Prediction method of thermal conductivity of nanofluids based on radial basis function

Abstract

Accurately predicting the thermal conductivity of nanofluids under various thermodynamic conditions is of great importance to promote the industrial application of nanofluids. Unfortunately, the accuracy and applicability of the current theoretical or empirical models cannot meet the demand, due to the inherent complex of nanofluids. In this study, an intelligent model, named radial basis function artificial neural network (RBF–ANNs), is developed to predict the thermal conductivity of nanofluids under various conditions. Five parameters including nanoparticle volume concentration, temperature, nanoparticle diameter, thermal conductivity of nanoparticle and thermal conductivity of base fluid are selected as the input variables. A total of 1444 experimental data samples are collected to optimize the structure of model. The RBF model is compared with six theoretical models and three intelligent models through statistical and graphical analyses. Also, trend analysis and sensitivity analysis are conducted to evaluate the influencing mechanism of nanoparticle concentration, temperature, nanoparticle size, thermal conductivities of base fluids and nanoparticle on the thermal conductivity of nanofluids. Meanwhile, the quality of the experimental data is evaluated by means of leverage algorithm. Results indicate the superiority of the RBF, especially when the data size is large. The overall correlation coefficient (R2), average absolute relative deviation (AARD%) and root-mean-squared error of the developed model are 0.9931, 2.715 and 0.0316, respectively. Among the five input parameters, the volume fraction of nanoparticles has the greatest impact on the thermal conductivity of nanofluid. The results of outlier detection demonstrate that the proposed RBF model and data samples are statistically valid.

This is a preview of subscription content, log in to check access.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10
Fig. 11
Fig. 12
Fig. 13
Fig. 14

References

  1. 1.

    S.U.S. Choi and J.A. Eastman. Enhancing thermal conductivity of fluids with nanoparticles, presented at international mechanical engineering congress and exhibition, November 1995, San Francisco, CA.

  2. 2.

    Yang L, Xu JY, Du K, Zhang XS. Recent developments on viscosity and thermal conductivity of nanofluids. Powder Technol. 2017;317:348–69.

    CAS  Article  Google Scholar 

  3. 3.

    Arthur O, Karim MA. An investigation into the thermophysical and rheological properties of nanofluids for solar thermal applications. Renew Sustain Energy Rev. 2016;55(4):739–55.

    CAS  Article  Google Scholar 

  4. 4.

    Bahiraei M, Hangi M, Saeedan M. A novel application for energy efficiency improvement using nanofluid in shell and tube heat exchanger equipped with helical baffles. Energy. 2015;93:2229–40.

    CAS  Article  Google Scholar 

  5. 5.

    Rostami Z, Rahimi M, Azimi N. Using high-frequency ultrasound waves and nanofluid for increasing the efficiency and cooling performance of a PV module. Energy Convers Manag. 2018;160:141–9.

    CAS  Article  Google Scholar 

  6. 6.

    Kleinstreuer C, Feng Y. Experimental and theoretical studies of nanofluid thermal conductivity enhancement: a review. Nanoscale Res Lett. 2011;6:229.

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  7. 7.

    Zhang HY, Wang SX, Lin YX. Stability, thermal conductivity, and rheological properties of controlled reduced graphene oxide dispersed nanofluids. Appl Therm Eng. 2017;119:132–9.

    CAS  Article  Google Scholar 

  8. 8.

    Kumar N, Sonawane SS, Sonawane SH. Experimental study of thermal conductivity, heat transfer and friction factor of Al2O3 based nanofluid. Int Commun Heat Mass Transfer. 2018;80:1–10.

    CAS  Article  Google Scholar 

  9. 9.

    Alade O, Oyehan TA, Popoola IK, Olatunji SO, Bagudu A. Modeling thermal conductivity enhancement of metal and metallic oxide nanofluids using support vector regression. Adv Powder Technol. 2018;29:157–67.

    CAS  Article  Google Scholar 

  10. 10.

