You are on page 1of 3

EE5327, Fall 2016

Project 2: Neural/Fuzzy Neural System ID


Due December 13, 2016
Please turn in this paper with your exam:
Pledge of honor: On my honor I have neither given nor received aid on this project
except from the TA or class instructor.

Signature: _______________________________________________

The van der Pol equation can be written as:

1 = 2
2 = (1 12 )2 1

Implement this nonlinear system, using a fixed time step of 0.01 seconds, a stop time of
100 seconds, and an initial condition of [x1, x2]=[1.8, 1]. Use the following signal
generator to create the value of mu used in the simulation.

Correctly implemented, the phase plot of the system looks like the following.
Your job as an engineer is to develop a method to estimate the value of mu from only the
state time history information (i.e. the estimator cannot be an explicit function of time)
and the estimator should work in real-time (in the simulation). Develop the estimator
using either neural or fuzzy neural methods and using as many tapped delays (or delays)
as you need. The estimator has the requirement to correctly identify the value of mu
to within +/-0.25 90% of the time over each mu interval. Transients or spikes in the
estimates are acceptable. Verification that your design meets the requirement is computed
when you feed the system an input signal that reverses the order of the training input, i.e.
mu is 2.5 from 0 to 20 seconds, mu is 2 from 20 to 40 seconds, mu is 1.5 from 40 to 60
seconds, mu is 1 from 60 to 80 seconds, and mu is 0.5 from 80 to 100 seconds.

Your project report should document the simulation, the method used to create the
estimator, and the final design that shows you meet the requirement.

As an example, heres a screenshot of one of my estimate attempts that appears to meet


the requirement.

You might also like