Title: Novel multi-step-based process modelling and prediction of thermodynamic properties of steam using machine learning

Authors: Ashwani Kharola; Kiran Sharma; Vishwjeet Choudhary

Addresses: Department of Mechanical Engineering, Graphic Era Deemed to be University, Dehradun, India ' Department of Allied Sciences, Graphic Era Deemed to be University, Dehradun, India ' Department of Mechanical Engineering, Birla Institute of Technology and Science (BITS), Pilani, India

Abstract: This study investigates different machine learning techniques for precise prediction of various thermodynamic properties of steam. Calculating interpolated values of different thermodynamic properties is a time consuming and tedious task. Therefore, different machine learning models are proposed in the study. The techniques considered for prediction include feed-forward neural network (FFNN), nonlinear autoregressive neural network (NARNN) and adaptive fuzzy inference system (ANFIS). An optimal FFNN model has been designed by varying the number of neurons (n) in hidden layer, i.e., n = 10, 20, 30, 40 and 50. Additionally, a novel NARNN model has been proposed in this study having 10 neurons in hidden layer for predicting different thermodynamic properties of steam through multi-step ahead predictions. The study further proposes an optimal ANFIS model designed considering the appropriate shape and number of membership functions. The performance of proposed techniques was finally measured in terms of different error indicators for better validation. [Submitted 27 December 2022; Accepted 24 August 2024]

Keywords: process modelling; multi-step ahead prediction; thermodynamic properties; steam; optimisation; prediction; feed-forward neural network; FFNN; nonlinear autoregressive neural network; NARNN; adaptive fuzzy inference system; ANFIS.

DOI: 10.1504/IJMR.2024.143417

International Journal of Manufacturing Research, 2024 Vol.19 No.3, pp.239 - 265

Accepted: 24 Aug 2024
Published online: 19 Dec 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article