| 摘要 |
<p>The present invention provides a rolling model optimization device capable of connecting predicted values of adjacent models without gradient, comprising: a boundary value table for storing boundary values representing boundaries of a plurality of models; a model formula connection boundary value change part for changing the boundary values stored in the boundary value table according to a rolled actual value; a model formula priority table for storing priority of the plurality of models; an evaluation part for determining the model suitable for a rolling requirement based on the boundary values obtained from the boundary value table and the priority obtained from the model formula priority table in the case of each input calculation condition; a model switching part for switching the model to the determined model; a learning calculation part for calculating a learning coefficient based on a difference between the calculated predicted value and the actual value for each model in the plurality of models; and a model calculation part for calculating the predicted value of the model by utilizing a model formula of the model switched by the model switching part and the learning coefficient calculated by the learning calculation part.</p> |
