.png){kind=icon}
.png){kind=icon}
Return
Multitask Maximum Likelihood Identification for ARX Model With Multisensor
.png){kind=icon}
PRE
AI
DOI：10.1109/TIM.2022.3173636.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In order to improve the identification performance by exploiting the knowledge of multisensor, a multitask (MT) maximum likelihood (ML) identification algorithm is proposed via designing an MT identification criterion function to make full use of all sensor information. The proposed method not only yields MT estimates of model parameters but also provides an explicit solution of the weights by solving an equality-constrained optimization problem. We next prove that it is unbiased and can improve the identification precision compared with the traditional single-task method. Then, to deal with the case of plug-and-play sensors, the recursive form of the proposed algorithm is presented for online MT identification, and its uniform convergence is analyzed. Furthermore, to quantify the effect of sensor attributes on the identification performance, an analytical relationship is derived between the identification accuracy and the sensor attributes (the number of sensors, measurements size, and sensor noise). Finally, a numerical example and a continuous fermenter example are provided to demonstrate the advantage and capability of the proposed algorithms.
Keywords:
Maximum likelihood (ML)
multisensor
multitask (MT) learning
system identification
Journal
IF:
5.9
Papers:
1.9W
Citations:
5.8W