含连续／离散变量结构优化中的神经网络与变尺度模拟退火方法

传统的优化方法难于有效地处理含有连续／离散混合变量优化问题，本文介绍了一种改进的变尺度模拟退火方法并与人工神经网络能量函数模型相结合，用于求解含连续／离散设计变量的工程结构优化问题，较好地解决了模拟退火技术用于工程结构优化时选取具有全局性的初始点困难及迭代次数较多的弱点。算例表明，该方法可以使模拟退火算法从局部最优的陷阱中跳出，最后求出整体最优解。

Neural Network and Rescaled Simulated Annealing Method for Structural Optimization with Mixed Continuous/Discrete Design Variables

The traditional optimization methods have difficulty to deal with the optimization problems with continuous/discrete design variables. A kind of improved rescaled simulated annealing(RSA) me-( thods) combined with energ y function model of artificial neural network(ANN) is presented for the optimal design of a structural system with mixed variables. It is very difficult to fin d a good start point and many iteration times are needed when simulated anne aling(SA) is used for structural optimization, RSA can improve the weakness of SA significantly. The results of the examples demonstrate that this kind of met hod may make SA skip from local optimization results trap, and get the best opti mal results.