https://github.com/fedebenelli/dgann

Neural Network that estimates the type of failure of a Electric Transformer, based on DGA analysis. Trained with Duval's public dataset
https://github.com/fedebenelli/dgann

Last synced: 2 months ago
JSON representation

Neural Network that estimates the type of failure of a Electric Transformer, based on DGA analysis. Trained with Duval's public dataset

• Host: GitHub
• URL: https://github.com/fedebenelli/dgann
• Owner: fedebenelli
• Created: 2020-08-11T11:08:44.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2020-08-14T17:29:54.000Z (over 4 years ago)
• Last Synced: 2024-10-14T03:46:12.177Z (3 months ago)
• Language: Python
• Size: 11.1 MB
• Stars: 7
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# DGANN

> Transformer Oil Dissolved Gas Diagnostic (DGA) assisted by neural networks

This respository consists in some scripts, a simple neural network module and a database of DGA provided by Duval

## `extract_data.py`

Extract the data from DGA done on faulted transformers, classifing them by type of fault, then saves it to a .csv file

## `NeuralNetwork.py`

Module that contains the Neural Network (NN), it consists in a simple backprop NN with a couple added modules using a RELU function.

## `train_nn.py`

Script used to train the Neural Network, after it ends training it shows two pictures "convergence.svg" and "precision.svg". Before it ends it saves the NN object to a file called `nn.obj`

The first one shows the loss value for each iteration

![](./images/convergence.svg)

The second one shows the correlation between the real data (y axis) and predicted data on the NN (x axis)

![](./images/precision.svg)

## `predict_nn.py`

Main program script that loads the `nn.obj` object and then it asks the user for data to load and returns a dictionary with the estimated probability for each fault.