A Novel Immune Algorithm for Multiparty Multiobjective Optimization
A new 'immune algorithm' outperforms existing methods for complex problems with competing objectives, like drone path planning.
A team of researchers led by Kesheng Chen and Wenjian Luo has introduced a novel AI optimization algorithm designed for a critical but underexplored class of problems: Multiparty Multiobjective Optimization Problems (MPMOPs). Published in IEEE Transactions on Emerging Topics in Computational Intelligence, their Multiparty Immune Algorithm (MPIA) addresses scenarios where multiple decision-makers (DMs)—such as different departments, stakeholders, or autonomous agents—have distinct and often competing objectives. Traditional multiobjective algorithms seek a single compromise, but MPIA is engineered to find a solution set that satisfies each party's unique Pareto front as closely as possible, a far more complex challenge.
The core innovation of MPIA lies in two key strategies. First, an 'inter-party guided crossover' uses non-dominated sorting ranks from different DM perspectives to guide genetic operations, maintaining population diversity across all parties. Second, an 'adaptive activation strategy' uses a novel 'multiparty cover metric' (MCM) to intelligently select which individuals proceed to the next generation. This biologically-inspired 'immune' approach allows the algorithm to dynamically focus its search power where it's needed most.
In rigorous testing, MPIA was benchmarked against both standard multiobjective evolutionary algorithms (MOEAs) and cutting-edge multiparty-specific algorithms. The tests included synthetic problems and a real-world 'biparty unmanned aerial vehicle path planning' scenario, where two entities have different optimal route priorities. The experimental results, detailed in the paper, demonstrate that MPIA consistently outperforms these existing state-of-the-art methods, proving its superior capability in navigating the trade-offs inherent in multi-stakeholder optimization.
- Solves Multiparty Multiobjective Optimization Problems (MPMOPs), where multiple decision-makers have competing goals, outperforming prior state-of-the-art algorithms.
- Uses a novel 'immune' approach with inter-party guided crossover and an adaptive activation strategy based on a new multiparty cover metric (MCM).
- Validated on real-world biparty UAV path planning, showing practical utility for logistics, robotics, and resource allocation systems with conflicting stakeholder needs.
Why It Matters
This enables more effective AI for real-world systems where multiple agents or stakeholders—like in supply chains or drone fleets—must negotiate optimal but conflicting outcomes.