Decentralized security federal learning method and system

The invention relates to a decentralized security federal learning method and system. The method comprises the following steps: deploying decentralized security federal learning at related terminal equipment; each terminal device initializes a local model and data; after initialization, each terminal device trains a local model, the trained terminal devices enter a consensus stage of federation certification, and a block weight is fight for through the accuracy degree and workload of model parameters generated by training; all the terminal devices participating in the consensus determine global model parameters through the consensus, update the global model and output blocks; and each terminal device obtains the global model parameters from the block chain and enters the next round of learning. According to the method, the consensus algorithm and the federated learning task computing power are reasonably distributed, normal operation of the Internet of Things terminal equipment is ensured, and the high partic