摘要
为解决管道系统的地震模拟试验中由于某些测点的加速度测量计损坏导致加速度试验数据测量错误的问题,提出使用神经网络对错误的数据进行预测的方法。首先利用同一工况下测得完整加速度时程曲线的数据分别训练BP、MEA-BP神经网络,不断调整网络结构使预测精度在可接受范围内,并验证网络的适用性,最后利用训练好的神经网络预测未测量完全的加速度数据,补全加速度响应时程曲线。研究结果表明利用神经网络对振动试验数据进行预测是有效的、可行的,且MEA-BP神经网络对结构响应的预测精度高于BP神经网络。
In order to solve the problem of incomplete acceleration measurement caused by the damage of accelerometers at some measuring points in the seismic simulation test of pipeline system, a method of using neural network to predict the incomplete data is proposed.First of all, under the same working condition, measured point data of complete time curve is used to training the BP,MEA-BP neural network respectively.Then network structure is constantly adjusted to make the prediction accuracy in the acceptable range, and verify the applicability of the network.Finally the trained neural network is used to predict unmeasured acceleration data completely, and fill acceleration response time curve.The results show that it is effective and feasible to use the neural network to predict the vibration test data, and the prediction accuracy of MEA-BP is higher than that of BP.
作者
俞树荣
尹思敏
薛睿渊
李勇霖
YU Shurong;YIN Simin;XUE Ruiyuan;LI Yonglin(College of Petrochemical Technology,Lanzhou University of Technology,Lanzhou 73)
出处
《甘肃科学学报》
2022年第6期31-37,共7页
Journal of Gansu Sciences
关键词
地震台模拟试验
管道结构
思维进化算法
神经网络
数据预测
Seismic station simulation test
Pipeline structure
Mind evolution algorithm
Neural network
Data prediction
