Fireworks Algorithm (FWA) is a recently contributed heuristic optimization method that has shown a promising performance in applications stemming from different domains. Improvements to the original algorithm have been designed and tested in the related literature. Nonetheless, in most of such previous works FWA has been tested with standard test functions, hence its performance when applied to real application cases has been scarcely assessed. In this manuscript a mechanism for accelerating the convergence of this meta-heuristic is proposed based on observed wind inertia dynamics (WID) among fireworks in practice. The resulting enhanced algorithm will be described algorithmically and evaluated in terms of convergence speed by means of test functions. As an additional novel contribution of this work FWA and FWA-WID are used in a practical application where such heuristics are used as wrappers for optimizing the parameters of a road traffic short-term predictive model. The exhaustive performance analysis of the FWA and FWA-ID in this practical setup has revealed that the relatively high computational complexity of this solver with respect to other heuristics makes it critical to speed up their convergence (specially in cases with a costly fitness evaluation as the one tackled in this work), observation that buttresses the utility of the proposed modifications to the naive FWA solver.