采用IncNet神经网络进行动态建模的仿真研究

由于IncNet神经网络的算法为串行学习算法,可利用实测的新数据不断对模型进行更新,因此,可以实现过程的动态建模.IncNet神经网络采用"统计新颖性"准则,有效地减少了用户预先定义的参数,而删除算法则可以使网络结构更加紧凑.以燃烧炉对象为例,采用IncNet神经网络进行过程动态建模,所建立的模型精度高,泛化能力强.在建模过程中,初始参数的设置对IncNet神经网络模型的结构和建模精度有着不同的影响,应当合理选择.

Simulation Study of Dynamic Modeling through Using IncNet Neural Network

The fact that IncNet neural network is a sequential learning algorithm makes it possible to set up a dynamic model through using constant actual data so as to establish process dynamic modeling.IncNet neural network which applies statistical "novelty" criterion can effectively reduce user pre-determining parameters of IncNet neural network.Besides,a more compact network structure can be obtained by employing pruning algorithm.The model of a gas stove,for example,which was established by means of IncNet neural network,is characteristic of high precision and striking generalization.In establishing such a model,the initial parameters that exercise different influences on the structure and precision of IncNet neural network should be properly chosen.