Researchers propose a new way to move floating offshore wind turbines so they block each other less. The method is model-free, meaning it does not rely on a detailed, central model of the whole wind farm. It instead uses a distributed online optimization scheme that operates with local decision rules and updates over time to reduce the wake effect — the reduced wind speed behind a turbine that can lower the output of downstream turbines.

To build the controller, the team combined a distributed optimization framework with physics-informed knowledge of wake dynamics. In other words, the controller does not depend on a precise numerical model of the whole farm, but it does use basic, physically motivated information about how wakes form and move. This helps the system deal with two important practical challenges: the underlying control problem is nonconvex (it has many possible configurations and local optima), and gradient information — the exact slope of performance with respect to turbine positions — may be unavailable.

At a high level, the method lets turbines or local controllers make positioning decisions online and in a distributed way rather than sending all data to a central planner. Because it is model-free, the approach is intended to be more feasible in practice when a full, accurate model of a floating wind farm is hard to obtain. The physics-informed elements guide the local updates so the distributed optimization can be effective even without gradient data.

The authors tested their approach in mid-fidelity simulations and compared it to a conventional, centralized, model-based control baseline. According to the reported results, the distributed model-free method achieved slightly better performance in reducing wake overlap, while using much less computational effort than the centralized approach.