aeolus header

Newsletter 20 January 2011

Well into the third year of the project, focus is now on validation and site-experiments related to the wind farm control results. Access to the Thanet wind farm has been provided by Vattenfall and site-experimentation is planned for spring-summer 2011.

Other recent results include:

  • Studies have shown that the statistical relation between wind speeds measured in points is weak. A new method where the turbines are used as wind measuring devises has therefore been developed. In effect this averages the wind speed over the rotor disc. This new type of wind speed measure is denoted effective wind speed (EWS). The EWS for neighboring turbines has been modeled using batch, recursive and adaptive system identification. This shows a better predictability compared to point wind speeds.

  • The implementation of both supervisory and reconfigurable control in software, and it's testing in the case study simulation, is in progress. A key aspect of any multivariable model based predictive control application is the derivation of an appropriate cost function for minimisation by the control algorithm. In the absence of any obtained a priori work in this area, much effort has been devoted to deriving an appropriate cost function that has regard for the turbine fatigue load (tower bending etc.) across the wind farm while also ensuring that the network power demand is met through power distribution over the individual available turbines.

  • Global performance bounds have been derived from distributed criteria using dual decomposition for dynamical systems. The decomposition is based on local cost functions for each turbine, together with prices to be paid by turbines that disturb the wind flow of their neighbours. Price functions have been constructed such that global performance guarantees can be recovered from local specifications for each turbine. Price based methods for distributed controls have been further developed in the context of Model Predictive Control (MPC). These methods require rather heavy communication between the turbines, since prices need to be exchanged and updated on-line. Hence, we are also developing less demanding methods, where prices are used only at a slower time-scale.

  • Simulation have been developed and implemented in Matlab that captures the main physical phenomena of a wind farm which includes the ambient wind, wake deficits, wake meandering, the wind turbine system, the wind farm controller, and a network operator. Combined they provide a unique wind farm simulation capability.

  • Models play an important role, but Aeolus also aims to demonstrate the developed algorithms on a real wind farm. To support this activity, a plan for assessment of relative performance of control strategies has been formulated. The work is formulated as a cost function for online validation of control performance.

  • To support dissemination in the wind turbine community a side-event was organized at the European Wind Energy Conference. Unfortunately, ash clouds cancelled the event. An event is scheduled for March 2011.

Newsletter 11 November 2009

After the review meeting in June 2009 hosted by Vestas Wind Systems A/S, the project has entered a stage where focus is on dynamic models and the control. Central to the project is a common understanding of models and the effort continues to develop the wind farm simulator to a level where it can serve as a benchmark. Financially the projected effort has been adjusted to reflect actual person month costs, adding a total of 55 person months to the project.

Work progress and achievements during the period

During the previous 6 months, the following results have been achieved, aimed at meeting the overall objectives of the project:

  • Quasi-static flow models: work on the quasi-static models is on-going and the model being scaled to provide flow information for large-scale wind farms and to allow calculations of the expected electrical power output and the expected mechanical loads.

  • Dynamic flow models: The notion of effective wind speed was introduced, and promising results on the ability to model and predict wind speeds based on effective wind speed was achieved.

  • Supervisory control: Model Predictive Control (MPC) was selected as an appropriate technology for the wind farm problem. Linearization of relevant models of turbines and flow. In order to evaluate the methodology on wind farms, models suited for control as well as a cost function are needed.

  • Decentralized control: The price based methods for distributed control have been further developed in the context of Model Predictive Control (MPC). In order to evaluate the methodology on wind farms, models suited for control as well as a cost function are needed.

  • Case study and dissemination: In order to strengthen the cooperation between partners a farm simulator capable of simulating different park configurations and turbines is being developed. A statement of work for the simulator has been defined by negotiations with the project partners. The statement of work analyses the current state of the simulator and specifies the capabilities and requirements for the final simulator.

Results

Public publications are available on the publications page.