In this paper, we propose FedTT, an effective, efficient, privacy-aware cross-city traffic knowledge transfer framework, which transforms the traffic domain of data from source cities and trains the traffic model on transformed data in the target city. First, we design a traffic view imputation method to complete and predict missing traffic data by capturing spatio-temporal dependencies of traffic view. Second, we utilize a traffic domain adapter to uniformly transform the traffic data from the traffic domains of source cities into that of the target city. Third, we propose a traffic secret transmission method to securely transmit and aggregate the transformed data using a lightweight traffic secret aggregation method. Fourth, we introduce a federated parallel training method to enable the simultaneous training of modules. Experiments using 4 datasets on three mainstream traffic prediction tasks demonstrate the superiority of the framework. In future work, we aim to extend FedTT to support a broader range of spatio-temporal prediction tasks.
Python.version = 3.9.21
Other dependencies are listed in requirements.txt.
All the experiments are conducted in the federated environment on four nodes, one as a server and the other three as clients, each equipped with two Intel(R) Xeon(R) CPU E5-2650 [email protected] 12-core processors, 128GB of RAM, and an internet speed of 100MB/s.
The datasets used in this paper are public and available in https://drive.google.com/drive/folders/1a4JNMJNWYGF4he7rlt_aYT3I4P4XX8FK?usp=sharing.
The running example of FedTT is as follows.
python main.py --address 'localhost' --batch_size 128 --clients_num 3--device 'cuda' --epoch 100 --frozen 5 --learning_rate 0.005 --model 'mlp' --task 'flow' --test_split 0.1 --time_step 12 --train_split 0.8 --val_split 0.1