The invention discloses a selection method for the economic goal-optimized self-adaptive backoff of a chemical process predictive control system, which comprises the following steps: determining an economic goal of an enterprise production unit, establishing a variable steady-state optimization constraint relationship of a control unit, determining production variable operating restrictions, obtaining an LQG performance index based on the relationship between control and output, adjusting weight to construct a discrete point set based on the relationship between control and output variance, fitting a nonlinear variance adjusting reference, solving an optimization proposition to determine optimal set values and optimal backoff values for control and output, and automatically refreshing thesettings of the predictive control device based on changes of production. The method can avoid the blindness of the experience set value of the enterprise production device predictive control goal, the conservatism of the manual backoff value setting and the imperfection and the infeasibility of the present MVC reference theoretical method in the physical production process, becomes a self-adaptive optimization and backoff strategy which can be practically put into service in the chemical process, raises the production yield, saves energy and reduces emission and improves economic benefits ofthe enterprise to a greater extent.