Bidirectional verifiable federal learning method for privacy protection

The invention relates to a privacy-protection bidirectional verifiable federated learning method, which comprises the following steps of: based on an aggregation server and clients, introducing stable communication client analysis in combination with a trusted third party KGC, and executing federated learning for a to-be-trained target global model to obtain a trained target global model; the security of an intermediate model is guaranteed by adopting a dual mode, and a ciphertext state of a model aggregation result is realized through parameter design, so that an attacker cannot speculate client local model gradient information according to multiple rounds of aggregation results, and the problem of offline of the client is effectively solved while the privacy security of the client is effectively guaranteed; two-way verifiability is achieved through design, the ciphertext encryption and decryption function is directly achieved through the client side and the aggregation server, the possibility that encryptio