Hybrid AI Control Models for Real-Time Enterprise Recommendations
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 training data and integration of physics-based and data-based models, leading to inconsistent and inaccurate control actions.
Innovation Solution
A method that configures a data-based model using historical field data and a physics-based model using simulated data, with data reconciliation, feature extraction, and virtual sensors to validate and train both models, providing feedback for real-time predictions and control actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a data-based model is used to provide real-time controls and recommendations, then execution speed and accuracy in normal operating conditions are improved, but the model requires large volume of upfront data and long training period, and fails to cover out-of-range operating conditions
Solution Approach 1:
The patent applies preliminary action by pre-training the data-based model using synthetic data generated from physics-based models before deployment. This pre-training phase prepares the model to handle various operating conditions, reducing the need for extensive on-site data collection and training time while maintaining execution speed.
2Reliability
If a data-based model is trained to cover large and out-of-normal operating conditions, then model robustness is improved, but data quality requirements increase and model complexity increases
Solution Approach 1:
The patent uses physics-based models as an intermediary to generate synthetic training data that covers out-of-range operating conditions. This intermediary approach allows the data-based model to learn from realistic but artificially generated scenarios without requiring actual field data from rare conditions, reducing model complexity while improving robustness.
3Measurement precision
If a physics-based model is used to provide controls and recommendations, then data requirements are reduced and theoretical accuracy is improved, but significant run-time computation is required and the model does not account for realistic problems such as building deterioration
Solution Approach 1:
The patent merges physics-based models with data-based models in a hybrid architecture. The physics-based model provides theoretical accuracy and handles out-of-range conditions, while the data-based model provides fast execution for normal operations. This combination reduces overall computation time while maintaining theoretical accuracy where needed.
4Loss of information
If a physics-based model is used, then understanding of underlying physical principles is improved, but the model works in ideal scenarios and does not consider realistic problems such as building deterioration and corrosion
Solution Approach 1:
The patent implements feedback by training the data-based model on actual field data that includes realistic degradation patterns. The model learns from historical data containing building deterioration and corrosion effects, allowing it to provide accurate recommendations for real-world conditions while the physics-based model maintains theoretical understanding.
5Reliability
If a hybrid model integrating both data-based and physics-based models is used, then limitations of individual models are eliminated, but model integration complexity and data reconciliation requirements increase
Solution Approach 1:
The patent applies segmentation by clearly dividing the hybrid model into distinct physics-based and data-based components, each handling specific tasks. The physics-based model generates synthetic data and handles out-of-range conditions, while the data-based model handles normal operations. This segmentation reduces integration complexity by defining clear boundaries and interfaces between components.
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.