Large scale data exchange in line with laws and regulations and data ethical constraints is an important conditional basis for making data as a production factor. As an emerging technology，federated learning solves the privacy problem of data exchange，which has been highly concerned by the academic community. However，its specific method is not mature，and it is far from large-scale application. The cooperation of blockchain and federated learning simplifies the data distribution scenario，provides the traceability of learning process，and realizes the “pre-evaluation” of data transaction. Based on the role of trust media of blockchain，the trusted privacy computing platform with multi-party cooperation can be realized by recording the training parameters，model data，data call process，etc. On the premise of not exposing specific data，through the data sharing of neural network model and gradient，the knowledge value contained in the data can be transferred to break the existing data island and build the data value chain.