Axial Compressive Capacity Prediction of CFRST Columns Based on PSO-BP Neural Network
Authors:
- Yunlong ZhangYZYunlong Zhang
- Yuzhou HeYHYuzhou He
- Guofeng DUGDGuofeng DU
- Juan ZhangJZJuan Zhang
Open Access
- 1 September 2024
- journal article
- Published by China Science Publishing & Media Ltd. in Industrial Construction
- Vol. 54 (9) , 141-148
Abstract
The traditional back propagation (BP) neural network has some defects in predicting the axial compressive capacity of concrete-filled rectangular steel tube (CFRST), such as system instability, slow convergence speed and difficult selection of hyperparameters, which will affect the stability of the prediction model and the accuracy of the prediction results. In order to improve the traditional BP model to achieve better prediction effect, particle swarm optimization algorithm (PSO) was applied to BP prediction model, and a CFRST axial compressive capacity prediction model PB7-7-1 based on PSO-BP neural network was proposed. The results showed that the fluctuation range of the predicted values of the PB7-7-1 model was substantially reduced compared with that of the traditional BP model, in which the absolute relative error (ARE) of the predicted values of 45% of the components was within 5%, and the ARE of 80% of the components was within 10%; prediction accuracy of the PB7-7-1 model had been improved by 30.79%, and the average ARE of its predictive values was only 6%. This showed that the PB7-7-1 model based on PSO-BP neural network had a significant improvement in the stability and accuracy of prediction results of CFRST axial compressive capacity compared with traditional BP network. In addition, according to the weight and bias of the hidden layer and output layer of PB7-7-1 model, the prediction formula of CFRST axial compressive capacity was constructed. Finally, SHAP machine learning interpretation algorithm was used to analyze the importance and contribution of each input parameter to the axial compressive capacity.Keywords
This publication has 6 references indexed in Scilit:
- Practical ANN prediction models for the axial capacity of square CFST columnsJournal of Big Data, 2023
- Fast and improved backpropagation learning of multi‐layer artificial neural network using adaptive activation functionExpert Systems, 2020
- A review of velocity-type PSO variantsJournal of Algorithms & Computational Technology, 2016
- Particle swarm optimization: Hybridization perspectives and experimental illustrationsApplied Mathematics and Computation, 2011
- Axial load behaviour of high-strength rectangular concrete-filled steel tubular stub columnsThin-Walled Structures, 2005
- Behavior of Centrally Loaded Concrete-Filled Steel-Tube Short ColumnsJournal of Structural Engineering, 2004