UC知道神经网络方面的翻译，望个位英语能人帮下忙，麻烦了
来源：百度知道 编辑：UC知道 时间：2024/05/27 12:07:42
After evaluating a large number of NNs, the optimal
ANN architecture has four layers with neurons of 8 and
5 in the first and the second hidden layers, respectively
(Fig. 4). The scaling and activation functions for all four
layers were linear [-1; 1], tanh15, symmetric logistic,
and symmetric logistic, respectively [see Eqns (3)–(8)].
Both learning rate and momentum were 0.1, and the
initial weights were 0.3 for all neurons in the ANN. The
observed and predicted values of the performance
indices of rice drying are plotted in Fig. 5. The
prediction error analysis is given in Table 1. The MRE
for six outputs varies from 2.0% (for Mf ) to 8.3% (for
Kc) (Table 1). Only Kc predictions have MRE > 5%. The
other five predictions are within 55% MRE. This
indicates the reliability of ANN predictions. Figure 5
indicates that ANN predictions are following the
observed data closely for all six outputs. Larger errors
ANN architecture has four layers with neurons of 8 and
5 in the first and the second hidden layers, respectively
(Fig. 4). The scaling and activation functions for all four
layers were linear [-1; 1], tanh15, symmetric logistic,
and symmetric logistic, respectively [see Eqns (3)–(8)].
Both learning rate and momentum were 0.1, and the
initial weights were 0.3 for all neurons in the ANN. The
observed and predicted values of the performance
indices of rice drying are plotted in Fig. 5. The
prediction error analysis is given in Table 1. The MRE
for six outputs varies from 2.0% (for Mf ) to 8.3% (for
Kc) (Table 1). Only Kc predictions have MRE > 5%. The
other five predictions are within 55% MRE. This
indicates the reliability of ANN predictions. Figure 5
indicates that ANN predictions are following the
observed data closely for all six outputs. Larger errors
在评估了大量的NN后，最理想的ANN结构由四层构成，在第一和第二隐藏层中分别有八和五个神经元（图4）。这四层的测量和活化功能分别为线性[-1；1]，tanh15，逻辑对称，逻辑对称[见Ehns(3)-(8)]。对于ANN中的所有神经元来说，比率和动量都为0.1， 原始重量为0.3。观察到的和预测的稻子干燥(rice drying)的性能指标如图5所示。对于预测的误差的分析在表格1中。六个输出量的MRE差别在2.0% 和8.3%之间（对于Kc）(表格1)。只有Kc的预测是MRE大于5%的。其余的五个预测在55%以内。这表明了ANN预测的可靠性。图五表明对于六项输出量来说ANN预测严格遵循了观察到的数据。在尝试19中，所有的输出量都有大的误差。这可能由大的实验误差造成。对于能量消耗E预测来说，二十六组实验数据中的三组存在大的误差。相似地，谷粒爆裂（kernel cracking）Kc预测中，四组数据出现大的误差；最后的湿度含量Mf预测中，两组数据出现大的误差。如果去除第19组数据，预测误差将被进一步降低。对于预测误差的分析和预测与观察到的性能指标的作图（图5），我们可以清楚地看到预测和观察到的数值之间的吻合。
。。。我尽力了，如果有你看不懂的地方，你发信息问我吧。。。我可以用大白话表达一下我想说的是什么。。。还有，我把翻译得不确定的单词写在括号里了。