Neural Network Speed Profile Prediction for Hybrid Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle route planning systems lack accuracy in predicting speed profiles, relying solely on historical data and static information, which can deviate from actual driving patterns, limiting the effectiveness of energy management and fuel efficiency in hybrid electric vehicles.
Innovation Solution
Implementing a neural network-based system that classifies route segments into driving zone types based on historical data, generating and updating predicted speed profiles using real-time traffic and driving behavior, allowing for dynamic energy management and improved route planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If historical data and static information are used for speed prediction, then the system is simple to implement, but the prediction accuracy deteriorates due to deviation from actual driving patterns
Solution Approach 1:
The system continuously compares predicted speed profiles with actual measured speed profiles from GPS and vehicle sensors, using the deviation information to update and refine the neural network model parameters, thereby improving prediction accuracy over time while maintaining system simplicity
Solution Approach 2:
The neural network model automatically learns and adapts to actual driving patterns by processing real-time vehicle data and historical information itself, without requiring external manual calibration or complex intervention, enabling the system to self-improve prediction accuracy
2Measurement precision
If neural network-based dynamic prediction is implemented, then speed prediction accuracy is improved through continuous learning, but device complexity increases
Solution Approach 1:
The route is divided into multiple segments with different driving zone types (e.g., urban, highway, rural), and separate neural network models are trained for each segment type, allowing the system to manage complexity through modularization while maintaining high prediction accuracy for diverse driving conditions
Solution Approach 2:
Neural network models are pre-trained offline using historical driving data before deployment, so that during actual vehicle operation, the system only needs to select the appropriate pre-trained model and input real-time data, avoiding the computational complexity of real-time training while maintaining high prediction accuracy
3Use of energy by moving object
If static route planning is used, then computational resources are conserved, but energy management effectiveness deteriorates due to inability to adapt to actual driving patterns
Solution Approach 1:
The system performs partial updates of the neural network model by only adjusting parameters for segments where prediction deviations are detected, rather than continuously retraining the entire model, thereby reducing computational energy consumption while maintaining energy management effectiveness
Solution Approach 2:
The system uses lightweight neural network models that can be quickly trained and updated with minimal computational resources, allowing frequent model updates without significant energy cost, thus maintaining adaptability to changing driving patterns while conserving computational energy
Data Source
AI summary
A vehicle includes a powertrain having an electric machine and an engine. The vehicle also includes a controller programmed to operate the powertrain according to a predicted vehicle speed profile for a predetermined route segmented according to a group of driving zone types, wherein each driving zone type is associated with a different characteristic speed profile shape and vehicle location. The controller is further programmed to update the predicted segment speed profile in response to deviation between the predicted speed profile and a measured speed profile.