The invention provides an artificial pancreas self-adaption and active-disturbance-rejection controller based on a blood glucose variation trend. The controller comprises a tracking differentiator module, an extended state observer module, a non-linear feedback module and a constrained module. A non-linear feedback model of the non-linear feedback module is u=(-fhan(k1e1, k2e2, r2, a)-z3) / b0, wherein k1, k2 and a are blood glucose variation trend self-adaption parameters, r2 is named as controlled quantity gain, e1 and e2 are a signal of an error between a blood glucose concentration set valueand a blood glucose concentration estimated value and a signal of an error between a blood glucose concentration variation rate set value and a blood glucose concentration variation rate estimated value respectively, z3 is a total disturbance estimated value, and b0 is a known gain coefficient. According to the controller, all uncertain factor actions on a controlled object come down to unknown disturbances, and the disturbances are estimated by using input and output data of the object and compensated. Therefore, the controller algorithm has certain robustness to the disturbances, such as personal parameter inaccuracy, model uncertainty, eating disturbance and before-meal dose inaccuracy.