【作者】 韩锋；

【导师】 徐丙垠； 边敦新；

【作者基本信息】 山东理工大学 ， 电力电子与电力传动， 2008， 硕士

【摘要】 开关磁阻电动机(Switched Reluctance Motor,SRM)具有结构简单、工作可靠、效率高和成本较低等优点,在很多领域都显示出强大的竞争力,但是其位置传感器的存在不仅削弱了SRM结构简单的优势,而且降低了系统高速运行的可靠性,增加了成本,于是探索实用的无位置传感器检测转子位置的方案便成为开关磁阻电机驱动系统(SwitchedReluctance Motor Drive,SRD)研究的热点。SRM高度非线性的电磁特性决定了在精确的数学模型基础上实现无位置传感器控制十分困难,而人工神经网络的出现为解决这个问题提供了新的思路。BP(Back Propagation)神经网络是目前研究最多、应用最广泛的一种多层前馈神经网络,具有收敛速度快、非线性逼近能力强等优点。本文提出了一种利用改进的BP网络来实现SRM无位置传感器的转子位置检测方法,该方法以电机三相绕组的相电流、磁链作为输入,转子位置作为输出,建立SRM电流、磁链与转子位置之间的非线性映射,从而实现SRM无位置传感器的转子位置检测。在现阶段,为了保证神经网络的检测精度与收敛速度,神经网络的训练样本数据一般都是电机在某一特定条件下的运行数据,并不能反映电机的实际运行情况。针对这一情况,本文提出了一种训练数据实时更新的思想,并将其引入到神经网络的输入向量中。在神经网络训练过程中,根据电机的实际运行情况,将不同运行情况下的三相绕组的相电流、磁通、转子位置角作为训练样本,分批次对建立的神经网络模型进行训练,直到网络的检测精度趋于稳定并达到要求为止。仿真结果表明,本文所提出的改进的BP神经网络在电机不同的运行情况下都能够实现SRM转子位置的准确检测,从而实现了电机的无位置传感器控制。神经网络检测精度高,动态特性好,具有较好的自适应性和鲁棒性。更多还原

【Abstract】 Due to its simple construction, reliability, high efficiency and low cost, switched reluctance motor (SRM) has shown huge competitive power in many fields. But mechanical position sensors add to the cost, complexity and potential unreliability at high speed and this has motivated the investigation of sensorless position estimation. Because of its high nonlinear electromagnetism characteristic, the sensorless control based on accurate model of SRM is hard to be accomplished. In recent years, artificial neural network(ANN) technology has made a great progress, which gives a new method to accomplish position sensorless control of SRM.A novel indirect detecting method of SRM’s rotor postion based on BP(Back Propagation) neural network is presented in this paper. For the adopted network,the training data set is comprised of magnetization data of the SRM for which three-phase current and flux linkage are inputs and the corresponding position is the output, the BP neural network can build up a correlation among phase current, phase flux linkage and position, thereby facilitating elimination of the rotor position sensor.At the present stage, in order to ensure the forecast precision and the convergence rate of the neural network, the neural network’s training sample data is generally the electrical motor’s operation data under the specified condition which can not reflect the real running situation of the electrical motor. In view of this situation, this paper proposes a method based on training data real-time renewal, and introduces the idea to the input vector of neural network. In neural network’s training process, according to the real running situation of the electrical motor, takes the three-phase current, flux linkage and rotor position in different operation circumstances as training sample, to train the established neural network model in batches, until the network’s forecast precision tends to be stable and meet the requirement.The simulation results show that the improved BP neural networks proposed in the paper can realize SRM rotor position’s accurate prediction in case of different operation of the electrical motor,and realizing the motor’s control without position sensor. The neural network has high forecast precision, favorable dynamic characteristics, better adaptability and robustness.更多还原