AI-Powered Energy Demand Forecasting & Pattern Recognition

A 3-day, hands-on program to turn real-time Canadian grid (IESO/AESO) and weather data into reliable short- and mid-term demand forecasts using ML and deep learning. You’ll engineer features, train and compare models (RandomForest/XGBoost/LightGBM/LSTM/Transformers), evaluate with MAE/RMSE/MAPE, and deliver a 24-hour forecasting model plus an interactive “actual vs predicted” dashboard. Includes live instruction, recordings, materials, and a certificate.

Equip participants to turn real-time Canadian grid data (IESO/AESO) into reliable short- to medium-term energy demand forecasts using modern ML/DL, and to deploy insights via practical dashboards and APIs.

• Access and curate real-time IESO/AESO and weather streams; engineer time-series features.

• Build, tune, and evaluate ML baselines (RandomForest, XGBoost, LightGBM) for demand prediction.

• Apply deep learning (LSTM/GRU/Transformers) for multi-step forecasting (1h/24h/7d).

• Diagnose models with MAE/RMSE/MAPE and residual/feature-importance analyses.

• Detect peaks and anomalies to support grid operations and market decisions.

• Compare ML vs DL trade-offs and select production-ready approaches.

• Deliver a working 24-hour forecast model, an interactive “actual vs predicted” dashboard, and an integration plan for API deployment.

📅 Day 1 – Foundations: Energy Data Access & Feature Engineering

• Introduction to Canadian energy systems: IESO & AESO grid operations and market structures
• Accessing real-time energy APIs: Ontario demand, Alberta markets, and weather data sources
• Hands-on: Fetching & exploring IESO/AESO data using Python
• Feature engineering for time-series: temporal features (hour, day, season), weather integration, lagged demand
• Data preprocessing: handling missing values, normalization, outlier detection
• Hands-on: Building the initial energy dataset with Pandas/NumPy

📅 Day 2 – ML for Demand Prediction

• ML algorithms for energy forecasting: RandomForest, XGBoost, LightGBM
• Feature importance & descriptor engineering (temporal, weather, calendar effects)
• Hands-on: Training baseline models with Scikit-learn
• Model evaluation: MAE, RMSE, MAPE, directional accuracy
• Hands-on: Visualization (actual vs predicted, residuals, feature importance)
• Peak demand prediction & anomaly detection

📅 Day 3 – Deep Learning for Time-Series

• Introduction to deep learning for energy forecasting: LSTM, GRU, Transformers
• Sequence modeling and attention mechanisms
• Hands-on: Building an LSTM model with TensorFlow/PyTorch
• Multi-step forecasting (1-hour, 24-hour, 7-day ahead)
• Hands-on: Comparing classical ML vs deep learning performance
• Real-world deployment considerations & API integratio

🧰 Technical Stack

• • Data: IESO API, Supabase streaming endpoints

• ML: Scikit-learn, XGBoost, LightGBM
• DL: TensorFlow/PyTorch (LSTM, Transformers)
• Visualization: Plotly, Matplotlib, Recharts

• Energy analysts, data scientists, and ML engineers working with time-series data

• Grid/market engineers in utilities, ISO/RTOs, and smart-grid companies

• Energy trading desks, quant researchers, and analytics teams

• Senior undergraduates, postgraduates, and early-career professionals in data/energy domains

• Prerequisites: basic Python and introductory stats/ML (time-series familiarity helpful, not required)

• A working 24-hour-ahead demand forecasting model built on IESO/AESO + weather data

• An interactive dashboard visualizing actual vs. predicted load, residuals, and peak flags

• Comparative report showing ML vs. DL performance (MAE/RMSE/MAPE, directional accuracy)

• Reproducible data pipeline: API ingestion, preprocessing, feature engineering, and validation

• Skills in training and tuning RF/XGBoost/LightGBM and LSTM/Transformers for multi-step horizons

• Methods for peak-demand prediction, anomaly detection, and uncertainty communication

• A lightweight deployment plan (API endpoints, monitoring/drift checks, retraining cadence)

• All notebooks, code templates, and study materials.