    Hemmati-Sarapardeha A, Varamesh A, Huseinc MM, Karan K. On the evaluation of the viscosity of nanofluid systems: modeling and data assessment. Renew Sustain Energy Rev. 2018;81:313–29.

    Article  CAS  Google Scholar 

  11. 11.

    Nadooshan A. An experimental correlation approach for predicting thermal conductivity of water-EG based nanofluids of zinc oxide. Physica E. 2017;87:15–9.

    Article  CAS  Google Scholar 

  12. 12.

    Keyvani M, Afrand M, Toghraie D, Reiszadeh M. An experimental study on the thermal conductivity of cerium oxide/ethylene glycol nanofluid: developing a new correlation. J Mol Liq. 2018;266:211–7.

    CAS  Article  Google Scholar 

  13. 13.

    Shukla KN, Koller TM, Rausch MH, Fröba AP. Effective thermal conductivity of nanofluids: a new model taking into consideration Brownian motion. Int J Heat Mass Transf. 2016;99:532–40.

    CAS  Article  Google Scholar 

  14. 14.

    Chon CH, Kihm KD, Lee SP, Choi SUS. Empirical correlation finding the role of temperature and particle size for nanofluid (Al2O3) thermal conductivity enhancement. Appl Phys Lett. 2005;87(153107):1–3.

    Google Scholar 

  15. 15.

    Godson L, Raja B, Lal DM, Wongwises S. Experimental investigation on the thermal conductivity and viscosity of silver-deionized water nanofluid. Exp Heat Transf. 2010;23:317–32.

    CAS  Article  Google Scholar 

  16. 16.

    Hamilton RL, Crosser OK. Thermal conductivity of heterogeneous two-component systems. Ind Eng Chem Fundam. 1962;1:187–91.

    CAS  Article  Google Scholar 

  17. 17.

    Timofeeva EV, Gavrilov AN, McCloskey JM, Tolmachev YV. Thermal conductivity and particle agglomeration in alumina nanofluids: experiment and theory. Phys Rev Lett. 2007;76:061203–16.

    Google Scholar 

  18. 18.

    Mintsa HA, Roy G, Nguyen CT, Doucet D. New temperature dependent thermal conductivity data for water-based nanofluids. Int J Therm Sci. 2009;48:363–71.

    CAS  Article  Google Scholar 

  19. 19.

    Prasher RS, Bhattacharya P, Phelan PE. Thermal conductivity of nanoscale colloidal solutions (nanofluids). Phys Rev Lett. 2005;94:025901.

    PubMed  Article  CAS  Google Scholar 

  20. 20.

    Aminian A. Predicting the effective thermal conductivity of nanofluids for intensification of heat transfer using artificial neural network. Powder Technol. 2016;301:288–309.

    CAS  Article  Google Scholar 

  21. 21.

    Ahmadloo E, Azizi S. Prediction of thermal conductivity of various nanofluids using artificial neural network. Int Commun Heat Mass Transfer. 2016;74:69–75.

    CAS  Article  Google Scholar 

  22. 22.

    Bahiraei M, Heshmatian S, Moayedi H. Artificial intelligence in the field of nanofluids: a review on applications and potential future directions. Powder Technol. 2019;353:276–301.

    CAS  Article  Google Scholar 

  23. 23.

    Heshmatian S, Bahiraei M. Numerical investigation of entropy generation to predict irreversibilities in nanofluid flow within a microchannel: effects of Brownian diffusion, shear rate and viscosity gradient. Chem Eng Sci. 2017;172:52–65.

    CAS  Article  Google Scholar 

  24. 24.

    Bahiraei M, Heshmatian S. Optimizing energy efficiency of a specific liquid block operated with nanofluids for utilization in electronics cooling: a decision-making based approach. Energy Convers Manag. 2017;154:180–90.

    CAS  Article  Google Scholar 

  25. 25.

