Neural network approach to gas flow measurement: radial basis function networks in differential pressure method applications

Abstract

This paper investigates the application of radial basis function (RBF) neural networks for predicting gas flow rate based on differential pressure measurements across an orifice. Several RBF network architectures with varying numbers of hidden neurons were developed and tested to assess their predictive performance. The results demonstrate that all models achieved high accuracy, with determination coefficients exceeding 0.93 across training, testing, and validation datasets. Models with more hidden neurons exhibited improved approximation of the nonlinear relationship between input parameters such as pressure, differential pressure, temperature, and orifice diameter ratio and gas flow rate, resulting in lower residual standard errors. Error analysis showed that the predicted values consistently followed the ideal convergence trend, and the models-maintained robustness without signs of overfitting. The residual errors remained within ±7 for all ranges of input variables, including orifice diameter ratio and differential pressure, indicating acceptable accuracy for practical use. The study highlights the effectiveness of RBF networks in capturing complex physical dependencies and their suitability for implementation in intelligent measurement systems for gas flow. These findings support the use of RBF neural networks as a reliable and efficient tool for automated gas flow estimation, particularly in industrial environments where accurate and adaptive modeling is essential.