Contents lists available at ScienceDirect Applied Thermal Engineering journal homepage: www.elsevier.com/locate/apthermeng Research Paper Thermal comfort prediction in a building category: Artificial neural network generation from calibrated models for a social housing stock in southern Europe Rocío Escandón a, ⁎ , Fabrizio Ascione b , Nicola Bianco b , Gerardo Maria Mauro b , Rafael Suárez a , Juan José Sendra a a Instituto Universitario de Arquitectura y Ciencias de la Construcción, Escuela Técnica Superior de Arquitectura, Universidad de Sevilla, Av. de Reina Mercedes 2, 41012 Seville, Spain b Università degli Studi di Napoli Federico II, Piazzale Tecchio 80, 80125 Napoli, Italy HIGHLIGHTS • ANN generation for thermal comfort prediction based on real data mea- sured in-situ. • ANN for a large building stock: linear- type social housing in southern Spain. • Evaluation of the thermal perfor- mance of a free-running building ca- tegory. • The building category characteriza- tion shows a lack of indoor thermal comfort. • The developed ANN becomes an im- portant tool for testing retrofitting measures. GRAPHICAL ABSTRACT ARTICLE INFO Keywords: Social housing stock Thermal comfort Building performance simulation Sensitivity analysis Simulation model calibration Surrogate models ABSTRACT A significant part of the housing stock in southern Europe is obsolete and in need of extensive retrofitting to improve its energy performance and thermal comfort. However, before adequate retrofit measures can be proposed for this housing stock, the characterization of current building performance is fundamental. Although the simulation tools frequently used and widely accepted by the scientific community ensure accurate results, these require high computational times. The main aim of this paper is the development of a surrogate model to speed up the thermal comfort prediction for any member of a building category, ensuring high reliability by testing the entire simulation process with real data measured in-situ. To this end, an artificial neural network (ANN) is generated under MATLAB® environment using the data obtained from EnergyPlus simulations for linear-type social housing multi-family buildings in southern Spain, which were constructed in the post-war period. The developed ANN provides a regression coefficient between simulation targets and ANN outputs of 0.96, with a relative error between monitored and simulated data below 9%. A further result is that the building category characterization shows a general lack of suitable indoor thermal comfort conditions, thereby showing the great need for effective retrofit strategies. https://doi.org/10.1016/j.applthermaleng.2019.01.013 Received 2 October 2018; Received in revised form 13 December 2018; Accepted 7 January 2019 ⁎ Corresponding author. E-mail address: rescandon@us.es (R. Escandón). Applied Thermal Engineering 150 (2019) 492–505 Available online 08 January 2019 1359-4311/ © 2019 Elsevier Ltd. All rights reserved. T