Thermal model for a power transformer - Synthetic temperature profile generation and AR model order study

This repository contains the Python scripts for the generation of a synthetic temperature profile for the previously defined regions of a transformer, simulating its thermal behavior over a full year based on its thermal data collected over a single day. Additionally, the optimal parameters of the AR model for the proposed prediction algorithm is studied.

---

Overview

This project implements a set of Python scripts (based on an previous MATLAB implementation of the project) to generate and analyze synthetic annual temperature values in various zones of an electrical transformer, starting from its temperature values for a specific day. This is done by modeling the transformer as a thermal system with random anomalies and malfunctions. The goal is not to create a model that perfectly matches the real-world system, but one that exhibits behavior qualitatively similar to that of an electrical transformer.

Subsequently, using this synthetic data, the performance of an autoregressive (AR) algorithm based on Artificial Intelligence and designed for detecting thermal novelties in the transformer's components is tested. This is the primary objective for using this data. Beforehand, the optimal order of this algorithm is studied for its best performance.

Associated article link for more detail:

---

Repository Structure

In the main directory, we find:

├── README.md — Main project documentation.
├── ar_lag — Scripts for studying the optimal order of the AR prediction model.
├── environment.yml
├── predictions — Implementation of the anomaly prediction algorithm.
└── synthetic_model — Creation of the synthetic annual temperature profile.

The directory synthetic_model contains the scripts for creating the synthetic annual temperature profile. The directory ar_lag contains the scripts for studying the optimal order of the AR prediction model. The directory predictions contains the scripts to run the anomaly prediction algorithm. Inside each directory, there is a README.md file with more detailed information about its contents and usage.

---

Core Features

• Synthetic generation of temperature data for different regions of a transformer over the course of a year.
• Export of results in formats compatible with the Python language.
• Configurable parameters, such as the system's thermodynamic parameters.
• Reproducible environment using environment.yml to ensure compatibility across different machines.
• Validation of thermal novelty detection algorithms that alert of potential faults in the transformer.

---

Usage

Create Conda Environment

To replicate the environment used for synthetic data generation:

conda env create -f synthetic_model/environment.yml
conda activate resisto_syn

(Replace resisto_env with the name defined in your environment.yml if it is different.)

Execution

In the file synthetic_model/python_code/configuration.py, replace the variable ruta with the path to the directory where the repository is located.

From the python_code/ directory:

python main.py

The results will be automatically saved in the output/ directory.

To execute the anomaly detection algorithm:

conda env create -f predictions/environment.yml
conda activate resisto_pred

python predictions/simulation.py

---

Acknowledgements

This work is funded by Universidad de Granada and Endesa Distribución under the Endesa-UGR chair in Artificial Intelligence. Besides, this research is part of the PID2022-137451OB-I00 and PID2022-137629OA-I00 projects, funded by the CIN/AEI/10.13039/501100011033 and by FSE+, and the C-ING-183-UGR23 project, cofunded by Consejería de Universidad, Investigación e Innovación and by the European Union under the Andalusia ERDF Program 2021-2027.

---