Novel Hybrid Algorithm for Training Feed-Forward Neural Networks

Strategic Insights -

A team of researchers from Delhi Technological University has proposed a novel hybrid Adaptive Particle Swarm Optimization-Back-propagation (APSOBP) algorithm for training feed-forward neural networks to identify nonlinear dynamical systems. The algorithm combines the strengths of Particle Swarm Optimization and back-propagation to improve the overall solution quality. Experimental results demonstrate that the hybrid algorithm outperforms traditional Particle Swarm Optimization and back-propagation algorithms in terms of convergence, accuracy, and robustness.

Key Takeaways:

  • The APSOBP algorithm uses Particle Swarm Optimization to optimize network weights, followed by back-propagation to fine-tune the optimized weights.
  • The approach prevents early convergence by dynamically adjusting Particle Swarm Optimization parameters based on a performance index.
  • Convergence analysis using Lyapunov stability theory is conducted to ensure the proposed algorithm converges to a stable solution.
  • The hybrid algorithm is evaluated on three benchmark nonlinear problems to validate its effectiveness.
  • The research demonstrates that the APSOBP algorithm outperforms traditional Particle Swarm Optimization and back-propagation algorithms in terms of convergence, accuracy, and robustness.
  • The authors, Shobana R, Rajesh Kumar, and Bhavnesh Jaint, are from the Department of Electrical Engineering at Delhi Technological University.
  • The research is published in the journal ISA Transactions.

Statistics:

  • The APSOBP algorithm is evaluated on three benchmark nonlinear problems.
  • The experimental results demonstrate a significant improvement in convergence, accuracy, and robustness compared to traditional algorithms.
  • The performance index used to dynamically adjust Particle Swarm Optimization parameters is calculated as the difference between the fitness value of the global best solution across consecutive iterations.

Sources:

  • NewsRx. Report Summarizes Mathematics Study Findings from Delhi Technological University (Feedback-based optimization of feed-forward neural network for the modeling of complex nonlinear dynamical systems using novel APSOBP algorithm). Journal of Engineering. October 20, 2025; p 2606.
  • Feedback-based optimization of feed-forward neural network for the modeling of complex nonlinear dynamical systems using novel APSOBP algorithm. ISA Transactions, 2025.