基于Hopfield神经网络的结构优化算法研究

Research on Structural Optimization Algorithm Based on Hopfield Neural Network

利用Hopfield神经网络并结合模拟退火算法,对甘蔗收获机械台架结构进行了优化。建立了神经网络系统能量函数与优化问题目标函数之间的对应关系、神经网络演化过程与优化问题寻优过程之间的对应关系、神经网络系统到达平衡点与优化问题最优解之间的对应关系。采用改进的惩罚算子以提高神经网络的收敛速度,经过12次迭代后,优化目标下降17.5%,且应力小于190MPa,表明该优化方法可充分利用设计资源,得到全局最优解。算例证明该算法高效可靠,切实可行,有较强的工程实用性。

The structural optimal problem of a sugarcane harvester was solved through the method of Hopfield neural network and simulated annealing. The corresponding relationships between neural network and optimal problem were built, such as neural network energy function and objective of optimal problem,neural network evolving process and searching process of optimal design, neural network equilibrium point and solution of optimal problem and so on. The improved published operator was used to accelerate neural network convergence. After 12 iterations, the object is decreased by 17.5 %, and the constraints are within the range of 190MPa, which shows the design resources can be used sufficiently and the whole optimal result can be received. An example proves this method to be an effective and reliable way and available in engineering project.