The invention discloses a cold-rolled sheet shape control self-learning method based on an optimum algorithm, which includes the steps: by means of gradually and iteratively computing through iterative formulas to approximate an optimum value, enabling an L1 CPU (central processing unit) sheet shape control self-learning program in online application to timely obtain a precise learning sample; and optimizing efficiency factors Pj<u+1> [i] of all plate shape control mechanisms including a back-up roll inclining mechanism, an intermediate roll bending mechanism, a working roll bending mechanism and an intermediate roll traversing mechanism of a rolling mill according to the learning sample, wherein the efficiency factors gradually tend to the optimum value along with increase of self-learning times, and in the self-learning process, the efficiency factors do not increase sharply, therefore computing precision of action regulating quantity of all the sheet shape control mechanisms is improved. By the aid of the cold-rolled sheet shape control self-learning method, optimized schedule of the efficiency factors of all the sheet shape control mechanisms is accelerated, and the advantage of the method on improving the computing precision of the efficiency factors of all the sheet shape control mechanisms is given full play, so that the sheet shape control program is benefited to improve the computing precision of the action regulating quantity of all the sheet shape control mechanisms in real time.