Translation. Region: Russian Federal
Source: State University of Management – Official website of the State –
At the VII All-Russian scientific and practical conference “Digital transformation of management: problems and solutions” held in April at the State University of Management, young scientists from the Center for Management of Engineering Projects of the State University of Management presented a promising development of an innovative hybrid decision support system (DSS) in logistics.
GUU postgraduate students Nikita Akinshin and Vladimir Kutkov drew attention to the lack of efficiency of DSS used in logistics and developed their own solution that combines the power of a cascade of specialized machine learning (ML) models with the interpretive capabilities of large language models (LLM).
Modern logistics is characterized by huge volumes of heterogeneous data, high demand uncertainty, and the need to coordinate multiple participants in real time. Traditional analytical tools are ineffective in highly dynamic situations and are based on static models. The results of analysis of advanced machine learning models are difficult to interpret and require highly qualified employees. Large language models are incapable of accurate calculations, lack industry logic, and can make unreliable conclusions.
The key element of the new decision support system (DSS) is a multi-level architecture that combines all the capabilities of new technologies. This structure is implemented for the first time, although its individual components are already being used in real market conditions.
At the first level, a cascade of several specialized ML models analyzes operational, logistics and economic data – from demand and arrival time forecasting to route optimization and cost assessment.
At the second level, a meta-model is connected – a kind of analytical brain of the system, which collects the conclusions of all ML components obtained at the first level, analyzes the relationships and dependencies, identifies bottlenecks, assesses risks and forms a complex request (prompt) for the LLM model.
At the third level of LLM, having received this “smart” prompt and interacting with infographics on BI platforms to obtain visualizations, synthesizes a deep, yet human-readable analytical report.
As a result, information graphs are displayed on the screen with an assessment of the current state of affairs for the task under consideration and options for increasing the efficiency of its solution.
This approach allows companies to quickly obtain a comprehensive picture, and employees to understand complex dependencies without having to delve into the technical details of how ML algorithms work and make timely, informed management decisions.
“The meta-model is the highlight of our development. It acts as an experienced logistics analyst who first understands the situation, identifies all the interrelations, and only then formulates the task for LLM so that it can generate a truly useful, meaningful report for the employee,” explains Nikita Akinshin.
It is also important that the new hybrid decision support model can perform tasks at all management levels, i.e. take on the roles of different employees of logistics companies. Mechanics and suppliers will receive reports on the technical condition of the transport fleet, middle managers will be able to build optimal routes, and managers will be able to more accurately forecast annual revenue.
“After receiving information from the DSS, further decisions will be made by employees, while the system’s analytics are advisory in nature. But in the near future, from 3 to 5 years, the decision-making process may also become automated. It is only necessary to settle moral and ethical issues and regulate the legal framework,” says Vladimir Kutkov.
The development of the employees of the engineering center of the State University of Management can be used not only in logistics, but also adapted for other sectors of the economy.
A scientific publication with a detailed description of the development is currently being prepared for release.
Please note: This information is raw content directly from the source of the information. It is exactly what the source states and does not reflect the position of MIL-OSI or its clients.