AN IMPROVED APPROACH FOR SOLVING HIERARCHICALLY COUPLED CONSTRAINED OPTIMIZATION PROBLEM IN SIMULTANEOUS OPTIMIZATION OF NEURAL NETWORK STRUCTURE AND WEIGHTS

Authors

  • Manik Rajora Georgia Institute of Technology
  • Pan Zou Georgia Institute of Technology
  • Steven Y. Liang Georgia Institute of Technology

DOI:

https://doi.org/10.23055/ijietap.2021.28.2.4261

Keywords:

Hierarchical Coupled Constrained Optimization Problem (HCCOP), Neural Network Structure and Weights Optimization, Simultaneous Optimization, Neural Network Prediction Model,

Abstract

Neural Networks (NN) structure is commonly determined before the training of its weights using a trial-and-error-based approach by an expert of the problem under consideration since there are no clear guidelines for selecting an optimal NN structure. The trial-and-error-based method can be very time-consuming and can also overlook possible optimal combinations due to the large amount of NN structure combinations available. To cope with these issues, the definitions developed by Zou et al. are used to identify the NN structure and weight optimization problem as a hierarchically coupled constrained optimization problem, and an improved version of the algorithm is developed and applied to find the optimal NN structure and weight values simultaneously. The proposed approach is then employed to build NN prediction models (referred to as NN-HCCOP) for a variety of case studies to test its validity. The results of the NN-HCCOP model are compared with five other prediction methods i.e., adaptive neuro-fuzzy inference system (ANFIS), ANFIS-firefly algorithm (ANFIS-FA), classical NN, regression analysis, and gaussian process regression analysis. The comparison results show that NN-HCCOP model can provide higher prediction accuracy in the majority of the testing scenarios compared to models built by other techniques.

Published

2021-11-23

How to Cite

Rajora, M., Zou, P., & Liang, S. Y. (2021). AN IMPROVED APPROACH FOR SOLVING HIERARCHICALLY COUPLED CONSTRAINED OPTIMIZATION PROBLEM IN SIMULTANEOUS OPTIMIZATION OF NEURAL NETWORK STRUCTURE AND WEIGHTS. International Journal of Industrial Engineering: Theory, Applications and Practice, 28(2). https://doi.org/10.23055/ijietap.2021.28.2.4261

Issue

Section

Production Planning and Control