Aggregate Production Planning Model And Intelligent Algorithm Under Double Uncertain Environments With Random And Fuzzy

There exist a great deal of uncertainties such as randomness, fuzziness and fuzzyrandomness in the fields of management science, computer science, system science, information science, engineering, etc. Many decisions in these fields need to be made under the uncertainties. Uncertain programming is a powerful tool to handle the decision-making problem. Aggregate production planning (APP) problem decides the products in every periods to meat the market demand by adjust the production level, inventory level, stockout level, etc. This dissertation proposes many uncertain programming models of the aggregate production planning problem under the double uncertain environment mixed with random and fuzzy, and solved them by intelligent algorithms based on simulation. The contents are described as follows:Under the double uncertain environment mixed with random and fuzzy, product demand, product cost, labour force level, capability of machine, capability are all be characterised as double uncertain variable such as random fuzzy, fuzzy random, birandom, bifuzzy. Also, set up a set of decision models under different decision rule.In many cases, it is very difficult or impossible to obtain the exact values of the uncertain functions with fuzzy random, random fuzzy, bifuzzy, birandom variables. Therefore, it is very necessary to estimate these values by simulations. This dissertation proposes the simulation perturbation stochastic approximation(SPSA) algorithms based on simulations(fuzzy random simulation, random fuzzy simulation, bifuzzy simulation, birandom simulation) to solve the double uncertain programming models. The algorithms can converge fast to the local optimal solutions. In many real optimization problems, due to the limitations on the resources employed in the problems, a local optimal solution may be accepted.In the SPSA algorithms based on simulations, much time will be spent on the simulations. Therefore, this dissertation designs the SPSA algorithms combining simulations with neural network (NN). First, simulations are used to generate a set of input-output data for the uncertain functions. Then the data are employed to train an NN, which is embedded in the SPSA algorithms. The algorithms can converge faster to the local optimal solutions than the SPSA algorithms based on simulations.For the optimization problems where the global optimal solution is needed, this dissertation designs hybrid optimization algorithms based on simulations. The algorithms integrate simulations, NN, genetic algorithm (GA), and SPSA. First, simulations are used to generate a set of input-output data for the uncertain functions. Then the data are employed to train an NN, which is embedded in GA and the SPSA algorithm. GA is employed to search the optimal solution in the entire solution space. Both the chromosomes in the initial population and the new chromosomes produced by the crossover operation and mutation operation in each generation are improved by the SPSA algorithm. Finally, the SPSA algorithm is used to improve all chromosomes in the population after GA is completed. The chromosomes whose fitness is biggest is regarded as the optimal solution. The algorithms have both the global search ability of GA and the strong convergence property of the SPSA algorithm. Numerical examples illustrate the effectiveness of the proposed algorithms.The innovation of the dissertation is setting up a set of uncertain programming models under the double uncertain environment, which can enrich the modeling theory under uncertain environment and advance the capability of the enterprise with productions.