中图分类号：TP273.1 文献标志码：A

Position compensation control of maglev ball system by fuzz inference and Jordan neural network

LI Xiaoru， CHEN Shisong， HUANG Zhiwen （SchoolofMechanical Engineering， UniversityofShanghai for Scienceand TechnologyShanghai 2ooo93，China）

Abstract： To address the issue of poor dynamic performance in control systems caused by the output uncertainty of insufficiently trained Jordan neural network （JNN），a novel position compensation control method of maglev ball based on fuzzy inference （FI） and JNN was proposed. A three-module control framework was designed， consisting of a basic control module， a JNN control module， and a FI module. The basic control module adopted a highly adaptable PID controler to provide baseline control performance. The JNN control module performed real-time identification and compensation for the maglev bal system. The FI module dynamically adjusted the output of the neural network controller to suppress the uncertainty introduced by insufficiently trained JNN. The experimental results show that， compared with the traditional neural network compensation control method， the proposed method reduces overshoot by 39.79% and 60.61% and shortens the settling time by 19.52% and 48.47% when tracking step and square wave signals. The proposed method significantly enhances the dynamic performance of the control system while maintaining steady-state accuracy.

Keywords： fuzzy inference; Jordan neural network; position compensation control; maglev ball

近年来，具有无摩擦、低噪声、少污染等特点的磁悬浮技术逐渐应用于多个领域，如磁悬浮车辆、磁悬浮电机、磁悬浮轴承和磁悬浮球[1]等。磁悬浮球系统的复杂非线性、强时变性导致对其进行实时高精度控制的难度大，故该系统常被用于进行各种先进控制方法的测试与验证。学者们针对该系统提出了许多控制方法[2-3]，如PID（proportional-integral-derivative）控制、鲁棒控制、模糊控制、LQR（linearquadratic regulator）控制。但这些传统控制方法在实际应用时，难以满足较高的精度要求，具有一定的局限性。

神经网络具有强大的自学习能力，被证明能够逼近任意非线性函数，并被成功应用于非线性系统的实时控制中，如汽车系统、机器人系统及电液伺服系统等复杂非线性控制系统。朱坚民等[4]使用神经网络在线实时补偿PID控制量，显著提升了磁悬浮球位置控制系统的稳态性能，验证了在传统控制方法中引入神经网络在线补偿的有效性。谢莉莉等[5]基于RBF（radial basis function）神经网络和滑模变结构原理设计了一种变结构控制器，仿真结果表明，小球能够较快达到稳定位置，控制系统具有较好的稳态特性和抗干扰特性。沈昕璐等在经典PID控制器的基础上，引入模糊神经网络对球杆系统进行在线辨识和控制补偿，仿真和实验结果均表明，该策略使得控制系统具有较好的稳态性能。Wei等[提出了一种基于反向传播神经网络的在线前馈补偿控制器，降低了磁悬浮球系统位置控制的稳态误差。Sun等提出了一种基于RBF神经网络的监督控制方法，在磁悬浮系统的跟踪控制中取得了优异的稳态性能。……