|
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.
The membership values are then multiplied weights attached to
the connection the input unit the hidden unit. This value multiplied with the connection
weight between the hidden unit and the output unit.
A special monotonic membership function which represents the conclusion of
the rule stored each hidden unit. The ARIC architecture was applied cart-pole
balancing and was shown that the system able solve this task [8]. feedforward neural
network with one hidden layer, which receives the system state its input and an
error signal from the physical system additional information [8].21
This architecture has the opportunity combine the advantages both
neural networks and fuzzy controllers. The fuzzy labels control rules are set for
each rule locally.1)
The hidden unit the fuzzy inference network represent the fuzzy rules, the
input units the rule antecedents, and the output unit represents the control action,
that the defuzzified combination the conclusions all rules (output hidden
units). ARIC uses
monotonic membership functions only.
ASN multilayer neural network representation fuzzy controller. predefining the fuzzy IF-THEN rules the
system learns faster than standard neural control system, because has not to
learn from scratch.
. fact,
it consists two separated nets, where the first one the fuzzy inference part and
the second one neural network that calculates p[t, 1], measure of
confidence associated with the fuzzy inference value u(t 1), using the weights of
time and the system state time stochastic modifier combines the
recommended control value u(t) the fuzzy inference part and the called
„probability“ value and determines the final output value the ASN [8]:
1,,)´( ttptuotu (4. the input layer, the system state variables are fuzzified [8].
The weight changes are determined reinforcement procedure that uses
the output the ASN and the AEN. The minimum those values its
final input [8]. The output value then
calculated weighted avarage all rule conclusions [8]. The crisp output value belonging the
minimum membership value can easily calculated the inverse function (thanks
to the monotonicity this function). Better state
have characteristically higher reinforcements. ARIC made feedforward neural networks, the Action-State
Evaluation Network (AEN) and the Action Selection Network (ASN). The network
output v[t, t´] viewed prediction future reinforcement that depends the
weights time and the system state time (which can t+1).
The AEN tries forecast the behavior the system
