Monitoring Device Waveform Analysis Reduces Training Data Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As the number of monitoring target electrical devices increases, the number of training feature amounts required for inferring power consumption also increases significantly, leading to a substantial processing load due to the need for numerous combinations of training feature amounts and sum training feature amounts.
Innovation Solution
A monitoring device and method that acquires unit-specific waveform data to infer operation states of electrical devices by analyzing differences in feature amounts over time, reducing the need for extensive training feature amounts by focusing on changes and combinations of operation states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of monitoring target electrical devices increases, then the monitoring coverage and precision improve, but the number of training feature amounts and processing load increase significantly
Solution Approach 1:
The patent segments the monitoring process into two distinct stages: a first inference unit that infers operation state changes based on waveform data differences, and a second inference unit that infers specific power consumption based on the operation state changes. This segmentation allows the system to handle multiple electrical devices without requiring all possible combinations of training feature amounts, as each stage processes information independently and builds upon the previous stage's results.
Solution Approach 2:
The patent performs preliminary action by first inferring operation state changes (on/off transitions) before inferring specific power consumption values. The first inference unit processes waveform data to determine when devices are turned on or off, and this preliminary classification is then used by the second inference unit to determine power consumption. This preliminary action reduces the complexity by pre-filtering the data before the main inference process.
2Measurement precision
If comprehensive training feature amounts are prepared for all electrical devices, then the inference accuracy improves, but the processing load and computational complexity increase substantially
Solution Approach 1:
The patent divides the inference process into two functional segments: the first inference unit handles operation state change detection using waveform data differences, while the second inference unit handles power consumption inference using the results from the first unit. This segmentation allows the system to achieve accurate power consumption measurement without processing all possible combinations of training feature amounts simultaneously, as each segment processes a subset of the data independently.
Solution Approach 2:
The first inference unit performs preliminary action by detecting operation state changes (when devices are turned on or off) before the second inference unit calculates power consumption. This preliminary classification reduces the processing load by pre-identifying which devices are active, allowing the second unit to only process power consumption data for relevant devices rather than all devices in the system.
3Measurement precision
If sum training feature amounts for all combinations of electrical devices are prepared, then the monitoring accuracy improves, but the device complexity and data requirements increase exponentially
Solution Approach 1:
The patent segments the training data into device-specific training data stored in a storage unit, eliminating the need to prepare sum training feature amounts for all possible combinations of electrical devices. The first inference unit processes waveform data differences to infer operation state changes for individual devices, and the second inference unit uses these results to determine power consumption. This segmentation reduces the quantity of training data from exponential combinations to individual device profiles.
Solution Approach 2:
The system performs preliminary action by first inferring operation state changes for each device individually using waveform data, before aggregating this information to determine overall power consumption. This preliminary individual analysis replaces the need to prepare comprehensive sum training feature amounts for all device combinations, reducing the data quantity while maintaining accuracy through a two-stage inference process.
Data Source
AI summary
A monitoring device (10) includes a unit-specific waveform data acquisition unit (11) that acquires waveform data in a unit in which electrical devices are installed; a first inference unit (13) that infers change in operation states of at least some of the electrical devices based on a first monitoring difference group including at least one of at least one kind of feature amount extracted from waveform data of a difference between waveform data of a first timing and waveform data of a second timing in the waveform data, and a differences of at least one kind of feature amounts extracted from the waveform data of the first timing and the waveform data of the second timing, and training difference information regarding a difference between a first operation state and a second operation state of each of the electrical devices; and a second inference unit (14) that infers an operation state of each of the electrical devices based on an inference result of the first inference unit (13).