非线性输出器反馈式神经网络 UPS 曲线预测-《计算机技术与发展》

文章信息/Info

作者:: 韩潇影1 ; 林婧婧2 ; 王雅萍1 ; 许霞1 ; 刘峰民1; 1. 甘肃省气象信息与技术装备保障中心,甘肃兰州 730020;
2. 兰州区域气候中心,甘肃兰州 730020

Author(s):: HAN Xiao-ying1 ; LIN Jing-jing2 ; WANG Ya-ping1 ; XU Xia1 ; LIU Feng-min1; 1. Gansu Meteorological Information and Technical Equipment Support Center,Lanzhou 730020,China;
2. Lanzhou Regional Climate Center,Lanzhou 730020,China

Keywords:: recurrent neural network; nonlinear output; uninterruptible power supply curve prediction

摘要:: 针对神经网络计算中中间层规模过大问题以及简单线性输出方式不能有效利用权值输出的高维信息的缺点,采用非线性输出装置代替传统反馈式神经网络计算时简单的线性输出单元,将其应用于反馈式神经网络模型。通过实验证明,使用了非线性输出器后,改进后的反馈式神经网络的性能要优于传统的人工神经网络。不间断电源电池在经过长时间放置或者长期使用之后,都会出现不同程度的电荷衰减现象。采用基于非线性输出器的反馈式神经网络来表达这一非线性系统,利用这种模型来有效掌握电池的工作状态,为电力检修提供了较大帮助。实验结果表明,在相同中间层规模下,相比于传统中间层计算模型,改进后的反馈式神经网络预测成功率提高了 2% 以上。

Abstract:: To solve the problem that the middle layer is too large in the calculation of neural network and the disadvantage of simple line aroutput method which cannot effectively use the weight value to output high-dimensional information,the nonlinear output device replaces simple linear output unit for traditional recurrent neural network computing,applied to recurrent neural network model. Through a series of experiments,after using the nonlinear output, the performance of the improved recurrent neural network is better than that of the traditional artificial neural network. The charge attenuation of the uninterruptible power supply battery will occur in different degree after long time placement or long time use. A recurrent neural network based on nonlinear output device is used to express the nonlinear system. Using this model to effectively master the working state of the battery,it has provided a great help for the electricity maintenance work. Experiment shows that in the same middle layer scale,compared with the traditional middle layer calculation model,the success rate of the improved recurrent neural network prediction increased by more than 2% .