Mining high average-utility itemsets is one of the hotspots in the current research. Aiming at the problem that the high average-utility itemsets mining algorithm generates a large number of meaningless candidate itemsets, which results in high memory consumption and long running time, the dMHAUI algorithm is proposed. Firstly, the algorithm defines the integration matrix Q, and proposes four compact average-utility upper bounds based on vertical database representation and three effective pru- ning strategies. Secondly, the information needed for high average-utility itemsets mining is stored in the IDUL structure tree, and the improved diffset technique is used to quickly calculate the average- utility and upper bound of itemsets. Finally, the high average-utility itemsets are obtained by recursively calling the search algorithm. Simulation results show that the dMHAUI function has better performance than the EHAUPM algorithm and the MHAI algorithm in terms of running time, join operation number and scalability.