中文摘要：

在传统滑模控制中，通常根据实测电流和观测器的电流估计值之差设计滑模面，但这往往会造成趋近滑模面较慢和相位滞后问题。以电流和磁链作为RBF神经网络滑模观测器的输入，把转子角速度作为未知量，采用RBF神经网络最小参数自学习算法，使神经网络滑模控制器的输出作为系统位置控制输入信号，通过参数的估计代替网络权值的调整，减少了计算量，采用指数型趋近律有效降低滑模控制的抖振现象。通过与传统滑模仿真试验相比，比较结果表明该方法可以提高系统的响应速度，实现转子位置的准确估算，降低抖振，改善系统的动态性能。

英文摘要：

In the traditional sliding mode control, the new sliding mode surface was designed according to the difference between the measured and the observer-based estimated currents, which could cause to reach the sliding surface slowly and phase lag. The stator currents were taken as the input of the RBF neural network sliding observer, and the rotor angular velocity as unknown variables. Since the network minimal parameters learning algorithm was adopted, so that the neural network sliding controller output was regarded as the position control system input signal. Instead of the network weights adjustment by the parameters estimating, which reduced the amount of computation. Exponential reaching law was used to reduce the vibration of the system. Comparing with the traditional sliding mode through, the results verified that the proposed control method offered faster speed response, accurately estimating the rotor position, reducing buffeting and improving the dynamic performance.