Machine teaching with method of moments for mechanical systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing control systems for mechanical systems, such as HVAC systems, often operate with fixed setpoints that do not adapt to variable demands or environmental changes, leading to inefficiency, excessive wear, and potential damage, while machine learning models struggle to explore a full range of setpoint values due to limited training data.
Innovation Solution
A machine learning model is trained using delta values derived from fixed setpoints and environmental fluctuations, with a method of moments converting these delta values back to absolute setpoints, and a regression brain is used to simplify on-site control, leveraging states like outside air temperature and load to generate optimal setpoints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If fixed setpoint values are used for training machine learning models, then the system operates with stable control, but the model cannot explore the full range of possible setpoint values leading to suboptimal performance
Solution Approach 1:
The patent applies dynamics by transforming fixed setpoint values into dynamic delta values that fluctuate within the allowable range. This is achieved by computing delta setpoints as differences between fixed setpoints and varying reference values, enabling the machine learning model to explore diverse setpoint configurations while maintaining system stability through the method of moments conversion back to absolute values.
Solution Approach 2:
The patent changes the parameter representation from absolute fixed setpoint values to relative delta values. By substituting fixed setpoints with delta setpoints that vary within bounds, the training data gains diversity and enables broader exploration of the setpoint space, while the method of moments ensures the transformed parameters remain physically meaningful.
2Loss of information
If operators experiment with different setpoint values to train models, then more diverse training data is obtained, but the experiments are disruptive, expensive, and time-consuming
Solution Approach 1:
The patent applies preliminary action by pre-computing delta setpoints from existing fixed setpoints and allowable ranges before training begins. This preprocessing step creates diverse training data without requiring actual system experiments, saving time and avoiding disruption while still providing the model with varied setpoint configurations to learn from.
Solution Approach 2:
The patent creates synthetic training data by copying and transforming existing setpoint information into delta representations. Instead of conducting physical experiments to generate diverse data, the system generates virtual delta setpoints mathematically from the fixed setpoints and their allowable ranges, providing sufficient training diversity without real-world experimentation.
3Ease of manufacture
If fixed setpoints are used in the training data, then data collection is simple, but the model output is limited to a narrow range of setpoint values
Solution Approach 1:
The patent segments the setpoint information into two components: the fixed base setpoint and the variable delta component. By separating these elements and training on their difference (delta setpoint), the system maintains the simplicity of collecting fixed setpoint data while enabling the model to generate a broader range of output values through delta variations.
Solution Approach 2:
The patent adds a dimensional transformation by converting absolute setpoint values into relative delta values. This dimensional change from absolute to relative space allows the model to explore variations around fixed setpoints, effectively expanding the output range without complicating the data collection process.
Data Source
AI summary
The techniques disclosed herein enable utilizing a full range of setpoint values to control a mechanical system. A machine learning model is trained with states collected from the mechanical system. Some of the states may have little to no variation, limiting exploration of possible setpoint values when training the model. To enable a more thorough exploration of possible setpoint values, the states are augmented with a fluctuating delta value that is derived from a fixed setpoint value. For example, a delta outside air temperature may be computed by subtracting outside air temperature, which fluctuates, from a fixed chilled water setpoint. A method of moments computation converts delta values inferred by the model back into absolute values. The absolute values are used to compute a regression equation that is usable by the mechanical system to compute a setpoint action for a given set of input states.