The invention relates to a modeling method for a rural power grid distribution transform weight overload short-term pre-warning model. The modeling method comprises the following steps: collecting original data of relevant variables required by modeling; designing and calculating characteristic variables of the pre-warning model, and making comparison on a recall ratio and a precision ratio, so asto obtain an optimal variable combination; after screening, carrying out regression fitting by virtue of logistic again, and optimizing, so as to obtain the pre-warning model; substituting actual data into the pre-warning model, predicting so as to obtain distribution transform weight overload probability lists, and carrying out sequencing; carrying out dynamic modeling on a change relation between new data distribution and temperature in future so as to establish a loess model, adjusting a probability value produced through logistic regression, so as to acquire a reliable prediction result;selecting the corresponding number of probability lists as weight overload probability lists in the current week; and carrying out prediction verification on the pre-warning model established throughhistorical data by virtue of data in the current year, directly utilizing the pre-warning result if the pre-warning result is good, and otherwise, optimizing the pre-warning model. According to the modeling method, the accurate distribution transform weight overload short-term pre-warning model can be established.