List of Abstracts 2023: SESSION D - AI-BASED LOGISTICS APPLICATIONS

In domestic road operations, the demand for increased efficiency and performance is on the rise. To enhance operational efficiency and minimize costs, it becomes crucial to generate dynamic pricing for routes through digital media. The primary objective, in line with our developed model, is to boost the profit margin by estimating freight prices on our digital platform based on factual data. These estimates are then transmitted to the operational units to reduce current freight costs. Our digital platform serves as a nexus for suppliers and customers in land transportation. This is especially critical in freight transportation, where ensuring economical and reliable cargo finding processes, matching cargo owners with suitable carriers, and swiftly generating accurate pricing for routes is paramount. We've established a digital learning system for freight pricing estimation using machine learning techniques and a mathematical model. We evaluated 41 regression models to select the artificial intelligence models for our prediction algorithm. Four regression models that demonstrated the best performance were chosen to build our artificial intelligence system. Acknowledging the combination algorithm developed by our company, the system operates within a continuous learning cycle and selects the most effective artificial intelligence models at specific intervals. Additionally, the mathematical model's purpose is to complement the results of artificial intelligence by using actual prices from similar routes to prevent potential errors. The degree to which the mathematical model refines the artificial intelligence's results is determined by our developed decision-making intelligent algorithm. The hybrid model's results were tested on real expeditions, confirming its success.

Logistic process automation is a crucial factor driving the rapid evolution of modern business and logistics. Large Language Models (LLM) are becoming more widespread, and logistics systems are essential. The presentation's main objective is to assess how well LLM models are used in logistics, especially when acquiring documents like waybills. The overview will evaluate these models' efficiency and adaptability while highlighting their potential for streamlining logistics procedures. Will also take into account how advanced LLM models might improve the acquisition process of documents, particularly in the case of waybills, by enhancing the effectiveness and accuracy of logistical data management. With a concentration on the use of LLM, the research will offer perceptions into recent developments and innovative solutions in logistic process automation comparative analyses using comparative analyses.

In the rapidly evolving landscape of logistics and warehousing, efficient inventory management is one of the most important issues. Forklifts, as indispensable tools in these environments, play an essential role, especially in the movement of goods. This paper introduces an innovative application of the You Only Look Once (YOLO) deep learning algorithm in the realm of visual warehouse management, specifically tailored for forklift operations. YOLO, as a a state-of-the-art object detection algorithm applied to warehouse environment monitoring can detect, classify, and track various items and obstacles in real-time. Our work presents the results of field trials and demonstrates the substantial improvements in warehouse management achieved through the implementation of YOLO-based visual perception of forklifts. The application not only streamlines inventory control but also enhances safety, reduces operational costs, and ensures compliance with industry regulations. The proposed innovative approach offers a holistic solution for real-time inventory monitoring, safety enhancement, and operational efficiency. It represents a significant stride towards the realization of smart, autonomous warehouses, promising to revolutionize the industry.

This paper presents a multi-pallet loading problem. It is assumed that products are heterogenous, and pallet loading units can be stacked creating sandwich loading units (SNDW). The research task is to support the optimal composition of products on pallet units using the available space at the maximum level and at the same time to fulfill the constraints, including the maximum weight, height and stability of multi-pallet unit. The aim of the research is to create the minimum number of SNDW under the specific constraints, requirements for multi-pallets, products and a distribution process. The paper discusses the procedure for creating SNDW with the application of simulation-based approach enriched with an original knowledge exploration procedure using Dominance-based Rough Sets Theory (DRST) and resulted rules generation. A simulation environment was applied to verify the proposed approach, and its implementation was carried out in a distribution company from the FMCG industry. The computational experiments were carried out on a set of 700 picking orders. On that basis a set of decision rules was created and used to evaluate new picking orders resulting in 83% effectiveness of their classification for potential improvement within multi-pallet loading process.

Turfgrass management represents a critical aspect of landscaping, sports fields, and golf course maintenance. The efficient logistics of turfgrass, encompassing tasks such as mowing, fertilization, and irrigation, can significantly impact its health and aesthetics. In recent years, remote sensing technologies and machine learning have emerged as powerful tools for optimizing turfgrass logistics. This work presents a comprehensive study on the application of machine learning algorithms to enhance the management of turfgrass through remote sensing data. The proposed approach leverages various remote sensing techniques, including satellite imagery, unmanned aerial vehicles (UAVs), and ground-based sensors, to gather high-resolution data on turfgrass health, moisture levels, and growth patterns. These data sources feed into a machine learning pipeline, comprising data preprocessing, feature engineering, and algorithm selection, to develop predictive models for turfgrasses. Our findings demonstrate that machine learning models, when trained on remote sensing data, can accurately predict turfgrass parameters that can be used for ensuring continuous improvement in turfgrass logistics. The integration of machine learning into turfgrass logistics not only enhances resource utilization but also reduces environmental impact by minimizing unnecessary inputs. We present case studies from various landscapes, including sports fields and golf courses showcasing the practicality and adaptability of our approach.