|
Kategorie: Diplomové, bakalářské práce |
Tento dokument chci!
Práce popisuje základy principu funkčnosti neuronů a vytvoření umělých neuronových sítí. Je zde důkladně popsána struktura a funkce neuronů a ukázán nejpoužívanější algoritmus pro učení neuronů. Základy fuzzy logiky, včetně jejich výhod a nevýhod, jsou rovněž prezentovány. Detailněji je popsán algoritmus zpětného šíření chyb a adaptivní neuro-fuzzy inferenční systém. Tyto techniky poskytují efektivní způsoby učení neuronových sítí.
For the selection
of the operation mode, the variable mode has set 'anfis' 'nnv' by
commenting and uncommenting the individual lines.
The purpose the first part read the parameters the test voice recordings. The usage this toolbox greatly
increases the processing speed case the processor multi-cored there are
more computers available. Subsequently, some clusters
are merged together based preset distance criterion, then the new cluster centers
are calculated. This method tries to
make use existing patterns simplify the input part the network.The next part the code brings into effect the actual
learning the network.
The main program the script file spust. For the purpose this
work, the practical usage ANFIS heavily limited its high demands on
processing power for the case higher number inputs and second level neurons. Its listing included Appendix. Firstly, the given network has created. This approach significantly increases the maximal
number usable inputs the system. The program
itself can separated into parts: ANFIS using the Matlab's Fuzzy Logic Toolbox
('anfis') and the individually built neural network based the backpropagation
algorithm presented Chapter ('nnv'). These clusters are used for simplification
of the input side the network.
. Its iterations are run parallel increasing the
computing speed. Therefore, special function was
used for the creation these functions and network nodes which analyses the input
data and searches for existing clusters it.
mode='anfis';
%mode='nnv';
The data reading implemented the wavload function, which receives as
input parameters the path the directory containing training files and the number of
output parameters. Parfor part of
the parallel processing toolbox.m.
The basic task network creation takes into account all combinations inputs and
membership functions.
This program serves for probing ANFIS well neural networks. THE REALIZATION THE PROGRAM
This chapter serves demonstrate the program built Matlab. The subtractive
clustering initially assumes all data points clusters.
The function genfis2 creates Sugeno-type FIS structure. the beginning this script, the parallel processing toolbox is
initialized the command matlabpool open.32
7.
In the case the mode set 'anfis', the parfor cycle used for the creation
and learning three ANFIS networks, each for one output variable. this case means very high number created
membership functions and second level neurons. For the creation of
input rules, the subtractive cluster analysis method used