| 摘要 |
A method for the development of neuro-fuzzy networks consists in that set of values of prototype object or process characteristics is used for the determination of neuro-fuzzy network parameters. Values of characteristics of diagnosed articles are measured, base of transactions is formed of measured characteristics, then discretization of value range of numerical characteristics is performed, and sets of characteristics is generated with support level not lower than extreme value with the result that frequently occurred sets are formed, on the basis of those content rules with required significance level is synthesized. Thereafter rules are combined with neighbouring limits of certain variables and redundant rules are removed. A base of content rules is created of obtained content rules, on the basis of this one a neuro-fuzzy network structure is synthesized, and values of membership function parameters and neuro-fuzzy network weight coefficients are determined. Then values of diagnostic characteristics for prototype object are measured and fed to synthesized neuro-fuzzy network, and decision is made on value of forecasting parameter on the basis of output value of constructed neuro-fuzzy network. |
