Finding Faults in 3-phase electrical lines using lossless reconstruction
The problem of finding faults in oscillating data in the form of 3-phase current and voltage values is considered. While these kind of data have high resolutions, we give a solution to multivariate changing point detection, framed as an anomaly detection problem from predetermined cycles in the data.
This solution utilizes signal reconstruction loss with autoencoders and 1D convolutional layers for fault detection in electrical line data. The service is packaged as a docker container and implemented in python. Tensorflow 2.10 is utilized for the machine learning components implementation.
This solution was developed under the I-NERGY project.
We provide a docker container to easily build and run an autoencoder model to discover faults in 3-phase electrical lines (comprising the current values and voltage values). The model utilizes a 6-column csv file with of phase 1 to phase 3 current and voltage values , respectively.. To build the Docker container we provide the following files:
- Dockerfile
- inference.py
- requirements.txt
The inference file takes a csv file and makes sequences of 100 cycles - padding the remainder with previous cycle values.
The scripts were implemented in tensorflow.