Adaptively learning surrogate model for predicting building system dynamics from system identification model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep neural network (DNN) models require a large amount of historic operational data to train, leading to delays in applying predictive control for building systems like HVAC, resulting in lost energy savings and increased computational resources needed for training.
Innovation Solution
A system identification model is used to generate initial training data, allowing a DNN-based surrogate model to be trained quickly with less complex parameters, reducing the time to value and computational power required, and enabling adaptive learning with real operational data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep neural network model is trained using conventional methods with large amounts of historic operational data, then the model achieves accurate predictions of building system dynamics, but the training time is excessively long and computational resources are heavily consumed
Solution Approach 1:
The system performs preliminary actions by using a system identification model to generate synthetic training data before actual model training begins. This preliminary data generation step enables the surrogate model to be trained quickly with less complex parameters, reducing training time while maintaining prediction accuracy through adaptive learning with real operational data later
Solution Approach 2:
A surrogate model is introduced as an intermediary between the complex deep neural network and the actual building system. The surrogate model with less complex parameters serves as a simplified representation that can be trained faster, while still capturing the essential dynamics for accurate predictions
2Measurement precision
If a deep neural network model is trained using conventional methods with large amounts of historic operational data, then the model achieves accurate predictions of building system dynamics, but the computational resources required for training are excessively high
Solution Approach 1:
The system extracts only the essential dynamics and patterns from the building system operation data using a system identification model. This extraction process creates a simplified representation that captures the core behavior without requiring all the computational power needed for training a full deep neural network on raw data
Solution Approach 2:
The system creates a copy or surrogate model with less complex parameters that replicates the essential behavior of the building system. This surrogate model can be trained more efficiently with fewer computational resources while maintaining adequate prediction accuracy for control applications
3Productivity
If a deep neural network model is trained from scratch with limited historic operational data, then the training process can be completed quickly, but the model lacks sufficient learning data to achieve accurate predictions
Solution Approach 1:
The system performs preliminary data generation using a system identification model to create synthetic training data before the surrogate model training begins. This preliminary action ensures that sufficient training data is available even when historic operational data is limited, enabling both quick training and accurate predictions
Solution Approach 2:
The system is self-sufficient in generating its own training data through the system identification model when external historic data is insufficient. The system uses available data to train the identification model, which then generates the training data needed for the surrogate model, eliminating the need for large external datasets
Data Source
AI summary
Systems and methods for training a surrogate model for predicting system states for a building management system based on generated data from a system identification model are disclosed herein. The system identification model is used to generate predicted system parameters of a zone of the building based on historic data from operation of the building equipment. The surrogate model is trained based on the predicted system parameters from the system identification model. Predicted future parameters of the variable state of the building are generated using the surrogate model. The surrogate model is re-trained based on new operational data from the building equipment. An updated series of predicted future parameters is generated using the re-trained surrogate model.