铸造工艺分析的神经网络方法_ppt.pdf

METRO MEtallurgical TRaining On-line Artificial neural networks in analysis of foundry processes Marcin Perzyk WUT Education and Culture Definition of artificial neural network (ANN) Artificial neural network (ANN) is a complex mathematical relationship, the structure of which imitates structure and data processing in cerebral cortex of mammals, including humans. Neuron (network knot) Synapses transfer values of variables and contain model parameters – synapses weights. Synapse (connection of knots, Neurons perform mathematical sometimes network output) operations on variables and weights. METRO – MEtallurgical TRaining On-line Copyright © 2005 Marcin Perzyk - WUT 2 Artificial neural networks Basic advantages • Ability to learn and to generalise the acquired knowledge. ANNs are able to find regularities in situations of large number of variables of various types. Such regularities often cannot be detected by senses of analysts or other mathematical models. • ANN is resistant to noise in data as well as errors appearing in some weights, i.e. incorrectly determined individual model parameters. • Fast data processing, sometimes online. METRO – MEtallurgical TRaining On-line Copyright © 2005 Marcin Perzyk - WUT 3 Artificial neural networks General characteristics ANNs are learning systems type models. Values of model parameters (network weights) are determined form results of observations (training