基于模糊推理决策神经网络前馈补偿值的直流伺服电机位置跃变控制

Position Leap Control of DC Servo Motor Based on Neural Network Feedforward Compensation Quantity Decided by Fuzzy Inference

摘要: 针对由于传统PID串级控制结构的局限性以及欠训练神经网络前馈控制器输出的控制量的不确定性导致的直流伺服电机位置跃变控制动态性能欠佳的问题，提出了一种基于模糊推理决策神经网络前馈补偿值的控制方法。该方法包括三个控制子模块：基础控制模块、神经网络控制模块、以及模糊决策模块。基础控制模块为传统PID控制，保证控制初期整个系统的稳定以及为神经网络控制模块提供在线学习样本；神经网络控制模块通过在线学习被控对象的动态逆模型后对PID控制器进行前馈补偿；模糊决策模块根据电机位置的实时跟踪情况输出决策因子，用于自适应决策神经网络控制模块在欠训练时的前馈输出量，提升直流伺服电机位置跃变控制时的动态品质。仿真和实验结果表明：提出的方法在不牺牲稳态精度条件下，显著提升直流伺服电机位置控制的快速跟随性能，减少了控制系统的超调量和调节时间，具有较好的动静特性和较强的鲁棒性。

Abstract: To overcome the poor dynamic performance of DC servo motor position leap control caused by the limitations of the conventional PID cascade control structure and the uncertainty control effect by the under-trained neural network feedforward controller, a control method based on fuzzy infer-ence decides neural network feedforward compensation control quantity is proposed. The method consists of three control submodules: basic control module, neural network control module and fuzzy decision module. The basic control module is conventional PID control, which ensures the sta-bility of the system at the initially stage of control and provides online learning samples for the neural network control module. The neural network control module performs feedforward com-pensation on the PID controller through learning the dynamic inverse model of the controlled ob-ject online. The fuzzy decision-making module outputs decision factor according to the real- time tracking situation of the motor position, which is used to adaptively decide the feedforward control quantity of the neural network control module when it is under-trained, and to improve the dy-namic quality of the position leap control of the DC servo motor. Simulation and experimental re-sults show that the proposed method significantly improves the fast-following performance of DC servo motor position control without sacrificing the steady-state accuracy, reduces the overshoot and settling time, has great dynamic and steady-state performance and strong robustness.