Machine Learning and Data Science for Power Systems
ELEC97077 teaching, materials, learning outcomes, and coursework.
This module introduces machine-learning and data-science methods for modern power-system operation, forecasting, security, and data-driven decision-making. It combines five hours of lectures with a one-hour tutorial and assessed coursework.
Course materials
Learning outcomes
By the end of the module, students should be able to:
- Distinguish supervised and unsupervised learning tasks.
- Implement and tune linear regression, logistic regression, support-vector machines, principal-component analysis, and k-means.
- Evaluate models using MSE, F1, TPR, FPR, FNR, and TNR.
- Diagnose generalisation, overfitting, model capacity, and regularisation.
- Conduct hyperparameter tuning with an appropriate validation pipeline.
- Explain the role of time-series foundation models and use pretrained models such as Amazon Chronos for energy forecasting.
Indicative topics
- ML for power systems: data-driven and model-driven approaches; operational use cases.
- Supervised learning: linear and logistic regression; support-vector machines.
- Unsupervised learning: principal-component analysis and k-means.
- Performance assessment: regression and classification metrics.
- Generalisation: capacity, bias–variance trade-offs, and regularisation.
- Model selection: train/validation/test splits and grid search.
- Special topics: No Free Lunch intuition, foundation models, time-series forecasting, and energy-data sources.
Coursework outcomes
The coursework connects lecture theory with hands-on analysis of real grid data. Students learn to:
- construct complete ML pipelines for power-system applications;
- collect and preprocess irregularly sampled power-system data;
- use zero-shot, autoregressive, and covariate-informed forecasting methods;
- build DC power-flow measurement matrices;
- perform weighted least-squares state estimation;
- design chi-square bad-data detection with a controlled false-positive rate;
- train classifiers for false-data injection attack detection; and
- present algorithms, equations, and experimental methodology clearly.