The characterisation of parameters of building energy models based on in-situ sensor information is generally performed after the measurement period, using all data in a single batch. Alternatively, on-line parameter estimation proposes updating a model every time a new data point is available: this establishes a direct link between external events, such as the weather, and the identifiability of parameters. The present study uses the Sequential Monte Carlo method to train a lumped building energy model (RC model), and thus estimate a Heat Loss Coefficient, and other parameters, sequentially. Results show the direct impact of solicitations (solar irradiance and indoor heat input) on this estimation, in real time. The method is validated by comparing its results with the Metropolis-Hastings algorithm for off-line parameter estimation. (C) 2019 Elsevier B.V. All rights reserved.