TL;DR
This paper introduces a novel "distributionally robust" optimal control method for wind turbines that minimizes flap actuation power while guaranteeing that critical blade loads remain within safe probabilistic bounds. This approach offers a more principled way to handle atmospheric turbulence and achieve significant load reduction compared to traditional LQG controllers.
What Problem Does This Paper Address?
As wind turbines grow larger, their blades become more susceptible to extreme loads from atmospheric turbulence and gusts. These loads can lead to safety concerns and increased structural degradation. While active control methods like pitch actuation exist, they are often limited to lower frequencies. Trailing-edge flaps offer a promising solution for addressing higher-frequency loads, but designing effective controllers that balance flap power consumption with load mitigation is challenging. Existing methods often treat load constraints as tuning parameters rather than fundamental design objectives, and they may not adequately account for the inherent uncertainty in atmospheric turbulence.
Methods
The authors propose a "distributionally robust" optimal control synthesis approach. This method is framed as a constrained linear quadratic Gaussian (LQG) control problem. The core innovation lies in its ability to minimize a cost function (flap actuation power) while simultaneously enforcing probabilistic constraints on blade loads (root bending moment, tip deflection, and torsion).
Key aspects of the methodology include:
- Distributionally Robust Control: This framework allows for a more general model of atmospheric turbulence than the standard Gaussian assumption. It focuses on ensuring constraints hold for a set of possible turbulence distributions, defined by their mean and covariance. This makes the control policy more robust to uncertainties in the turbulence model.
- Constrained LQG: The approach synthesizes a controller that is optimal in a quadratic sense (minimizing actuation power) but also explicitly incorporates constraints on blade loads. This is achieved by solving a semi-definite program (SDP) to find the optimal controller gains.
- Aeroservoelastic Modeling: The wind turbine's dynamics are modeled using the SHARP framework, which couples a structural model (composite beam) with an aerodynamic model (Unsteady Vortex Lattice Method - UVLM). This provides a comprehensive representation of the turbine's behavior.
- Model Reduction: Due to the high dimensionality of the full system model, model reduction techniques (balanced truncation) are employed to obtain a smaller, computationally tractable model for controller synthesis.
Datasets
The paper uses a standard NREL 5-MW reference wind turbine model for its simulations and evaluations. No new datasets are introduced.
Key Insights and Ideas
- Principled Handling of Turbulence: The distributionally robust approach offers a more theoretically sound way to deal with the inherent uncertainty of atmospheric turbulence, moving beyond simple Gaussian assumptions.
- Guaranteed Load Bounds: The controller is designed to guarantee that blade loads remain within specified probabilistic bounds by construction, rather than relying on post-hoc tuning. This is a significant advantage for safety-critical applications.
- Trade-off Optimization: The method explicitly balances the minimization of flap actuation power with the strict enforcement of load constraints, providing a clear and quantifiable trade-off.
- Simpler Tuning: The authors suggest their approach requires less tuning than traditional LQG methods because the design objectives and constraints are directly incorporated into the optimization problem.
Key Results
The paper demonstrates that the proposed optimal control policy (K*) achieves significant reductions in blade loads compared to a naively tuned LQG controller (K_LQG) for the same actuation power.
- Load Reduction: For a stationary operating condition (ω_G = (0;0; 1.3) rad/s), K* reduced the root bending moment (My) by approximately 35% and the tip deflection (qz) by about 50% compared to K_LQG, while meeting all probabilistic constraints.
- Constraint Satisfaction: The proposed controller successfully met all probabilistic constraints (e.g., P{−M_y ≤ My(k) ≤ M_y} ≥ 1 − ε) for the specified probability levels.
- Actuation Power: The K* controller achieved these load reductions with comparable or even lower actuation power compared to K_LQG, which often failed to meet the constraints.
- Robustness: The performance improvements were shown to be consistent across different turbulence levels and operating conditions.
Limitations and Future Work
- Model Assumptions: The UVLM model relies on assumptions of incompressible flow, attached flow, and small angles of attack, which may not hold under all extreme conditions.
- Computational Complexity: While the SDP formulation is efficient, synthesizing controllers for very high-dimensional systems can still be computationally intensive, necessitating model reduction.
- Future Work: The authors suggest that further investigation into the impact of actuator dynamics and more complex turbulence models could enhance the controller's performance. Extending the approach to handle multiple turbines in a wind farm is also a potential future direction.
Summary updated