The goal of this project is to make a Federated Algorithm in order to obtain a classification on two labels : Location and faultLabel. I am working with four different datasets. They are composed as follows :
- Divided into four groups :
- df1 : with shape (968,51)
- df2 : with shape (968,51)
- df3 : with shape (968,51)
- df4 : with shape (968,51)
We obtained this dataset by injecting fault with different resistances on four different zones in the IEEE-13 distribution network with renewable energy. The columns are :
- voltage : signal measured
- measloc : zone from which the voltage has been measured (between 1 and 4 (IEEE-13))
- locLabel : zone in which the fault has been injected (1-4)
- resistance : shows the different number of resistances
- faultLabel : represents the type of the faults (11 types : ABC, ABCG, AB, AC, BC, ABG, ACG, BCG, AG, BG, CG)
In order to see what would have changed using another kind of prediction, I decided to use the same Split Neural Network, but instead of computing two different accuracies, i chose to add a combination of fault and location.
Here below you can see the results
At this point, i need to have a baseline, in order to judge the goodness of the results. So for this task, i chose to use an algorithm for single locations, where i calculate the accuracy of the prediction for each location. To make it more clear, i'll show here below an example of the output of the code :
Here below there is a comparation between al the results obtained tuning the hyperparameters