Machine-Learning PID Autotuning for Dynamic Set-Point Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing PID controller tuning methods are inadequate for real-time adjustments in dynamic environments, such as building automation, where weather conditions and occupancy changes affect temperature control, leading to suboptimal performance and potential discomfort for end-users.
Innovation Solution
The approach incorporates reinforcement learning with preprocessing steps to produce PID controllers that include integral action, allowing for self-tuning based on control error and time differences, ensuring steady-state offset-free tracking and optimal control actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional PID controller tuning methods are used, then the controller structure remains simple, but the controller cannot adapt to dynamic environmental changes such as weather conditions and occupancy variations
Solution Approach 1:
The patent applies dynamics by transitioning from static PID parameter tuning to dynamic reinforcement learning-based parameter adjustment. The controller continuously learns and adapts PID parameters (Kp, Ki, Kd) in real-time based on environmental conditions, occupancy changes, and control performance, making the system dynamically responsive rather than statically fixed
Solution Approach 2:
The patent implements self-service through autonomous self-tuning capability. The reinforcement learning agent automatically adjusts PID parameters without requiring manual intervention from control engineers or building automation specialists. The system learns from historical data and real-time feedback, performing its own optimization and adaptation tasks
2Ease of operation
If manual PID tuning is performed by technicians, then some control performance can be achieved, but it requires specialized knowledge and cannot be adjusted in real-time
Solution Approach 1:
The system performs self-tuning automatically without requiring technician intervention. The reinforcement learning agent continuously optimizes PID parameters based on real-time performance metrics and environmental conditions, eliminating the need for specialized commissioning knowledge while maintaining reliable control
Solution Approach 2:
The patent implements continuous feedback loops where the reinforcement learning agent monitors control performance (temperature deviations, energy consumption, comfort metrics) and uses this feedback to automatically adjust PID parameters. This closed-loop learning process ensures both ease of operation and reliable performance through ongoing optimization
3Productivity
If PID parameters are fixed during commissioning, then installation is quick, but the controller cannot optimize performance under varying conditions
Solution Approach 1:
The patent enables dynamic parameter adaptation where PID gains are no longer fixed but continuously adjusted based on real-time conditions. The reinforcement learning system processes incoming sensor data and environmental information to dynamically recalibrate controller parameters, maintaining both quick initial deployment and ongoing optimization capability
Solution Approach 2:
The system performs preliminary setup quickly with default or rough-tuned parameters, then continuously improves performance through automated reinforcement learning. The preliminary commissioning phase is expedited, while the system autonomously performs the optimization work that would otherwise require extensive manual tuning under various conditions
Data Source
AI summary
An approach for auto tuning a PID controller that may incorporate determining set-points for controlled variables, obtaining controlled variables from a controlled process module, taking a difference between the set-points and the controlled variable as inputs to a proportional, integral and derivative (PID) controller, calculating a loss or reward from the difference between the controlled variables and the set-points, and achieving set-point tracking. The loss or reward needs to be an increasing or decreasing function of a control error value. Also incorporated may be adding loss or reward components based on controlled variables time difference or control action time difference, which may effect a self-tuned performance of the PID controller.