Stochastic Model Predictive Control of Charging Energy Hubs with Conformal Prediction
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911128874909696 |
|---|---|
| author | Fernández-Zapico, Diego Hofman, Theo Salazar, Mauro |
| author_facet | Fernández-Zapico, Diego Hofman, Theo Salazar, Mauro |
| contents | This paper presents an online energy management system for an energy hub where electric vehicles are charged combining on-site photovoltaic generation and battery energy storage with the power grid, with the objective to decide on the battery (dis)charging to minimize the costs of operation. To this end, we devise a scenario-based stochastic model predictive control (MPC) scheme that leverages probabilistic 24-hour-ahead forecasts of charging load, solar generation and day-ahead electricity prices to achieve a cost-optimal operation of the energy hub. The probabilistic forecasts leverage conformal prediction providing calibrated distribution-free confidence intervals starting from a machine learning model that generates no uncertainty quantification. We showcase our controller by running it over a 280-day evaluation in a closed-loop simulated environment to compare the observed cost of two scenario-based MPCs with two deterministic alternatives: a version with point forecast and a version with perfect forecast. Our results indicate that, compared to the perfect forecast implementation, our proposed scenario-based MPCs are 13% more expensive, and 1% better than their deterministic point-forecast counterpart |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_00685 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Stochastic Model Predictive Control of Charging Energy Hubs with Conformal Prediction Fernández-Zapico, Diego Hofman, Theo Salazar, Mauro Systems and Control This paper presents an online energy management system for an energy hub where electric vehicles are charged combining on-site photovoltaic generation and battery energy storage with the power grid, with the objective to decide on the battery (dis)charging to minimize the costs of operation. To this end, we devise a scenario-based stochastic model predictive control (MPC) scheme that leverages probabilistic 24-hour-ahead forecasts of charging load, solar generation and day-ahead electricity prices to achieve a cost-optimal operation of the energy hub. The probabilistic forecasts leverage conformal prediction providing calibrated distribution-free confidence intervals starting from a machine learning model that generates no uncertainty quantification. We showcase our controller by running it over a 280-day evaluation in a closed-loop simulated environment to compare the observed cost of two scenario-based MPCs with two deterministic alternatives: a version with point forecast and a version with perfect forecast. Our results indicate that, compared to the perfect forecast implementation, our proposed scenario-based MPCs are 13% more expensive, and 1% better than their deterministic point-forecast counterpart |
| title | Stochastic Model Predictive Control of Charging Energy Hubs with Conformal Prediction |
| topic | Systems and Control |
| url | https://arxiv.org/abs/2504.00685 |