ANN
CONTROLLER
Figures are uploaded based on requisition to sabareeshwarank@gmail.com or howtousematlab@gmail.com
The usage of ANN controller in the matlab is described below with step by step diagrams.
Basically the controller
operates based on the error signal. Your system is first developed on the PI
controller base and then used ANN by the training of the neurons…..
Fig 2: shows the nftool i.e
Neural network fitting tool in MATLAB 2013a version….
The controller alters the
output quicker for the input data….
First we run the system with
the PI controller and we take datas from it i.e Variation in the output as per
the input..
How to get get datas from PI
controller…
Fig 1 explains how the datas
gathered from the PI controller…. The datas present infront of the controller
and after the controller is taken to the workspace (to MATLAB command window)..
We save that datas obtained for training the neurons….
Fig 1
The below diagram shows the training
process…. The nftool asks for the input and output to be trained… We will feed
the data’s obtained from the workspace.. As per our iteration time the number
of values are separated into 1000 periods…. i.e 1, 2, 3…., 1000..
Fig 2
After giving the input and output the
training process is to be started… For that we use Levenberg-Marqardt back
propogation training. On clicking train the following fig appears… From this we
can see that the input has “3” denoting three signals are given to the
controller…. Hiden layer consists of weights, bias and neurons… Normally
neurons are kept at 20 for efficient result. Again the output layer converts to
“3” as output from the controller..
The Epoch denotes the number of iterations
done…. i.e 1000 combination of input and output is created. If the training
stops in between 1000 iterations it should be trained again and again…
After doing training for several times the
epoch value reaches the 1000 showing the “Maximum epoch is reached”. From here
we can see the performance of the trained neural network..
The below fig shows the performance of the
trained neurons i.e how they adopted for the system…. The trained and tested
datas are to be on the best performance area… i.e reducing the error value…
The below fig shows how the training is done
at different stages…. On maximum iteration of 1000 the performance is
increased….
This is the histogram…. The minimum error
value denotes the maximum performance… The neurons trained to be at the zero error.
ERROR =SET VALUE – ACTUAL VALUE
The regression data’s is shown for the
training, validation and test. The neurons are trained for the data’s and those
data’s are validated and checked for the data’s to fit with the actual data’s…
This is the architecture of thr formed
network.. We can alter the hidden layers.. the three inputs are given to the
controller and the three outputs are obtained (for three phase)… If the system
is not suiting our system i.e not performing well we will alter the neurons no.
do yhe training process again.
After the training process we can generate
the network diagram and check the input , hidden and output layers.
On this click on the simulation diagram to
get the ANN CONTROLLER for our specified system. This controller is replaced in
the place of old PI controller…
These are the developed neural networks for
our simulation… We can use this in place of the specific PI controller where we
taken the data’s to workspace.
This is the Solar and wind output DC
voltage…. Red one denotes the solar and pink one denotes the wind.
If we train the ANN it will
automatically starts to produce the output as per our training….
INPUT
|
OUTPUT
|
1
|
3
|
2
|
4
|
3
|
5
|
If the first three datas are taken as input and output
datas to train during the training the training the neural network
automatically adjusts for the untrained data as follows
|
|
8
|
10
|
20
|
22
|
Copy writted material of SABAREESHWARAN
