The performance of the well-known particle swarm optimization (PSO) method can be improved by minimizing the possibility of premature convergence in a local minimum. We can achieve this by modifying some of the particles with crossover and mutation operators used in genetic algorithms. However, the impact of genetic operators on the optimization process should depend on the current state of the PSO algorithm. In this article, we propose to use the neuro-fuzzy system to dynamically determine the strength with which these operators will affect the process of finding the optimal solution. Results obtained for well-known benchmark functions demonstrate the advance of the proposed method over the original PSO algorithm and its selected modifications.