Value-oriented policy distillation for interpretable wind farm power control

Max Bruninx; Timothy Verstraeten; Jan Decuyper; Ann Nowé; Jan Helsen

cs.AINov 5, 2026

Value-oriented policy distillation for interpretable wind farm power control

Authors: Max Bruninx, Timothy Verstraeten, Jan Decuyper, Ann Nowé, Jan Helsen

Organizations: Vrije Universiteit Brussel, Brussels, Belgium

Abstract

Wind farm power control can help the system maintain balance and avoid blackouts. In Europe, this can be achieved through implicit balancing, where a wind farm adjusts its power setpoint in anticipation of the imbalance price using information published by the Transmission System Operator. Thereby, the wind farm helps restore system balance while also reducing its own imbalance costs resulting from forecast errors. However, implicit balancing is a challenging problem since the settlement price is unknown in advance and can be impacted by the pursued control strategy. Reinforcement learning methods could help to solve this problem, but lack the interpretability needed to be deployed on real wind farms. Therefore, we propose a policy distillation framework using explainable boosting machines (EBMs), a variant of generalized additive models, to obtain a transparent control policy. To increase the value captured by the EBM model, we introduce a value-oriented perspective to policy distillation inspired by research in the field of energy forecasting. Furthermore, we suggest to include expert knowledge during model design, such that the control strategy is also justifiable to stakeholders. Using real data of the Belgian grid and an offshore wind farm, we find that our value-oriented and expert-based controller can reduce imbalance costs by 82%, while outperforming a price threshold controller as well as a policy distillation baseline using decision trees.

Figures & tables

	MAE	RMSE	$R^{2}$
DT (depth=2)	0.21	0.29	0.38
DT (depth=3)	0.21	0.27	0.45
EBM (full)	0.18	0.23	0.61
EBM (expert)	0.19	0.24	0.57
EBM (expert + value)	0.19	0.24	0.57
LightGBM	0.15	0.19	0.72

Table 1: Performance evaluation metrics across the interpretable models and a LightGBM baseline. Mean absolute error (MAE) and root mean squared error (RMSE) are normalized by the capacity of the wind farm. Lower scores indicate a better fit for MAE and RMSE, while higher scores are better for

R^{2}

	Mean	Min	Max	P05	P95
No control	3.22	-73.01	211.77	-6.55	22.13
Price threshold	1.47	-129.73	197.53	-8.39	16.30
DT (depth=2)	1.21	-105.74	193.24	-9.72	16.01
DT (depth=3)	0.75	-116.04	193.24	-13.17	15.74
EBM (full)	0.36	-99.00	193.24	-14.13	15.56
EBM (expert)	0.63	-73.01	193.24	-13.89	16.30

Table 2: Observed imbalance costs (+) or revenues (-) for the different policies over the evaluation period. The results are presented in € per MW installed per settlement period.

Figure 1: Visualization of the learned shape function for the latest published imbalance price estimate

\hat{\lambda}^{\mathrm{B}}_{t}

. The score illustrates the effect the variable has on the final prediction. A negative score implies that this variable reduces the power setpoint and vice versa.

Figure 2: Visualization of the learned shape function for the interaction effect between the system imbalance forecast $\hat{SI}^{\mathrm{ISP}}_{t}$ and the minute of the settlement period $t^{\mathrm{ISP}}$ . A negative score implies that this interaction effect reduces the power setpoint and vice versa.

Figure 3: Example of local explanation of the interpretable policy visualized as a waterfall chart. The control action in this example is to reduce the power setpoint by

35.2

MW compared to the trading schedule.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

	Method	MAE	RMSE	$R^{2}$
SDT (depth=2)	Mean	0.20	0.26	0.50
SDT (depth=3)	Mean	0.18	0.24	0.56
SDT (depth=2)	Max	0.26	0.33	0.18
SDT (depth=3)	Max	0.24	0.31	0.27

Appendix

Table 3: Performance evaluation metrics for the SDT models. The inference method is specified in the ‘Method’ column (cf., below). Mean absolute error (MAE) and root mean squared error (RMSE) are normalized by the capacity of the wind farm. Lower scores indicate a better fit for MAE and RMSE, while higher scores are better for

R^{2}

Power Systems

Power systems research studies the planning, operation, control, and protection of electrical grids and their integration with energy resources.

