Module 4 · Project 3: Electricity Demand Forecasting (time series with RNN/LSTM)

OBJECTIVE: forecast the next 24 hours of electricity demand from the past week, and learn how neural networks handle sequences (RNN, LSTM, GRU) while building it. WHAT WE COVER (in project order): • What makes time-series data different (order matters, no random split) • Simple baselines every forecast must beat • Preparing sequence data: time split, scaling, sliding windows, 3D input shape • RNN concepts: memory, vanishing gradients, LSTM gates, GRU • Building and training an LSTM in PyTorch • Evaluation: MAE, RMSE, MAPE, error by hour, comparison with the baselines and your ML model • Deployment: Gradio forecasting app AFTER THIS MODULE YOU CAN: • Turn any time series into training windows for a neural network • Build and train an LSTM/GRU in PyTorch • Judge honestly whether deep learning beats simpler methods DATASET: Spain hourly electricity demand + weather 2015–2018 (the same data you cleaned in the EDA module — no new cleaning needed). TOOL: PyTorch.