Hybrid System State Estimation for Non-Intrusive Load Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current non-intrusive load monitoring systems lack accurate disaggregated reporting of individual appliance energy consumption and rely on cumbersome multiple metering devices or require extensive training data, often missing physical dynamics and inducing false positives/negatives.
Innovation Solution
A hybrid system models appliances with continuous dynamics as discrete states, using state estimation techniques to determine active appliances without an event detector, directly accounting for physical dynamics and leveraging transition probabilities for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple metering devices are distributed throughout the building to measure individual device consumption, then measurement precision is improved, but device complexity and installation cost increase
Solution Approach 1:
The patent segments the total power consumption signal into individual appliance contributions through signal processing and pattern recognition algorithms. By analyzing transient patterns, steady-state characteristics, and load signatures in the aggregate power signal, the system virtually separates individual appliance measurements without physical separation of measurement devices.
Solution Approach 2:
The patent introduces an intermediary processing layer (signal analysis algorithms, pattern recognition systems, and computational models) that acts as a mediator between the single-point measurement and individual appliance consumption data. This intermediary transforms the aggregate measurement into disaggregated information through computational techniques.
2Productivity
If event detectors are used to detect appliance transitions, then productivity is improved, but reliability deteriorates due to false positives and missed events
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously monitors power signal characteristics, compares them against known appliance patterns, and adjusts its detection thresholds and classification decisions based on accumulated evidence. The system uses historical data and pattern matching to refine its detection accuracy over time, reducing false positives while maintaining rapid detection capability.
Solution Approach 2:
The patent changes detection parameters dynamically based on the operating context, appliance types detected, and signal characteristics observed. By adjusting detection thresholds, time windows, and pattern matching criteria based on real-time conditions, the system maintains high productivity while adapting to reduce false positives and missed events.
3Measurement precision
If classification algorithms use extensive features from signal characteristics, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent extracts only the most discriminative features from the power signal that are essential for appliance classification, rather than processing all possible signal characteristics. By identifying and extracting key transient patterns, steady-state signatures, and load characteristics that uniquely identify appliances, the system achieves high classification accuracy with reduced computational complexity.
Solution Approach 2:
The patent applies partial action by focusing on the most critical signal features and time periods that provide sufficient classification accuracy without analyzing every aspect of the power signal. The system processes only the necessary portion of the signal characteristics needed for reliable appliance identification, avoiding unnecessary computational overhead.
4Measurement precision
If training data is collected to train classification algorithms, then measurement precision is improved, but loss of time increases due to tedious training process
Solution Approach 1:
The patent performs preliminary action by pre-programming classification algorithms with prior knowledge of typical appliance power signatures, transient patterns, and load characteristics. Instead of requiring extensive on-site training data collection, the system comes pre-loaded with reference patterns and classification rules that can be quickly adapted to specific buildings with minimal training overhead.
Solution Approach 2:
The patent enables self-service by allowing the classification system to automatically learn and adapt to building-specific patterns through continuous operation, without requiring manual training intervention. The system uses unsupervised learning techniques and pattern recognition to automatically refine its classification accuracy over time as it observes actual appliance usage patterns in the building.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for non- intrusively monitoring a load including a plurality of appliances includes retrieving a plurality of mathematical models for modeling operation of a respective subset of the appliances. A value of a respective operational parameter (X) is predicted for each of the subsets of appliances. An output (Y) of the load is measured. A respective value of each of the operational parameters is calculated based on the mathematical models and the outputs of the load. The predicting, measuring and calculating steps are repeated until a metric pertaining to a difference error between the measured output and the predicted output calculated from the operational parameter and the mathematical models is equal to or below a threshold for one of the subsets of appliances. It is decided that the one subset of appliances is currently operating whose metric pertaining to a difference error is equal to or below the threshold.