A unique multi-gradient identification approach is suggested to identify the FIR model. Traditional gradient algorithm has less computation cost but converges slowly. The new approach, which uses stacked gradients rather than the original single gradient, incorporates the concept of multi-gradient to speed things up. New difficulties arise when figuring out how long multi-gradient algorithms should stack. The strong Wolfe condition is used to overcome this problem. The stacking length selected by the proposed method can guarantee that the cost function declines at a reasonable pace. The suggested algorithm was validated by numerical experiments.