基于神经网络的非线性模型辨识

在分析了影响多层前馈神经网络泛化性能各项因素的基础上,应用BP网络对一个微型锅炉非线性对象进行了模型辨识,以建立该系统的预测模型.在辨识过程中注意采用泛化方法解决样本数据采集和网络结构确定方面的问题,利用贝叶斯正则化方法训练神经网络,以保证在满足训练精度的要求下,网络还具有较好的泛化性能.通过选取一组数据对辨识结果模型进行测试,结果表明所辨识出的对象模型能够较好地表现出对象的动态行为,且具有较好的泛化性能.

Nonlinear Model Identification Based on Neural Networks

The generalization ability of neural networks can be improved by using network architecture optimisation and regularization methods . On studying the generalization theory of feedforward neural networks, a nonlin ear mini-boiler system was identified with a BP neural network to construct a pr edictive model. Generalization methods were used to solve problems on collecting sample data and training the network. Bayesian model comparison framework was a pplied to choice of weight decay terms. A set of data were used to test the per formance of the model, the result indicates that the neural network can behave with the dynamic characteristics of the boiler and good generalization abilitie s.