In recent years, large-scale numerical simulations played an essential role in estimating the effects of explosion events in urban environments, for the purpose of ensuring the security and safety of cities. Such simulations are computationally expensive and, often, the time taken for one single computation is large and does not permit parametric studies. The aim of this project is to facilitate real-time and multi-query calculations by developing a non-intrusive Reduced Order scheme. Reduced Order Models (ROM) allow us to reduce the computational time employed by a single simulation by reducing the dimension of the system. The scheme hereby proposed is based on a combination of traditional methods, such as the Proper Orthogonal Decomposition (POD), and deep learning techniques. In the case of blast waves, the parametrised PDEs are time-dependent and non-linear and represent a transient and fast event. For such problems, the Proper Orthogonal Decomposition (POD), which relies on a linear superposition of modes, cannot approximate the solutions efficiently. Therefore, we add a nonlinear reduction step based on Autoencoders (AE), which are a type of artificial neural network. The efficacy of this method is shown in an example consisting of an explosion happening in the vicinity of a building. We show that the deep-learning based ROM introduced can reconstruct the solutions efficiently and performs better than the traditional POD method. In cooperation with Joint Research Centre of European Commission, Ispra, Italy.