experiments are conducted on 83,155 Sub-AIG samples constructed from publicly available open-source Verilog RTL repositories, which are synthesized and further converted into AIG graph samples through a standard EDA flow. The results indicate that PhaseConeGNN achieves more accurate gate-pair functional similarity estimation than the compared baselines, with 25.30% MAE reduction and 25.15% RMSE reduction over the best-performing baseline. These results indicate that PhaseConeGNN can learn phase-consistent and structure-robust gate-level functional representations for AIGs from open-source RTL designs. The source code is available at https://anonymous.4open.science/r/PhaseConeGNN-3607.