younesassini1998-ai/global-renewable-energy-forecasting-cnn-lstm: First stable research release

Abstract

🚀 Release v1.0.0 — Official Research Implementation This is the initial official release providing the complete, reproducible source code and experimental pipeline for the research paper: "A Hybrid CNN-LSTM Framework for Global Renewable Energy Forecasting: Integrating Socio-Economic Drivers, Explainability, and a Case Study on Morocco" 📌 Key Highlights & Features Empirical Dataset Pipeline: Preprocessing and linear interpolation pipeline built for global macroeconomic and energy data (1990–2023, ~200 countries, 6,596 time-series sequences) from Our World in Data (OWID). Model Architectures & Benchmark: End-to-end implementation of: Proposed Hybrid CNN-LSTM model (Spatial-temporal feature extraction) Bi-LSTM (Bidirectional Long Short-Term Memory) GRU (Gated Recurrent Unit) Random Forest (Statistical Baseline) Temporal Rigor: 5-Fold expanding window Time Series Cross-Validation (TimeSeriesSplit) evaluating temporal generalization across historical periods. Explainable AI (XAI): Global feature attribution analysis using SHAP (KernelExplainer) quantifying the predictive weight of demographic, economic, and energy intensity variables. Hypothesis Testing: Non-parametric Wilcoxon signed-rank test confirming statistical significance ($p < 10^{-60}$) of model gains. Regional Application: Dedicated validation on Morocco's national energy trajectory ($R^2 = 0.8650$). 📊 Summary of Experimental Results | Metric / Experiment | Value | | :--- | :--- | | Global Test Set $R^2$ (CNN-LSTM) | 0.9951 | | Global MAE / MAPE | 0.0007 / 10.14% | | 5-Fold Time Series Cross-Validation ($R^2$) | 0.8481 ± 0.1546 | | Morocco Case Study ($R^2$) | 0.8650 | | Wilcoxon Significance vs. Baselines | $p < 10^{-60}$ (Significant) | 📦 What's Included in This Release notebooks/energy_forecasting_pipeline.ipynb: Complete executable workflow (preprocessing, training, evaluation, SHAP, cross-validation). requirements.txt: Environment dependencies specification. README.md: Reproduction guidelines and documentation. LICENSE: MIT License. 📄 Citation If you use this codebase or model architecture in your research, please cite: @misc{renewable_energy_forecasting_2026, author = {Younes Assini}, title = {A Hybrid CNN-LSTM Framework for Global Renewable Energy Forecasting}, year = {2026}, publisher = {Zenodo}, version = {v1.0.0}, doi = {10.5281/zenodo.XXXXXXX}, url = {https://github.com/YOUR_USERNAME/global-renewable-energy-forecasting} }