Abstract: An induction motor for driving a sucker rod pump can be powered by an SCR voltage regulator.The load on the motor is varied all the time.Therefore,the modeling of the above system is quite complicated using traditional modeling methods.But it can be achieved by using a dynamic recurrent Artificial Neural Network (ANN),namely Elman ANN,based on the operation characteristics of the system.A dynamic back-propagation (BP) learning method with an inertia item is applied to ANN training,which has some advantages over the traditional BP method with resonance and slow convergence rate.The trained ANN makes it possible to regulate the voltage cross the motor with the load changes on it.The ANN configuration,the modeling procedure and the dynamic BP learning method were described.The selection methods of the parameters for system modeling were also discussed.The simulation results show that the results of this method have a relative small error that is as good as expected.

Key words: sucker rod pump, voltage regulator, artificial neural network, non-linear modeling

摘要： 根据电动机拖动抽油机运行的特点,阐明了采用神经网络方法对该系统建模的必要性.基于带有回归单元的Elman神经网络,对拖动抽油机的变载荷三相异步电动机的晶闸管三相调压器系统进行了建模.采用了一种带惯性项的动态反向传播学习算法,克服了通常的BP(back propogation反向传播)算法振荡和收敛速度慢的弱点,使抽油机系统随负载变化时对电动机实现调压控制.对Elman神经网络的结构运用方法,以及惯性项的动态反向传播学习算法作了较详细的介绍,对由晶闸管三相调压器构成的拖动系统建模所选向量参数进行了说明.实例表明,利用该方法迭代后的学习结果更容易将误差减小至期望值.

关键词: 抽油机, 调压器, 神经网络, 非线性建模