    Bahiraei M, Heshmatian S, Keshavarzi M. Multi-criterion optimization of thermohydraulic performance of a mini pin fin heat sink operated with ecofriendly graphene nanoplatelets nanofluid considering geometrical characteristics. J Mol Liq. 2019;276:653–66.

    CAS  Article  Google Scholar 

  26. 26.

    Longo GA, Zilio C, Ceseracciu E, Reggiani M. Application of Artificial Neural Network (ANN) for the prediction of thermal conductivity of oxide-water nanofluids. Nano Energy. 2012;1:290–6.

    CAS  Article  Google Scholar 

  27. 27.

    Vafaei M, et al. Evaluation of thermal conductivity of MgO-MWCNTs/EG hybrid nanofluids based on experimental data by selecting optimal artificial neural networks. Physica E. 2017;85:90–6.

    CAS  Article  Google Scholar 

  28. 28.

    Esfea MH, Bahiraei M, Hajmohammad MH, Afrand M. Rheological characteristics of MgO/oil nanolubricants: experimental study and neural network modeling. Int Commun Heat Mass Transfer. 2017;86:245–52.

    Article  CAS  Google Scholar 

  29. 29.

    Papari MM, Yousefi F, Moghadasi J, Karimi H, Campo A. Modeling thermal conductivity augmentation of nanofluids using diffusion neural networks. Int J Therm Sci. 2011;50:44–52.

    CAS  Article  Google Scholar 

  30. 30.

    Hojjat M, Etemad SGh, Bagheri R, Thibault J. Thermal conductivity of non-Newtonian nanofluids: experimental data and modeling using neural network. Int J Heat Mass Transf. 2011;54:1017–23.

    CAS  Article  Google Scholar 

  31. 31.

    Zhao NB, Wen XY, Yang JL, Li SY, Wang ZT. Modeling and prediction of viscosity of water-based nanofluids by radial basis function neural networks. Powder Technol. 2015;281:173–83.

    CAS  Article  Google Scholar 

  32. 32.

    Zhao NB, Li ZM. Viscosity prediction of different ethylene glycol/water based nanofluids using a RBF neural network. Appl Sci. 2017;7:409–25.

    Article  CAS  Google Scholar 

  33. 33.

    Sayahi T, Tatar A, Bahrami M. A RBF model for predicting the pool boiling behavior of nanofluids over a horizontal rod heater. Int J Therm Sci. 2016;99:180–94.

    CAS  Article  Google Scholar 

  34. 34.

    Zendehboudi A, Tatar A. Utilization of the RBF network to model the nucleate pool boiling heat transfer properties of refrigerant-oil mixtures with nanoparticles. J Mol Liq. 2017;247:304–12.

    CAS  Article  Google Scholar 

  35. 35.

    Barati-Harooni A, Najafi-Marghmaleki A. An accurate RBF-NN model for estimation of viscosity of nanofluids. J Mol Liq. 2016;224:580–8.

    CAS  Article  Google Scholar 

  36. 36.

    Murshed SMS, Leong KC, Yang C. Investigations of thermal conductivity and viscosity of nanofluids. Int J Therm Sci. 2008;47:560–8.

    CAS  Article  Google Scholar 

  37. 37.

    Patel HE, Sundararajan T, Das SK. An experimental investigation into the thermal conductivity enhancement in oxide and metallic nanofluids. J Nanopart Res. 2010;12:1015–31.

    CAS  Article  Google Scholar 

  38. 38.

    Żyła G, Fal J. Experimental studies on viscosity, thermal and electrical conductivity of aluminum nitride–ethylene glycol (AlN–EG) nanofluids. Thermochim Acta. 2016;637:11–6.

    Article  CAS  Google Scholar 

  39. 39.

    Yiamsawasd T, Dalkilic AS, Wongwises S. Measurement of the thermal conductivity of titania and alumina nanofluids. Thermochim Acta. 2012;545:48–56.

    CAS  Article  Google Scholar 

  40. 40.

