Association analysis for Binary time series is a difficult problem, because most of association rules lay emphasis on the relation among the items, but ignore the temporal correlation in the transaction database. Therefore, a new improved algorithm of mining association rules for binary time series is presented to make use of both the relationship among the items and the temporality of association. By using the proposed algorithm, the binary data is converted to common numerical value for representing the time-value implicitly, then clustering algorithm is combined with the association analysis, which improves the supports of most association rules. Several indicators are used to evaluate the results from the proposed algorithm. The experimental results on the prognosis dataset of stroke show the effectiveness of the proposed method.