A Study Hybrid Algorithm Solve Stochastic Expected Value Models And Chance-constrained Programming

There exist uncertainties in the fields of operation research, management science,information science, system science, computer science and engineering. In these fields manystrategies are made under such uncertainties. Uncertain programming is an effective way tosolve these strategy problems. Thus, the research on the build of uncertain programming andsolving methods are of great practical and theoretical value. This paper analyses and buildsstochastic expected value models and stochastic chance-constrained programming, and at thesame time, it puts forward a new hybrid algorithm based on the original method. The specificcontents are as follows:At first, this paper seeks to simulate random variable, solve Mathematical expectationprobability, Positive values, Pessimistic values, and produce input and output data for thetarget functions of stochastic expected value models andstochastic chance-constrainedprogramming by using Monte Carlo. It takes long time and much calculation using thismethod. In order to save time and reduce calculation, this paper, on one hand, makes aresearch on the method of reducing simulation times. On the other hand, it analyzes theconnection between simulation times and different parameters and obtains a suitable methodthat reduces simulation times. It also calculates mathematical expectation, probability,positive values, pessimistic values and testify the feasibility of this method by using selectivenumerical value test.Then, it tries to train a neural network to approach the uncertain function by using theinput and output data. In order to improve the approaching ability of the trained neuralnetwork, it needs to optimize hidden layer nodes, transfer function, train function, andlearning rate. While training the neural network, this paper adopts the rule that minimizesMSE to select parameters. Then it analyses and confirms hidden layer nodes, Transferfunction, train function, and learning rate by adopting the diagram obtained from numericalvalue test. And then it testifies the successfully trained neural network to prove that thenetwork can approach the uncertain function better.At last, this paper treats the target function and constraint function obtained from theneural network as the target function and constraint function in PSO. Then, the solutionobtained from PSO is just the optimizing solution of expectation model and random chanceconstrained programming. This paper, then, solves expectation model and random chance constrained programming by adopting the standard PSO and the improved PSO respectivelyand shows that the improved PSO outweighs the standard PSO.