    Esfe MH, et al. Experimental study on thermal conductivity of ethylene glycol based nanofluids containing Al2O3 nanoparticles. Int J Heat Mass Transf. 2015;88:728–34.

    Article  CAS  Google Scholar 

  41. 41.

    Mostafizur RM, Bhuiyan MHU, Saidur R, Abdul-Aziz AR. Thermal conductivity variation for methanol based nanofluids. Int J Heat Mass Transf. 2014;76:350–6.

    CAS  Article  Google Scholar 

  42. 42.

    Mostafizur RM, Saidur R, Abdul-Aziz AR, Bhuiyan MHU. Thermophysical properties of methanol based Al2O3 nanofluids. Int J Heat Mass Transf. 2015;85:414–9.

    CAS  Article  Google Scholar 

  43. 43.

    Mahbubul IM, Saidur R, Amalina MA. Influence of particle concentration and temperature on thermal conductivity and viscosity of Al2O3/R141b nanorefrigerant. Int Commun Heat Mass Transfer. 2013;43:100–4.

    CAS  Article  Google Scholar 

  44. 44.

    Pryazhnikov MI, Minakov AV, Rudyak V Ya, Guzei DV. Thermal conductivity measurements of nanofluids. Int J Heat Mass Transf. 2017;104:1275–82.

    CAS  Article  Google Scholar 

  45. 45.

    Esfe MH, Saedodin S, Mahian O, Wongwises S. Thermal conductivity of Al2O3/water nanofluids. J Therm Anal Calorim. 2014;117:675–81.

    Article  CAS  Google Scholar 

  46. 46.

    Garg J, et al. Enhanced thermal conductivity and viscosity of copper nanoparticles in ethylene glycol nanofluid. J Appl Phys. 2008;103:074301.

    Article  CAS  Google Scholar 

  47. 47.

    Jiang HF, et al. Temperature dependence of the stability and thermal conductivity of an oil–based nanofluid. Thermochim Acta. 2014;579:27–30.

    CAS  Article  Google Scholar 

  48. 48.

    Aberoumand S, Jafarimoghaddam A. Experimental study on synthesis, stability, thermal conductivity and viscosity of Cu–engine oil nanofluid. J Taiwan Inst Chem Eng. 2017;71:315–22.

    CAS  Article  Google Scholar 

  49. 49.

    Jiang WT, Ding GL, Peng H. Measurement and model on thermal conductivities of carbon nanotube nanorefrigerants. Int J Therm Sci. 2009;48:1108–15.

    CAS  Article  Google Scholar 

  50. 50.

    Esfe MH, et al. Experimental investigation and development of new correlations for thermal conductivity of CuO/EG–water nanofluid. Int Commun Heat Mass Transfer. 2015;65:47–51.

    Article  CAS  Google Scholar 

  51. 51.

    Ebrahimi S, Saghravani SF. Experimental study of the thermal conductivity features of the water based Fe3O4/CuO nanofluid. Heat Mass Transf. 2017;54:999–1008.

    Article  CAS  Google Scholar 

  52. 52.

    Esfe MH, Saedodin S, Mahian O, Wongwises S. Heat transfer characteristics and pressure drop of COOH − functionalized DWCNTs/water nanofluid in turbulent flow at low concentrations. Int J Heat Mass Transf. 2014;73:186–94.

    CAS  Article  Google Scholar 

  53. 53.

    Afrand M, Toghraie D, Sina N. Experimental study on thermal conductivity of water − based Fe3O4 nanofluid: development of a new correlation and modeled by artificial neural network. Int Commun Heat Mass Transfer. 2016;75:262–9.

    CAS  Article  Google Scholar 

  54. 54.

    Amiri A, Sadri R, Shanbedi M, et al. Performance dependence of thermosyphon on the functionalization approaches: an experimental study on thermophysical properties of graphene nanoplatelet − based water nanofluids. Energy Convers Manag. 2015;92:322–30.

    CAS  Article  Google Scholar 

  55. 55.

