Ashraf, WM;
Dua, V;
(2024)
A data-driven approach for constructing the prediction bounds on the output variables using a modified loss function and analysing information retained during development of the model.
In:
Computer Aided Chemical Engineering.
(pp. 3085-3090).
Elsevier
![[thumbnail of Ashraf_807Ashraf-ver1.pdf]](https://discovery.ucl.ac.uk/style/images/fileicons/text.png) |
Text
Ashraf_807Ashraf-ver1.pdf Access restricted to UCL open access staff Download (686kB) |
Abstract
The model-based simulated responses are plagued by prediction errors for models obtained by using machine learning (ML) techniques, thereby leading to the need to estimate the prediction bounds around the model-simulated responses. Here in this work, we propose a novel method utilizing the iterative values of the model parameters obtained during model development to construct the prediction bounds around the model-simulated responses using artificial neural network (ANN) model. The loss function of the ANN model includes the least-mean squared error and is also augmented with the standard deviation between the true and model-simulated responses. During the training and development of ANN model, the values of connection weights and the biases associated with the working layers of the ANN model are stored and at the end of the training these values are deployed to construct the prediction bounds around the model-simulated responses. The proposed methodology is applied to energy efficiency cooling and energy efficiency heating (bench-mark dataset from University of California - Irvine database for ML). The developed ANN model showed superior modelling performance having following predictive errors: energy efficiency cooling (RMSE_test = 1.40%) and energy efficiency heating (RMSE_test = 0.46%) compared with those of feedforward neural network reported in literature (RMSE_test = 1.63% and RMSE_test = 0.63% respectively). The width of prediction bounds made by the proposed technique is found to be comparable to those of input perturbation method. The proposed SWARM approach for drawing the prediction bounds can be applied to different real-life applications, facilitating the decision makers to incorporate these bounds for optimal decision making.
| Type: | Book chapter |
|---|---|
| Title: | A data-driven approach for constructing the prediction bounds on the output variables using a modified loss function and analysing information retained during development of the model |
| ISBN-13: | 9780443288241 |
| DOI: | 10.1016/B978-0-443-28824-1.50515-9 |
| Publisher version: | https://doi.org/10.1016/B978-0-443-28824-1.50515-9 |
| Language: | English |
| Additional information: | This version is the author accepted manuscript. For information on re-use, please refer to the publisher’s terms and conditions. |
| UCL classification: | UCL UCL > Provost and Vice Provost Offices > UCL BEAMS UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Chemical Engineering |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10196899 |
Archive Staff Only
 |
View Item |
