Evolutionary multitasking for bidirectional adaptive codec: A case study on vehicle routing problem with time windows

Abstract

The vehicle routing problem (VRP), an NP-hard problem of considerable complexity, pertains to the determination of optimal sequences of customer visits in an effort to minimize the total operational cost of vehicles. Conventional approaches to tackling VRP have primarily focused on optimizing each task independently, utilizing mathematical methodologies or evolutionary algorithms. Nevertheless, research exploring the application of the advanced concept of evolutionary multitasking (EMT) to address VRP remains limited. To enrich the body of knowledge in this field, our study presents an innovative EMT algorithm, dubbed EMT-RD. This algorithm employs a bidirectional adaptive codec to facilitate the transfer of information between real and discrete solutions. The EMT-RD algorithm permits the transfer of real and discrete solutions across varying tasks, utilizing an encoding rule that transforms information from real to discrete solutions for each task. Further, a decoding rule from discrete to real solutions generates search experiences, facilitating the transfer of knowledge between real solutions. The mechanism is capped off with an adaptive transfer strategy that instigates transfer operations across different tasks. Each task is tackled using a single-task optimization algorithm that employs an artificial bee colony algorithm. Through comprehensive experimentation, we found that our proposed EMT-RD algorithm outperforms competitors on 168 VRP with time windows multitasking instance pairs and four real-world problem pairs that were developed for this study. These results underscore the efficacy of the adaptive transfer strategy and bidirectional adaptive codec in addressing inter-task correlations when solving the VRP.