Researchers at the Tokyo University of Science have developed a two-layer optimization framework, termed CSPSO, that integrates chaotic search (CS) with particle swarm optimization (PSO) to solve complex logistics and scheduling problems. By using PSO to dynamically tune the control parameters of chaotic search, the method improves both the stability and solution quality of algorithms used for combinatorial optimization.
Addressing Instability in Chaotic Search Algorithms
Combinatorial optimization problems, such as logistics, scheduling, and network design, require identifying the best option from a massive set of discrete possibilities. As problem sizes grow, the number of potential solutions increases exponentially, rendering exhaustive searches impossible.
Chaotic search (CS) is a common heuristic used to navigate these search spaces. Unlike stochastic methods, CS relies on chaotic dynamics—deterministic but highly irregular trajectories—to explore the solution space thoroughly. This allows the algorithm to avoid getting trapped in local solutions. However, CS is sensitive to its initial control parameters. According to a study published in the journal Nonlinear Theory and Its Applications, IEICE (NOLTA) on July 1, 2026, even minor mismatches in these parameters can cause unstable search behavior.
The CSPSO Two-Layer Framework
To mitigate these instabilities, the research team—led by Professor Tohru Ikeguchi of the Tokyo University of Science—introduced a learning-based adaptive tuning method. This approach, known as CSPSO, creates a two-layer hierarchy:
- The Outer Layer (PSO): A swarm of particles is initialized, where each particle represents a candidate parameter vector for the chaotic search. The PSO algorithm evaluates the fitness of these parameters based on the results of the inner layer.
- The Inner Layer (CS): The chaotic search performs the optimization task using the parameters provided by the PSO layer.
This iterative process allows the system to adaptively tune chaotic excitation. By delegating the parameter-tuning burden to the swarm-based learning layer, the method reduces the need for manual calibration, which is often a significant bottleneck in applying chaotic neural networks to real-world problems.
Performance in Logistics Optimization
The researchers tested the CSPSO framework against the capacitated vehicle routing problem (CVRP), a logistics optimization problem where a fleet must meet customer demand while adhering to strict vehicle capacity limits.
The findings indicate that CSPSO provides higher robustness and superior solution quality compared to conventional CS and previous parameter-tuning approaches like CST, which relied on uniform updates based on global statistics. While the CSPSO method requires more computational time than CST, the researchers note that it offers a more practical, automated alternative to the difficult task of manually configuring chaotic neural networks.
This adaptive approach is expected to improve efficiency in fields ranging from factory production planning to complex information technology network design.
Keep reading