Aiming at the problems of unstructured data processing and insufficient dynamic correlation capture in the identification of global digital service trade barriers, this study proposes an intelligent analysis framework based on dynamic heterogeneous graph neural network. By integrating multi-source heterogeneous data such as WTO policy text, multinational enterprise operation data and digital service traffic logs, a 3D dynamic heterogeneous graph model containing policy nodes, enterprise entities and digital infrastructure is constructed to innovatively design an adaptive edge weight learning mechanism, and effectively capture the implicit correlation between tariff barriers and technical trade measures. Experiments show that the accuracy of barrier identification on OECD dataset is 92.7%, which is relatively traditional. In terms of path design, Dual-Channel GAT is developed, combining the timing characteristics of policy evolution and spatial dependence, and a decision tree containing three core paths: technical compliance optimization, cross-border data flow protocol adaptation, and digital tax system collaboration is generated. Empirical analysis shows that the framework successfully identifies 7 types of new digital trade barriers in cross-border cloud service scenarios, and the proposed intelligent negotiation algorithm improves the access efficiency of digital services by 58.3%. This study provides theoretical support for the construction of digital trade rules intelligent decision system, and its dynamic graph representation learning method has universal value for multi-dimensional analysis of complex economic systems.