Purpose Integrating production planning on unrelated parallel machines with the vehicle routing problem in modern logistics operations presents significant challenges. This study aims to address this integration to optimize operational efficiency and meet customer demands. Design/methodology/approach This study presents a comprehensive solution framework that combines exact and heuristic methods. First, it formulates a mixed integer programming (MIP) model and develop two tailored heuristics – a constructive heuristic and a neighborhood search heuristic – to generate high-quality solutions. Next, the study proposes a machine learning-based selection mechanism that predicts, for each problem instance, which heuristic will perform best. Finally, the study introduces two novel metaheuristics: a randomized variable neighborhood descent (RVND) algorithm and a hybrid framework (PPO-VND) that integrates variable neighborhood descent (VND) with the proximal policy optimization (PPO) reinforcement learning algorithm. Together, these contributions offer a robust, adaptive toolkit for solving complex instances across diverse settings. Findings Comparative tests demonstrate the superiority of our proposed methods over existing approaches, confirming their effectiveness in addressing the critical challenges posed by integrated production planning and vehicle routing in logistics operations. Research limitations/implications The proposed model was tested on controlled instances with limited problem sizes, which may not fully capture the complexity of large-scale real-world scenarios. Moreover, the performance of the machine learning framework depends on the quality and diversity of the training data. Future research could explore alternative production configurations, different routing policies and more dynamic environments, as well as assess scalability under realistic uncertainties such as demand fluctuations and stochastic delivery times. Practical implications This study offers a practical and adaptable approach for companies operating in complex logistics environments, enabling more efficient integration of production and distribution decisions. By applying artificial intelligence (AI) and hybrid algorithms, managers can obtain high-quality solutions in less time, reducing delays and operational costs. The model’s flexibility allows its application across various sectors – such as manufacturing, retail and transportation – supporting more agile and synchronized supply chains, particularly in contexts that demand rapid and customized customer responses. Social implications Improved logistics integration can positively impact society by reducing waste, optimizing resource use and enhancing delivery punctuality, benefiting both end customers and suppliers. On a broader scale, adopting intelligent solutions in supply chains may promote more sustainable and responsible practices. Furthermore, the diffusion of AI-based technologies can encourage professional upskilling, fostering new job opportunities in sectors focused on innovation and digital transformation. Originality/value This study highlights the key role of AI in logistics optimization, promoting adaptability and resilience in dynamic operational environments. These innovations significantly improve solution quality and computational efficiency.
Araujo et al. (Sat,) studied this question.