Project Showcase

Predictive Modeling: Energy Demand Forecasting

Building probabilistic forecasts to optimize grid operations across Portuguese municipalities.

Key Technologies

Python Feature engineering and probabilistic modeling with pandas and scikit-learn.
Prophet Hierarchical time-series forecasting with external regressors.
MLflow Experiment tracking, model registry, and deployment packaging.

GitHub Repository

DiogoRibeiro7/energy-demand-forecasting

Forecasting pipeline with reproducible experiments and deployment assets.

Stars: —
Forks: —
Open Issues: —

Live Demo & Visualizations

Launch Interactive Demo

Problem Statement

Municipal grid operators struggled to anticipate demand spikes during extreme weather events, leading to costly peaker plant activations and service disruptions.

Solution Approach

Aggregated smart meter telemetry, weather forecasts, and socioeconomic indicators.
Engineered lagged consumption features, holiday flags, and regional mobility indices.
Trained gradient boosted decision trees and Prophet ensembles with hyperparameter sweeps.

Results & Findings

The combined ensemble reduced mean absolute percentage error (MAPE) by 18% compared to the incumbent baseline and enabled proactive load shifting programs.

MAPE improvement

18%

Relative to previous autoregressive baseline across validation windows.

Peak demand warning lead time

4 hours

Average actionable notice before critical threshold breaches.

Carbon savings

6.5%

Reduction in emergency peaker usage measured quarter-over-quarter.

Challenges & Lessons Learned

Harmonizing disparate telemetry sampling rates required building a robust temporal alignment service and data quality audit dashboard.

Datasets & Sources

Portuguese smart meter telemetry (2021–2023)
ECMWF weather ensemble forecasts
Municipal census indicators sourced from national statistics offices

Model Performance

Forecast intervals calibrated within ±6% coverage error across municipalities. Residual diagnostics confirmed minimal autocorrelation after differencing and regressor tuning.

Interactive forecast explorer with interval coverage overlays.

Future Work & Improvements

Extend the model to include real-time grid telemetry and reinforcement learning-based demand response optimization.

Collaboration & Contributions

Interested utilities and research partners can open issues or reach out via dfr@esmad.ipp.pt to discuss pilot deployments.

Read Contribution Guidelines

Exploratory Analysis: Urban Energy Patterns
Visual analytics companion detailing consumption clusters.

BibTeX

@misc{ predictive-modeling-energy-demand,
  title = { Predictive Modeling: Energy Demand Forecasting },
  author = { Diogo Ribeiro },
  year = { 2026 },
  publisher = { DataLog | Data Science \& Research Theme },
  url = { https://diogoribeiro7.github.io/analytics-blog-jekyll/portfolio/predictive-modeling-energy-demand/ }
}

RIS

TY  - COMP
TI  - Predictive Modeling: Energy Demand Forecasting
AU  - Diogo Ribeiro
PY  - 2026
PB  - DataLog | Data Science & Research Theme
UR  - https://diogoribeiro7.github.io/analytics-blog-jekyll/portfolio/predictive-modeling-energy-demand/
ER  -

EndNote

%0 Computer Program
%T Predictive Modeling: Energy Demand Forecasting
%A Diogo Ribeiro
%D 2026
%I DataLog | Data Science & Research Theme
%U https://diogoribeiro7.github.io/analytics-blog-jekyll/portfolio/predictive-modeling-energy-demand/

Open science & reproducibility badges

These badges highlight the transparency practices applied to this work. Hover or focus on each badge to learn more about the criteria.

Open Data Dataset and code repository published with permissive license. Public repository, DOI issued, README with reproduction steps.
Reproducible Workflow Containerized environment and automated tests provided. Continuous integration pipeline with reproducibility checks.
Transparent Peer Review Peer review reports archived with DOI and linked to article. Open peer review statement and archived reports on Zenodo.