부산대학교 도서관

The traditional Raft algorithm has such an issue as "vote snatching" among candidates that results in abnormal elections in the absence of a majority. To address this issue, a random forest-based method is introduced for the identification of authorized nodes. First, the attributes of authorization are introduced to the original Raft algorithm to be the label of authorized nodes, i.e., only authorized nodes are eligible for candidacy. Second, the random forest algorithm is introduced to classify the authorized nodes according to their term, broadcasting duration, timeout duration, and other characteristics to divide all nodes into preferred nodes and ordinary nodes. In the end, the main consensus group makes up of preferred nodes adopted integral accumulation to top out the preferred node with the highest integral as the authorized nodes. According to the results, the optimized Raft algorithm avoids abnormal elections and outperforms the original algorithm in election efficiency. The method offers an orderly mechanism of node operation for the Consortium Blockchain.