The aim of the work is to study various issues of the theory and practice of using neural network technologies in agricultural production, as well as to develop and implement a neural network model for predicting crop yields (using the example of winter wheat yields). This article is about promising areas of application of neural networks in agriculture such as pattern recognition and their classification, diagnostics, clustering, forecasting, monitoring by using machine vision, optimization and optimal control, robotics. The problems of using neural networks to predict crop yields were considered. Factors of crop productivity were analyzed. A neural network was built to predict crop yields (in the case of winter wheat). In this case, the following stages of constructing a neural network model were implemented: determining the architecture of the neural network, its software implementation using the PyTorch framework, Pandas and Matplotlib libraries, and interpretation of the results obtained using MS Excel tools. The calculated value of the average absolute percentage error of the MAPE forecast for the training set was 1.93%, for the test set it was 2.17%, which indicates a high level of model approximation. The greatest correlation with the yield of winter wheat has such parameters as the maximum soil moisture during the formation of the flag leaf is Pearson correlation coefficient 0.776, the maximum soil moisture during the heading period is 0.775, the amount of precipitation is 0.772. It is noted that the problem of comprehensive digitalization of agro-industrial production is currently extremely relevant, which makes neural network modeling very popular in terms of its tasks and goals
Keywords
forecasting, agriculture, modeling, prospects, neural networks, Neural networks, PyTorch