Control engine system and method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Building automation systems (BAS) often operate sub-optimally due to the complexity and resource-intensive nature of model-based predictive control (MPC), which requires significant time and computational resources, creating a knowledge barrier for practitioners and concerns about liability and risk in live site implementations.
Innovation Solution
A control engine system and method that utilizes empirical rule extraction heuristics, including a data preprocessing engine, random forest generator, tree selector, and BAS code generator, to generate a supervisory decision tree for BAS controllers, enabling simpler configuration and deployment of MPC algorithms while maintaining energy savings and explainability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If model-based predictive control (MPC) is implemented in building automation systems, then control performance and energy savings are improved, but system complexity and computational requirements increase significantly
Solution Approach 1:
The patent extracts the core decision-making logic from complex MPC algorithms and isolates it into a supervisory controller that generates simplified decision rules. These rules are then implemented in standard BAS controllers, separating the computational burden from the control execution layer while maintaining energy optimization benefits
Solution Approach 2:
The patent transforms MPC's continuous mathematical optimization parameters into discrete decision rules with if-then logic. By changing the parameter representation from continuous mathematical variables to discrete control rules, the system achieves comparable energy savings with significantly reduced computational requirements suitable for standard BAS controllers
2Manufacturing precision
If detailed MPC algorithms are used, then control precision and optimization performance are improved, but ease of operation and explainability deteriorate
Solution Approach 1:
The patent introduces a supervisory controller as an intermediary layer between the detailed MPC algorithm and the BAS controller. This intermediary extracts essential decision logic into human-readable if-then rules, making the control strategy explainable to operators while maintaining the precision benefits of MPC through the supervisory decision framework
Solution Approach 2:
The patent creates simplified copies of the MPC decision-making logic in the form of supervisory decision trees. These copies capture the essential control intelligence in an interpretable format that maintains precision for critical decisions while being easily understandable by building operators
3Measurement precision
If historical time series data is collected and processed for MPC training, then model accuracy is improved, but loss of time and computational resources increase
Solution Approach 1:
The patent performs preliminary data processing and model training offline before deployment. Historical time series data is collected and processed in advance to generate the supervisory decision rules, which are then deployed to the BAS controller. This preliminary action separates the time-consuming data processing from real-time operation, maintaining model accuracy while reducing implementation time
Solution Approach 2:
The patent uses a partial approach to data processing by selecting only the most relevant features and historical periods for training the supervisory controller. Rather than processing all available historical data, the system identifies and uses the critical subset needed for accurate decision rules, reducing processing time while maintaining sufficient model accuracy
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure provides a control engine system (1) connected to a BAS (6), the control engine system (1) includes a data preprocessing engine (2), a random forest generator (3), a tree selector (4), and a BAS code generator (5). The data preprocessing engine (2) receives a data from the BAS (6) and performs preprocessing on the data and generating a training data accordingly. The random forest generator (3) receives the training data and generates a random forest accordingly. The tree selector (4) receives the random forest for selecting a final decision tree. The BAS code generator (5) receives the final decision tree and encodes a supervisory decision tree according to the final decision tree, the BAS code generator (5) outputs the supervisory decision tree to the BAS (6).