Hammerstein模型的改进新型神经动力学辨识方法及其在混合建模中的应用

Hammerstein Model Identification Method Based on the New ImprovedNeural Dynamics and Its Application to Hybrid Modeling

  • 摘要: 推导了多输入多输出Hammerstein模型的矩阵格式,并提出了一种改进的新型神经动力学算法.应用此算法,可同时辨识出Hammerstein模型的多组未知参数, 提高了收敛精度和速度.首先,对改进的新型神经动力学算法进行了参数的收敛性分析.之后,推导了基于Hammerstein模型的混合模型, 并利用其建立实际模型与机理模型之间的偏差模型,具有很好的补偿效果.由于改进的新型神经动力学方法可以在线调整Hammerstein模型参数, 所以混合模型可以准确地模拟复杂过程在大范围内的动态行为.实验表明该方法的合理性和有效性.

     

    Abstract: The matrix format of multi-input multi-output Hammerstein model is deduced, and a kind of the new improvedneural dynamics algorithm is proposed. The algorithm can be used to identify many groups of unknown Hammertein modelparameters, which improves accuracy and convergence rate. Firstly, the parameters’convergence of the new improved neuraldynamics algorithm is analyzed. Then a new hybrid model based on Hammerstein model is deduced to build an error modelbetween actual system and mechanism system. The hybrid model has good compensating effect. Since the new neural dynamicsmethod can adjust Hammerstein model parameters online, the hybrid model can be used to simulate dynamic behaviorof complex processes in a large scale exactly. The rationality and efficiency of the presented method are demonstrated bysimulation experiment.

     

/

返回文章
返回