Sudden Increases in Computational Noise During Simulation

Question

tkarg 2022-1-13

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1627980-sudden-increases-in-computational-noise-during-simulation

编辑： Rangesh 2023-9-30

question.slx

In my Simulink model, a signal of interest should occur to be zero. Due to computational errors, it has some noise, I understand that. The thing I do not understand is that this computational noise level is suddenly increased multiple times during the simulation. As the picture below shows, at least 4 update in the noise amplitude can be counted. If the increase in the noise amplitude was continuous, it could have been explained as accumulation of errors, but here I see sudden changes as if Simulink decides to increase computational error tolerance.

The model generating this result seems like this:

I am using variable-step solver (ode45). Duration of the simulation is 100000 seconds. I tried disabling absolute tolerance auto scaling but it did not change anything. Also, I suspected accumulation of error at the sinusoidal wave, but the reason is not that either.

I wonder why this happens and how to avoid it. I have attached the simulink model for reproducability.

2 个评论
显示无隐藏无

Ankit 2022-1-20

can you please add here once 2018b version also?

Mathieu NOE 2022-1-20

hello

I would prefer to run a fixed step solver - can you try this ?

IMHO, I would go for discrete models + fixed step solver

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Rangesh 2023-9-29

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1627980-sudden-increases-in-computational-noise-during-simulation#answer_1322079

编辑：Rangesh 2023-9-30

Hi tkarg,

I understand that you want to reduce the noise in the Simulink computations. To reduce noise in simulations, consider using a fixed step solver instead of a variable step solver. Variable step solvers can introduce more noise when there are numerical instabilities or rapid transitions in the system. By using a fixed step solver, you can mitigate these issues and achieve more stable and accurate results.

Attached below is the result generated by your model obtained using fixed step solver: