Abstract: Confronted by the present state that the rules for hot metal temperature forecasting are made merely on the base of the experience of blast furnace (BF) experts, a new approach to the rules established through association rules mining from BF data was put forward. The algorithm of multidimensional time series rules mining was improved. The improved algorithm, which bases on the fuzziness of both subsequence and suhsequence interval, avoids the influence of "time border sharpness" on the result of mining. The algorithm was applied to No. 1 BF at Wuhan Iron and Steel Group Corporation, and its effects turned out to be satisfactory.