Because this method is a non - linear , non - differentiable and combinatorial - constrained optimization problem , mathematical optimal approaches are not suitable for solving this problem . the genetic algorithm based on natural selection and natural gene mechanism is presented in this paper . it simulates the network structure optimization by simple encoding technique and propagating mechanism
12.
To circumvent the non - differentiable difficulty caused by the positive homogeneously function that is involved in an equivalent unconstrained formulation for general inequality constrained optimization problems , we turn to the classical lagrangian function and redefine m ( x ) by a conic optimization problem with the lagrangian as the objective function
