| 基于EKF-RNN算法的抗震棒材性能预报模型研究 |
| |
| 引用本文： | 顾力平. 基于EKF-RNN算法的抗震棒材性能预报模型研究[J]. 机械设计与制造, 2011, 0(11) |
| |
| 作者姓名： | 顾力平 |
| |
| 作者单位： | 南通职业大学电子信息工程学院,南通,226007 |
| |
| 基金项目： | 南通市应用研究计划(K2010065) |
| |
| 摘 要： | 以抗震棒材的C、Si、Mn、P四种化学元素及冷却速度、终轧温度为输入;抗拉强度、断面收缩率作为输出建立了热轧抗震棒材力学性能预报模型。提出了一种基于扩展卡尔曼滤波的回归神经网络权值训练算法(EKF-RNN),对抗震棒材性能预报模型进行权值训练。并与随机梯度法训练回归神经网络权值的算法进行比较,仿真结果表明,随机梯度法存在局部极小值、收敛速度慢,而扩张卡尔曼滤波算法很好地解决了这些问题,并得到了比较满意的结果,更具优越性。
|
| 关 键 词： | 扩展卡尔曼滤波 回归神经网络 系统辩识 抗震棒材 性能预报 |
Study on prediction models of anti-knock steel bar quality based on EKF-RNN algorithm |
| |
| GU Li-ping. Study on prediction models of anti-knock steel bar quality based on EKF-RNN algorithm[J]. Machinery Design & Manufacture, 2011, 0(11) |
| |
| Authors: | GU Li-ping |
| |
| Affiliation: | GU Li-ping(School of Electronics & Information Engineering,Nantong Vocational College,Nantong 226007,China) |
| |
| Abstract: | A prediction model for hot rolling anti-knock steel bar property has been established with 4 chemical elements(C,Si,Mn,P),the cooling rate,the maximum temperature as input,and the tensile strength,the percentage of area reduction as output.Then an algorithm of recurrent neural network weight based on extended kalman filter trains(EKF-RNN)is proposed to carry out the weight training for model of predicting anti-knock steel bar property,which will be compared with the EKF-RNN algorithm trained by random gradi... |
| |
| Keywords: | Extended kalman filter Recurrent neural network System identification Anti-knock steel bar Property prediction |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
