Knowledge graph were introduced into recommender systems to alleviate their cold-start and data sparsity problems, but the privacy risks associated with centralised storage approaches are increasingly prominent. Federated recommendation effectively mitigates the privacy problem through distributed training, but most of the existing methods ignore higher-order information modelling. To this end, this paper proposes Federated Knowledge Collaborative Recommendation (FedKCRec) model, which fuses knowledge graph and federated learning to locally construct a collaborative knowledge graph and extract higher-order semantic information on the client side, and server-side aggregation of model parameters to achieve global optimisation. The model also introduces two privacy protection mechanisms to guarantee data security during training. Experiments on three publicly available datasets demonstrate that FedKCRec is effective and outperforms the state-of-the-art federated recommendation models in existence.