Addressing the time-sensitive and multi-constrained characteristics of multi-Unmanned Ground Vehicle (UGV) firefighting task allocation problem in forest fires, this paper constructs a multi-objective optimization model with time-varying task states. To tackle challenges such as large discrete decision space, complex constraints, and dynamic fire changes inherent in this problem model, an improved Non-dominated Sorting Genetic Algorithm II (NSGA-II) is proposed. The algorithm employs integer encoding to match tasks with vehicles, and introduces dynamic crowding distance sorting and an adaptive mutation strategy to enhance solution convergence and distribution diversity. Simulation experiments in various scale scenarios demonstrate that the proposed algorithm outperforms comparative algorithms, in terms of superiority, diversity, and stability. It can significantly shorten the firefighting cycle and reduce resource consumption, providing an intelligent multi-UGV task allocation solution for mitigating forest fire losses, thus validating its practical value in forest fire suppression scenarios.