Autoencoder based Consensus Mechanism for Blockchain-enabled Industrial Internet of Things

Abstract

Conventional blockchain technologies developed for cryptocurrency applications involve complex consensus algorithms which are not suitable for resource constrained Internet of Things (IoT) devices. Therefore, several lightweight consensus mechanisms that are suitable for IoT devices have been proposed in recent studies. However, these lightweight consensus mechanisms do not verify the originality of the data generated by the IoT devices, so false and anomalous data may pass through and be stored in the ledger for further analysis. In this work to address the data originality verification problem, we propose an autoencoder (AE)-integrated Chaincode (CC)-based consensus mechanism in which the AE differentiates normal data from anomalous data. The AE is invoked through the CC once a transaction is initiated; the result returned from the AE to the CC is stored in the ledger. We have conducted a case study to train and test the AE model on the IoTID20 dataset. Also, Minifabric (MF) is used to implement the CC and illustrate the CC operation that stores only original IoT data. Moreover, the performance has been shown for the CC in terms of latency and throughput.