Adaptive time-series based wind farm optimization (bibtex)

by Schmidt, Jonas, Vollmer, Lukas, Requate, Niklas and Meyer, Tobias

Abstract:

One widely used approach in wind farm control is based on separate sector-based optimizations for given inflow conditions per sector. The aim of these optimizations is to maximize power generation by reducing power of upwind turbines or by offsetting their yaw angle to reduce the wake effect on downwind turbines. Commonly, the optimization problem is solved by assuming steady-state wind conditions (wind speed, direction, turbulence intensity, etc). During actual wind farm operation, this assumption does not hold. Instead, continuous adaptation of all turbines is required. In many approaches, the wind farm adaptation itself is based on an evaluation of inflow conditions and a selection of the best-fitting optimal solution. Therefore, the yawing dynamics are neglected inside of the optimization and only considered in a separate subsequent step (Bossanyi 2018; Doekemeijer 2020). Failures or planned work in a wind farm imply changes in the wind farm configuration. Failures can include turbines running in local greedy control, e.g. due to a communication system fault that disconnects them from the wind farm controller. Planned work on turbines, e.g. maintenance, leads to a shutdown of individual turbines. These temporary changes must be considered. Adaptation dynamics are mostly limited by the yawing dynamics of the wind turbines. For a sufficiently slow time step between two consecutive optimal solutions, yawing dynamics can be neglected. However, for faster adaptation cycles they must be considered. Another problem that is particularly difficult to resolve with a static sector-based optimization is the symmetry problem for wake steering, where a wake can be steered away from a downwind turbine to the same effect by either misaligning the upwind turbine with a yaw offset to the right or to the left. In sector-based optimization, these solutions are equal. However, in operation, a situation might arise where a left-steering solution is best for one time step, but a right-steering solution is best for the following time step. Such sudden flipping changes in yaw angle should be avoided. We propose to solve these issues by periodically updating the turbine parameters in regular time steps, e.g., 10 minutes, cf. Fig. 1. Each parameter update is the result of an optimization that takes the current wind farm state into account and avoids large jumps in parameter space. This is realized by incorporating a penalty for large deviation from the current values. A crucial ingredient for our approach is the short-term prediction of the wind conditions for the next time step. In this work a very simple local weather prediction model is considered, based on the set of previously monitored conditions at each turbine. During each step, the wind farm model is reconfigured to closely model the actual wind farm in this time instant. For turbines which are shut off or which are running in local greedy control, their behavior is included in the model. The optimization of the next parameter step is then run based on the predicted upcoming wind conditions and the reconfigured wind farm model. For each wind prediction time step, the energy-optimal solution is computed independently. Our approach is similar to Model-Predictive Control and could be enhanced by a yaw dynamics model in the future. This proof-of-principle type paper demonstrates the functionality and the benefit that arises from the described approach based on widely tested in-house simulation tools.

Reference:

Schmidt, J.; Vollmer, L.; Requate, N.; Meyer, T.: Adaptive time-series based wind farm optimization. Conference presentation, 2021. (Slides available at: http://www.tobi-meyer.de/Schmidt_2021.pdf)

Bibtex Entry:

@misc{Schmidt_2021, howpublished = {Presentation}, type={Conference presentation}, abstract = {One widely used approach in wind farm control is based on separate sector-based optimizations for given inflow conditions per sector. The aim of these optimizations is to maximize power generation by reducing power of upwind turbines or by offsetting their yaw angle to reduce the wake effect on downwind turbines. Commonly, the optimization problem is solved by assuming steady-state wind conditions (wind speed, direction, turbulence intensity, etc). During actual wind farm operation, this assumption does not hold. Instead, continuous adaptation of all turbines is required. In many approaches, the wind farm adaptation itself is based on an evaluation of inflow conditions and a selection of the best-fitting optimal solution. Therefore, the yawing dynamics are neglected inside of the optimization and only considered in a separate subsequent step (Bossanyi 2018; Doekemeijer 2020). Failures or planned work in a wind farm imply changes in the wind farm configuration. Failures can include turbines running in local greedy control, e.g. due to a communication system fault that disconnects them from the wind farm controller. Planned work on turbines, e.g. maintenance, leads to a shutdown of individual turbines. These temporary changes must be considered. Adaptation dynamics are mostly limited by the yawing dynamics of the wind turbines. For a sufficiently slow time step between two consecutive optimal solutions, yawing dynamics can be neglected. However, for faster adaptation cycles they must be considered. Another problem that is particularly difficult to resolve with a static sector-based optimization is the symmetry problem for wake steering, where a wake can be steered away from a downwind turbine to the same effect by either misaligning the upwind turbine with a yaw offset to the right or to the left. In sector-based optimization, these solutions are equal. However, in operation, a situation might arise where a left-steering solution is best for one time step, but a right-steering solution is best for the following time step. Such sudden flipping changes in yaw angle should be avoided. We propose to solve these issues by periodically updating the turbine parameters in regular time steps, e.g., 10 minutes, cf. Fig. 1. Each parameter update is the result of an optimization that takes the current wind farm state into account and avoids large jumps in parameter space. This is realized by incorporating a penalty for large deviation from the current values. A crucial ingredient for our approach is the short-term prediction of the wind conditions for the next time step. In this work a very simple local weather prediction model is considered, based on the set of previously monitored conditions at each turbine. During each step, the wind farm model is reconfigured to closely model the actual wind farm in this time instant. For turbines which are shut off or which are running in local greedy control, their behavior is included in the model. The optimization of the next parameter step is then run based on the predicted upcoming wind conditions and the reconfigured wind farm model. For each wind prediction time step, the energy-optimal solution is computed independently. Our approach is similar to Model-Predictive Control and could be enhanced by a yaw dynamics model in the future. This proof-of-principle type paper demonstrates the functionality and the benefit that arises from the described approach based on widely tested in-house simulation tools.}, author = {Schmidt, Jonas and Vollmer, Lukas and Requate, Niklas and Meyer, Tobias}, year = {2021}, date = {2021-05-27}, title = {Adaptive time-series based wind farm optimization}, address = {online}, note = {Slides available at: \url{http://www.tobi-meyer.de/Schmidt_2021.pdf}}, series = {Wind Energy Science Conference} }

Powered by bibtexbrowser