Probabilistic Machine Learning for EV Owner Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods fail to reliably distinguish electric-vehicle (EV) owners from non-EV owners using power consumption data due to the inability to account for idiosyncratic changes in EV charging patterns, leading to inaccurate disaggregation models.
Innovation Solution
A computer-implemented method and system that uses probabilistic machine-learning models and pattern matching techniques to analyze power load-curve data, identifying characteristic increases and decreases in power consumption associated with EV charging, and inferring EV ownership status through the comparison of usage interval data with reference characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional disaggregation models are used to identify EV owners from power consumption data, then the system can process power usage data, but the identification accuracy is low due to inability to account for idiosyncratic EV charging patterns
Solution Approach 1:
The patent transforms the approach by changing the parameters analyzed - instead of using traditional disaggregation models, it analyzes specific temporal patterns in power consumption data (load curve characteristics, usage interval data) and applies probabilistic machine learning models that can capture the idiosyncratic charging behaviors of EV owners, thereby improving identification accuracy and model reliability
Solution Approach 2:
The patent replaces traditional mechanical disaggregation methods with probabilistic machine learning models that can learn and adapt to complex, non-linear EV charging patterns from training data, enabling more accurate identification of EV owners by substituting deterministic models with data-driven probabilistic approaches
2Measurement precision
If probabilistic machine-learning models are used to analyze power load-curve data, then EV ownership status can be accurately inferred, but the computational complexity and data processing requirements increase
Solution Approach 1:
The patent applies preliminary action by using pre-trained probabilistic machine learning models that have been trained on labeled training data sets containing load-curve data from users with known EV ownership status. This pre-training allows the models to be deployed for inference without requiring complex real-time training, reducing computational complexity during actual EV owner identification while maintaining high accuracy
Solution Approach 2:
The patent segments the problem into distinct phases: (1) model training phase using labeled training data, and (2) inference phase using the configured model on new usage interval data. This segmentation allows complex computational work to be done offline during training, while runtime analysis becomes more efficient and manageable
Data Source
AI summary
The subject disclosure relates to methods and systems for identifying and classifying electric-vehicle (EV) owners. Methods of the subject technology can include steps for generating an initial model based on a plurality of load-curve characteristics, and training the initial model using a training data set to produce a configured model. In some implementations, the methods can also include steps for determining a probabilistic classification for each of a second plurality of users by analyzing load-curve data associated with the second plurality of users using the configured model. Systems and computer readable media are also provided.