基于Hebbina监督学习算法的神经网络变距控制

Neural network variable-pitch control based on Hebbina supervised learning algorithm

变距控制是变速恒频风力发电机组的核心技术之一,由于变距系统具有纯滞后非线性的特性,采用常规的PID算法已无法满足控制目标的要求.由1 MW变速恒频风力发电机组实际运行情况可知,其控制器PID参数需要不断在线整定,为此设计了基于Hebbina监督学习机理的神经网络变距控制算法.根据机组运行的实际数据进行离线学习,确定Hebbina监督学习算法的学习速率iη,然后进行在线整定,以保证风力发电机组处于最佳运行状态.给出了1 MW风力发电机组采用常规PID算法和神经网络变距控制算法的仿真对比结果,从中可以看出后者的动态特性和稳态特性明显优于前者,对于实际应用将起到指导作用.

Variable-pitch control is one of key technologies in variable-speed and constant-frequency（VSCF） wind turbine. Because of hysteretic nonlinear characteristics, variable-pitch system adopting traditional PIE） algorithm can＇t meet the requirements of control object. From the practical operation situation of 1 MW VSCF wind turbine, it can be seen that PID controller parameters need to be adjusted online all the time, the neural network variable-speed control based on Hebbina supervised learning algorithm was proposed. Learning offline on the basis of the practical data of wind turbine operation gives the learning rate ηi of Hebbina supervised learning algorithm, and then adjusting online is performed in oMer to make the wind turbine stay in the best running situation. The simulated results of 1 MW wind turbine generator system adopting traditional PID were compared with those adopting Hebbina supervised learning algorithm. It can be seen that the dynamic and steady characteristics of the latter are obviously superior to those of the former. The present conclusions will play important roles in real operations.