See all 239 papers

Noise Steering: Exploiting Wake Steering and Acoustic Directivity for Noise-Constrained Wind Farm Control via Reinforcement Learning
Oct 27, 2026Martín de Frutos, Baptiste Corban, Claire Bizon Monroc +4Optimal Control RL Control
ETSIAE-UPM - School of Aeronautics, Universidad Politécnica de Madrid, Plaza Cardenal Cisneros 3, Madrid, E-28040, Spain · Inria, 48 rue Barrault, Paris, 75013, France · DI ENS, Ecole Normale Supérieure, PSL Research University, Paris, France +3
Accelerating Reinforcement Learning for Wind Farm Control via Expert Demonstrations
Apr 13, 2026Marcus Binder Nilsen, Julian Quick, Tuhfe Göçmen +2RL Control Offline-To-Online RL
Department of Wind and Energy Systems, Technical University of Denmark, Roskilde, Denmark
Load constrained wind farm flow control through multi-objective multi-agent reinforcement learning
Apr 13, 2026Teodor Åstrand, Marcus Binder Nilsen, Iasonas Tsaklis +3Multi-Agent Reinforcement Learning Constrained RL
Department of Wind and Energy Systems, Technical University of Denmark, Roskilde, Denmark

Oct 27, 2026J/K move · Enter open · S save

Noise Steering: Exploiting Wake Steering and Acoustic Directivity for Noise-Constrained Wind Farm Control via Reinforcement Learning

Martín de Frutos, Baptiste Corban, Claire Bizon Monroc, Ana Bušić+3

ETSIAE-UPM - School of Aeronautics, Universidad Politécnica de Madrid, Plaza Cardenal Cisneros 3, Madrid, E-28040, Spain · Inria, 48 rue Barrault, Paris, 75013, France · DI ENS, Ecole Normale Supérieure, PSL Research University, Paris, France+3

Optimal Control RL Control

Open paper AskPDF arXiv

Existing noise mitigation strategies for wind farms usually consist in reducing turbines power output, without accounting for the wake interactions between them and the noise emission directivity. This leads to over-conservative policies and unexploited wind power. Yet, turbine yaw can simultaneously affect wake interactions and the direction of noise emission, offering an additional degree of freedom for farm-level noise control. We therefore propose Noise Steering, a model-free reinforcement learning (RL) strategy that jointly controls turbine yaw angles and power setpoints to maximise farm power while satisfying a local noise constraint in a defined region. To enable training at a reduced computational cost, a low-fidelity acoustic model is coupled with the steady-state FLORIS simulator. The RL agent effectively learns how to redirect wakes and noise directivity away from the protected region while redistributing power across the farm. On a six-turbine test farm, the policy satisfies the 40 dB noise constraint throughout the full range of turbulent winds, with an average power deficit of only 2% relative to greedy unimpeded control, and up to 7.7% at high wind speeds. In comparison, the industry-standard noise-reduced operation (NRO) baseline here generates an average power loss of 17%. When deployed in the higher-fidelity FASTFarm simulator, the policy largely retains its performance under steady and turbulent inflow, despite differences in simulation fidelity. These results demonstrate the potential of combining yaw control and power derating through RL to improve the power-noise trade-off in wind farm operation.

Abstract

Power Systems

Explore similar work

Noise Steering: Exploiting Wake Steering and Acoustic Directivity for Noise-Constrained Wind Farm Control via Reinforcement Learning

Accelerating Reinforcement Learning for Wind Farm Control via Expert Demonstrations

Load constrained wind farm flow control through multi-objective multi-agent reinforcement learning