Hybrid Plant State Prediction for Impurity-Rich Recycling Feedstock
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In chemical recycling plants, it is challenging to execute operation assistance using a unified physical model due to variations in raw materials, which degrades prediction accuracy and validity, making it difficult to manage the processing of waste plastics with high impurity levels.
Innovation Solution
An information processing device that combines a physical model with machine learning models to predict and assist in the operation of chemical recycling plants, using optical spectrum data and hybrid models to complement the physical model's limitations, thereby enhancing prediction accuracy and safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a single physical model is used to match respective variations in raw materials, then the model can be simplified and easier to operate, but accuracy of prediction is degraded
Solution Approach 1:
The patent combines multiple specialized machine learning models (each trained for specific raw material types) with a unified physical model. This hybrid approach merges the strengths of both: the physical model provides a simplified operational framework while the multiple ML models collectively handle various raw material variations, thereby maintaining ease of operation while improving prediction accuracy.
Solution Approach 2:
The patent creates a universal system that can handle multiple types of raw materials (waste plastics, biomass, etc.) by employing multiple machine learning models that are selectively applied based on the input material type. This multi-functional approach allows the system to maintain a single operational interface while adapting to diverse materials, resolving the contradiction between operational simplicity and prediction accuracy.
2Device complexity
If a single physical model is used for all variations, then device complexity is reduced, but validity and safety of operation assistance are decreased
Solution Approach 1:
The patent segments the prediction system into multiple specialized machine learning models, each trained on data for specific raw material types or process conditions. This segmentation allows each model to specialize in particular variations, improving reliability and validity for each specific case while the overall system structure remains manageable through modular design.
Solution Approach 2:
The patent creates a composite modeling approach by integrating multiple machine learning models with a physical model. This composite system combines the interpretability and simplicity of physical models with the adaptive accuracy of data-driven ML models, achieving both reduced operational complexity and enhanced reliability through the synergistic combination of different modeling paradigms.
3Measurement precision
If multiple machine learning models are used to handle raw material variations, then prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent implements a dynamic model selection mechanism that automatically selects or activates appropriate machine learning models based on the type of raw material being processed. This dynamic approach allows the system to use multiple specialized models only when needed, maintaining high prediction accuracy while avoiding the constant complexity of managing all models simultaneously, as the system adapts its complexity to the specific operational context.
Data Source
AI summary
An information processing device includes a controller that: acquires a first prediction result that indicates a state of a plant based on a physical model; acquires other prediction results that indicate a state of the plant, based on machine learning models that are generated based on data concerning the plant; and outputs information concerning a state of the plant based on the first prediction result and the other prediction results.