Neural Network Technologies in Predicting the Operating Status of Agricultural Enterprises

Abstract

All agricultural facilities in Russia are currently going through digital transformation. However, the process needs a unified approach for the entire agricultural sector. Neural network methods have already proved extremely effective in various areas of IT. The authors used neural networks to analyze statistic data and assess the performance of agricultural infrastructure. This study involved technical data from the production cycle of agro-industrial enterprises, namely packaging and greenhouses. The data obtained were analyzed using artificial neural networks. The procedure included identifying a set of factors that described an agro-industrial complex or some of its properties that corresponded to a specific task. These data were used in planning and making managerial decisions. The program identified five factors that described the state of an agricultural enterprise. These factors were used to build a model while its elements served as output data for the neural network. The model calculated the future state of the object. Trials were run on a limited data set on a multilayer perceptron. The neural network showed reliable results for a small data set. The root mean square error of was 0.1216; the mean modulus deviation was 0.0911. In this research, modern neural network technologies demonstrated good prospects for the domestic agro-industrial complex as a method of control, management, and dispatching. However, specific operational patterns require further studies.