A hybrid technique for global optimization of hierarchical systems
Cang-Pu Wu · 2002
A hybrid technique which is a combination of neural networks devised for solving dynamic programming problems and an improved genetic algorithm is proposed to deal with a class of global optimization problems of bilevel hierarchical systems. A specific augmented Lagrangian multiplier is embedded in neural network model for dynamic programming to deal with both equality and inequality constraints. Under certain assumptions, it is proved that the neural network devised can solve the lower level optimal control problems accurately. An improved genetic algorithm is adopted which adjusts some of the parameters of the operations involving selection, crossover, and mutation to exploit the features of the hierarchical structure and thus enhance the efficiency of the overall optimization procedure. The main advantages of the technique include a great reduction of computation time and attaining a global optimum for the overall hierarchical system with objective functions in a wide range.