Thermostat Schedule Learning Using Preference-Weighted Setpoints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing smart device control systems, such as thermostats, often rely on rules-and-exceptions-based learning approaches that can produce inefficient or erratic schedules due to overemphasis on initial interactions, leading to unsuitable temperature setpoints that fail to accurately reflect user preferences.
Innovation Solution
Implementing a preference function that maps relative weights to device settings based on user behavior, allowing for the generation and optimization of automated schedules that adapt to user preferences over time, with weights adjusted and decayed to prioritize recent behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a rules-and-exceptions-based learning approach is used to generate automated schedules, then the system can learn from user interactions over time, but it may produce errant results and inefficient schedules due to overemphasis on initial interactions and inability to handle scenarios that don't match defined rules
Solution Approach 1:
The patent transforms the discrete rules-and-exceptions approach into a continuous preference function that maps temperature settings and times to preference weights. This parameter transformation allows the system to handle any user interaction scenario continuously rather than relying on predefined discrete rules, thereby improving reliability while maintaining adaptability.
Solution Approach 2:
The patent replaces the mechanical rules-based system with a preference function-based system that uses weighted scoring. Instead of following rigid if-then rules, the system evaluates candidate schedules against a preference function that continuously adapts to user behavior, substituting a flexible mathematical model for a rigid rule-based mechanism.
2Adaptability or versatility
If the rules-and-exceptions-based learning approach includes a large number of rules and exceptions to cover all scenarios, then it may handle more cases, but the device complexity increases significantly
Solution Approach 1:
The preference function serves as a universal model that can handle any user interaction scenario without requiring specific rules for each case. A single preference function with temperature and time parameters can evaluate any candidate schedule, replacing the need for multiple specialized rules and exceptions with one multi-functional evaluation mechanism.
Solution Approach 2:
The patent converts the complex rule-based system into a parameter-based preference function where temperature settings and times are mapped to preference weights. This parameter transformation simplifies the system by replacing numerous discrete rules with a continuous mathematical model that naturally handles all scenarios through parameter variation.
3Extent of automation
If the rules-and-exceptions-based learning approach is used, then automated schedule generation is achieved, but energy consumption increases due to inefficient temperature setpoints
Solution Approach 1:
The preference function learns from actual user interactions and adjusts temperature preferences accordingly. By continuously incorporating feedback from user behavior (manual adjustments, occupancy patterns, weather responses), the system optimizes temperature setpoints to match actual user preferences, thereby reducing energy waste from inappropriate heating or cooling.
Solution Approach 2:
The patent implements a dynamic preference function that evolves over time based on user interactions rather than using static rules. The preference weights are continuously updated to reflect changing user behaviors and patterns, allowing the automated schedule to adapt dynamically and improve energy efficiency as it learns from actual usage data.
Data Source
AI summary
In an embodiment, an electronic device may include storage containing processor-executable instructions and a current setpoint schedule, and a processor configured to execute the instructions. The instructions may cause the processor to control an environmental control system based at least in part on the current setpoint schedule, and to determine the current setpoint schedule by detecting user behavior that indicates satisfaction with setpoints, based at least in part on the detected user behavior, determining a preference function that maps weights indicating user preferences for setpoints, determining candidate setpoint schedules, scoring the candidate setpoint schedules against the weights of the preference function to obtain first scores, where candidate setpoint schedules that best fit the weights of the preference function have the highest first scores, modifying the first scores to obtain second scores, and selecting the candidate setpoint schedule with the highest second score as a new current setpoint schedule.