Machine Learning-Based Batch Processing for Calibration of Model and Noise Parameters

Non-Gaussian or non-whiteness of noise sources often occurs in many digital avionics systems. Incorrect modeling of the system degrades the performance of parametric model-based estimators and controllers. To calibrate the model and noise parameters, this paper proposes a machine learning-based batc...

Full description

Saved in:
Bibliographic Details
Main Author: Lee, Kyuman
Format: Conference Proceeding
Language:English
Subjects:
Aerospace electronics
batch processing
Batch production systems
Calibration
Machine learning
Mathematical models
modeling error
non-whiteness noise
Simulation
Training data
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Non-Gaussian or non-whiteness of noise sources often occurs in many digital avionics systems. Incorrect modeling of the system degrades the performance of parametric model-based estimators and controllers. To calibrate the model and noise parameters, this paper proposes a machine learning-based batch processing approach. We first mathematically formulate a state augmentation system containing three types of noise: color noise, state-dependent noise, and correlation noise. Next, we define accessible process and measurement residuals to create the training data set. Finally, we propose offline batch processing that recursively utilizes a machine learning technique to calibrate the model and noise parameters. Simulation results under various conditions validate the calibration performance of the proposed approach.
ISSN:2155-7209
DOI:10.1109/DASC58513.2023.10311101