A Machine-Learning Based Approach for Data-Driven Identification of Heating Dynamics of Buildings' Living-Spaces

Modeling the heating dynamics of a given living-space of a real building remains a challenging engineering-science problem because of the quite large number of diverse kinds of involved parameters and their usually nonlinear interdependency. However, the need of such living-spaces' heating dynamics modeling appears as a foremost requirement for designing adaptive controllers scheming the complex behaviors of nowadays' smart buildings. In this context and through considering the above-mentioned complex dynamics' modeling within the slant of “Time-Series Prediction” paradigm, in this paper we propose a Machine-Learning-based data-driven approach for overcoming difficulties inherent to the aforementioned challenging engineering-science problem. The proposed approach takes advantage from the nonlinear autoregressive exogenous (NARX) model's capabilities in time-series' forecasting and the Multi-Layer Perceptron's (MLP) learning and generalization skills. The proposed approach has been applied for living-space heating dynamics identification of a fully automated four-floor real building located at Senart Campus of University Paris-Est Creteil (UPEC). The obtained results assessing the accuracy of the investigated Machine-Learning-based approach are reported and discussed.