A Preliminary study of relationships between thermal conductivity and petrophysical parameters in Hamra Quartzites reservoir, Hassi Messaoud field (Algeria) Ahmed Ali Zerrouki a,∗ , Yves Geraud b , Marc Diraison c , Kamel Baddari d a Univ Ouargla, Fac. des Hydrocarbures, des énergies Renouvelables et Sciences de la terre et de l'univers, Lab. Géologie du Sahara, 30000, Ouargla, Algeria b Ecole Nationale Supérieure de Géologie, CNRS UMR7359-GéoRessources, Université de Lorraine, CREGU, 2 Rue du Doyen Marcel Roubault, TSA 70605, F-54518, Vandoeuvre-lès-Nancy, cedex, France c Institut de Physique du Globe de Strasbourg (IPGS), CNRS UMR7516, Université de Strasbourg, EOST, 5 Rue René Descartes, 67084, Strasbourg cedex, France d Laboratoire LIMOSE, Département de Physique, Faculté des Sciences, Université M׳Hamed Bougara, 2 Avenue de l'indépendance, 35000, Boumerdès, Algeria A R T I C L E I N F O Keywords: Thermal conductivity Hamra quartzites reservoir Petrophysical parameters Cement RBF neural network A B S T R A C T In geothermic studies, the thermal conductivity (TC) is an essential parameter needed to calculate heat flow in rocks. It is obtained generally with high accuracy from laboratory measurements. The methodology proposed in this paper is to estimate the TC, which depends on many parameters; such as mineralogy, porosity, shape of voids and nature of contact between grains, based on linear and nonlinear relationships. In order to predict the thermal conductivity from other petrophysical parameters in the Hamra Quartzites reservoir, we have measured porosity, density and permeability, for dry samples taken from core wells. The correlation coefficients (R) were calculated between thermal conductivity and other petrophysical parameters for all samples. The results show that, the correlation coefficient is moderate between TC and the porosity, weak between TC, density and permeability. To improve these correlations, samples were classified into cemented and uncemented sets. A minor improvement on the correlation coefficients is noted between TC, density and porosity in uncemented samples, with values equal 0.51 and 0.73, respectively. The application of Radial Basis Function (RBF) neural networks, using density, porosity and permeability as inputs and thermal conductivity as output, permit us to predict the thermal conductivity with high precision. The correlation coefficient between TC estimated by the RBF neural network is the same as that measured in laboratory equaling 0.983. 1. Introduction The thermal conductivity is a fundamental parameter in geothermic studies. It is an essential factor in the calculation of heat flow in fundamental and applied geothermal studies. This parameter is defined as the quantity of heat transmitted due to the unit temperature gradient (Popov et al., 1999; Lide, 1998). Divided-bar, Line-source and Optical scanning are principal laboratory techniques used to measure this parameter. The thermal conductivity of rocks is influenced by several intrinsic and extrinsic factors, such as mineralogical composition, texture, grains size, degree of crystallization, porosity, density, pressure, cracks, nature of fluid, fluid saturation, rock lamination and temperature (Jumikis, 1966; Harmathy, 1970; Fayette et al., 2000; Sevostianov, 2006; Abdulagatova et al., 2010; Alishaev et al., 2012). The TC can be obtained by the following principal methods: (i) from empirical relationships based on laboratory measurements, which relate thermal conductivity and other petrophysical properties measurements of rocks (Özkahraman et al., 2004; El Sayed, 2011; Duchkov et al., 2014), (ii) from wells log data, by searching the major minerals composition of the rock, then derive indirectly the thermal conductivity from them (Demongodin et al., 1991; Hartmann et al., 2005, 2008) and (iii) recently some authors have used artificial intelligence to predict thermal conductivity for sandstone (Goutorbe et al., 2006; Singh et al., 2007; Vaferi et al., 2014; Gitifar et al., 2014). Many studies are carried out to search relationships between thermal conductivity and petrophysical properties in different rock types. For example, the relationships between thermal conductivity, porosity, P and S-waves velocity in core samples of sandstone were investigated by El Sayed (2011), Haffen et al. (2013) and Esteban et al. (2015). Sukanta Roy (2014) elaborated the effect of water saturation on thermal conductivity in sedimentary rock samples and in metamorphic and igneous rocks. The relationships between thermal conductivity and other petrophysical parameters with double porosity rock characteristic were established in carbonates (Kazatchenko et al., 2006). The equations in these works have empirical forms, and are valid generally in the studied areas. If the mineralogical compositions of rock samples change in other areas, these empirical relationships yield weak results. Radial Basis Function (RBF) neural network is a nonlinear mathematical approach. It is a particular class of multi-layer neural networks. The popularity of RBF neural networks compared to other kinds of neural networks concerns: (i) their capacity to approximate function problems, (ii) their simple and reduced architecture