Fedchain: Decentralized Federated Learning and Blockchain-Assisted System for Sustainable Irrigation

Abstract

The conventional Internet of Things-based soil moisture monitoring system for irrigation decision making suffers from huge network traffic, high latency and energy consumption, and compromise in data security. To overcome these challenges, this paper proposes a blockchain-assisted decentralized federated learning strategy, Fedchain for irrigation decision making based on edge-cloud computing. A peer-to-peer network is formed among the edge servers. The edge servers have their local datasets, which are analyzed locally using Long Short-Term Memory (LSTM) network, and the model parameters are exchanged between the peer nodes. The model is updated accordingly. For data security purposes, blockchain is used. The LSTM model updates are recorded in the InterPlanetary File System by the local user, and a unique Content Identifier is generated for data retrieval, and it is stored in the blockchain as a transaction. A Distributed Hash Table in the file system maps the Content Identifier in the blockchain to the stored data in the file system, ensuring effective retrieval. The results show that Fedchain achieves above 99% prediction accuracy, and reduces latency and energy consumption by ~78% than the edge-cloud framework without federated learning. The use of blockchain reduces the mining cost by ~78% than the competing methods.