A Novel Penalty Strategy Finite-Time Convergent-Differential Neural Network for Solving Nonlinear and Nonconvex Optimization Problems

Nonlinear nonconvex optimization problem (NNOP) holds significant importance across various scientific fields as a common problem in engineering applications. To achieve fast convergence and strong robustness within a limited time when solving NNOPs, a novel penalty strategy finite-time convergent-differential neural network (PS-FT-CDNN) is presented. The equality constraints in NNOPs are transformed into penalty terms in the objective function through penalty functions. Then, Based on the design method of convergent differential neural networks, PS-FT-CDNN is designed to solve NNOPs with multiple constraints. The results of computer simulations confirm that the proposed PS-FT-CDNN achieves faster error convergence and demonstrates superior robustness compared to existing neural networks when addressing NNOPs with equality and inequality constraints.