.png){kind=icon}
返回
An Expectation Maximization Sample Transfer Identification Method for Dynamic Systems Under Nonideal Data
.png){kind=icon}
PRE
AI
DOI：10.1109/TIM.2025.3643055.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
This article proposes an expectation maximization sample transfer identification (EM-STI) algorithm to address the parameter identification problem in dynamic systems with nonideal data. Traditional identification methods usually require a large amount of high-quality ideal data that follows the same distribution. However, in industrial measurement scenarios, challenges such as sensor noise, environmental complexity, and sensor data loss often make such datasets difficult to obtain. To solve the above problems, the EM-STI algorithm constructs a transfer identification (TI) framework, and models the distribution discrepancy between the source and target systems as latent variables, and iteratively optimizes parameters using the EM algorithm to improve the reliability and accuracy of the model. The core of the algorithm includes establishing a unified probabilistic framework, explicitly modeling distribution differences as latent variables, inferring the latent variable relationships through the expectation step, calculating the log-likelihood expectation, and updating the parameters in the maximization step to reduce distribution discrepancy. The algorithm also provides a theoretical guarantee of convergence. The simulation experiments on the mass-spring-damping system and the pH process system, as well as the industrial case study on the demand prediction of magnesium furnaces, have proved that this algorithm has excellent estimation accuracy under nonideal data conditions such as high noise and data scarcity, and has strong robustness against non-Gaussian noise.
Keyword:
Vectors
Noise
Accuracy
Heuristic algorithms
Data models
Pollution measurement
Parameter estimation
Covariance matrices
Mathematical models
Iterative algorithms
Expectation maximization (EM)
nonideal data
system identification
transfer identification (TI)
transfer learning
期刊
IF:
5.9
论文数:
2.0W
被引数:
5.8W
机构
引用论文
分享收藏Improved estimation of carbon dioxide and methane using machine learning with satellite observations over the Arabian Peninsula.png)
SCIENTIFIC REPORTS
OAAI
SCIENTIFIC REPORTSIF3.9
OAAI分享收藏Identification of ARX Model with Multi-Gaussian NoisesBai W., Guo F., Chen L., Hao K., and Huang B. (2019), 多高斯噪声下ARX模型的辨识，2019年第19届控制、自动化与系统国际会议（ICCAS），613-618.
PREAI
分享收藏分享收藏A Comprehensive Expectation Identification Framework for Multirate Time-Delayed Systems多速率时滞系统的综合期望识别框架
PREAI
分享收藏
查看更多内容