An Evaluation of a NN-Based Model for the Prediction of Foundation Heat Transfer From Basements
Abstract In this paper, a feed-forward artificial neural network module is presented to predict seasonal variations of foundation heat transfer from conditioned basements. The training data for the NN-based module were obtained from a detailed solution of the ground-coupled problem. Input variables for the NN module include foundation geometric dimensions, insulation configuration, indoor and outdoor temperatures, and soil thermal properties. The paper discusses the network architecture and the training and testing procedures. The predictions of the NN-based module are compared to a correlation-based method for a set of basement configurations. The main conclusion of the paper is that NNs can predict seasonal variation of building foundation heat transfer with high accuracy and little effort for model development.