Analysis of the possibilities of self-adaptive neural networks with search behavior in agro-ecological systems
Аналіз можливостей нейронних мереж, що самоадаптуються, з пошуковою поведінкою в агроеко-логічних системах. release_fmrbc6sohnc4tci2t7mq5eig6e

by Brovarets Oleksandr, Kyiv Cooperative Institute of Business and Law, Yuiry Chovnyuk, National University of Life and Environmental Sciences of Ukraine

Published in Mathematical machines and systems by Institute of Mathematical Machines and Systems Problems of the NAS of Ukraine.

2020   p125-133

Abstract

The complexity of environmental problems facing modern science in connection with the deterioration of the environmental situation on the planet and the growing dynamics of ongoing processes are con-stantly growing (this primarily concerns agro-ecological systems). At the same time, the flexibility and accuracy of ecological models created by traditional mathematical methods, as well as the speed of their construction, in practice often does not live up to expectations. The same can be said about the engi-neering management of agro-ecological facilities. Among the most flexible and effective ways to solve such problems, neural network models and neurocomputers are highlighted. However, the concepts un-derlying the construction of modern neural network training algorithms impose serious limitations on the potential range of application of neuroinformatics achievements in solving environmental problems of agricultural systems. The number of scientific publications with radically new results is steadily decreasing and existing developments are beginning to "spread" across applications. The manifestation of such trends indicates that the main potential of the ideas that caused the next progress in this most important bionic direction and the creation of the 6th generation of computers – neurocomputers are exhausted. Modern advances in neuroinformatics, based on the use of supervisor algorithms, are mainly associated with the possibility of using hidden layers of neurons (not connected to the input and output), which provided high adaptive capabilities of neural networks, and universality based on the ability to train a neural network to solve a precisely posed problem. In this paper, the main limitations inherent in modern approaches to limiting neural networks are indicated, and the concep of constructing a new type of training neural network and network algorithms is proposed. Some non-traditional oppor-tunities provided by the proposed concept are described. The conceptual foundations of the development of the neuroinformation system for controlling the electrotechnical complex of the information and technical system for local operational monitoring are proposed.
In application/xml+jats format

Archived Files and Locations

application/pdf  533.7 kB
file_snacvidu3nbw3ggsb4xecuhivq
www.immsp.kiev.ua (publisher)
web.archive.org (webarchive)
Read Archived PDF
Preserved and Accessible
Type  article-journal
Stage   published
Year   2020
Journal Metadata
Not in DOAJ
Not in Keepers Registry
ISSN-L:  1028-9763
Work Entity
access all versions, variants, and formats of this works (eg, pre-prints)
Catalog Record
Revision: 828a27b2-68bf-46e8-9c47-cb3dd9809be2
API URL: JSON