Federal learning recommendation method based on space-time consistency in mobile environment

The invention relates to a federal learning recommendation method based on space-time consistency in a mobile environment. According to the method, the federal learning process is optimized by mainly utilizing the space-time consistency of tracks, and the method comprises track completion, client clustering and weighted aggregation. Trajectory completion can effectively improve the training performance of a local model under the condition that data is limited, a clustering strategy can effectively promote information sharing between similar clients, and weighted aggregation can reduce the problem of model drift. According to experimental verification, the method provided by the invention realizes higher model performance and better convergence effect than all baselines while ensuring the privacy and security of the user. Compared with the optimal baseline, the SCFL performance is improved by 84.26%, and the convergence speed of the model is greatly improved. 本发明涉及提出了一种移动环境下基于时空一致性的联邦学习推荐方法。该方法主要利用轨迹的时空一致性来优化联