Abstract: The stochastic expected value model belongs to a class of stochastic programming problems，which has wide application backgrounds，in order to search an algorithm which can solve this problem effectively，in the paper，random simulation is used to produce training samples for BP neural network to approximate the stochastic function.And a hybrid intelligent algorithm for stochastic expected value models combined particle swarm optimization with BP neural network for approximation of the fitness function and checking feasibility of solution is presented.Finally，the simulation results of two examples are given to show the correctness and effectiveness of the algorithm.