Hierarchical Predictive Control for Noisy Nonlinear Plant Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive control systems face challenges in constructing accurate models for complex, nonlinear systems due to noise and the difficulty in learning the relationship between inputs and outputs, particularly when dealing with unidirectional causal relationships between forecast and controlled variables.
Innovation Solution
A predictive control system that includes a model generator to train subprocess and main process models independently using historical data, leveraging the hierarchical structure of the plant, where forecast variables affect controlled variables but not vice versa, and a controller to execute predictive control processes using these models to optimize plant operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If system identification is performed using traditional black-box or gray-box methods to generate predictive models, then the model can capture complex nonlinear relationships, but the model accuracy deteriorates due to noise in the data and difficulty in learning input-output relationships
Solution Approach 1:
The system divides the plant into multiple subprocesses and generates separate predictive models for each subprocess rather than a single monolithic model. This segmentation allows each subprocess model to be trained on more focused data, improving accuracy while maintaining the ability to capture complex relationships through the composition of multiple specialized models.
Solution Approach 2:
The system introduces forecast variables as intermediary elements between manipulated variables and controlled variables. These forecast variables serve as mediators that capture the unidirectional causal relationships in the system, improving model accuracy by explicitly representing the flow of influence through the plant while maintaining adaptability to complex nonlinear behaviors.
2Adaptability or versatility
If a single comprehensive predictive model is used for the entire plant, then the model can handle all processes, but the device complexity increases and model training becomes more difficult due to noise and intertwined relationships
Solution Approach 1:
The plant is segmented into multiple subprocesses with individual predictive models for each. This reduces the complexity of each individual model while maintaining comprehensive plant coverage through the aggregation of subprocess models. The hierarchical structure organizes the complexity in a manageable way.
Solution Approach 2:
The system introduces a hierarchical dimension to the model structure, organizing models at different levels (subprocess level and plant level). This dimensional organization allows the system to manage complexity by distributing computational tasks across multiple levels rather than concentrating all complexity in a single model.
3Ease of manufacture
If traditional system identification methods are used without considering unidirectional causal relationships, then the model can be simpler to construct, but the model accuracy deteriorates because it fails to capture the true input-output relationships
Solution Approach 1:
Forecast variables are introduced as intermediary elements that explicitly represent unidirectional causal relationships. This approach maintains relative simplicity in model construction while significantly improving prediction accuracy by capturing the true flow of influence from manipulated variables through forecast variables to controlled variables.
Solution Approach 2:
The system adds a temporal and causal dimension to the model structure by introducing forecast variables that represent future states. This dimensional enhancement allows the model to capture causal relationships more accurately without excessively complicating the construction process.
Data Source
AI summary
A predictive control system for a plant includes a model generator and a controller. The model generator is configured to train a subprocess model representing a subprocess of the plant using historical values of forecast variables and manipulated variables. The model generator is further configured to train a main process model representing a main process of the plant using historical values of controlled variables, forecast variables, and manipulated variables. The controller is configured to execute a predictive control process using the subprocess model and the main process model to control operation of the plant. The predictive control process includes using the subprocess model to predict future values of the forecast variables, using the main process model to predict future values of the controlled variables based on the predicted future values of the forecast variables, and controlling operation of the plant based on the predicted future values of the controlled variables.