Main Title: Adaptive Cognitive Q-Learning–Based Security Model for Blockchain Protocols: Simulation and Experimental Evaluation on a Private Ethereum Network.

Abstract

The rapid growth of blockchain technologies has enabled decentralized applications based on smart contracts and distributed consensus. However, the increasing number of attacks exploiting protocol logic and network dynamics highlights the limitations of traditional, static security mechanisms. This study proposes an adaptive cognitive security model based on a Q-learning agent to enhance the protection of blockchain protocols. The agent is designed to analyze transaction behavior, assess risk levels, and dynamically select appropriate countermeasures. The proposed approach is evaluated through a dual experimental framework combining large-scale simulation using SimPy and execution on a private blockchain environment implemented with Ganache. Experimental results show a detection rate of approximately 70%, no observed false positives, a response time close to one second, and a very low operational gas cost. These results demonstrate that reinforcement learning can effectively improve the adaptability and responsiveness of blockchain security mechanisms while preserving network performance and economic viability. The study confirms the potential of cognitive and adaptive approaches for building more resilient and autonomous blockchain security systems.