Deep learning for optimal dispatch of automatic generation control in a wind farm

Abstract

As a wind farm participates in automatic generation control (AGC), it should trace the real-time AGC signal from the independent system operator. To achieve a high responding performance, the real-time AGC signal should be rapidly distributed to multiple wind turbines (WTs) via an optimal dispatch. It is essentially a non-linear complex optimization due to the wake effect between different WTs. To solve this problem, a deep learning is employed to rapidly generate the dispatch scheme of AGC in a wind farm. The training data of deep learning is acquired from the optimization results of different anticipated tasks by genetic algorithm. In order to guarantee a reliable on-line decision of deep learning, the error of the regulation power command is corrected via an adjustment method of rotor speed and pitch angle for each WT. The effectiveness of the proposed technique is evaluated by a wind farm compared with multiple optimization methods.