Hybrid Feedback Control with Limited ML Output for Stable Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Control devices using machine learning controllers face reliability issues due to the possibility of generating abnormal control commands when encountering situations significantly different from the learning data, particularly in large-scale deep neural networks, making it difficult to predict their responses.
Innovation Solution
A feedback control device that combines a first control unit using PID control with a second control unit employing machine learning, where the machine learning unit includes a limiter to restrict the output range of the manipulated variable, ensuring stability and accuracy by adding the outputs from both units to the controlled object.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a controller based on machine learning is used to improve control performance, then control accuracy can be enhanced, but reliability deteriorates due to the possibility of generating abnormal control commands for situations greatly different from learning data
Solution Approach 1:
The controller is divided into two independent control units: a first control unit using classic control theory (e.g., PID control) and a second control unit using machine learning. Each unit operates independently to generate control commands, allowing the system to leverage the strengths of both approaches while mitigating their individual weaknesses.
Solution Approach 2:
The outputs from the first control unit and the second control unit are combined through an adder to generate the final control command. This merging approach allows the system to benefit from both classic control reliability and machine learning accuracy, creating a hybrid controller that achieves superior overall performance.
2Adaptability or versatility
If a large-scale deep neural network is used to improve control adaptability, then the controller can handle more complex situations, but reliability worsens because it becomes difficult to grasp how the controller responds to input
Solution Approach 1:
The control system is segmented into a machine learning-based second control unit that handles adaptability and a classic control-based first control unit that provides predictable baseline control. This segmentation allows the complex neural network to be used for adaptability while the classic controller ensures predictable behavior.
Solution Approach 2:
The first control unit acts as an intermediary that provides a predictable control baseline. By combining its output with the machine learning unit's output, the system maintains predictability while incorporating adaptive capabilities from the neural network.
3Measurement precision
If only a machine learning controller is used to improve control accuracy, then precision can be enhanced, but stability worsens due to potential abnormal outputs in unexpected situations
Solution Approach 1:
The first control unit provides beforehand cushioning by generating stable control commands based on proven classic control theory. This creates a safety net that prevents the system from becoming unstable when the machine learning unit produces abnormal outputs in unexpected situations.
Solution Approach 2:
By merging the outputs of the stable first control unit and the precise second control unit, the system achieves both stability and precision. The combination ensures that even when one unit produces suboptimal output, the other unit compensates to maintain overall system stability.
Data Source
AI summary
A feedback control device that takes information regarding a control deviation between a measured value and a desired value of a controlled object as input, and outputs a manipulated variable for the controlled object, includes: a first control unit that takes information regarding the control deviation as input, and outputs a manipulated variable for the controlled object; a second control unit that takes information regarding the control deviation as input, and that includes a learning control unit in which a parameter for outputting a manipulated variable for the controlled object is determined by machine learning; and an adder that adds a first manipulated variable output from the first control unit and a second manipulated variable output from the second control unit. A manipulated variable from the adder is output to the controlled object, and the second control unit includes a limiter that limits the second manipulated variable.