This invention relates to the field of hot rolling technology, and in particular to a method and apparatus for self-learning correction of a finishing mill model. The method includes: determining whether the actual rolling force of the finishing mill is effective; if not, acquiring the rolling force setting value, the rolling force measurement value, and the rolling force recalculation value of the finishing mill; determining whether the rolling force setting value, the rolling force measurement value, and the rolling force recalculation value meet preset conditions; if so, updating the self-learning of the finishing mill model, thereby effectively avoiding the situation where the loop tension in the rolling head range between the stands of the finishing mill is too large and the actual rolling force detected in the corresponding area is too small, resulting in the self-learning result of the finishing mill model being opposite to the expected direction, thus improving rolling stability.