How This Workflow Works
This workflow predicts hourly energy demand by aligning and preprocessing timestamps from historical data, then generating lagged consumption features as predictors and training a deep neural network to obtain forecasts. It evaluates the model's performance by comparing predicted and actual demand using visualizations and scoring metrics.
Key Features:
- Align timestamps and generate lagged consumption features to use as predictors
- Train a deep neural network to model complex dependencies in the data
- Forecast future hourly energy demand automatically
- Visualize and assess prediction accuracy with clear plots and scoring metrics
Step-by-step:
1. Engineer Usage-based Predictors:
The workflow aligns timestamps and creates lagged features from historical energy usage. These engineered predictors help the model learn how past consumption influences future demand.
2. Train Deep Learning Model:
A deep neural network, specifically a Long Short-Term Memory (LSTM) model, is trained using the engineered features. This model is designed to capture sequential patterns and dependencies in time series data, improving forecast accuracy.
3. Generate and Evaluate Predictions:
The trained model produces hourly demand forecasts for both in-sample (historical) and out-sample (future) periods. The workflow compares these predictions to actual values using scoring metrics to assess model performance.
4. Visualize and Share Insights:
The workflow creates line plots that display actual versus predicted energy demand over time. These visualizations help users quickly identify trends, deviations, and the overall effectiveness of the forecasting approach.