Hybrid Building Prediction Models for Low-Data Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Building management systems face challenges in predicting relationships between physical plant inputs and outputs, as existing regression models are less accurate without sufficient data, while artificial neural networks require extensive data sets for accurate predictions.
Innovation Solution
A building management system that combines regression and artificial neural network modeling techniques, using a database with a trust region identifier, regression model predictor, distance metric calculator, and combined prediction calculator to determine accurate predictions by leveraging both models' strengths, with methods like weighted averages or Kalman filters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If regression models are used for predictions, then immediate predictions can be made without sufficient data, but prediction accuracy deteriorates
Solution Approach 1:
The system dynamically switches between regression models and ANN models based on data availability. Regression models are used during the pre-operational stage when data is scarce, while ANN models are activated during the operational stage when sufficient data has been collected, optimizing both response time and accuracy at different stages
Solution Approach 2:
The system changes the modeling approach parameter based on data sufficiency conditions. When data is insufficient, regression models with fewer parameters are used; when data is sufficient, ANN models with more parameters are deployed, allowing the system to adapt its complexity to the available information
2Measurement precision
If artificial neural network models are used for predictions, then prediction accuracy improves, but data requirements increase
Solution Approach 1:
The system performs preliminary data collection and evaluation during the pre-operational stage to determine when sufficient data has been accumulated. Only after this preliminary assessment confirms data sufficiency does the system transition to using ANN models, ensuring data requirements are met before deploying the more data-intensive modeling approach
Solution Approach 2:
The system dynamically adjusts its data requirements based on the operational stage. During the pre-operational stage, regression models require minimal data; during the operational stage, the system transitions to ANN models that leverage the accumulated data corpus, optimizing the balance between data requirements and prediction accuracy
3Device complexity
If a single modeling technique is used, then system complexity is reduced, but prediction reliability deteriorates
Solution Approach 1:
The system merges regression modeling and ANN modeling into a unified hybrid framework. The combined prediction calculator integrates outputs from both model types, leveraging the immediate responsiveness of regression models and the high accuracy of ANN models to achieve superior overall reliability
Solution Approach 2:
The system introduces a combined prediction calculator as an intermediary component that reconciles predictions from both regression and ANN models. This mediator synthesizes the strengths of both approaches, using weighted combinations or ensemble methods to produce more reliable predictions than either model could achieve alone
Data Source
AI summary
A method for initiating and automatically improving model-driven operations in a low-data scenario includes creating a regression model using pre-operation data prior to initiating the model-driven operations, using the regression model to initiate and perform the model-driven operations during an operational stage, collecting operational data during the operational stage, creating a first artificial neural network model using the operational data, transitioning from using the regression model to perform the model-driven operations to using the first artificial neural network model to perform the model-driven operations responsive to the operational data satisfying a first sufficiency threshold.