    Esfe MH, Saedodin S, Mahian O, Wongwises S. Thermophysical properties, heat transfer and pressure drop of COOH − functionalized multi walled carbon nanotubes/water nanofluids. Int Commun Heat Mass Transfer. 2014;58:176–83.

    Article  CAS  Google Scholar 

  56. 56.

    Duangthongsuk W, Wongwises S. Measurement of temperature—dependent thermal conductivity and viscosity of TiO2—water nanofluids. Exp Thermal Fluid Sci. 2009;33:706–14.

    CAS  Article  Google Scholar 

  57. 57.

    Said Z, et al. Performance enhancement of a flat plate solar collector using titanium dioxide nanofluid and polyethylene glycol dispersant. J Clean Prod. 2015;92:343–53.

    CAS  Article  Google Scholar 

  58. 58.

    Suganthi KS, Vinodhan VL, Rajan KS. Heat transfer performance and transport properties of ZnO–ethylene glycol and ZnO–ethylene glycol–water nanofluid coolants. Appl Energy. 2014;135:548–59.

    CAS  Article  Google Scholar 

  59. 59.

    Pastoriza-Gallego MJ, Lugo L, Cabaleiro D, Legido JL, Piñeiro MM. Thermophysical profile of ethylene glycol–based ZnO nanofluids. J Chem Thermodyn. 2014;73:23–30.

    CAS  Article  Google Scholar 

  60. 60.

    Lee GJ, et al. Thermal conductivity enhancement of ZnO nanofluid using a one − step physical method. Thermochim Acta. 2012;542:24–7.

    CAS  Article  Google Scholar 

  61. 61.

    Jeong J, et al. Particle shape effect on the viscosity and thermal conductivity of ZnO nanofluids. Int J Refrig. 2013;36:2233–41.

    CAS  Article  Google Scholar 

  62. 62.

    Broomhead DS, Lowe D. Radial basis functions, multi-variable functional interpolation and adaptive networks. Compl Syst. 1988;2:321–55.

    Google Scholar 

  63. 63.

    Yu L, Lai KK, Wang S. Multistage RBF neural network ensemble learning for exchange rates forecasting. Neurocomputing. 2008;71:3295–302.

    Article  Google Scholar 

  64. 64.

    Zendehboudi A, Saidur R. A reliable model to estimate the effective thermal conductivity of nanofluids. Heat Mass Transf. 2018;55(2):397–411.

    Article  Google Scholar 

  65. 65.

    Ayatollahi S, Hemmati-Sarapardeh A, Roham M, Hajirezaie S. A rigorous approach for determining interfacial tension and minimum miscibility pressure in paraffin-CO2 systems: application to gas injection processes. J Taiwan Inst Chem Eng 2016: https://doi.org/10.1016/j.jtice.2016.02.013.

    Article  Google Scholar 

  66. 66.

    Jung S, Kwon S. Weighted error functions in artificial neural networks for improved wind energy potential estimation. Appl Energy. 2013;111:778–90.

    Article  Google Scholar 

  67. 67.

    Mohammadi H, Eslamimanesh A, Gharagheizi F, Richon D. A novel method for evaluation of asphaltene precipitation titration data. Chem Eng Sci. 2012;78:181–5.

    CAS  Article  Google Scholar 

  68. 68.

    Mohagheghian E, Zafarian-Rigaki H, Ghahfarrokhi YM, Hemmati-Sarapardeh A. Using an artificial neural network to predict carbon dioxide compressibility factor at high pressure and temperature. Korean J Chem Eng. 2015;32(10):2087–96.

    CAS  Article  Google Scholar 

  69. 69.

    Bruggeman DAG. Berechnung verschiedener physikalischer Konstanten von heterogenen Substanzen. I. Dielektrizitätskonstanten und Leitfähigkeiten der Mischkörper aus isotropen Substanzen. Ann Phys. 1935;416:636–64.

    Article  Google Scholar 

  70. 70.

