摘要
通过对永磁无刷直流电机的无位置传感器检测原理和小波网络特性的分析,提出了基于小波神经网络的永磁无刷直流电机无位置传感器控制新方法.该方法构建小波网络模型,采用梯度下降法对网络进行训练.网络训练分为离线训练和在线训练,由离线训练初步确定网络隐层节点的小波平移因子、尺度因子及网络输出层权值,以滤波和逻辑处理后的网络输出信号为教师对网络输出层连接权进行在线调整.从而由电机的相电流、端电压映射出电机的换相信号,取代了传统的位置传感器.最后仿真及实验结果表明,该方法能得到准确的永磁无刷直流电机的换相信号.
The principle of position sensorless control for PM brushless DC motors （BLDCM） and the characteristics of wavelet neural network （WNN） are analyzed. As a result, a novel control method based on WNN for BLDCM is proposed. In this method a wavelet neural network model is built, through a gradient descent error algorithm, the network updates its parameters. The WNN is trained both offline and online. In the offline training, the scale factor, the displacement factor of wavelet function and the weights between the hidden layer and the output layer are obtained. While in the online learning, the weights of WNN are updated based on the output signals after filtering. By mapping the terminal voltages and phase currents to communication signals, the network can replace the traditional position sensors. Simulation and experimental results show that exact commutation signals can be obtained using the proposed method.
出处
《天津大学学报》
EI
CAS
CSCD
北大核心
2007年第2期190-194,共5页
Journal of Tianjin University(Science and Technology)
基金
天津市科技攻关计划重大资助项目(05ZHGCGX00100)
天津市应用基础研究重点资助项目(043802011)
关键词
永磁无刷直流电机
小波函数
神经网络
无位置传感器控制
梯度下降法
PM brushless DC motor
wavelet function
neural network
position sensorless control
gradient descent error algorithm
