磁流变阻尼器神经网络模型的建立及优化

Establishment and Optimization of the Magnetorheological Damper Neural Network Model

磁流变阻尼器在振动控制中有广阔的应用前景,建立其精确的力学模型是取得良好控制效果的关键因素之一。文中对某磁流变阻尼器进行了动力学性能测试;并通过统计学方法确定了该阻尼器正向、逆向模型的BP神经网络结构;针对传统遗传神经网络(GA-BP)早熟和收敛速度慢的问题,提出一种结合适应度线性变换、自适应交叉和变异概率的改进遗传神经网络(IGA-BP)算法;在此基础上,分别用BP神经网络、GA-BP神经网络和IGA-BP神经网络对阻尼器正向、逆向非参数化模型进行辨识。研究结果表明:文中提出的改进遗传神经网络算法收敛速度更快,模型精度更高,该非参数化模型能更准确地反映磁流变阻尼器的动力学特性。

As a smart vibration control device,there is a broad application prospect for the magnetorheological(MR)damper.And establishing its precise mechanical model is one of the key factors for excellent control.In this paper,the dynamic perfor⁃mance of MR damper is tested.And the neural network structure of the forward and reverse models are determined by statistical methods.Then,aiming at the problem of early maturity and slow convergence speed of traditional genetic neural network(GABP),an improved genetic neural network(IGA-BP)algorithm with linear transformation of fitness,adaptive crossover and mu⁃tation probability is proposed.Finally,The BP neural network,GA-BP neural network and IGA-BP neural network are used to identify the forward and reverse models of the damper,respectively.The results show that the IGA-BP algorithm proposed in this paper has faster convergence speed and higher model accuracy,and the models can better reflect the dynamic characteristics of the MR damper.