Aiming at the characteristics of small sample and multiple variation trends of inertial platform error coefficient, an improved GM (1, 1) model is proposed. The model uses the improved background value and operator transformation of original data. The error caused by the “straight instead of curving” is eliminated by the method of background value reconstruction. Operator transformation method can weaken the randomness of the original data and enhance the monotonicity of it. Therefore, the improved GM (1, 1) model has a wider scope of application and higher forecast accuracy. Instance analysis result shows that: Compared with the traditional models, the improved model has higher accuracy of forecast and better fitting degree, which is more accurate to reflect the variation trend of inertial platform error coefficient.