.png){kind=icon}
.png){kind=icon}
Return
Multi-output regression using polygon generation and conditional generative adversarial networks
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2022.117288.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper proposes an innovative multi-output regression method that processes and converts the numeric data variables into representative images (polygons) to build accurate predictive models in industrial applications with several dependent variables (responses). In this method, polygon images are generated from both the inputs and outputs of numeric data. The images representing the data inputs are then translated into those representing the outputs by training a conditional generative adversarial network (cGAN). The output images of the trained cGAN are then mapped into the outputs to get back the predicted numeric values. The advantage of the proposed method is that it makes use of the breakthrough of deep generative modeling to learn the true distribution of complex data, which is difficult to determine in many industrial applications. This is attributed to the fact that the generated polygons express all interrelationships between the data variables in the form of trustworthy representational images used to train the cGAN model. The performance of the proposed method was validated successfully using a complex industrial dataset acquired from a black liquor recovery boiler (BLRB) in a Kraft pulp & paper mill located in Canada. Three key performance indicators (KPIs), one economic and two environmental, are used as regression outputs in the BLRB dataset. The results of the proposed method demonstrate better performance than other comparable machine learning regression methods.
Keywords:
Generative modeling
Conditional generative adversarial network
(cGAN)
Polygon generation
Multi -output regression
Deep learning (DL)
Hamiltonian cycles
Image processing
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W