A Data-driven Active Power Distribution Strategy of Wind Farm AGC Substation Based on Critic Algorithm

Abstract

In order to cope with the dispatching challenges of a high proportion of renewable energy power system, it is a necessary requirement for wind farms to participate in automatic generation control (AGC). Due to the different operating states and control strategies of each wind turbine, how to effectively realize the distribution of AGC instruction in wind farm stations is a key issue. And the current mainstream idea is to adopt a simple average distribution strategy or distribute according to the wind speed as the weight. However these strategy cannot take full advantage of the regulation endowment of the unit and the station, resulting in large regulation error and pitch angle action. Moreover, the unit model is complicated and difficult to build accurately. So as to improve the performance of wind farm in response to dispatching instructions, a data-driven active power distribution strategy of wind farm AGC sub-stations based on critical algorithm to evaluate the unit regulation performance is proposed. Because it is hard to build a precise model of wind farm, a data-driven algorithm is introduced to establish the performance evaluation model. First, the data of input wind speed, pitch angle and rotor speed of the unit is collected, and the critic algorithm is used to qualify the weight of each index affecting the adjustment performance reflecting the active power control capability of the units. Then, the weight distribution of the field's active power command based on the regulation performance of units is completed, and the rolling optimization distribution is continuously carried out in one AGC control cycle in consideration of the regulation error of the previous stage. Finally, the simulation of the active power allocation strategy of the field with 5 permanent magnet wind turbines is experimented on the MATLAB/Simulink platform. The outcomes show that compared with the average allocation strategy, the proposed algorithm can reduce the control error by 12% at most, and with the weight distribution based on wind speed, the average action degree of pitch angle can be reduced by 20% at most, which has a better effect.