Session: 01-11-03: Wind Energy

Paper Number: 110799

110799 - Multi-Objective Cooperative Control of Wind Farm Using a Double-Layer Machine Learning Framework 

The critical issues in reducing the cost of wind energy lie in two aspects, maximizing power production and extending the effective lifetime of the wind farm by reducing the fatigue load. Notwithstanding, higher power capture and lower fatigue load sometimes can not be realized simultaneously, as the wake deflection of the upstream turbine may bring more wind turbulence to the downstream flow fields, thus leading to more fatigue loads on the downstream turbines. Therefore, highly efficient cooperative control calls for a reasonable optimization scheme coupling the power improvement and lifetime extension, satisfying the power demand while reducing the fatigue load as much as possible. A novel double-stage scheme will be proposed to resolve the control optimization problem. Firstly, the baseline coordinated control action is determined in stage 1 to maximize the overall power output, regarding the optimal solution as the baseline power to feed stage 2. Then the joint set of yaw-offset angle is further adjusted to extend the average lifetime of the wind farm as much as possible with the extra constraint on the acceptable sacrifice of the baseline power output. A double-layer machine learning framework is employed as the optimization strategy for both the power and lifetime extension factor in the respective stage. The proposed double-stage coupling optimization scheme will be deployed to the 5-turbine row to perform the multi-objective cooperative control collaborating with the ANN yawed wake model. Firstly, Different allowable power knockdown coefficients are applied to the power constraint to determine the optimal value, representing the maximum lifetime extension per unit percentage of baseline power sacrifice. Then based on the above optimal value, parametric analysis is performed on the inflow condition to investigate its influence on the tradeoff between power and lifetime in the multi-objective control strategy. It is discovered that the ANN yawed wake model performs well in the lifetime prediction. The proposed double-stage multi-objective cooperative control framework can effectively extend the average lifetime of the wind farm at a small amount of power tradeoff. Moreover, it brings more benefit to the service life of the wind farm at the expense of the limited power tradeoff under the relatively high or low inflow velocity, but the advantage is somewhat weakened under the moderate inflow velocity. In addition, it performs desirably in the relatively high inflow turbulence intensity, obtaining a considerable lifetime extension at the cost of the minor power compromise, but incurs an unsatisfactory lifetime gain under low inflow turbulence.

Presenting Author: Shanghui Yang The University of Hong Kong

Presenting Author Biography: Yang Shanghui is a Ph.D. student in the Department of Civil Engineering, The University of Hong Kong. Her research is on the structural dynamic analysis of offshore wind turbines considering soil-pile-structure interaction, computational fluid dynamics (CFD) analysis of offshore wind turbines, large-scale wind and wave field prediction, and large-scale wind farm optimization.

Authors:

Shanghui Yang The University of Hong Kong
Xiaowei Deng The University of Hong Kong