Abstract: The establishment and adjustment of the judgment matrix is very important in Analytic Hierarchy Process（AHP） algorithm. The paper establishes judgment matrix using the three-scale method after improving Satty’s 1 to 9 scaling method. Considering the reciprocal matrix necessary and sufficient conditions， the paper proposes the adjustment method when judgment matrix is inconsistent to improve the consistency of judgment matrix and reduce the amount of computation. When dealing with the load balancing of Web cluster system， the method can be used to determine the weights of the four major categories of parameters affecting the server load. On the basis of the original data, the following experiments show that the improved algorithm of AHP is more efficient when judgment matrix is inconsistent.

Key words: Judgment matrix, 0~2 scales, optimization and adjustment of consistency, Analytic Hierarchy Process（AHP）, load balancing

摘要： 判断矩阵的建立和调整是AHP算法的关键。对Satty的1~9标度方法进行改进，引入3标度法建立判断矩阵。从正互反矩阵充分必要条件出发，提出判断矩阵不一致时的调整方法，以提高判断矩阵的一致性和减少计算量。该方法应用于Web集群系统负载均衡中，确定影响服务器负载的四大类参数的权重。实验表明，改进的AHP算法在判断矩阵不一致时，在尊重原始数据的基础上，调整判断矩阵的计算量减少，判断矩阵的一致性更好，同时可以提高负载均衡系统的性能。

关键词: 判断矩阵, 0~2标度, 一致性优化调整, 层次分析法, 负载均衡