Vehicle Path Training Using Real-World Driving Profile Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods face challenges in predicting and optimizing vehicle emissions across various driving cycles, as existing technologies struggle to generate representative speed profiles that account for real-world driving conditions, leading to potential compliance issues and inefficient emission management.
Innovation Solution
A computer-implemented method using a machine learning system to generate driving histories by training a generator and discriminator with neural networks, optimizing parameters to produce speed profiles that mimic real-world distributions, thereby enabling more accurate emissions prediction and optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a vehicle drive system is optimized for critical driving cycles with extreme speed profiles, then emissions compliance for these critical cycles is improved, but emissions for less critical but more frequent driving cycles deteriorate, leading to higher overall emissions in real operation
Solution Approach 1:
The patent applies parameter changes by transforming the optimization approach from focusing on extreme speed profiles to using realistic speed profiles that reflect actual driving distributions. The machine learning system generates speed profiles with statistically representative characteristics (mean, standard deviation, skewness) that match real-world driving patterns, thereby optimizing emissions across the full spectrum of driving conditions rather than prioritizing critical but rare cycles
Solution Approach 2:
The patent uses copying by generating synthetic driving histories that replicate the statistical distribution of real-world driving patterns. The machine learning system creates virtual speed profiles that mirror actual driving behavior, allowing comprehensive emissions optimization without requiring exhaustive testing of all possible driving cycles
2Adaptability or versatility
If the inspector freely chooses driving cycles within wide limits for emissions testing, then the ability to assess various driving conditions is improved, but the complexity of predicting emissions during vehicle development increases
Solution Approach 1:
The patent applies universality by creating a machine learning system that handles multiple driving cycle types and conditions through a single unified framework. The system generates speed profiles that can represent various driving scenarios (urban, rural, motorway, mixed) and can be applied to different vehicle types, eliminating the need for separate prediction methods for each driving cycle category
Solution Approach 2:
The patent transforms the complex problem of predicting emissions across diverse driving cycles by changing the approach from cycle-specific analysis to distribution-based generation. By parameterizing speed profiles using statistical characteristics (mean, standard deviation, skewness) rather than individual cycle specifications, the system simplifies emissions prediction while maintaining versatility across all permissible driving cycles
Data Source
Figure 1
AI summary
A computer-implemented method for training a machine learning system to generate vehicle driving profiles is presented, comprising the following steps: - initial routes are selected from a first database of driving routes, - a generator of the machine learning system receives the initial routes as input and generates corresponding initial driving profiles for each initial route, - a second database stores driving routes and their corresponding driving profiles recorded during operation, - second routes and their corresponding second driving profiles recorded during operation are selected from the second database, - a discriminator of the machine learning system receives as input pairs from one of the initial routes with their corresponding first generated driving profile and pairs from second routes with their corresponding second driving profile recorded during operation.- Depending on the input variables, the discriminator calculates outputs that characterize, for each pair of input variables, whether it is a pair with a first, generated driving profile or a pair with a second driving profile recorded during operation. - Depending on the outputs of the discriminator, an objective function is optimized that represents a distance between the distribution of pairs with first, generated driving profiles and the distribution of pairs with second, recorded driving profiles.