Hybrid Vehicle Driving Pattern Prediction Using Event-Specific Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting a driver's behavior in hybrid vehicles fail to accurately reflect individual driving patterns, leading to inefficient predictions due to the averaging of diverse driving styles and events, resulting in suboptimal engine and motor control.
Innovation Solution
A method and system for predicting a driving pattern in hybrid vehicles that involves acquiring current driving information, determining upcoming events and driving styles, and generating acceleration/deceleration predictions using pre-learned models, incorporating sensors and machine-learning algorithms to tailor predictions to specific drivers and events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single prediction model is generated based on acquisition of driving data of a plurality of drivers and learning of the drivers' patterns, then the prediction system can be simplified and easier to implement, but prediction accuracy greatly varies depending on drivers' peculiarities and cannot accurately predict individual driving behavior
Solution Approach 1:
The patent divides the single prediction model into multiple event-specific prediction models (first prediction model for speed camera events, second prediction model for tollgate events, third prediction model for intersection events). Each model is specialized for a particular type of event, allowing accurate prediction of individual driving behavior for different situations while maintaining manageable system complexity through modular structure.
Solution Approach 2:
The patent applies different prediction models based on the specific type of upcoming event detected. Instead of using a uniform model for all situations, the system selects and applies the appropriate specialized model (first, second, or third prediction model) according to the event type, ensuring locally optimized prediction accuracy for each event category.
2Adaptability or versatility
If learning of driving pattern is conducted based on samples derived from multiple drivers with different behaviors (e.g., driver A decelerates in advance before speed camera, driver B decelerates suddenly), then more diverse driving patterns can be covered, but the calculated prediction result merely corresponds to an intermediate value that is a simple mean value and is not suitable for either driver
Solution Approach 1:
The patent segments the learning process into event-specific learning tasks. Each prediction model (first, second, third) is trained separately on driving data corresponding to its specific event type. This allows the system to capture the full range of individual driving behaviors for each event type without averaging them out, as each model learns the peculiarities of drivers' responses to that specific event type.
Solution Approach 2:
The patent changes the learning parameters by creating separate prediction models for different event types rather than using a single model with averaged parameters. Each model has its own learned parameters specific to its event type, allowing the system to adapt to diverse driving patterns while maintaining high prediction accuracy for individual drivers in each situation.
3Productivity
If a prediction model is generated from one driver's data (e.g., driver who decelerates suddenly at tollgate but in advance at speed camera), then the model can be simpler and faster to compute, but it cannot accurately predict driving behavior for different types of events
Solution Approach 1:
The patent segments the prediction task into multiple specialized models, each trained on data from multiple drivers for a specific event type. This segmentation allows each model to be computationally efficient (dealing with one event type at a time) while collectively providing comprehensive coverage and high accuracy across all event types through the ensemble of specialized models.
Data Source
AI summary
The present disclosure provides a hybrid vehicle and a method of predicting a driving pattern in the same. The method includes: acquiring current vehicle driving information, determining an upcoming event and a driving style based on the current vehicle driving information, and generating an acceleration/deceleration prediction value based on a prediction model corresponding to the upcoming event and the driving style selected from a plurality of pre-learned prediction models.