Deep Federated Adaptation: An Adaptative Residential Load Forecasting Approach with Federated Learning

. 2022 Apr 24;22(9):3264. doi: 10.3390/s22093264

Algorithm 1: DFA: Deep Federated Adaptation

Require: Data from different houses

{D_{1}, D_{2}, \dots, D_{N}}

Ensure: Adaptative forecasting models

f_{u}

1: Build an initial global model

f_{G}

with public datasets using (7)

2: repeat

3: Distribute

f_{G}

to all computing devices

4: Use local data to train

f_{u}

based on

f_{G}

with (8)

5: Devices upload

f_{u}

to the model server

6: Average models with (9)

7: Update the global model

f_{G} = {f_{G}}^{'}

8: until

|Θ_{G} - {Θ_{G}}^{*}| < ε

9: Use (12) to optimize the convergent global model

f_{G}

, get adaptative models

f_{u}