Some Key Techniques Of Engineering Optimization Design Grid Platform And Their Applications

The grid is to provide distributed computing infrastructure for advanced science and engineering problems by coordinated resource sharing and task solving in dynamic, multi-institutional virtual organizations. By means of grid the engineers can access computers, equipments, software, data base and other resources directly, but not files exchange only. At present, the grid techniques has developed very quickly and become a hot topic in high-performance computation field.Optimization design plays very important role in the engineering applications, but it often involves huge computational efforts and requires powerful computing environment. The grid is a good choice for solving complex optimization design problems, because it integrates the massive idle resources to super-powerful environment. However, the distributed, dynamic and heterogeneous characteristics of grid environments put difficulties in engineering applications of the grid. Moreover, the serial and some parallel algorithms can not be run on the grid platform in a straightforward manner. Three main problems are encountered in the applications of grid to the modern engineering designs: (1) the physical and mathematical models for the engineering problems are even more complicated and difficult to optimize directly, so that some general analysis programs have to be used as black-box; (2) the highly heterogeneous and dynamic essence of the grid leads to the great difficulties for the grid resources distribution; (3) the researchers did not make efforts to develop the typical applications for engineering optimizations based on the grid.In order to solve the complex optimization problems by using grid computing, two black-box optimization algorithms are proposed, which are an evolution grid algorithm and a surrogate function grid algorithm. For the surrogate function grid algorithm, a Kriging model is adopted to build the approximate relationship of objective and design variables with a Modified Rectangular Grid (MRG) sampling method, and the optimization iterations are based on the Kriging approximate relationship. The evaluation grid algorithm is based on the genetic algorithm, and some techniques are integrated in order to obtain good accuracy and efficiency in the grid environments, such as multi-population genetic strategy, entropy-based searching technique. In both of grid algorithms, independent meta-tasks are conducted in a collaborative way and computing pool technique is adopted to balance the computational loads.Employing the Analytic Hierarchy Process (AHP) method, a multi-Qos model is proposed to evaluate the quality of grid service such as computing capacity, price, transfer performance, security and reliability of grid resources and user's benefits. A Resource Monitoring and Load Adjusting Based Scheduling Algorithm(RMLABSA) model for independent tasks is presented in this work. The resource monitoring is aimed to find the real-time information before the task distribution and the real service ability during the computing. Integer programming method is adopted to carry out the distribution and adaptive adjusting technique is used to overcome the defaults of performance prediction for dynamic resources. The computational examples in this work show that the RMLABSA model balances the load well.Four-layer Engineering Optimization Design Grid (EODG) platform is constructed for solving the complicated engineering optimization problems, which includes grid resources layer, grid middleware layer, optimization design & scheduling layer and user interface layer. Optimization algorithm and black-box programs are sealed as necessary components and the GT4 standard is used to obtain safe, reliable, cheaper using of the massive idle resources. The EODG is successfully applied to the turbine foundation optimization, injection gate location optimization, and injection processing optimization.Some typical applications are implemented on the EODG employing the grid algorithms. The surrogate function grid algorithm is introduced to complete dynamic optimization design of the turbine engine foundation and processing optimization of injection of plastic production, and the evaluation grid algorithm is applied to injection design of gate location. The optimization results based on the grid show that the grid algorithms are very efficient and the grid platform is suitable for the engineering optimization designs.This dissertation is financially supported by the National Natural Science Foundation (10272030).