Title: Optimization method for artificial neural network
Application Number: 200410011350 Application Date: 2004.12.13
Publication Number: 1622129 Publication Date: 2005.06.01
Approval Pub. Date: Granted Pub. Date:
International Classifi-cation: G06N3/02,G06N3/12
Applicant(s) Name: Changchun Institute of Optical Machinery, CAS Address: 130031
Inventor(s) Name: Sun Jiashi
Attorney & Agent: liang airong
Abstract:
     The present invention relates to the optimized design of artificial nerve network, and solves the problems of providing theoretic reference for the determination of hidden layer of neure node number, etc. Experiment optimizing secondary universal rotary combined design method is adopted, regression equation of hidden layer neure node number and nerve network mean-square deviation is established based on practical engineering requirement and optimized in genetic algorithm to obtain the minimum value of nerve network mean-square deviation and the hidden layer neure node number of nerve network and thus the structure of the nerve network. The present invention lays the theoretical foundation for nerve network design and makes the optimized design of artificial nerve network scientific and practical.
Time: 10
<- Previous Patent:Digital image color correction me...   |  Next Patent:Parameter adjustment device ->