The ensemble prediction system (EPS) is an approach employed in meteorology to estimate forecast uncertainty of dynamical systems. In EPS, an ensemble of auxiliary simulations is launched along with the main prediction. Recently, an application with the EPS framework was proposed as a method that enables algorithmic tuning of parameters of large-scale models, in cases where high-CPU demands make usual iterative optimization impractical. The approach was aimed and tested for operational numerical weather prediction models, with a relatively small number of parameters and well-tuned initial values. Here, we present a new version of the approach as a general-purpose parameter estimation method for situations where effective parallel computing is available, but high-CPU requirements exclude the use of standard sequential approaches. We treat the problem as a stochastic optimization task and employ an evolutionary approach, the differential evolution as the optimizer. We demonstrate improved convergence properties, especially for strongly biased initial values or higher number of parameters. For parametric uncertainty quantification, the approach can be considered as a heuristic sampler of the parameter distributions.