Process Identification and Control

Sapienza University of Rome · DIAG · Academic year 2026/2027 · Prof. Francesco Liberati

The three parts of the course around one process: identification turns measured data into a model, model predictive control turns the model into an optimal plan over a prediction window, and reinforcement learning turns experience into a policy. All three produce the control applied to the process, whose measurements feed all three back: data to identify with, state to decide on, state and reward to learn from.

News

About the course

How do we control a process optimally while respecting its constraints, and where do the models we need come from? The course answers in three steps:

  • Model Predictive Control (MPC): we know a model and solve a finite-horizon optimal control problem at every step, as an approximation of the ideal infinite-horizon constrained problem.
  • Reinforcement Learning (RL): we learn value functions and policies from data, approximating the same ideal problem.
  • Process identification: we build models and estimate states from measured data, closing the loop with the controllers seen in the first two parts.

Theory is always paired with examples and numerical experiments in Julia and Python. See the program and the schedule.

Logistics

Lessons (official schedule)

Day Time Classroom
Monday 16:00–19:00 28, San Pietro in Vincoli
Wednesday 13:00–16:00 29, San Pietro in Vincoli
Thursday 16:00–19:00 7, San Pietro in Vincoli

Office hours: Thursdays, 14:00–16:00, in presence or online. Send me an email in advance to book an appointment; for an online meeting I will reply with the link.

Contact: liberati@diag.uniroma1.it

Study materials: for every lesson, the lesson diary lists the blackboard notes and the exact sections to study. The lessons of this edition are not recorded, but the recordings of the previous edition, and the other materials of past editions, remain available on the previous course website. All references are free and openly accessible whenever possible.