Blockchain-Enabled Parallel Learning in Industrial Edge-Cloud Network: a Fuzzy DPoSt-PBFT Approach

Abstract

Recently, parallel reinforcement learning (PRL) based Industrial Internet of Things (IIoT) edge-cloud resource scheduling has elicited escalating attention. However, with the scale of IIoT expands, there are several challenges in the existing researches: 1) large number of parallel servers slows down the convergence rate of PRL; 2) malicious parallel server affects resource allocation efficiency. In order to solve the above efficiency and security problem, blockchain-based approaches are introduced in PRL based resource allocation problem. However, traditional consensus algorithm in blockchain is not suitable for resource allocation and is inefficient. Thus, in this article, based on a novel fuzzy delegated proof of state and practical byzantine fault tolerance (fuzzy DPoSt+PBFT) consensus algorithm, we propose a blockchain-enabled collaborative parallel Q-learning (CPQL) approach to address the above challenges. To be specific, we first construct an edge-cloud collaborative architecture for executing the diversity intelligence IIoT applications. Then, we propose a CPQL algorithm for edge-cloud resource allocation and choosing the optimal number of parallel edge servers to speed up the Q-table training. In the Q-table aggregation process in CPQL, a fuzzy DPoSt+PBFT algorithm is designed for secure CPQL training and efficient consensus. Experimental results show the superior performance of the proposed approach. And the proposed approach has great potential in IIoT resource allocation problem.