    Wasp FJ. Solid–liquid slurry pipeline transportation. Berlin: Trans. Tech; 1977.

    Google Scholar 

  71. 71.

    Vatani A, Woodfield PL, Dao DV. A survey of practical equations for prediction of effective thermal conductivity of spherical–particle nanofluids. J Mol Liq. 2015;211:712–33.

    CAS  Article  Google Scholar 

  72. 72.

    Baghban A, Habibzadeh S, Ashtiani F. Toward a modeling study of thermal conductivity of nanofluids using LSSVM strategy. J Therm Anal Calorim. 2018. https://doi.org/10.1007/s10973-018-7074-5.

    Article  Google Scholar 

  73. 73.

    Meybodi M, Naseri S, Shokrollahi A, Daryasafar A. Prediction of viscosity of water-based Al2O3, TiO2, SiO2, and CuO nanofluids using a reliable approach. Chemometr Intell Lab Syst. 2015;149:60–9.

    CAS  Article  Google Scholar 

  74. 74.

    Mehrabi M, Pesteei S, Pashaee T. Modeling of heat transfer and fluid flow characteristics of helicoidal double-pipe heat exchangers using Adaptive Neuro-Fuzzy Inference System (ANFIS). Int Commun Heat Mass Transfer. 2011;38:525–32.

    Article  Google Scholar 

  75. 75.

    Selimefendigil F, Öztop H. Magnetic field effects on the forced convection of CuO-water nanofluid flow in a channel with circular cylinders and thermal predictions using ANFIS. Int J Mech Sci. 2018;146–147:9–24.

    Article  Google Scholar 

  76. 76.

    Aminossadati S, Kargar A, Ghasemi B. Adaptive network-based fuzzy inference system analysis of mixed convection in a two-sided lid-driven cavity filled with a nanofluid. Int J Therm Sci. 2012;52:102–11.

    CAS  Article  Google Scholar 

  77. 77.

    Tatar A, Barati-Harooni A, Najafi-Marghmaleki A, et al. Predictive model based on ANFIS for estimation of thermal conductivity of carbon dioxide. J Mol Liq. 2016;224:1266–74.

    CAS  Article  Google Scholar 

Download references

Acknowledgements

This work was supported by Yunnan Provincial Department of Education Science Research Fund Project (Grant No. 2018JS551, 2019J0025) and Scientific Research Foundation of Kunming Metallurgy College (Grant No. Xxrcxm201802).

Author information

Affiliations

Authors

Corresponding authors

Correspondence to Hui Huang or Xiaobo Long.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendices

Appendix 1

See Tables 6 and 7.

Table 6 Details of the experimental data used in this study
Table 7 The weights and biases of the optimized RBF model

Appendix 2

Multilayer perceptron artificial neural networks (MLP–ANNs)

MLP has been widely applied for solving complex nonlinear problem with high accuracy. An MLP contains one input layer, one output layer and at least one hidden layer. Each layer consisted of numerous units (neuros) connected to the neurons in the following layers. The difference between RBF and MLP is the way that the neurons process the information [10]. The inputs to each neuron are multiplied by the weights (w), and the results are added with each other and with the bias (b) [21]. Then, the resultant value is obtained through a suitable activation function (f):

$$y_{\text{j}} = f\left( {\sum\limits_{i = 1}^{n} {w_{\text{ji}} x_{\text{i}} + b_{\text{j}} } } \right)$$
(10)

where xi is the ith input vector, wji is the connection weight between the jth neuron in hidden and the ith neuron in input layer, n is the number of neurons in hidden layer, yj is the output of the jth neuron, bj is the bias of jth neuron, and f represents the transfer function. Many researchers suggested that linear (purelin) function was suitable for the output layer, while the tangent sigmoid (tansig) function was suggested for the hidden layer due to their higher ability [10, 21, 64].

