Modeling concrete strength with high-order neural networks. Issue 8 (November 2016)

• Record Type: Journal Article
• Authors: Tsai, Hsing-Chih
• Is Part Of: Neural computing & applications. Volume 27:Issue 8(2016)
• Page Start: 2465
• Page End: 2473
• Publication Date: 2016-11
• Subjects: High-order neural networks -- Soft computing programming -- Soft computing tuning -- Concrete strengths -- Particle swarm optimization; Neural networks (Computer science) -- Periodicals; Neural circuitry -- Periodicals; Artificial intelligence -- Periodicals; Neural Networks (Computer) -- Periodicals; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux nerveux -- Périodiques; Intelligence artificielle -- Périodiques; 006.32
• DOI: 10.1007/s00521-015-2017-6
• Languages: English
• ISSNs: 0941-0643

Abstract

Abstract The present study designs a three-layer high-order neural network (HONN) to predict and program the 7- and 28-day strength values of concrete cylinders. The prediction results and programmed formulae generated by the network are similar in form to polynomials. Results show that the three-layer HONN interprets training patterns significantly better than current linear neural networks. Particle swarm optimization is applied to HONN parameter learning as well as to the work of pruning HONN structures to avoid overfitting and increase formula concision. To extend the uses of HONN programming, this paper further applies HONN to tune Abrams' laws. Results demonstrate that HONN programming delivers good prediction accuracy using a novel programmed formula. Furthermore, HONN tuning significantly improves Abrams' laws. Finally, the programmed formulae that were developed for the present study provide insight into the impact of input parameters on HONN programming and tuning.