System Parameter Estimation with Projection Algorithm

1 reply · 2K views · last reply
rfrederic
Messages
6
Reaction score
0
I am recently trying to identify system parameters with projection algorithm, but faced a problem, and the dynamic model is the following:
[itex]\ddot{y}(t)+a\cdot\dot{y}(t)=b\cdot e(t)[/itex]

The true value of [itex]a[/itex] is [itex]2.8[/itex], [itex]b[/itex] is [itex]0.1[/itex].

While inputing volt [itex]e(t)=12sin(2\pi t)+5sin(2t)[/itex], I can get a good result showed in 1c_PA_r0.1_a0.jpg, and both of a and b are convergent to the true value.

But while [itex]e(t)=24[/itex], the result showed in doesn't seem right.

My question is: will [itex]e(t)[/itex] affect the convergence of parameters? And why?
By the way, I had also tried to use the least square to identify parameters while [itex]e(t)=24[/itex], the result also seems good.
 
Attachments
  • Five-panel MATLAB plot: true a (2.8) and estimated a rise in steps over 0 to 5 s; true b (0.1) and estimated b settle after a spike near 0.6 s; sinusoidal theta-hat and theta plots; and an error plot with spikes near 0.6 s and 1.5 s.
    1c_PA_r0.1_a0.jpg
    45.2 KB · Views: 534
  • Six-panel plot grid of parameter estimation vs time over 0 to 5 s: true a (2.8) flat at top-left with estimate a at 0, true b (0.1) and estimate b rising then decaying to 0, theta and theta_hat converging, theta'' and theta_hat'' decaying, and an error curve dipping below zero then settling at 0.
    1b_PA_r0.1_a0.jpg
    39.7 KB · Views: 444
rfrederic said:
will e(t)e(t)e(t) affect the convergence of parameters? And why?
Why not? Different e(t) are different cases. There is no reason to expect that all cases are equally easy to identify.