Optimization algorithm used for training the model plays a key role in the performance of MLP. Through training algorithm, the values of weight and bias are adjusted until the minimum deviation between predicted and experimental values is obtained. Backpropagation (BP) is the most common algorithm for training the MLP network. The inputs are feed-forward propagating in neural network. The weights and biases between neurons are optimized according to the errors between the predicted and experimental data, during the backward propagation process. Among the BP training algorithm, Levenberg–Marquardt backpropagation (LM) algorithm is selected to train the MLP model [21]. For now, there is no general method to determine the number of hidden layers in MLP and the number of neurons in hidden layer, and it is usually determined by trial-and-error method. For a complex system with larger amount of data, a MLP with two hidden layer was found to be more appropriate [10, 64]. Through the comprehensive calculation, the best structure of LM–MLP with the lowest error is found to be 5–12–8–1. The values of weights and biases are listed in Tables 8 and 9.

Table 8 The weights and biases for LM-MLP (part 1)
Table 9 The weights and biases for LM-MLP (part 2)

Least square support vector machine (LSSVM)

With the development of artificial intelligence (AI), support vector machine (SVM) has been developed to solve complex problems (classification, nonlinear regression, pattern recognition, etc..) with high accuracy and efficiency [10, 72, 73]. In an SVM, the inputs are mapped to a high-dimensional feature space using nonlinear mapping function. However, the efficiency of conventional SVM is less when a wide range of experimental data is being used, due to the inequality constraints. To overcome the difficulties of SVM, a modified version of SVM, named least square support vector machine (LSSVM), has been proposed. In LSSVM, regression error is added to the optimization constraints [10], and by this way, the convergence speed of LSSVM increases. LSSVM is suitable for solving the nonlinear problems, especially when the number of data is large.

In LSSVM, the penalized function is defined as

$$Q_{\text{LSSVM}} = \frac{1}{2}w^{\text{T}} w + \frac{1}{2}\gamma \sum\limits_{i = 1}^{n} {e_{\text{i}}^{2} }$$
(11)

subjected to the following equality constraints:

$$y_{\text{i}} = w^{\text{T}} g(x_{\text{i}} ) + b + e_{\text{i}} ,\quad i = 1,2,3, \ldots ,n$$
(12)

where ei is the error variable, γ is the regularization constant which shows the summation of regression errors, g (x) is the mapping function, w is the regression weight, b is the bias, n is the number of data, and superscript T is the transpose matrix.

The weight coefficient (w) is expressed as

$$w = \sum\limits_{i = 1}^{n} {\alpha_{\text{i}} x_{\text{i}} }$$
(13)

The Lagrangian multiplier (αi) is expressed as

$$\alpha_{\text{i}} = 2\gamma \cdot e_{\text{i}}$$
(14)

Considering the dependent and independent parameters in LSSVM is linearly separable, the equation will be modified as

$$y = \sum\limits_{i = 1}^{n} {\alpha_{\text{i}} } x_{\text{i}}^{\text{T}} x + b$$
(15)

Then, the Lagrange multiplier (αi) can be obtained as:

$$\alpha_{\text{i}} = \frac{{y_{\text{i}} - b}}{{x_{\text{i}}^{\text{T}} x + (2\gamma )^{ - 1} }}$$
(16)

For nonlinear regression problems, the Kernel function is introduced to combine with Eq. (B6)

$$y = \sum\limits_{i = 1}^{n} {\alpha_{\text{i}} } K(x_{\text{i}} ,x) + b$$
(17)

where the Kernel function K (xi, x) is defined as:

$$K(x_{\text{i}} ,x) = g(x_{\text{i}} ) \cdot g(x)^{\text{T}}$$
(18)

One of the most common used Kernel functions in LSSVM is Gaussian radial basis function because it has lower tuning parameters and less numerical adversity [73]. In this work, Gaussian Kernel is used for developing the LSSVM, as shown below

$$K(x_{\text{i}} ,x) = \exp \left( { - \frac{{\left\| {x_{\text{i}} - x} \right\|^{2} }}{{\sigma^{2} }}} \right),\quad i = 1,2,3, \ldots ,n$$
(19)

where σ2 is a tuning parameter that should be optimized by optimization algorithm during LSSVM training.

