Scribd
Upload
151 views10 pages

Assignment - 1: Modeling and Simulation of Dynamic System

The document summarizes two problems involving modeling and simulating dynamic systems using neural networks. Problem 1 uses a multilayer perceptron with 1 hidden layer of 20 neurons to model a system, achieving a mean squared error of 0.664386 after 1200 epochs. Problem 2 uses a radial basis function network with 10 hidden neurons and 2 outputs to model another system, achieving a mean squared error of 0.00437211 after 1200 epochs. The initial and updated center points for the 10 radial basis functions are provided.

Uploaded by

VinaykumarJatoth
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
151 views10 pages

Assignment - 1: Modeling and Simulation of Dynamic System

The document summarizes two problems involving modeling and simulating dynamic systems using neural networks. Problem 1 uses a multilayer perceptron with 1 hidden layer of 20 neurons to model a system, achieving a mean squared error of 0.664386 after 1200 epochs. Problem 2 uses a radial basis function network with 10 hidden neurons and 2 outputs to model another system, achieving a mean squared error of 0.00437211 after 1200 epochs. The initial and updated center points for the 10 radial basis functions are provided.

Uploaded by

VinaykumarJatoth
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 10

Modeling and Simulation of Dynamic System

ASSIGNMENT - 1

Submitted by:
VINAY KUMAR JATOTH
120102069

PROBLEM-1:
Two multilayered networks N1, N2 are considered.
The No. of Hidden layers in N1 & N2 = 1.
No. of Neurons in each Hidden layer of N1 & N2 = 20.
Activation function assumed is Signum.
Inputs were initialized namely xd1, xd2, u1, u2 and by the concept of Back
propagation Mean square errors were calculated for 1200 epochs.
Assumed Learning rate is 0.1 .
The MSE error at the end of 1200th epoch is 0.664386.

Following is the graph for MSE vs Epochs:

< MSE vs Epochs plot for 1200 epochs >

The graphs for xd1 (desired output1) vs x1 (network output1) are as follows:

(a) < (unfilled)-DESIRED OUTPUT-1 & (filled)-NETWORK OUTPUT-1 >

(b) < (light)-DESIRED OUTPUT-1 & (dark)-NETWORK OUTPUT-1 >

The graphs for xd2 (desired output2) vs x2 (network output2) are as follows:

(c) < (unfilled)-DESIRED OUTPUT-2 & (filled)-NETWORK OUTPUT-2 >

(d) < (light)-DESIRED OUTPUT-2 & (dark)-NETWORK OUTPUT-2 >

PROBLEM-2:
RBF Network with 10 Neurons in 1 Hidden layer and 2 output layers is
considered. These 10 Neurons are initialized with some random centers.
Inputs namely xd1, xd2, u, and initial weights are initialized.
The Learning rate for Weights assumed (n2) = 0.2.
The Learning rate for Centers assumed (n1) = 0.2.
Activation function assumed is Gaussian.
Then by Gradient descent algorithm both the Weights & Centers are updated.
The epoch wise errors for 1200 epochs are collected and plotted in Mat lab.
The MSE error at the end of 1200th epoch is 0.00437211.

< MSE vs Epochs plot for 1200 epochs >

The centers initialized were 10 each with three dimensions as there are three
inputs x1, x2, & u:
C1 = (0.1, 0.2, 0.3)
C2 = (0.2, 0.4, 0.6)
C3 = (0.3, 0.6, 0.9)
C4 = (0.4, 0.8, 1.2)
C5 = (0.5, 1, 1.5)
C6 = (0.6, 1.2, 1.8)
C7 = (0.7, 1.4, 2.1)
C8 = (0.8, 1.6, 2.4)
C9 = (0.9, 1.8, 2.7)
C10 = (1, 2, 3)

Checked for 20 centers but getting undefined values nan as centers.

The Updated centers or the centers after 1200 epochs are as follows:
C1 = (1.68254, 0.608334, -1.0707)
C2 = (0.16651, -0.270035, -1.54906)
C3 = (-1.48972, -0.0193725, -0.45873)
C4 = (2.37209, 1.67309, 1.27476)
C5 = (2.08608, 3.0689, 1.21935)
C6 = (0.515312, 0.416489, 1.54837)
C7 = (-0.610708, -1.58019, 1.25789)
C8 = (-0.78486, 1.91765, 1.40235)
C9 = (1.11499, 1.33142, 0.977151)
C10 = (1.36538, 1.15421, 1.22994)

Following are the images of the INITIAL & FINAL CENTERS using matlab
function, Stem3:

< Initial centers >

< Final centers >

The graphs for xd1 (desired output1) vs x1 (network output1) are as follows:

(a) < (unfilled)-DESIRED OUTPUT-1 & (filled)-NETWORK OUTPUT-1 >

(b) < (light)-DESIRED OUTPUT-1 & (dark)-NETWORK OUTPUT-1 >

The graphs for xd2 (desired output2) vs x2 (network output2) are as follows:

(c) < (unfilled)-DESIRED OUTPUT-2 & (filled)-NETWORK OUTPUT-2 >

(d) < (light)-DESIRED OUTPUT-2 & (dark)-NETWORK OUTPUT-2 >

You might also like