First，a definition of the residual capacity of the battery with variable current and its modified formula under the actual working condition are proposed in this paper． Next，the power curve of the battery of a parallel hybrid excavator is scaled down to the range of a single battery，and the scaled curve is taken as the input signals of the tested battery． Then，the data samples of the residual capacity are obtained from the input signals，and they are analyzed by using several grey prediction algorithms． On this basis，a fragmentation prediction scheme is proposed，which adopts the direct grey prediction model of the cumulative-average operator at the charge stage and the grey prediction model of the variable weight buffer factor at the discharge stage． The results show that the proposed method can fully and effectively improve the accuracy of estimating the residual capacity of the battery in hybrid devices．