An Uncertainty-Aware Data-Driven Predictive Controller for Hybrid Power Plants

TL;DR

This work tackles the challenge of coordinating wind, solar, and storage in hybrid power plants under weather uncertainty by employing an uncertainty-aware data-driven predictive controller based on subspace predictive control (SPC). It encodes HPP dynamics from data, provides forecasts of component power, and proactively manages battery charge/discharge to support peak demand, demonstrated with real-world load profiles. The approach uses a data-driven predictor where future outputs $y_N$ arise from past measurements via $y_N = S^* y_{T_{ini}} u_{T_{ini}} u_N$, with probabilistic wind-energy constraints and a horizon of $N=20$ samples sampled every 20 s; optimization is solved with CasADi on standard hardware. Findings indicate good closed-loop tracking and useful forecasts of HPP power output, supporting potential real-world deployment and market participation, with future work on incorporating real weather forecasts and higher-fidelity models.

Abstract

Given the advancements in data-driven modeling for complex engineering and scientific applications, this work utilizes a data-driven predictive control method, namely subspace predictive control, to coordinate hybrid power plant components and meet a desired power demand despite the presence of weather uncertainties. An uncertainty-aware data-driven predictive controller is proposed, and its potential is analyzed using real-world electricity demand profiles. For the analysis, a hybrid power plant with wind, solar, and co-located energy storage capacity of 4 MW each is considered. The analysis shows that the predictive controller can track a real-world-inspired electricity demand profile despite the presence of weather-induced uncertainties and be an intelligent forecaster for HPP performance.

Figures (9)

• Figure 1: The control scheme of the hybrid power plant (HPP) considered in this work. The HPP supervisory controller ($HSC$) attempts to track a load reference $P_r$ by selecting optimal setpoints (superscript $u$) for the underlying plant components. Measured outputs (superscript $y$) are summed to obtain the overall HPP output $P_l$. $\mathbf{C}$ and $\mathbf{G}$ respectively represent controller and system dynamics, for each plant component.
• Figure 2: A sample load-tracking result. A constrained feedback-optimization based controller generates input-output data for multiple load references $P_r$ortmann2022online.
• Figure 3: Day-long forecasts for available wind and solar power to be used in the simulation study in Section \ref{['sec:results']}. Uncertainty added to the wind-speed translates to uncertainty in available wind power ($P^{\max}_{w,un}$).
• Figure 4: Open-loop uncertainty-aware setpoint selection and tracking for the wind farm. $P^{y,ol}_w$ is the open-loop response of the wind farm to setpoints $P^{u,SPC}_w$ of the predictive controller, in the presence of uncertainty in available wind power $P^{max}_{w,un}$. The error of the predicted output $P^{y,SPC}_w$ with respect to $P^{y,ol}_w$, normalized by the mean of $P^{y,SPC}_w$, is 6.5%.
• Figure 5: Open-loop setpoint selection and tracking for the solar farm and battery. $P^{y,ol}_s$ and $P^{y,ol}_b$ are respective open-loop responses of the solar farm and battery to setpoints $P^{u,SPC}_s$ and $P^{u,SPC}_b$ of the predictive controller. The error of the predicted solar farm output $P^{y,SPC}_s$ with respect to $P^{y,ol}_s$, normalized by the mean of $P^{y,SPC}_s$, is 8.5%. For the battery, this value is 10%.