Bayesian Regression for Vehicle Control Unit Parameter Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining critical variables in vehicle control units, such as exhaust gas temperature and emission values, are either expensive or lack accuracy, especially for complex combustion processes, and do not provide reliable control strategies.
Innovation Solution
Implementing non-parametric data-based Bayesian regression, specifically using Gaussian processes, to predict variables in real-time, which reduces prior knowledge requirements and computational effort, and allows for more accurate control by considering variance in output variables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If characteristic maps or simplified physical models are used to determine critical parameters, then the implementation cost is reduced, but the predictive accuracy and reliability deteriorate
Solution Approach 1:
The patent changes the mathematical approach from traditional characteristic maps or simplified physical models to Bayesian regression methods. This parameter change in the calculation methodology enables accurate prediction of critical combustion parameters while maintaining real-time computational feasibility, thus achieving both accuracy and cost-effectiveness
Solution Approach 2:
The patent replaces complex physical modeling approaches with a data-driven Bayesian regression system. This substitution eliminates the need for complex mechanical/physical models while achieving superior predictive accuracy through statistical learning from measurement data
2Measurement precision
If non-parametric data-based Bayesian regression is implemented, then the predictive accuracy and reliability improve, but the computational effort and memory requirements increase
Solution Approach 1:
The patent performs preliminary computation during an offline calibration phase, pre-calculating and storing necessary parameters and models. This preliminary action transfers computational burden from real-time operation to offline setup, enabling fast real-time predictions with reduced computational effort during actual vehicle operation
Solution Approach 2:
The patent segments the computational process into offline calibration phase and online real-time phase. The offline phase handles intensive computation for model training and parameter identification, while the online phase executes lightweight predictions, thus distributing computational load effectively
3Device complexity
If traditional characteristic curves are used, then the device complexity is low, but the ability to handle multidimensional relationships deteriorates
Solution Approach 1:
The patent transitions from one-dimensional characteristic curves to multidimensional Bayesian regression models. This dimensional expansion enables the system to capture complex multidimensional relationships between multiple input parameters and output variables, significantly improving adaptability while maintaining computational feasibility through the offline-online computation strategy
Data Source
Figure 1~2a
Figure 2b~3b
Figure 4~5
AI summary
The invention relates to a control device in a vehicle, having means for calculating at least one output parameter for a controller of functions of the vehicle during operation of the vehicle on the basis of at least one input parameter determined during operation. The control device thereby comprises means for carrying out the calculation of the output parameter using a Bayesian regression of training values determined for the output parameter and the input parameter prior to operation.