There are multiple python libraries for conducting System Identification.
| Library Name | Installation | Link | Videos |
|---|---|---|---|
| SIPPY | https://github.com/CPCLAB-UNIPI/SIPPY | ||
| SysidentPy | pip install sysidentpy | https://github.com/wilsonrljr/sysidentpy | |
| PySINDy | pip install pysindy | https://github.com/dynamicslab/pysindy | https://www.youtube.com/watch?v=SfIJiuJ38W0&list=PLN90bHJU-JLoOfEk0KyBs2qLTV7OkMZ25 |
| PyDSTool | https://pydstool.github.io/PyDSTool/FrontPage.html |
SysIdentPy
PySINDy
PySINDy and MPC There is only one example of PySINDy being used in an MPC context, which is more social systems oriented.
Repo where they have used PySINDy for modelling process dynamics: https://github.com/ihmstefanini/dynamic-modeling-pysindy/blob/main/src/fit.py
It seems that PySINDy hasn’t been used for process dynamics much, which is what we’re interested in.
Questions about PySINDy
- How do you account for a systems’ reaction to unputs ? Do we need to account for these?
- Do you need to give it basis functions? Candidate models? What do you need to give it beyond state data?
- Sample rates & noise seem to be something to be aware of when working with this. Are there ways to reduce? Could use an SNR value, which could be a sensor attribute.
- Can you add a column of one particular value to represent a constant?