Abstract

The thermal conductivity of Fe₃O₄/water nanofluid was forecasted using two methods of artificial neural network (ANN) along with response surface method (RSM). For ANN methods, the optimal neurons number and for RSM, the usefulness of several predicting function was specified using R-square criteria, and margin of deviation (MOD). It was found that $R^2$ for ANN was 0.999 while for RSM, this figure was 0.998. The mean square error for the former and latter methods was 0.00038 and 0.0013, respectively. Taking into account 0.964% and 1.895% for ANN and RSM, it was concluded that ANN efficacy was superior to RSM. Moreover, ANN was able to predict all points with a MOD below 1%, while 70% of data points in the RSM technique have a MOD of less than 1%.

摘要

_{Fe 3} O ₄ /水纳米流体的热导率预测采用人工神经网络（ANN）和响应面法（RSM）两种方法。对于 ANN 方法，最佳神经元数量和对于 RSM，使用 R 平方标准和偏差边际 (MOD) 指定几个预测函数的有用性。结果发现$R^2$ANN 为 0.999，而 RSM 为 0.998。前一种方法和后一种方法的均方误差分别为 0.00038 和 0.0013。考虑到 ANN 和 RSM 的 0.964% 和 1.895%，得出的结论是 ANN 功效优于 RSM。此外，ANN 能够预测 MOD 低于 1% 的所有点，而 RSM 技术中 70% 的数据点的 MOD 低于 1%。