Open-Source Python Seeq Packages for Sysid and Optimization Based PID Tuning

This study presents new open-source Python packages for system identification (SysID) and PID tuning, developed as Jupyter Notebook add-ons to the Seeq software. Seeq is a web-browser-based commercial platform for data exploration, cleansing, inferential building, and other common data analysis. The new additions, SysID and PID tuning packages, build upon the Seeq Workbench software with advanced data analysis and modeling capabilities, tailored for dynamic model identification and control tuning.

The SysID app has six model identification options including ARX, ARIMAX, FIR, transfer function, subspace, and neural network models. It supports a diverse array of linear and nonlinear models, like linear time-series, state space, nonlinear differential and algebraic equations, continuous and integer variables, and machine-learned models, based on the Python Gekko modeling platform. In addition to the traditional format of models, the SysID app also enables constrained SysID, Transformer Neural Networks, Transfer learning, and Physics-Informed Neural Networks.

In a second contribution, a PID tuning app uses an optimization-based PID retuning process. This method leverages historical setpoint and load disturbance data to refine the PID controller tuning parameters more effectively. The exhaustive search method avoids local minima in an attempt to improve tuning by exploring a predefined PID parameter search space. The reliability of the tuning is enhanced by simulating conditions to be as close as possible to realistic operating conditions, creating a digital twin of the PID controller's characteristics. The presentation includes a discussion of a potential future work to combine the SysID and PID tuning apps with identification to drive more accurate process models from open-loop or closed-loop data for PID retuning.