Aiming at the shortcomings of low precision in high dimension, slow convergence speed and precocity in the basic firefly algorithm, the firefly algorithm with oscillation, constraint and natural selection mechanism is proposed. Firstly, the second-order oscillation factor is introduced to balance the influence of the previous generation of individuals on the current generation of individuals, so as to prevent the individuals of fireflies from falling into local extremum. Then, the constraint factor based on sigmoid function is added to dynamically adjust the moving distance of individual, so as to avoid the situation that the individual of firefly caused by excessive disturbance near the theoretical optimal value leads to the reduction of precision in the later stage of the algorithm. Finally, natural selection based on the decreasing trend of reciprocal Gaussian integral is used to keep individual diversity and accelerate the convergence speed of the algorithm. The convergence and time complexity of the improved algorithm are proved by theoretical analysis. Through the simulation of 10 standard functions with multiple dimensions and different characteristics, the test results show that the optimization accuracy and convergence speed of the OCSFA(natural selection firefly optimization algorithm with oscillation and constraint) are obviously improved. Especially in the case of high dimension, the theoretical optimum can still be found for almost all functions, which better solves the problem of the unsuitability of the firefly algorithm for high dimension solution.