In the context of intelligent justice constructing, using artificial intelligence technology to improve judicial efficiency and fairness has become an important research direction. Crime charge prediction, as an important task of intelligent justice, aims to accurately determine the corresponding crime based on the case description. However, most existing methods only focus on the semantic level and ignore that syntactic understanding can reduce misjudgments caused by semantic biases. Therefore, this paper proposes a multi-chain perception crime charge prediction method, MPCCP. This method aims to achieve a deeper semantic and syntactic understanding of case texts through graph neural networks. It designs a multi-chain perception mechanism to explore the potential connections between information. The mechanism captures the complex relationships between criminal behaviors through long-distance propagation and multi-step inference on semantic and syntactic information and introduces a rule engine to effectively predict the principal and subordinate crime charges that overlap in cases. Experimental results show that this method has achieved significant performance improvements in the crime charge prediction task.