基于GRU-NN预测模型的压电作动器MPC-KAN控制方法

MPC-KAN Control Method for Piezoelectric Actuators Based on GRU-NN Prediction Model

摘要:

为了提高压电作动器(PEAs)的轨迹跟踪性能,提出了一种基于门控递归单元(GRU)神经网络(NN) 预测模型的Kolmogorov-Arnold网络前馈模型预测控制(MPC-KAN)。与神经网络逆模型控制不同,该方法使用 GRU-NN正向建模,并根据模型预测结果调整模型预测控制(MPC)的输出。首先,根据线性化模型选择GRU-NN 的训练输入特征,并训练该网络。然后,为了提高优化效果和缩短优化时间,将麻雀搜索算法(SSA)用作MPC优化器,并建立Kolmogorov-Arnold网络(KAN)以替代SSA优化。该方法的有效性在PEAs平台上得到验证,与传统方法相比,控制精度提高了约30%。

Abstract:

To improve the trajectory tracking performance of piezoelectric actuators (PEAs), this study proposes a Kolmogorov-Arnold network feedforward model predictive control (MPC-KAN) based on a gated recurrent unit (GRU) neural network (NN) prediction model. Unlike neural network inverse model control, this method uses GRU-NN forward modeling and adjusts the model predictive control (MPC) output based on the model prediction results. First, this study selects the training input features of GRU-NN based on a linearized model and trains the network. Then, to improve optimization performance and shorten optimization time, the sparrow search algorithm (SSA) was used as the MPC optimizer, and a Kolmogorov-Arnold network (KAN) was established to replace the SSA optimization. The effectiveness of this method has been verified on the PEA platform. Compared with tradi tional methods, the control accuracy has been improved by approximately 30%.