A Multi-Agent Collaborative Dynamic Election Method for Blockchain Consensus Nodes

Abstract

Fault detection in mainstream blockchain consensus protocols such as Practical Byzantine Fault Tolerance (PBFT), Raft, and Delegated Proof of Stake (DPoS) still relies on heartbeat timeouts, a passive paradigm that reacts only after a node has fallen silent. Acting earlier requires the degrading node itself to raise the alarm, which opens an attack surface the timeout paradigm does not have: a declaration can be fabricated. This paper proposes a multi-agent collaborative dynamic election method that closes that surface with an unforgeable physical-evidence gate and builds a proactive pipeline of trend detection, evidence-based validation, and pre-triggered re-election upon it. A claim is admitted only when neighbors confirm the Receive Error Counter growth recorded by their own controllers and the accused node's own Transmit Error Counter is independently elevated, the latter being what identifies which node caused the errors. The method is evaluated over a Controller Area Network (CAN) bus under bursty, distance-coupled, and duty-cycled electromagnetic interference. A validator relying on bus-wide evidence alone admits a false accusation in 100% of trials even with no adversary present, whereas the proposed gate holds that rate at zero across the whole tested Byzantine range and refuses a fabricated self-report in every attempt. Acting on a measured error-accumulation trend rather than on silence, the method detects in 1287 ms and recovers in 1689 ms, roughly three-quarters below PBFT and one-third below Raft, the strongest baseline, and it precedes the Bus-off failure point in 96 of 100 trials. A cooling observation period limits role oscillation, the whole pipeline costs three orders of magnitude less diagnostic bandwidth than centralized polling, and a four-way comparison attributes the refusal to the corroboration requirement rather than to the layer split.