Mixed Optimization Technique for Large-Scale Water-Resource Systems

Abstract

Dealing with significantly large design problems for water-resource systems a mixed optimization procedure based on network linear programming and the subgradient method will be described. Using a linear problem formulation, the procedure uses network linear programming as a subproblem that assumes the knowledge of design variables. Since inside its domain, the global objective functions is a convex piecewise linear function, a subgradient method is used to obtain the direction of the improvement of design variables at each iteration using the solutions of the network subproblem. The mixed technique permits an efficient evaluation of the design variables in order to reach a good approximation of the global objective function optimum. The solution technique performs well in the purely linear case and, moreover, allows some kinds of nonlinearities in the cost functions of design variables.