Chapter 12: Model Fidelity, Surrogates, and Data-Driven CCD

Obtaining useful models without making optimization impossible¶

A useful CCD model preserves decision-driving couplings while remaining inexpensive, differentiable, and robust enough for repeated optimization. Fidelity is a managed hierarchy.

Learning objectives¶

After completing this chapter, you should be able to:

1. explain and apply high-fidelity simulation;

2. explain and apply training data;

3. explain and apply dynamic surrogate;

4. explain and apply optimization;

5. formulate and verify the chapter methods on a floating wind turbine modeled in OpenFAST and with reduced, LPV, and derivative-function surrogates.

Mathematical lens¶

The recurring quantities are high-fidelity $F_H$, surrogate $F_L$, design $p,c$, and model error $e_M$:

Running example¶

The recurring example is a floating wind turbine modeled in OpenFAST and with reduced, LPV, and derivative-function surrogates. Retaining one system prevents apparent improvements from being caused by changed physics, information, loads, or metrics.

Recommended workflow¶

1. sample high fidelity.

2. identify reduced model.

3. train surrogate.

4. validate trajectories.

5. optimize and confirm.

Chapter map¶

1. Analysis Models Versus Optimization Models

2. Control-Oriented Modeling

3. Linearization and Local Models

4. Reduced-Order Models

5. Response-Surface Surrogates

6. Dynamic Surrogate Models

7. Derivative-Function Surrogate Models

8. Linear Parameter-Varying Models

9. Gaussian Processes and Neural Networks

10. Multi-Fidelity Optimization

11. Model-Error Assessment and Trust

12. High-Fidelity Validation