Federal learning model optimization method and device, medium and program product

The invention discloses a federated learning model optimization method and device, a medium and a program product, and the method comprises the steps: obtaining multi-attribute data collected by an edge end, and calculating a data size factor, a data quality factor and a data distribution characteristic factor of the edge end, thereby obtaining the credibility of the edge end; selecting a credible edge end to update parameters of the global model; sending the first global parameter to the trusted edge end, so that the trusted edge end performs training of a local model and obtains a local model parameter; according to the average error of the local model and a preset global model learning rate, calculating to obtain a dynamic learning rate; according to the first global parameter, the dynamic learning rate, the credibility of the credible edge end and the local model parameter, a second global parameter is obtained through calculation, and parameter updating of the global model is achieved. According to the e