Hybrid Process Simulation Integrating Data-Driven Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Process simulation models in industrial plants face challenges in modeling and predicting key performance indicators that cannot be readily measured in real-time or derived from first principles, limiting optimization and control capabilities.
Innovation Solution
Integration of a data-driven model, such as a machine learning soft sensor model, within the process simulation system to predict target process variables, augmenting first principles models and enabling predictions for variables not measurable in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If first principles models are used for process simulation, then physical and chemical accuracy is maintained, but prediction capability for unmeasurable process variables is limited
Solution Approach 1:
The patent combines first principles models with data-driven models to create a hybrid simulation system. The first principles model provides physical/chemical accuracy for measurable variables, while the data-driven model predicts unmeasurable process variables by learning from historical data patterns. This merging allows the system to maintain physical accuracy while extending prediction capability to variables that cannot be directly measured.
Solution Approach 2:
The data-driven model acts as an intermediary between the first principles model and the unmeasurable process variables. Instead of directly measuring unmeasurable variables, the system uses the data-driven model to infer their values from measurable variables and historical data patterns, thereby bridging the gap between what can be measured and what needs to be predicted.
2Adaptability or versatility
If data-driven models are integrated into process simulation, then prediction capability for unmeasurable variables is enhanced, but model complexity increases
Solution Approach 1:
The hybrid model is segmented into two distinct components: a first principles model for physical/chemical calculations and a data-driven model for pattern recognition. Each component handles specific types of predictions, allowing the system to maintain manageable complexity by dividing the modeling task into specialized modules rather than creating a single monolithic complex model.
Solution Approach 2:
The data-driven model serves multiple functions: it predicts unmeasurable process variables, captures non-linear relationships, and handles cases where first principles models are insufficient. By making the data-driven component multi-functional, the system avoids the need for separate specialized models for each prediction task, thereby reducing overall system complexity.
3Measurement precision
If historical process data is used to train data-driven models, then prediction accuracy for target variables is improved, but data processing requirements and computational time increase
Solution Approach 1:
The data-driven model is trained in advance using historical process data before the actual simulation runs. This preliminary training allows the model to capture patterns and relationships from past data, so that during actual operation the model can make predictions quickly without requiring real-time data processing, thereby reducing training time impact on operational performance.
Solution Approach 2:
Instead of processing all historical data in real-time during simulation, the system creates a simplified representation (copy) of historical patterns through the trained data-driven model. This copied knowledge allows the system to make accurate predictions without reprocessing the full historical dataset during each simulation run, significantly reducing computational time requirements.
Data Source
AI summary
Embodiments of the disclosure provide for data-driven predictions within a process simulation system. Some embodiments, generate a data-driven model configured to output first model-predicted data associated with at least one process of an industrial plant. The first model-predicted data may include a predicted value for each of one or more target process variables associated with the at least one process. The data-driven model may be integrated within a process simulation model. The process simulation model may be configured to simulate the execution of the at least one process at one or more operating conditions of a plurality of operating conditions. The process simulation model may be deployed for use.