Goal-Conditioned Resource Allocation With Hierarchical Offloading and Reliable Consensus for Blockchain-Based Industrial Digital Twins

Abstract

In the current technological landscape, digital twins (DTs) are critical enablers for enhancing communication efficiency, data processing and on-line monitoring with virtual copies in industry network environments. However, heterogeneous devices and sensitive data breaches intensify challenges in security and management. Rapidly changing business requirements further exacerbate these issues, as traditional algorithms struggle to adapt to dynamic industrial demands. Simultaneously, overloaded edge servers, ultra-reliable low latency communications (URLLC), and limited resources make real-time decision-making even more difficult. Hence, we propose a hierarchical offloading and resource allocation framework for blockchain-based industrial D2D DT (OR-BIDT), which addresses these challenges by providing offloading and allocation strategies that protect data privacy and reliable communication. Then, we propose an R-DPoS consensus mechanism that optimizes node selection by introducing a voting mechanism with transmission reliability and computation frequency to improve the security of block verification. For problems requiring optimization over a goal space rather than the simple linear weighted sum in OR-BIDT, we design a goal-conditioned reinforcement learning (GCRL) approach with locality sensitive hashing-based experience replay (LSHER) to accomplish efficient experience returns. Simulations show that the critical and actor networks of our proposed algorithm converge 71.43% and 14.29% faster than the benchmark method, respectively.