Wednesday, 28 September 2016
14:30 - 16:00 Advanced control strategies for wind plants
Turbine technology  
          

Modern control design methods like non-linear model predictive control can take account of complex dynamics and actuator constraints, while making use of richer information input from advanced sensors such as LIDARs. They contribute to the coordinated control of entire wind farms while meeting new operational demands from the electricity system. The session addresses advanced state estimation methods, an important component of such controllers. You will hear about field test results from a controller using wind preview from a LIDAR sensor as well as changes in design and certification guidelines needed to account for the use of LIDARs.

You attended this session?

Please give us feedback

 

Learning objectives

• Advanced non-linear state estimation techniques for wind turbines that run in real time;
• Non-linear model predictive control that rung in real time and improve performances;
• A LIDAR-assisted flatness-based controller validated in field tests;
• Requirements for design and certification of wind turbines with LIDAR assisted control;
• A wind farm control framework that can mitigate wake effects and provide grid balancing services.

Co-chair(s):

Bastian Ritter Industrial Science GmbH, Germany
Co-authors:
Bastian Ritter (1) ^{F P} Axel Schild (2) Ulrich Konigorski (3)
(1) Industrial Science GmbH, Darmstadt, Germany (2) IAV, Gifhorn, Germany (3) Darmstadt University of Technology, Darmstadt, Germany

Making nonlinear state estimation techniques ready for use in industrial wind turbine control systems

Introduction

To meet the steadily increasing industrial requirements, wind turbine controllers have reached a mature level of closed-loop performance with respect to energy produc-tion and load reduction. These controllers are nonlinear, complex and build heavily upon design heuristics gained from practical experiences. Unfortunately, the latter significantly complicates direct extension of field-proven controllers to new ad-vanced turbine designs without major modifications.
On the other hand, it is even possible to outperform such strong heuristics with advanced multivariable controllers, if only precise information about the dynamic wind turbine state were available. Such information is indeed generated by modern powerful estimation algorithms which handle most practical challenges satisfactorily.


Approach

The presented work falls into the category of turbine technology and focuses on leveraging the potential of modern control methodologies for improving wind turbine operation. It discusses and assesses the industrial applicability of nonlinear Bayesian estimation techniques to reconstruct online important state information. The reliability, precision and ubiquitous availability of this information paves the path for employing powerful control algorithms like model predictive control (MPC). In contrast to existing works, emphasis is put on demonstrating real-time feasibility, assessment of practical observability for nonlinear hi-fidelity models and evaluation of benefits from additional non-standard measurements.

Main body of abstract

Thanks to the enormous advances in optimization theory and computational power available on today's industrial control hardware, powerful approaches like MPC are about to become reality in many industrial sectors. Especially the wind energy sector with its moderate requirements in terms of sampling times is predestined for the deployment of such optimization-based concepts.

A major obstacle for the reliable utilization is the requirement for accurately known full dynamic state, external disturbances and critical system parameters. Nonlinear Bayesian estimators like Unscented Kalman Filters (UKF) provide a suitable tool to generate this information online with a hi-fidelity control-oriented turbine model covering the relevant drive-train, tower and blade dynamics, and the known history of measurements from standard instrumentation. Despite the existing rich body of theory and free-of-charge academic tools, state estimation has so far received only little attention in the wind energy community.

As a central contribution, this paper demonstrates that nonlinear estimation based on a monolithic UKF can be conducted with sufficient precision in real-time on modern embedded industrial controllers and standard instrumentation like speed sensors, IMUs and wind anemometer. The filter reconstructs accurately the desired state information based on realistic simula-tion data according to IEC-64100 scenarios and provides in addition important design-relevant turbine loads like tower-base and blade-root bending moments. A thorough field-test is yet to be conducted.

Moreover, the practical (local) observability of the employed model is assessed thor-oughly which is a very important step prior to observer design, often neglected in many existing publications though. The paper closes with a summary of further critical issues of nonlinear estimators for wind turbine applications.



Conclusion

The investigation of powerful recursive nonlinear algorithms for state estimation of wind turbines proves to be real-time feasible already. This is demonstrated by estimation results for a standard 5-MW reference turbine and hardware testing. Incorporation of non-standard measurements is recommended since blade root sensors show a significant potential for both improved estimation quality and control performance. The body of research yet provides all required features to successfully tackle practical challenges like robustness of implementation and constraint-handling. The main challenge remaining is the integration of these features into an industry-ready platform which is an ideal field of activity for premium technology suppliers.


Learning objectives
This contribution will convey the following messages:

• Nonlinear state estimation for wind turbine is real-time feasible on embedded controllers and is thus ready for use
• Monolithic model-based estimators are suboptimal for practical implementation. It's all about developing the right estimation architecture.
• Choosing the right modeling-detail and thoroughly studying model properties like observability are prerequisites for a successful application