电液伺服系统的神经网络建模方法研究

Neural networks modeling of electrohydraulic systems

针对电液伺服系统固有的流量-压力特性等非线性因素使得采用传递函数等传统方法难以获得电液伺服系统的精确模型的问题,详细研究了电液伺服系统的神经网络建模方法。研究了两种最常见的神经网络,即多层感知器神经网络和径向基函数神经网络,采用5种典型学习算法构造了3种多层感知器神经网络和2种径向基函数神经网络,并结合自动定深电液伺服系统的工程实例,详细分析了这5种神经网络在电液伺服系统中的建模性能。研究结果表明,采用正交最小二乘算法的径向基函数神经网络最适合电液伺服系统的建模。

Aiming at the problem that construction of an accurate model of an electrohydraulic system based on traditional linear methods remains a difficult task due to its nonlinear characteristics including flow/pressure relation, etc, the paper presents a thorough study on modeling of electrohydraulic systems using different types of neural networks. The two widely used neural networks, i.e. the multilayer perceptron neural network （MLPNN） and the radial basis function neural net- work （RBFNN） were investigated, and three MLPNNs and two RBFNNs were constructed using their five typical training algorithms. All these techniques were then applied to an automatic depth control electrohydraulic system, and the model- ing performance of the five networks in the electrohydraulic system was analyzed. The results clearly indicated that the ra- dial basis function neural network with the orthogonal least square training algorithm is prior to other neural networks for modeling of electrohydraulic systems.