Overview
Agent-based PID Control
AUTO-PILOT is an agent-based framework for autonomous PID tuning. By training a reinforcement learning agent to interact with a process simulator, AUTO-PILOT learns to adjust PID parameters in real-time to optimize control performance.
Capabilities
From Units to Full Plants
In this software, we connect to a demethanizer column that is simulated in Aspen HYSYS. However, the framework is flexible and can be applied to any process simulator, unit operation, or full plant.
To train a surrogate model, data must be collected for different combinations of PID parameters, set points, and disturbances.
Customization
Specific Control Policies Can Be Learned
Control agents can be trained to have different performance objectives, such as:
- Rapid response with minimal rise time while maintaining acceptable stability and overshoot.
- Slower, smoother system response that prioritizes stability and minimizes abrupt changes in output.
- Tight control around the setpoint with minimal steady-state error and reduced oscillations.
- Using the Integral of Squared Error (ISE) criterion to minimize the accumulated squared tracking error over time.
- And more. You can customize the objectives!
Software
Easy and Intuitive GUI
AUTO-PILOT features an easy-to-use graphical interface that allows users to set up simulations, build surrogate models, train agents, and deploy agents in real time without needing to write code.