Optimizing distributed resource allocation using GAG-ASGRU in blockchain-enabled federated reinforcement learning systems

Abstract Resource Allocation (RA) among devices is important as cloud computing rapidly expands, requiring many devices to share their resources. Prior models use various approaches for RA but become less effective as traffic demand increases. This paper proposes Blockchain (BC)-Centric Federated Reinforcement Learning (FRL) for RA, utilizing Galois AMSGrad-ArcTan Softplus Gated Recurrent Units (GAG-ASGRU). First, mobile nodes are implemented. Then, sensed data are sent to the RA phase. Next, network attributes are extracted from data requests. The Crossover Mutation Entropy-based Artificial Gorilla Troop Optimization Algorithm (CME-AGTOA) is used for load balancing. From a dataset containing historical transmission data, features are extracted and reduced using Exponential Fisher’s Linear Discriminant Analysis (E-FLDA). The reduced features are used to train GAG-ASGRU. The trained system tests incoming data requests to identify suitable resources for allocation. If additional response time is required, the request must be authenticated and transmitted over a global network to ensure a prompt response. Authentication uses an Interpolation-centric Algorithm for Secure Hashing 512 (I-ASH512) to generate a Hashcode (HC) and ensure BC verification. Finally, RA is performed by evaluating the HC value with the BC server. The proposed system achieves 98% RA and an Energy Consumption (EC) of 1024 mJ, demonstrating its reliability and better performance compared to previous techniques.