Creation of neuron network productivity of lucerne in Steppe zone of Ukraine

Cite this article: Zaporozhchenko, V. Y., Shepel, A. V., & Tkachuk, A. V. (2019). Creation of neuron network productivity of lucerne in Steppe zone of Ukraine. Agrology, 2(1), 47‒50. doi: 10.32819/2617-6106.2018.14017 Abstract. In arid conditions of the Steppe zone of Ukraine for obtaining stable yields of lucerne and observance the conditions of resource-saving, it is important to know from what factors the value of the yield of lucerne depends on. According to the results of the conducted research, an agroecological model of the productivity of growing crop on irrigated lands of the Ukrainian Steppe has been formed. For the carrying out research, the method of artificial neuron networks was used. Creating an agroecological model of lucerne production using neuron networks consists of the following phases: search of data; preparation and normalization of data; choice of type of neuron network; experimental choice of network characteristics; experimental choice of parameters; obtaining an artificial neuron network for modeling the productivity of lucerne; checking of adequacy of the model; adjustment of parameters, final training. As a result of the research it was found that artificial neuron networks are fundamentally different from traditional methods of statistical data analysis. In the capacity of main elements of the system are taken: the sum of effective temperatures above +5 °С; amount of atmospheric precipitation; solar lighting duration; irrigation norms; depth of soil tillage; fertilization and plant protection. The article presents a constructed neuron network with architectural parameters. It has been established that among the significant number of natural and agrotechnical factors affecting the productivity of crops of lucerne, the greatest influence is carried out by atmospheric precipitation and, in our case, water-saving irrigation norms. Among the investigated factors there are a high degree of pair and multiple correlations. It is proved that the components of architecture contain different compositions of multilayered perceptrons, radial-basic functions, and also linear components.