The Study On Stochastic Multi-Objective Supplier Selection Problem

This dissertation provides a disruptive risk management strategy for supplierselection problem. Considering the case of selecting multi suppliers for various prod-ucts under risk control management, this strategy can aford the most reasonablepatterns of supply portfolio and order quantity. Since the parameters of cost, qualityand lead time usually present randomness in realistic selection process, we wouldemploy stochastic programming theory to deal with the stochastic supplier selectionproblem quantitatively.This thesis first formulates a non-linear multi-objective stochastic optimizationprogramming, which maximizes three service levels about cost, quality and leadtime, respectively. Then, two types of methods to handle the objective functionsare adopted. The first method is the main objective approach that receives a single-objective probability constrained programming model by making the probabilitiesabout the total quality and total lead times satisfy some given service levels aschance constraints while regarding the cost probability as the single objective. Thesecond classical method is goal programming which involves preemtive level in eachobjective and reformulates a stochastic supplier selection goal programming even-tually by minimizing three negative deviations. In the case of random parametersfollowing joint normal distributions and more general continuous distribution, crispanalytic expressions for probability are found by standard transformation and sam-ple approximation approach, resulting in equivalent and deterministic formulationsand an approximate mixed-integer programming problem. In the view of complex-ity in solving the equivalent optimization models, we propose an improved hybridmemetic algorithm to obtain feasible solutions. Finally, a case about the selection ofauto parts supporting suppliers is proposed to illustrate the selection process, wherethe memetic algorithm is utilized on numerical examples and proves its efectivenessby algorithm comparison.The major new results of this thesis contain the following four aspects:(i)A non-linear multi-objective stochastic optimization programming is formulated by maximizing three service levels about cost, quality and leas time;(ii) The probabil-ity constrained model and stochastic supplier selection goal programming problemare received through two types of multi-objective programming methods;(iii) In thecases that parameters are multivariate normal vectors or general continuous vari-ables, the proposed models are converted into their deterministic equivalent ones;(iv) An improved hybrid memetic algorithm is designed to solve the proposed mod-els, and some numerical examples are provided to illustrate the modeling ideas andthe efectiveness of the proposed hybrid algorithm.