Considering the problem of the low inference accuracy of the extended Belief Rule Base (BRB) which was proposed by Liu, etc (LIU J, MARTINEZ L, CALZADA A, et al. A novel belief rule base representation, generation and its inference methodology. Knowledge-Based Systems, 2013, 53: 129-141), an improved method of rule-base construction and inference was proposed. This approach was based on the method of Liu's rule-base construction, and a new generation method of rule antecedents and a new calculation method of rule weights were provided. Subsequently, in order to avoid activating so many unnecessary rules, the 80/20 rule was introduced to improve the strategy of rule activation. Then an integrated construction and inference methodology of belief rule-base was formed. Finally, in order to validate the accuracy and efficiency of the new approach, the case study in pipeline leak detection was provided. The experimental results show that the proposed approach not only can keep lower time-consumption, but also can make the Mean Absolute Error (MAE) of system be reduced to 0.17342. This proves that the new approach has high accuracy and efficiency.