Machine learning and statistical modeling for forecasting electricity demand and detecting anomalous consumption patterns in smart grid demand response programs.
This repository contains research experiments developed during my PhD research in applied machine learning for cyber-physical energy systems.
Demand Response (DR) programs rely on accurate prediction of electricity consumption patterns to balance supply and demand in smart grids.
However, consumption data may be affected not only by natural behavioral variation but also by cyber-attacks targeting communication protocols used in smart grid networks.
Protocol-level attacks such as MODBUS, TCP/IP, and WiMAX can introduce anomalies in power consumption data, leading to:
- incorrect demand forecasts
- financial losses for grid operators
- reduced reliability of energy systems
Traditional statistical forecasting alone often fails to capture these attack-induced deviations.
This project investigates how machine learning can improve demand response analytics through:
- seasonal demand forecasting
- detection of anomalous consumption patterns
- correlation of anomalies with network protocol attacks
- improved reliability of smart grid demand response systems
The research pipeline includes:
- preprocessing and cleaning of smart grid demand response datasets
- feature engineering on temporal and behavioral attributes
- seasonal time-series modeling using ARIMA / SARIMA
- machine learning models for anomaly detection
- deep learning models for temporal pattern learning
- comparative evaluation of statistical, ML, and deep learning approaches
Statistical Models
- ARIMA
- SARIMA
Machine Learning Models
- Random Forest
- Support Vector Machines (SVM)
Deep Learning Models
- LSTM for modeling long-term temporal dependencies
- ARIMA and SARIMA models successfully captured seasonal demand patterns
- Machine learning models identified abnormal consumption patterns linked to network protocol anomalies
- LSTM models improved detection of subtle time-dependent anomalies
Programming
Python
Machine Learning
scikit-learn, TensorFlow, PyTorch
Data Processing
NumPy, Pandas
Visualization
Matplotlib, Seaborn
data/ -> datasets src/ -> model implementations results/ -> experiment outputs assets/ -> figures and plots
data/ -> datasets src/ -> model implementations results/ -> experiment outputs assets/ -> figures and plots
pip install -r requirements.txt
Run experiments python src/main.py
This project was developed during my PhD research in Computer Engineering focusing on machine learning for cyber-physical systems and smart grid analytics.
Relating Network Behavior to Demand Response during DDoS Attack in the Smart Grid
Future Technologies Conference (FTC), 2023
Testbed for Evaluating Smart Grid Behavior in Demand Response Scenarios
ICUMT 2022
Google Scholar
https://scholar.google.com/citations?user=2XswkUcAAAAJ
- Time-series forecasting
- Machine learning for energy systems
- anomaly detection in cyber-physical systems
- feature engineering
- reproducible research workflows
Dr. Rajesh Manicavasagam
Research Software Engineer | Applied Machine Learning | Scientific Computing
GitHub
https://github.com/rmanicav