DQMMBSC: design of an augmented deep Q-learning model for mining optimisation in IIoT via hybrid-bioinspired blockchain shards and contextual consensus

Single-chained blockchains are highly secure but cannot be scaled to larger IIoT (Internet of Industrial Things) network scenarios due to storage costs, communication costs, verification complexity, and delay needed for reading and archiving the chains. To overcome these issues, a wide variety of sharding models are proposed by researchers, but most of these models either generate a larger number of shards, compromise on quality of service (QoS), or reduce security under a larger number of attacks. These issues were minimised in this text via design of an augmented deep Q-learning model for mining optimisation via hybrid-bioinspired blockchain shards and contextual consensus. The model initially collects minimal IIoT information sets from the network-under-deployment, and uses them to select miner nodes. This is done by enforcing two novel consensus models, which are proof-of-spatial trust (PoST) and proof-of-temporal trust (PoTT), that enable the network to identify miner nodes with good spatial and temporal efficiency levels. The selected miner nodes are reconfigured via a bacterial foraging Optimisation (BFO) process, which assists in selecting communication-specific encryption and hashing techniques for security of underlying blocks. The secured blocks are then stored on a set of spatially optimised blockchain shards. Shard configurations are decided by a deep Q-learning-based genetic algorithm (DQGA), which assists in deciding optimal shard length and merging temporally unused shards. These decisions allow the proposed model to reduce mining delay by up to 8.3% when compared with existing sharding models. The proposed DQMMBSC model was also tested under Finney, Man-in-the-Middle, and 51% attacks, using differential attack patterns. It was observed that the proposed model showcased 3.5% lower delay, 4.9% lower energy consumption, 6.5% higher throughput, and 1.9% higher packet delivery ratio even under large-scale attacks, thereby making the model useful for heterogeneous network scenarios.