Simulation, optimal control, and control co-design of wind and marine turbines using derivative function surrogate models

Abstract

Researchers in industry and academia are investigating different approaches to identify cost-effective, system-level optimal designs of wind and marine turbines, which would lead to wide-scale adoption of these technologies. Central to these efforts are open-source computational modeling tools and reference models of these systems. Researchers typically propose improvements to the reference models, implement these changes using the modeling tools, and compare the performance to the baseline performance. As a part of the design process, researchers have investigated the impact the controls have on the performance of these systems. A controller is necessary to ensure that these systems can produce power and minimize loads in offshore environments. In addition to the controllers, researchers are utilizing design optimization studies to efficiently explore the design space associated with these systems, including identification of optimal designs and non-intuitive trade-offs. Additionally, control co-design (CCD) studies, where both the physical aspects and the controller are optimized simultaneously, have been identified as a potential pathway to minimize the cost of energy associated with these systems. Since wind and marine turbines are at different technology readiness levels, the key focus of the research being carried out for them is different. A key requirement for marine turbines is to develop open-source modeling tools and reference models. Recently, OpenFAST, an open-source tool used for wind turbines, has been extended for the simulation of marine turbines. An open-source reference model of a floating marine turbine has been developed as well. However, to effectively simulate this model using OpenFAST, a controller is required. For wind turbines, there are multiple open-source reference models available. A key research focus is to formulate and solve multi-objective optimization studies with a focus on the controller, to explore the trade-offs between multiple conflicting objectives. To aid this, design tools built on OpenFAST have been developed. However, a main drawback of utilizing OpenFAST for design optimization studies is the computational cost. Depending on the system, it can take anywhere between 10 minutes to 7 hours to simulate OpenFAST. Design optimization studies require several hundred function evaluations to identify the optimal design. Therefore, utilizing OpenFAST directly can be ineffective. The core contribution of this dissertation is to address this issue. To address this issue, different data-driven low-fidelity or surrogate modeling approaches are explored. A variety of approaches are considered, ranging from system identification approaches to recurrent neural networks. In addition to these, an approach called a derivative function surrogate model (DFSM), which has been utilized in recent studies to approximate the dynamic response of wind turbines, is also explored. A novel approach is developed to construct a DFSM that could be used for simulation and control optimization studies. Low-fidelity models using these approaches are constructed, and the trade-offs associated with each approach are explored. Three factors inform the trade-offs between these models. The simulation time using these models, how extendable these modeling approaches are towards modeling the system response for different wind speeds, and the variance in response of these models for different signals. Results show that the proposed DFSM approach balances computational time and model accuracy better than the system identification and deep learning based models. The DFSM approach is tested for five different systems, which include four different floating wind turbine systems and a marine turbine system. The DFSM results in nearly 12 to 700 times speedup in terms of simulation time, in comparison to the high fidelity model across these five systems. Then, different case studies are carried out that explore the efficacy of the DFSM for optimal control and multi-objective CCD studies. The key takeaways from these studies can be summarized as follows. From the open-loop optimal control study carried out for a marine turbine system, it was found that for a 5% increase in the generated power, there is a corresponding 15% increase in the tower base fatigue loads. Using a multi-fidelity optimization approach for closed-loop optimal control studies, it is possible to identify a set of controller parameters that are within 0.5% of the high-fidelity maxima, with nearly a quarter of the number of high-fidelity function calls. And finally, a multi-objective CCD study is carried out to identify plant and controller parameters that balance conflicting objectives of tower-base fatigue load minimization and improving power quality for floating wind turbines. The study considers a single plant variable, namely the column spacing of the semisubmersible platform, and the controller parameters correspond to the above-rated blade pitch controller. A nested optimization approach is used to solve this problem, and for every iteration of the inner loop, a multi-objective controller optimization problem is solved. The identified point has just 2% higher fatigue damage loading than the optimal case, while just having 15% higher value of power quality metric.