ML techniques for State Estimation

This repo contains the code of my Master's Thesis. Specifically, it consists in exploring different techniques(Explanable AI, Physics Informed NN, ...) to perform State Estimation

Current stage

Still pretty far from the finish line. Let's say project is at 10%

Dataset

In order to be able to run the script, you first need to download the dataset at: https://drive.google.com/drive/folders/1Rn1Tnv0XAM1oODwcPImpoSrmGZTdzQrO?usp=sharing. The experiments have been taken with the file named data_for_SE_case118_for_ML.mat, contained in the MLP folder of the Drive repository. The downloaded .mat file must be inside /case118.

What and How to run?

• script.py is meant to provide an idea about how shap values can be computed and what kind of explanations can be generated;
• script_retraining_with_SHAP.py retrains the model by exploiting some information related to the shap values, similarly to what is described here: Utilizing Explainable AI for improving the Performance of Neural Networks

How to run script.py

• First, you need to general the model (.pth):
  • python script.py --train True
• Then, you can load the trained model and use it to generate the shap values
  • python script.py --shap_values True
• Finally, you can load the trained model and the shap values to generate explanations
  • python script.py

How to run script_retraining_with_SHAP.py

• First, you need to general the model (.pth):
  • python script_retraining_with_SHAP.py --train True
• Then, you need to apply the retraining procedure (described above)
  • python script_retraining_with_SHAP.py --retrain_time True
• Finally, you can load the retrained model and perform again the tests
  • python script_retraining_with_SHAP.py --test_retrained True