HYBRID NEURAL NETWORK MODEL FOR OPTIMIZATION OF CONFECTIONERY PRODUCTION PARAMETERS ON A DIGITAL PLATFORM

Abstract

This article examines the problem of low operational efficiency in confectionery enterprises due to manual planning and unstable raw material properties. Rapid technological advances currently require new approaches to solving problems at various levels. Confectionery production, despite the specific nature of the production process and the specific ingredients used, requires a modernized approach to the supply chain, processing data on raw materials and the final product itself. This area of activity also requires digitalization, particularly in terms of data collection and processing, analytical materials, and business analysis and forecasting. Manual assessment and analysis are no longer sufficient, as today's demands accelerated processes, more flexible forecasting options, and the development of optimistic scenarios. Digitalization of processes and the use of neural networks offer the opportunity to significantly accelerate data processing, perform calculations, predict developments, and streamline management decisions, all within the context of lean management. The development of a hybrid neural network model integrated into a digital platform, enabling real-time optimization of confectionery production parameters, is a pressing scientific and practical challenge.

The authors propose a mathematical formulation of the optimization problem. They propose a hybrid neural network model integrating a convolutional neural network for product defect recognition and a recurrent network for predicting process parameters. The developed model is implemented within a digital platform combining data collection, forecasting, and optimization modules. The results of a pilot implementation at a confectionery factory are presented, confirming a 15–20 % reduction in defects and a 12–18 % reduction in equipment downtime.