Related Experiment Videos
Stacking ensemble learning for PCC voltage prediction in SEIG-ELC based off-grid micro-hydro systems
Shalini Sinha1, Mrinal Kanti Rajak2, Rajen Pudur1
1Department of Electrical Engineering, National Institute of Technology Arunachal Pradesh, Jote, Itanagar, 791123, Arunachal Pradesh, India.
Scientific Reports
|July 18, 2026
Summary
Accurate voltage prediction is crucial for off-grid systems. This study develops an ensemble learning model to forecast point-of-common-coupling voltage in self-excited induction generator systems, improving stability and control.
Area of Science:
- Electrical Engineering
- Renewable Energy Systems
- Machine Learning Applications
Background:
- Voltage stability is a significant challenge in off-grid systems, particularly those utilizing self-excited induction generators (SEIGs).
- Rural micro-hydro settings face frequent load variations and limited generation capacity, exacerbating voltage control issues.
- Accurate prediction of point-of-common-coupling (PCC) voltage is essential for system control and equipment protection.
Purpose of the Study:
- To develop and evaluate a data-driven predictive model for forecasting PCC voltage in a self-excited induction generator (SEIG) with an electronic load controller (ELC).
- To address the challenge of maintaining voltage stability under sudden load changes in off-grid micro-hydro systems.
- To quantify prediction uncertainty and demonstrate the potential of ensemble learning for enhanced voltage control.
Main Methods:
- An experimental setup utilizing an STM32F407VG microcontroller-based ELC was employed to collect real-time operational data.
- A stacking ensemble (SE) learning model was developed, integrating extreme gradient boosting, Gaussian process regression, and support vector regression, with ridge regression as the meta-learner.
- Model performance was rigorously compared against individual base learners using statistical metrics and significance tests (Wilcoxon signed-rank, Diebold-Mariano).
Main Results:
- The proposed stacking ensemble model achieved superior predictive performance with minimal error compared to individual machine learning models.
- Prediction uncertainty was effectively quantified using a probabilistic Gaussian process, providing confidence bounds for the estimated PCC voltage.
- Data augmentation techniques were explored but found to be unnecessary, confirming the sufficiency of the original dataset for accurate predictions.
Conclusions:
- Ensemble learning models demonstrate significant potential for improving voltage prediction accuracy in SEIG-ELC systems.
- The developed model can support advanced control strategies for enhancing voltage stability in off-grid micro-hydro applications.
- Accurate voltage forecasting is key to ensuring reliable operation and safety in decentralized power systems.