To now, there are two tuning parameters, namely σ2 and γ, which should be optimized to find the global optimum of the problem. In this study, cross-validation is employed for optimizing the parameters of LSSVM. The optimum values of σ2 and γ are 0.4886 and 46.821, respectively.

Adaptive neuro-fuzzy inference system (ANFIS)

Adaptive neuro-fuzzy inference system (ANFIS) is the integration of ANNs and fuzzy logic algorithms, and it has been known as a useful tool for solving intricate problem in many fields [22]. The advantages of ANNs and fuzzy logic algorithms are mutually complemented, and the deficiencies in either of the two methods are minimized by the other method [74]. The combination process is carried out through introducing fuzzy if–then rules and specific functions called membership functions. These MFs are subjected to tuning processes by utilizing the ability of ANNs. The ANFIS structure is composed of introductory part and concluding part, which are linked by a set of rules. For a simple Takagi–Sugeno ANFIS with two input variables (x1 and x2), the model can be constructed by the following rules:

$$\begin{aligned} {\text{Rule}}\;1\;{\text{If}}\left( {x_{1} \;{\text{is}}\;A_{1} } \right)\;{\text{and}}\;\left( {x_{2} \;{\text{is}}\;B_{1} } \right)\;{\text{then}}\;y_{1} = p_{1} x_{1} + q_{1} x_{2} + r_{1} \hfill \\ {\text{Rule}}\;2\;{\text{If}}\left( {x_{1} \;{\text{is}}\;A_{2} } \right)\;{\text{and}}\;\left( {x_{2} \;{\text{is}}\;B_{2} } \right)\;{\text{then}}\;y_{2} = p_{2} x_{1} + q_{2} x_{2} + r_{2} \hfill \\ \end{aligned}$$

where Ai and Bi are the fuzzy sets of corresponding input parameters. yi is the output of the system. pi, qi and ri are designing parameters which are obtained from learning.

Briefly speaking, the ANFIS structure consists of five distinct layers with each layer comprised of neurons, which forms a multilayer network [75]. The first layer is an adaptive layer in which fuzzy formation takes place. In the second layer, the input signal values in each neuron are multiplied by each other and fuzzy rules are performed. In the third layer, the firing strength of rules is normalized. The fourth layer is the conclusive part of fuzzy rules which is obtained from multiplication of normalized firing strength. The fifth layer contains single node, which is responsible for calculating the overall output of network. A more detailed information about ANFIS can be found in Mehrabi et al. [74].

The structure identification of ANFIS involves (1) selecting the input parameters, (2) input space partitioning, (3) choosing the appropriate number and the kind of membership functions for inputs, (4) creation of fuzzy rules, (5) premise and conclusion parts of fuzzy rules and (6) selection of the initial parameters for membership functions [76]. The ANFIS employs training methods developed from an ANNs for finding the suitable fuzzy rules and fuzzy membership functions. Considering the situation of data used for modeling as well as the capability of the network, subtractive clustering method is used to construct the ANFIS due to its independency to dimension of problem and faster training process [77]. The optimum value of influence radius (RD) is 0.4542, and Gaussian membership function is selected as it is associated with the minimum value of RMSE. The adaptive and consequent parameters set in ANFIS are obtained via the integration of backpropagation algorithm with least squares method. Details of the proposed ANFIS for predicting keff are listed in Table 10.

Table 10 Details of the proposed ANFIS for predicting keff

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Zhang, S., Ge, Z., Fan, X. et al. Prediction method of thermal conductivity of nanofluids based on radial basis function. J Therm Anal Calorim 141, 859–880 (2020). https://doi.org/10.1007/s10973-019-09067-x

Download citation

Keywords

  • Effective thermal conductivity
  • Nanofluid
  • Radial basis function
  • Neural network
  • Sensitivity analysis
  • Outlier detection