Introduction. The study is motivated by the need to justify the efficiency of localizing the production of high-protein dairy products capable of withstanding a single exposure to sub-zero temperatures during transport to the consumer.

Purpose. To develop a balanced mathematical model of raw material flows for the production of cryostable curd cheese from milk, enabling the scientifically grounded selection of formulation and processing parameters for production in the sub-Arctic regions of the Russian Far East.

Materials and Methods. The study employs classical material balance equations for milk separation and established patterns of protein transfer during curd production. An invariant mathematical model was developed and implemented as software in the Wolfram Mathematica 10.2 environment using principle of mass conservation, Raoult's second law (with the van't Hoff correction), linear programming, the Ostwald - de Waele power law, and the Atwater system.

Results. Two scenarios for utilizing the cream obtained during separation were analyzed: partial (Scenario A) and complete (Scenario B) utilization. Variables included the proportion of curd in the binary "curd–cream" system (T = 65…85 %) and the cream fat content ( =10…85 %). Non-linear relationships were established regarding the impact of factors T and  on the yields of curd cheese, whey, and secondary cream; on protein and fat mass fractions; and on the product's cryoscopic depression and dynamic viscosity. A fundamental transition boundary between the scenarios was identified (T = 65 % при = 44,48 %), along with a terminal region where solutions for Scenario B do not exist without the addition of milk protein concentrate.

Conclusion. The proportion of curd in the binary mixture is the key limiting factor of the technology, whereas cream fat content acts as a supplementary factor. The developed model and Powersoftware enable the targeted design of the formulation and production parameters for curd cheese with specified yield, composition, and cryostability.

Abstract. Due to the marked increase in consumer interest in wines with a protected geographical indication PGI and a protected designation of origin PDO, the development of reliable methods for confirming their geographical origin is becoming particularly relevant. Despite the widespread use of complex isotope ratio analysis, elemental profile, and machine learning methods in world practice, there are no scientifically sound classification models for Russian wines that take into account the characteristics of the country's main grape-growing areas. 

Purpose. To establish the possibility of classifying samples of domestic wines by their geographical origin based on a comprehensive analysis of the isotopic ratios of carbon, oxygen and hydrogen of ethanol and the elemental profile using machine learning methods. 

Materials and Methods. 63 samples of wines produced in the main wine-growing areas of Russia were selected. Of these, 20 samples are from the Krasnodar Territory, 32 samples from the Crimea, 7 samples from Dagestan, 4 samples from the Don Valley. The isotopic ratios δ13C, δ18O, and δD of the isolated ethanol were measured in these samples, as well as 71 mass concentrations of macro-, microelements, and rare earth metals. The Random Forest model was used in the study, precision, recall, and f1-score were used as metrics, and stratified k-fold cross-validation with 100 repetitions was used to obtain confidence intervals for metrics. 

Results. After training and validating the model, its average accuracy was 0.92 for Dagestan, 0.99 for the Don Valley, 0.82 for the Krasnodar Territory and 0.90 for the Crimea. The most important parameters for the model were the isotope ratios of the elements in the composition of ethanol molecules, as well as the mass concentrations of Li, B, Cs, Sr, W, Ni, and Sn. Among rare earth metals, the concentrations of Dy, Lu, Yb, Tm, Ho, Y, and La were the most important for the model. 

Conclusion. The results obtained show the prospects of applying machine learning methods to the tasks of authenticating domestic wine products. This model is the basis for classifying wines produced in more limited geographical areas, including wine-growing areas and specific terroirs.