In this paper, data mining based on the log of active decision engine is introduced to find the CUBE using pattern of analysis rules, which can be used as important reference information for materialized views selection. Based on it, a 3A probability model is designed, and the greedy algorithm, called PGreedy (probability greedy), is proposed, which takes into account the probability distribution of CUBE. Also view keeping rule is adopted to achieve better performance for dynamic view adjusting. Experimental results show that PGreedy algorithm can achieve better performance than BPUS (benefit per unit space) algorithm in real-time active data warehouses environment.