AN INTELLIGENT SYSTEM FOR DISH-LEVEL DIET PLANNING BASED ON AN OPTIMIZATION MODEL

Abstract

The paper addresses the problem of developing intelligent systems for personalized nutrition planning. Modern research in this field demonstrates a transition from classical formal diet models to hybrid architectures that combine two methodological paradigms: knowledge-driven and data-driven approaches. However, there is some methodological gap between them. Knowledge-driven models provide mathematical rigor and guarantee the satisfaction of nutritional and resource constraints, but they are usually limited in adaptability and personalization. In contrast, data-driven approaches, including modern generative models, demonstrate high flexibility and the ability to incorporate behavioral data, yet they do not provide formal guarantees of optimality and constraint satisfaction. This contradiction motivates the development of an integrated intelligent nutrition planning system that combines the advantages of both approaches. The objective of this study is to develop an intelligent dish-level nutrition planning system whose core is a formalized multicriteria diet optimization model. Unlike the classical diet problem, where optimization is performed over individual food products, the proposed approach models nutrition at the level of complete dishes, which improves the practical feasibility, interpretability, and usability of the resulting dietary plans. The mathematical model is formulated as a multicriteria optimization problem in which the decision variables represent the number of dish portions, while constraints reflect nutritional, energetic, logical, and temporal requirements. The proposed model is implemented within a multilayer system architecture consisting of a data layer, an optimization core, an intelligent decision-support layer, and a user interaction layer. The optimization core ensures mathematical correctness and computes optimal solutions, whereas the intelligent layer provides adaptation, personalization, and interpretation of results. The model is further extended to a dynamic form using a rolling planning horizon, allowing the diet plan to be updated as new data and user preferences become available. Computational experiments have demonstrated that changes in criterion weights lead to transitions between several stable optimal meal structures, reflecting the discrete nature of the considered multicriteria optimization problem.