Please take a moment to complete this survey below
Library's collection Library's IT development CancelThe development of artificial neural network control system is quite rapid.
A good parallel process ability, non linear mapping, self learning ability and self
generalization are factors which motivate its development as a smart control
system.
In this final project, a control system with an artificial neural network is
designed to control water temperature, based on IBM PC. The neural network
architecture used in this control system is back propagation with a supervised
learning method. In this supervised learning method, the input and output supply
is done explicitly so that each samples gives an output, resulted from the input
given. The neural network software consists of two processes, training and
running. In training process, network parameter will be determined, input layer,
hidden layer, output layer and learning parameter value, where training could be
done until the error wanted is achieved and the output produced will be employed
in the running process.
After doing the system respond system, optimal neural network parameter
are obtained, that is by using 0,0005 learning parameter value and 3 layers, each
of them consists of 2 node input layer, 40 node hidden layer and 1 node output
layer. The result of the test points out the steady state error of about 5% from the
setting point value.
The respond of the water temperature running system towards the neural
network is quite good. In order to improve it, several methods can be done such as
undertaking a more various network structure training, making training data which
have a larger variety of training pattern.