PGU-SGP: A Pheno-Geno Unified Surrogate Genetic Programming For Real-life Container Terminal Truck Scheduling

Abstract: Data-driven genetic programming (GP) has proven highly effective in solving combinatorial optimization problems under dynamic and uncertain environments. A central challenge lies in fast fitness evaluations on large training datasets, especially for complex real-world problems involving time-consuming simulations. Surrogate models, like phenotypic characterization (PC)-based K-nearest neighbors (KNN), have been applied to reduce computational cost. However, the PC-based similarity measure is confined to behavioral characteristics, overlooking genotypic differences, which can limit surrogate quality and impair performance. To address these issues, this paper proposes a pheno-geno unified surrogate GP algorithm, PGU-SGP, integrating phenotypic and genotypic characterization (GC) to enhance surrogate sample selection and fitness prediction. A novel unified similarity metric combining PC and GC distances is proposed, along with an effective and efficient GC representation. Experimental results of a real-life vehicle scheduling problem demonstrate that PGU-SGP reduces training time by approximately 76% while achieving comparable performance to traditional GP. With the same training time, PGU-SGP significantly outperforms traditional GP and the state-of-the-art algorithm on most datasets. Additionally, PGU-SGP shows faster convergence and improved surrogate quality by maintaining accurate fitness rankings and appropriate selection pressure, further validating its effectiveness.

Pheno-Geno Unified Surrogate Genetic Programming for Container Terminal Truck Scheduling

The paper introduces a novel approach to dynamic container terminal truck scheduling (DCTTS) through an advanced genetic programming (GP) algorithm termed Pheno-Geno Unified Surrogate Genetic Programming (PGU-SGP). This method is aimed at enhancing the efficiency and effectiveness of truck scheduling through an innovative surrogate model that integrates both phenotypic and genotypic characterizations of GP individuals. The focus is on improving the selection of surrogate samples and the accuracy of fitness predictions.

PGU-SGP addresses one of the primary bottlenecks of conventional data-driven GP approaches: the intensive computational demand of fitness evaluations needed for complex, real-world optimization problems. Existing methods often rely heavily on phenotypic characterization (PC) for estimating fitness, which can overlook vital genotypic differences among individuals. This oversight can degrade surrogate quality and, subsequently, algorithm performance due to insufficient capture of the search space.

Key Contributions and Methodology

The notable contributions of this research are threefold. Firstly, it proposes an innovative unified similarity metric that combines both PC and genotypic characterization (GC) — termed PGU distance — to guide the selection of surrogate samples. This dual focus ensures a comprehensive representation of individual similarities by capturing both behavioral and genetic features. Secondly, an effective method for representing GC is introduced, emphasizing node frequency to reflect genetic material distribution across individuals. This representation aids in providing a more nuanced understanding of genotypic diversity. Lastly, the empirical evaluation of the PGU-SGP algorithm demonstrates a substantial reduction in training time, approximating 76%, compared to traditional GP, while achieving comparable or superior performance within the same computational budget.

Empirical Findings

Quantitative analysis on real-life DCTTS scenarios reveals that with an equivalent training duration, PGU-SGP significantly outperforms both traditional GP and the state-of-the-art SGP_PC on most dataset instances. This is largely attributed to PGU-SGP's ability to maintain consistent fitness rankings and appropriate selection pressure through its improved surrogate model.

The algorithm was evaluated using four datasets, each consisting of distinct loading ratios and truck distributions. Results consistently showed that pursuant to the incorporation of GC into the selection mechanism, PGU-SGP presented faster convergence rates and superior quality of evolved heuristics. This validates the efficacy of integrating phenotypic and genotypic data in predicting the fitness landscape more accurately.

Implications and Future Directions

PGU-SGP offers substantial implications for real-world applications in logistics and operations research, particularly those characterized by dynamic and uncertain environments like DCTTS. The method’s scalability and ability to efficiently handle complex evaluations position it as a versatile tool for various combinatorial optimization problems. Furthermore, the research opens avenues for integrating adaptive mechanisms that dynamically balance the phenotypic and genotypic weights during evolution, potentially leading to even more refined optimization results.

In conclusion, by leveraging the combined strengths of phenotypic and genotypic data, this research paves a promising path for future developments in genetic programming and related fields, underscoring the value of holistic similarity metrics in surrogate-assisted evolutionary algorithms. As AI continues to evolve, the refinement of such surrogate models is anticipated to play a pivotal role in enhancing the practicality and applicability of automated scheduling systems in complex, dynamic sectors.