Data-Driven Controller Configuration for Building Energy Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Building management systems face challenges in optimizing energy consumption and comfort due to the complexity of parameters involved, with existing MPC approaches using simplified models that do not consider all relevant information and requiring significant modeling effort and calibration.
Innovation Solution
A computer-implemented method using two data-driven models: a first model trained with pre-known data sets to predict future target variables, and a second model trained with reinforcement learning to determine optimal control actions based on a reward function, balancing energy consumption and comfort goals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If simplified models are used in MPC approaches, then the control problem becomes more tractable, but the prediction accuracy and optimization performance deteriorate
Solution Approach 1:
The patent transforms the control approach by changing from traditional physics-based models to data-driven models that learn parameters directly from operational data. The first model learns system dynamics parameters, while the second model learns optimal control parameters, allowing accurate predictions without complex physical modeling.
Solution Approach 2:
The patent replaces traditional mechanical/MPC control systems with a two-stage machine learning system. The first model (state predictor) replaces physical system modeling, and the second model (control optimizer) replaces traditional control algorithms, achieving both simplicity and accuracy.
2Measurement precision
If detailed models considering all building parameters are used, then prediction accuracy improves, but modeling effort and calibration requirements increase
Solution Approach 1:
The patent enables the system to self-configure by automatically learning building parameters from operational data without requiring manual modeling or calibration. The data-driven models adapt to the specific building characteristics through training on historical data, eliminating the need for expert intervention.
Solution Approach 2:
The patent performs preliminary learning during a training phase using historical operational data before actual control deployment. This preliminary action allows the models to capture building-specific characteristics in advance, so no further calibration is needed during operation.
3Productivity
If traditional MPC approaches are used, then control optimization is achieved, but setup time and configuration costs increase
Solution Approach 1:
The patent replaces time-consuming manual MPC configuration with automated machine learning model training. The system learns optimal control strategies directly from data, eliminating the need for manual model development, parameter tuning, and validation that characterize traditional MPC setup processes.
4Productivity
If more building parameters and sensor data are considered, then control optimization improves, but system complexity increases
Solution Approach 1:
The patent merges multiple data sources and parameters into unified data-driven models. The first model integrates state variables, target variables, and control variables to predict system behavior, while the second model combines these predictions with optimization objectives to determine control actions, simplifying the handling of complex multi-parameter systems.
Data Source
AI summary
A computer-implemented method for configuring a controller for a technical system is provided. The controller controls the technical system based on an output data set determined by the controller for an input data set, wherein the method includes: training a first data driven model with training data including several pre-known input data sets and corresponding pre-known output data sets for the respective pre-known input data sets, where the first data driven model predicts respective future values of one or more target variables for one or more subsequent time points; training a second data driven model with the training data using reinforcement learning with a reward depending on the respective future values of the one or more target variables which are predicted by the trained first data driven model, where the trained second data driven model determines the output data set for the input data set within the controller.