Hyper-Heuristic Approaches for Combinatorial Optimization

Combinatorial optimisation problems arise in numerous domains, from logistics and scheduling to network design and resource allocation. Traditional metaheuristics are often tailored to a single problem type, requiring significant expertise to tune and adapt. Hyper-heuristics represent a higher level of abstraction, aiming to automate the selection or generation of heuristics that operate on a given problem. By orchestrating collections of low-level heuristics according to contextual information and learned patterns, hyper-heuristics seek robust performance across diverse instance sets without human intervention. Two main paradigms have emerged: selection hyper-heuristics, which choose from predefined heuristics at each step, and generation hyper-heuristics, which construct novel heuristics from components. Recent advances incorporate machine learning to capture search behaviour, transfer knowledge between problem domains and introduce additional layers of generalisation. These developments promise more adaptive and maintainable frameworks for tackling real-world combinatorial challenges such as vehicle routing, timetabling and resource scheduling.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Innovations in apprenticeship learning have been applied to selection hyper-heuristics for vehicle routing problems by observing expert search behaviour. A time-delay neural network records decisions made by multiple human-designed hyper-heuristics on training instances and automates the creation of a selection mechanism. Empirical tests on open vehicle routing benchmarks demonstrate that the machine-designed hyper-heuristic matches or outperforms existing expert systems, highlighting the potential of offline learning for generalisation across instance sizes.

Parameter tuning remains a key challenge for cross-domain optimisation. Recent work applies an automated racing procedure to tune both the parameters of a memetic algorithm and its embedded low-level heuristics across nine distinct single-objective problems. The adaptive racing strategy identifies configurations that deliver consistently strong performance, surpassing conventional manual or fixed tuning methods and illustrating how systematic, data-driven strategies can harmonise search behaviour across problem types.

To enhance flexibility and reusability, a multi-layer hyper-heuristic model has been proposed for job shop scheduling. Instead of combining low-level heuristics directly, this “squared” hyper-heuristic employs a second-level hyper-heuristic to select among primary hyper-heuristics, thereby adding an abstraction layer. Under varied testing scenarios, the model not only outperforms single-layer frameworks on diverse scheduling instances but also allows practitioners to adjust model complexity by adding or removing layers, laying the groundwork for transfer learning in combinatorial search.

Hyper-Heuristic Approaches for Combinatorial Optimization publication trend

The graph below shows the total number of articles in hyper-heuristic approaches for combinatorial optimization across all publications each year (not limited to Nature Index journals).

Technical terms

Combinatorial optimisation: the process of finding an optimal object from a finite set of objects, often under discrete constraints.

Hyper-heuristic: a high-level method that selects, generates or adapts heuristics to solve optimisation problems broadly.

Selection hyper-heuristic: a hyper-heuristic variant that chooses one of several predefined low-level heuristics at each iteration.

Low-level heuristic: a problem-specific operator that modifies a candidate solution during search.

Move acceptance criterion: a rule determining whether a newly generated solution is retained in the search process.

Memetic algorithm: an evolutionary approach combining population-based search with local refinement strategies.

Cross-domain optimisation: the application of a search method across multiple problem types rather than specialising for a single domain.

References

1. Constructing selection hyper-heuristics for open vehicle routing with time delay neural networks using multiple experts. Knowledge-Based Systems (2024).
2. An investigation of F-Race training strategies for cross domain optimisation with memetic algorithms. Information Sciences (2023).
3. Hyper-heuristics: A survey and taxonomy. Computers & Industrial Engineering (2024).
4. Beyond Hyper-Heuristics: A Squared Hyper-Heuristic Model for Solving Job Shop Scheduling Problems. IEEE Access (2022).