A Novel Consensus Algorithm Based on Segmented DAG and BP Neural Network for Consortium Blockchain

Currently, because of the excellent properties of decentralization, hard tamperability, and traceability, blockchain is widely used in WSN and IoT applications. In particular, consortium blockchain plays a fundamental role in the practical application environment, but consensus algorithm is always a key constraint. Over the past decade, we have been witnessing the obvious growth in blockchain consensus algorithms. However, in the existing consortium blockchain consensus algorithms, there is a limited characteristic of scalability, concurrency, and security. To address this problem, this work introduces a new consensus algorithm that is derived from a directed acyclic graph and backpropagation neural network. First, we propose a partitioned structure and segmented directed acyclic graph as data storage structure, which allows us to improve scalability, throughput, and fine-grained granularity of transaction data. Furthermore, in order to provide the accuracy of node credit evaluation and reduce the possibility of Byzantine nodes, we introduce a novel credit evaluation mechanism based on a backpropagation neural network. Finally, we design a resistant double-spending mechanism based on MapReduce, which ensures the transaction data are globally unique and ordered. Experimental results and security analysis demonstrate that the proposed algorithm has advantages in throughput. Compared with the existing methods, it has higher security and scalability.