In this paper, we propose a federated learning(FL)-based collaborative anomaly detection system. This system consists of multiple edge nodes and a server node. The edge nodes are in charge of not only monitoring and collecting data, but also to train an anomaly detection neural network classification model based on the local data. On the other hand, the server aggregates the parameters from the edges and generates a new model for the next round. This system structure achieves light weight transmission between the server and the edge nodes, and user privacy can be well protected since raw data are not communicated directly. We implement the proposed scheme in the practical system and present experimental results that demonstrate results competitive with those of state-of-the-art models.