This article proposes an improved k-means clustering-based co-evolutionary genetic algorithm, which preserves the individuals closest to the center in each cluster, performs traditional genetic algorithm operations (selection, crossover, mutation) on the original population, selects the optimal subset of individuals from the new population generated by the genetic algorithm, merges the individuals retained by K-means with the optimal individuals generated by the genetic algorithm, and forms a new generation population. This hybrid algorithm combines the global search capability of K-means with the local fine search capability of genetic algorithm. Finally, this article uses the Alpine and other function to test and analyze the optimization of the final algorithm. The improved algorithm can quickly jump out of local optima and converge to the global optimum.