Federal learning security aggregation method and system based on modular component homomorphism

The invention discloses a federal learning security aggregation method based on modular component homomorphism, and aims to ensure data privacy and security during multi-party joint training intermediate result exchange. According to the scheme, when multi-party joint training exchanges an intermediate result, the intermediate result is encrypted by using a homomorphic encryption algorithm at a client, then the encrypted intermediate result is sent to a server and aggregated by using a modular component homomorphic encryption algorithm, the aggregated result is returned to all clients, the aggregated result is decrypted at the clients, and finally, the intermediate result is subjected to multi-party joint training exchange. Through mutual cooperation of the server and the client in each iteration, an optimal global model is generated, it is ensured that an intermediate result is not leaked, a training task is completed, and through setting a modular cardinal number and redundant modular projection, the degree