Smart Meter Split-Model Training for Private Power Data Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current smart electric meters lack on-device intelligent data analysis capabilities, leading to issues such as user privacy concerns, data transmission congestion, and decision response delays due to insufficient memory, computing, and communication resources, hindering the effective use of distributed data for improved model performance.
Innovation Solution
An electric meter with an electric power data acquisition device, memory, and processor is equipped to train a part of an electric power data analysis model, with the assistance of an edge server, performing data processing and backward propagation to enhance data analysis efficiency and accuracy while preserving user privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If smart electric meters transmit collected data in bulk to centralized data management systems, then data management is simplified, but user privacy leakage and data transmission congestion occur
Solution Approach 1:
The patent introduces edge servers as intermediaries between smart electric meters and centralized data management systems. These edge servers perform preliminary data processing and analysis locally, filtering and aggregating data before transmission to the central system. This intermediary layer reduces the volume of transmitted data while maintaining data management simplicity, and prevents raw user privacy data from being exposed during transmission and storage.
2Loss of information
If smart electric meters transmit all collected data to centralized systems, then comprehensive data analysis is achieved, but data transmission congestion and decision response delays occur
Solution Approach 1:
The patent segments the data processing function into multiple levels: (1) Smart electric meters perform local data collection and preliminary processing; (2) Edge servers perform regional data aggregation and analysis; (3) Centralized systems perform overall coordination and long-term analysis. This segmentation enables critical decisions to be made at the edge level with minimal latency, while comprehensive analysis is performed at all levels, thus reducing transmission congestion and response delays without sacrificing analytical completeness.
3Extent of automation
If complex model training is performed on smart electric meters, then on-device intelligent analysis capability is improved, but memory, computing and communication resources become insufficient
Solution Approach 1:
The patent transitions the model training problem from a single-device constraint to a distributed multi-dimensional system. Instead of requiring full model training on resource-constrained smart meters, the system distributes different components of the training process across multiple dimensions: smart meters provide local data and perform lightweight processing, edge servers handle intermediate model training and aggregation, and centralized systems coordinate overall model development. This dimensional distribution enables intelligent analysis capability without overwhelming individual device resources.
4Measurement precision
If distributed data from smart electric meters is used to improve model performance, then model accuracy is enhanced, but user privacy concerns hinder data utilization
Solution Approach 1:
The patent extracts and removes sensitive personal information from raw electricity consumption data before using it for model training. The system processes only anonymized and aggregated features that preserve the statistical patterns necessary for accurate model training while eliminating identifiable user information. This extraction approach enables distributed data utilization for improved model performance while addressing user privacy concerns by removing harmful personal identifiers.
Data Source
AI summary
An electric meter, an edge server, and a system for electric power data analysis are disclosed. The electric meter includes an electric power data acquisition device for collecting electric power data; a memory for storing a first part of an electric power data analysis model; and a processor for training the first part, including: using the first part to process electric power data to generate a first representation of the electric power data; sending the first representation to an edge server; receiving a second representation of the electric power data generated by processing the first representation; using the first part to process the second representation to generate a first electric power data analysis result; using the first electric power data analysis result to perform backward propagation on the first part to generate an updated first part and a gradient of the second representation.