Telecom operators generate complex business systems to meet different needs of users, which causes difficulties for users to obtain relevant information. To address this problem, this paper proposes a KGQA method for complex queries in the telecom operator domain, which semantically understands the input query, converts the multi-constraint problem into a multi-constraint query graph, and thus obtains structured query statements. In the process of constraint binding, the beam search mechanism is used to effectively reduce the search space of relations and entities and improve the parsing efficiency. Finally, the model uses reinforcement learning based optimization to maximize the expected payoff in the absence of inference path annotations. Experiments on real dataset shows, compared with existing methods, the method proposed in this paper has better performance.