Parallel Computing in HP-OCP

Abstract

The most computationally demanding part of structural design optimization is the solution of the FE equations and design of the structural model. Therefore, there is a need for the implementation of strategies that can reduce the computational cost of each iteration and thus manage to achieve the same optimized result with considerable reduction in the optimization time. High Performance Optimization Computing Platform (HP-OCP) is an optimization software developed in C# programming language by ISAAR-NTUA and OptiStructre Ltd. [1] which provides a holistic optimization approach for civil engineering structures. It combines powerful derivative-based and derivative-free optimization algorithms like the Projected Quasi-Newton (PQN), Constrained Optimization by Linear Approximation (COBYLA), Latin Hypercube (LH), Differential Evolution etc. [2] integrated with different structural analysis software's like SAP2000, ETABS & SCIA Engineer utilizing their abilities in finite element analysis and most importantly different design codes into the optimization procedure. To deal with the computational demand deriving from this coupling of optimization algorithms and commercial structural analysis software's parallel computational procedures have been implemented to HP-OCP. These procedures were tested in real world civil engineering problems and produced very good results. Parallel strategies are implemented both at the level of the optimization algorithm, by exploiting the natural parallelization features of the evolutionary algorithms, as well as at the level of the repeated structural analysis problems that are required by the optimization algorithm. The numerical tests presented demonstrate the computational advantages of the proposed parallel strategies, which become more pronounced in large-scale optimization problems. 1

Full document