Hybrid AI Control Platform for Real-Time Model Reconciliation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hybrid models for enterprise applications, such as HVAC and energy efficiency, fail to provide real-time accurate and interpretable recommendations due to limitations in data-based and physics-based models, including inadequate training on feature extraction, virtual sensors, and handling of missing data, leading to inconsistent and inaccurate results.
Innovation Solution
A method integrating data-based and physics-based models by configuring data-based models with historical field data and physics-based models with simulated data, using feature extraction and virtual sensors to validate and reconcile data, and providing feedback for real-time predictions and control actions, leveraging machine learning and artificial intelligence techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a data-based model is used for real-time predictions, then execution speed is improved, but accuracy deteriorates under out-of-range operating conditions
Solution Approach 1:
The patent combines data-based models with physics-based models into a hybrid framework. The physics-based model provides physically consistent predictions for out-of-range conditions, while the data-based model handles normal operating conditions with high execution speed. This merging resolves the contradiction by leveraging the strengths of both approaches.
Solution Approach 2:
The patent introduces a data reconciliation layer as an intermediary that validates and harmonizes predictions from both data-based and physics-based models. This mediator ensures that predictions remain accurate across all operating conditions while maintaining the execution speed benefits of data-based models.
2Reliability
If a physics-based model is used for predictions, then reliability is improved, but computation time increases
Solution Approach 1:
The patent applies partial action by using physics-based models only when necessary (for out-of-range conditions or validation), rather than continuously. For normal operating conditions, the system relies on faster data-based models, thus reducing overall computation time while maintaining reliability where needed.
Solution Approach 2:
The patent segments the prediction process into different operational zones: data-based models handle normal operating ranges, while physics-based models handle out-of-range conditions. This segmentation reduces computation time by avoiding unnecessary physics-based calculations during normal operations.
3Adaptability or versatility
If a data-based model is trained on historical data, then adaptability is improved, but data quality requirements increase
Solution Approach 1:
The patent implements a feedback mechanism where predictions from the hybrid model are validated against physical principles and historical data. This feedback loop identifies and corrects data quality issues, allowing the system to maintain adaptability while reducing the stringency of data quality requirements through continuous validation and refinement.
4Adaptability or versatility
If existing hybrid models are used, then integration of data-based and physics-based models is achieved, but real-time accuracy deteriorates due to inadequate feature extraction
Solution Approach 1:
The patent performs preliminary feature extraction and data reconciliation before the main prediction process. By pre-processing and validating features in advance, the system achieves both real-time accuracy and proper model integration, resolving the contradiction between integration capability and prediction precision.
Data Source
AI summary
The present disclosure relates to development of a self-service artificial intelligence platform by integrating data-based model with physics-based model and vice-versa to generate real-time recommendations and control actions. Further, the present disclosure provides the system and method for at least one of data collection and preparation, developing a hybrid system/control model, and developing a physics-based model driven by data-based model and vice versa to generate real-time recommendations and control actions.