A Research Of ARPSO-BP In Structural Optimization Design

In this paper, a combined computational intelligence method based on Adaptive Random Particle Swarm Optimization (ARPSO) and Back Propagating (BP) Neural Network is applied to engineering structural optimization. Aiming at solving the problem that Particle Swarm Optimization (PSO) converges in global optimum effectively with a certain inertia weight in certain function, a random adaptive inertia weight is introduced. The benefit of ARPSO to optimization analysis is verified by several typical mathematical poly-peaks functions and engineering optimization problems as well. The results of simulation show that optimization process can be efficiently avoided being entrapped into local optimum with ARPSO and with higher efficiency. A group of BP neural networks are used instead of Finite Element Analysis, which accelerates the computation speed in structural analysis greatly. ARPSO method is also introduced to construct BP neural networks, which makes BP neural networks smart ones and be with adaptive abilities to solve different optimization problems. Finally, ARPSO method is used combined with BP neural network to optimize a turbine disk with fourteen design variables. The optimum result is proven to be good enough in comparison with FEM.A Graphical User Interface which is practicable is built with C++. An optimization problem with no more 50 design variables can be solved with it, which makes it practicable in solving engineering problems. The new method is convenient and efficient and will find wide use in structural optimization.