This paper presents a control-oriented reduced-order building model intended for optimal management of space-heating demand in district heating systems. The model is designed in accordance with the conclusions drawn from a preliminary study, also presented hereby, evaluating the impact of internal mass and the heating circuit inertia on short-term buildings thermal dynamics. The model parameters identification is carried-out using a hybrid Particle Swarm Optimization-Hooke-Jeeves search algorithm, for three buildings of different energy classes. The algorithm searches for an optimal set of parameters that minimizes the error between the model's output and historical data generated using a higher order building simulator written in Modelica language. Data used for the identification process is exclusively non-intrusive to the building indoors, and relies solely on measurements available at the substation level. We assess the quality of the identifications per building type. Results showed that the identified models have high prediction ability of the building indoor temperature dynamics, with a maximum absolute error less than 0.7°C. Implementation of optimal predictive space-heating control based on the identified models is in prospect.