Vehicle Controller Feedback Correction for ML Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle control systems using insufficiently trained machine learning models fail to maintain accurate control, leading to suboptimal performance.
Innovation Solution
A vehicle controller that inputs operating-condition parameters into a learning model, corrects control parameters using feedback values, and updates the model with training data to achieve desired operations, ensuring sufficient accuracy and reducing computational burdens.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning system is used to control a vehicle device, then the accuracy of control can be improved when the model is sufficiently trained, but the control accuracy deteriorates when the model is insufficiently trained
Solution Approach 1:
The patent implements feedback control by calculating a feedback value based on the difference between the actual operation evaluating value and the target value, then correcting the control parameter using this feedback value. This ensures that even with an insufficiently trained learning model, the control accuracy is maintained by continuously adjusting the control parameters based on actual system performance.
Solution Approach 2:
The patent performs preliminary training of the learning model using stored training data before actual vehicle operation. The controller determines whether the learning model is sufficiently trained by evaluating its performance on training data, and only when sufficient training is confirmed does the system rely on the learning model for control, otherwise it uses feedback control to compensate.
2Measurement precision
If feedback control is applied to correct control parameters, then control accuracy is maintained with insufficiently trained models, but computational load increases
Solution Approach 1:
The patent dynamically switches between different control strategies based on the training status of the learning model. When the model is sufficiently trained, the system uses only the learning model for control (lower computational load). When the model is insufficiently trained, the system combines the learning model with feedback control (higher computational load). This dynamic adaptation optimizes the balance between control accuracy and computational resources.
3Measurement precision
If the learning model is continuously updated with training data, then the model accuracy improves over time, but the system complexity increases
Solution Approach 1:
The patent implements self-service learning where the vehicle controller automatically collects operation evaluating values during normal operation, stores them as training data, and periodically retrains the learning model without requiring external intervention. The system autonomously determines when retraining is needed based on the accumulation of sufficient training data, thereby improving model accuracy while maintaining relatively simple system architecture.
Data Source
AI summary
A vehicle controller includes a processor configured to input an operating-condition parameter indicating an operating condition of a device into a learning model to calculate a control parameter indicating a control condition for causing the device to perform a desired operation corresponding to the operating-condition parameter; correct the control parameter so as to achieve the control condition, based on a feedback value depending on an operation evaluating value; control the device in accordance with the control parameter when a predetermined accuracy condition concerning accuracy of control of the device performed in accordance with the control parameter is satisfied, and controls the device in accordance with the corrected control parameter when the predetermined accuracy condition is not satisfied.