Simulator

Environment
Contains multiple Agents (Agent 1 ... Agent n)

Cooperation Module

Agent 1 ... Agent n
Inputs: state, observation
Outputs: action
Rewards

Reinforcement Learning Module
Online Network & Target Network
Prioritized Experience Replay
Knowledge-Integration Module
Double DQN: Q(s,a;θ)
Loss Function & Soft Update

Double-Adaptive VNS (DA-VNS)
Adaptive Shaking
Adaptive VND

Final Solution

Initial Solution

Explainable AI (XAI) Layer
• Attention visualization (MHA)
• Feature importance (SHAP/LIME)
• Operator contribution (DA-VNS)

Fig. 2. General architecture of the proposed framework integrating MHA-enhanced MARL, Double-Adaptive VNS (DA-VNS), and an Explainable AI layer for DMDEVRPTW.