模拟退火改进的神经网络算法及其在振动分析中的应用

Improved Simulated Annealing Neural Network Algorithm and Its Application in Vibration Analysis

将传统的反向传播算法(BP算法)神经网络模型结合模拟退火算法及最佳保留原则,提出一种改进的神经网络模型,并将改进之后的网络模型应用于对颗粒碰撞阻尼的分析。训练仿真结果显示:改进后的算法与传统的BP算法、LM算法相比具有更高的可靠性,更快的收敛速度,仿真结果与实验结果更接近。用训练好的模拟退火神经网络模型对颗粒碰撞阻尼的激振频率、填充率和振幅有效值等参数进行了仿真,得到了系统在低频阶段颗粒粒度、填充率和振幅有效值之间的关系。

The traditional back-propagation algorithm （BP algorithm ） neu- ral network model was improved combined with simulated annealing algorithm and the best retain principle. A new improved neural network model was proposed and applied to the analysis of particle collision damping. Training simulation results showed that the improved algorithm compared with the traditional BP algorithm, LM algorithm had a higher reliability, faster convergence speed. Using the trained simulated annealing neural network model, the excitation frequency, fill rate, RMS amplitude and other parameters of particle collision damping were simulated. The relationship between the particle size, fill rate and amplitude RMS of the system in the low-frequency phase was obtained.