arXiv:2610.26055v2 Announce Type: replace Abstract: Wind farms can provide flexibility by curtailing their output during periods of excess generation in the power grid. One way to achieve this is through implicit balancing, where curtailment is performed in anticipation of a negative imbalance price. By doing so, the wind farm helps reduce system imbalance while also lowering its own costs. However, implicit balancing is challenging, since imbalance prices are hard to predict and can also be affected by the control actions taken by the farm. To address both challenges, we use distributional soft actor-critic (distSAC) to learn a control policy that adjusts the power setpoint every minute. Using real data of the Belgian grid and an offshore wind farm, we show that our method can turn the imbalance costs caused by forecast errors into a small revenue. Moreover, distSAC outperforms a price threshold controller, often used in industry, and performs slightly better than soft actor-critic (SAC). Unlike SAC, distSAC can also learn risk-averse policies by optimizing a risk measure, such as conditional value at risk. These risk-averse policies were found to better account for their price impact, reducing overshoots by 16-27% while achieving similar cost savings. Last, we find that 1-minute setpoints lead to lower imbalance costs than 15-minute setpoints, regardless of the control policy.