Semi-asynchronous federated learning framework based on tree graph block chain

Federal learning based on a linear block chain is limited by a block chain performance bottleneck in a heterogeneous environment, so that the training efficiency is low. A directed acyclic graph block chain is introduced in a current solution to solve the problem, but the directed acyclic graph block chain sacrifices the verifiability of the block chain, an outdated model is difficult to process, and the convergence speed is low. The invention discloses a semi-asynchronous federated learning framework based on a tree graph block chain, and the bottom block chain structure design is a directed acyclic graph taking a block as a center, so as to support verifiable and semi-asynchronous training. In order to promote rapid convergence, a trunk chain generation algorithm is designed, topological sorting is carried out on a semi-asynchronous training process, and a client is guided to sample a proper model. In addition, a consensus mechanism is closely combined with federal learning, so that attacks aiming at the mo