Adaptive Real-Time Optimization for Dynamic Decision Environments: An Actor-Based Modular Framework for Continuous Operational Tuning

This paper presents a novel operational management framework designed for dynamic, real-time decision environments where static optimization models fail to integrate with live data streams and stakeholder feedback. We introduce an actor-based large neighborhood search (A-LNS) algorithm that enables continuous, responsive optimization by embedding a message-passing architecture directly into the metaheuristic loop. The approach allows for seamless parameter and constraint adjustments mid-execution critical for settings involving negotiation, evolving capacities, and shifting priorities. Using a multi-compartment multi-knapsack problem as an illustrative case, we demonstrate how the framework adapts to real-time changes in inclusion constraints, knapsack capacities, and item weights. Results show that A-LNS maintains solution feasibility and near-optimality without restarting, reducing convergence time and supporting interactive decision-making. The proposed architecture provides a foundation for building modular, scalable optimization systems that can be integrated with dynamic data sources and human-in-the-loop processes.