This paper proposes a flexible new method to improve the accuracy of SMA(shape memory alloy) self-sensing control, especially in low load, and broaden the application condition of it. SMA is a kind of intelligent material, it can be acted as an actuator(have big output force ratio) and a sensor at the same time in control system. For these advantages, the SMA has a lot of applications. Since the control model of SMA varies with the change of control variables, the accuracy of the self-sensing control is not high enough, especially under low pretightening. In this paper, we constructed the control model of SMA by using two base models to work out the changeable control model. Compared with traditional control method which use only the base model under certain pretightening force, the proposed new method is easier to apply and has less limit on pretightening force, and the accuracy also improved a lot. The SMA actuator can be applied onto a lot of micro mechanism, such as artificial muscle, micro wrist, micro finger, so this improvement have significance in practical use.