Artifical Fishswarm Algorithm is an optimization algorithm to achieve the global optimal result by imitates fishswarm's behavior such as prey, swarm and follow. The optimization problem exists in applying neural networks to coupling the regress of projection pursuit, and using AFSA to optimize it in the learning process so as to attain the optimal projection direction, threshold and the Hermite orthogonal polynomial coefficients.The paper describes the AFS algorithm for coupling the regress model of neural networks and projection pursuit. The simulation results indicate that this algorithm gives a satisfactory forecast effect.