Predictive Control with Forgetting Factor for Model Mismatch
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Model predictive control systems face performance degradation due to future prediction errors when the model used is significantly different from the actual control target or when unknown disturbances occur, leading to suboptimal control outcomes.
Innovation Solution
A control device that introduces a forgetting factor to asymptotically forget past predicted values, thereby reducing the adverse effects of future prediction errors, and includes a forgetting prediction time series storage unit to update prediction data efficiently, allowing for improved control performance even with limited calculation resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If model predictive control uses a fixed plant response model for future prediction, then the control system maintains simplicity in model management, but control performance deteriorates when the model significantly differs from the actual control target or when unknown disturbances occur
Solution Approach 1:
The patent applies dynamics by transitioning from a fixed plant response model to a dynamic model that evolves over time through online learning. The model parameters are continuously updated based on new input-output data, allowing the system to adapt to changing conditions while maintaining computational efficiency through incremental parameter updates rather than complete model re-identification.
Solution Approach 2:
The patent implements feedback by using the difference between actual controlled variable values and predicted values to update model parameters online. This feedback mechanism allows the system to continuously learn from prediction errors and adjust the plant response model accordingly, improving both reliability and adaptability simultaneously.
2Adaptability or versatility
If the control system continuously updates model parameters online, then the adaptability to changing conditions improves, but the calculation complexity and processing time increase
Solution Approach 1:
The patent applies partial action by selectively updating only the necessary model parameters based on the excitation level and prediction error, rather than performing complete model re-identification at every control step. This approach maintains adaptability while significantly reducing computational burden by focusing updates only where needed.
Solution Approach 2:
The patent implements parameter changes by transitioning from fixed model parameters to dynamically adjustable parameters that are updated online. The system modifies model parameters incrementally based on incoming data and prediction errors, enabling continuous adaptation without requiring complex computational resources for full model re-identification.
3Reliability
If the system forgets past predicted values asymptotically, then the impact of prediction errors is reduced and control performance is maintained, but the amount of stored prediction data decreases
Solution Approach 1:
The patent applies periodic action through the forgetting mechanism that systematically reduces the weight of past predicted values over time. This periodic forgetting allows the system to maintain stable control performance by preventing accumulation of prediction errors while retaining sufficient historical information for accurate modeling through the exponential weighting scheme.
Data Source
AI summary
A control device is configured to calculate a difference between a target value and a current value of a controlled variable for a control target; store predicted values obtained by forgetting a predicted value of the controlled variable that is predicted in a past time by a plant response function; calculate a corrected difference from the difference and a predicted value of the controlled variable after a predetermined lookahead length elapses, based on the predicted values; calculate a change in a value of a manipulated variable based on the corrected difference and a predetermined control gain; calculate a new value of the manipulated variable by adding the change of the value of the manipulated variable to the value of the manipulated variable; and output the new value of the manipulated variable to the control target so that a value of the controlled variable is caused to track the target value.