The article presents an agent-based demographic model of Russia designed to run on supercomputers. The technologies used in the model allow researchers to create an artificial society with the number of agents up to 109 and effectively parallelize the work of the simulator. The software package designed to implement the model combines separate subsystems written in programming languages of different levels. On the one hand, this provides effective load balancing between computing processes and messaging between agents (implemented in C++), and on the other hand, this simplifies the development of model blocks that implement the simulation of demographic processes (implemented in C#). The demographic processes in the model are simulated based on the actions of individual agents, taking into account their family ties, which they maintain by exchanging messages. Key features of the demographic agent-based models are the following: a) dynamic change in the size and composition of populations of agents – removal of part of the agents (their “death”) and the emergence of new ones (“birth”); and b) separation of actions performed at the simulation step in stages, each of which can cause the revision of the general settings that are specific to regions or groups of agents, and/or exchange of messages between agents.
In the course of computer experiments, the model has been tested on real data and has shown good results at testing for the following parameters: a) the quality of recreating the age-sex structure of the population for the country as a whole and in the regions with the use of the population of agents; b) the stability of the model and a low margin of error of the results of forecasting the main demographic indicators in comparison with the variants of Rosstat’s official forecast; c) efficiency of parallelization of the program code when running on supercomputers. The model is the basic one for an integrated regional simulation model that is currently being developed; however, the model can be useful as an independent forecasting tool
Keywords
agent-based modeling, simulation of demographic processes, supercomputer technologies, application of metis graph decomposition, demographic forecast for russia