Two-stage fuzzy stochastic programming with Value-at-Risk criteria

A new class of fuzzy stochastic optimization models - two-stage fuzzy stochastic programming with Value-at-Risk (FSP-VaR) criteria is built in this paper. Some properties of the two-stage FSP-VaR, such as value of perfect information (VPI), value of fuzzy random solution (VFRS), and bounds of the fuzzy random solution, are discussed. An Approximation Algorithm is proposed to compute the VaR by combining discretization method of fuzzy variable, random simulation technique and bisection method. The convergence of the approximation algorithm is proved. To solve the two-stage FSP-VaR, a hybrid mutation-neighborhood-based particle swarm optimization (MN-PSO) which comprises the Approximation Algorithm is proposed to search for the approximate optimal solution. Furthermore, a neural network-based acceleration method is discussed. A numerical experiment illustrates the effectiveness of the proposed hybrid MN-PSO algorithm. The comparison shows that the hybrid MN-PSO exhibits better performance than the one when using other approaches such as hybrid PSO and GA.