Журнал «Современная Наука»

Russian (CIS)English (United Kingdom)
 MOSCOW +7(495)-142-86-81

DEEP LEARNING IN NATURAL INFORMATION INTERPRETATION AND AGGREGATION TASKS

Koshelev Nikita Dmitrievich  (postgraduate student of the department "Design and Production of Radio equipment" Penza State University )

Novikov Kirill Sergeevich  (postgraduate student of the department "Design and Production of Radio equipment" Penza State University )

Yurkov Nikolay Kondratievich  (Doctor of Technical Sciences, Head of the Department "Design and Production of Radio equipment" in Penza State University )

Background. The increasing number of natural disasters related to climate change requires the development of modern risk assessment methods based on integrated environmental information. Traditional approaches to analyzing climatic and geophysical data (temperature, precipitation, seismic activity, etc.) often fail to identify hidden nonlinear dependencies between heterogeneous parameters. Deep learning provides new opportunities for automatically extracting complex spatiotemporal features from large amounts of data, but deep neural network models are "black boxes" and difficult to interpret. This paper proposes a research approach combining deep neural network architectures (convolutional neural networks and transformers) with a dynamic fuzzy DENFIS inference system for interpreting and aggregating natural information. Materials and methods. The task of integrating heterogeneous time series of natural indicators to assess the risk of natural events is formalized. A hybrid model has been developed where deep neural networks identify hidden patterns in data, and DENFIS aggregates them through adaptive fuzzy rules. Mathematical models of the components used are presented, including the transformer self-attention mechanism and fuzzy neural inference. Results. A series of computational experiments based on monitoring data of climatic indicators (precipitation, soil moisture, etc.) and geophysical measurements was carried out. The stages of data preprocessing, architecture and model training, evaluation criteria (accuracy, completeness, F1-measure) and visual tools for analyzing the results (error convergence graphs, classification error matrix, etc.) are described. Results. It is shown that the proposed hybrid model is more accurate (up to ~95%) than individual deep learning methods (~92%) and classical algorithms (up to ~80%). As part of the ensemble, DENFIS improves the identification of boundary cases and ensures interpretability of solutions by generating understandable rules (for example, if precipitation levels are high and the soil is saturated with moisture, then the risk of flooding is high). Conclusions. Combining deep neural network methods with fuzzy logic makes it possible to achieve high accuracy in predicting natural disaster risks while obtaining explicable models, which confirms the prospects of the proposed approach for early warning and decision support systems.

Keywords:deep learning, data interpretation, data aggregation, climate data, geophysical data, neuro-fuzzy systems, DENFIS, convolutional neural networks, transformers, risk assessment, natural disasters.
 

Read the full article …

Citation link:
Koshelev N. D., Novikov K. S., Yurkov N. K. DEEP LEARNING IN NATURAL INFORMATION INTERPRETATION AND AGGREGATION TASKS // Современная наука: актуальные проблемы теории и практики. Серия: Естественные и Технические Науки. -2025. -№06/2. -С. 105-113 DOI 10.37882/2223-2966.2025.06-2.16
LEGAL INFORMATION:
Reproduction of materials is permitted only for non-commercial purposes with reference to the original publication. Protected by the laws of the Russian Federation. Any violations of the law are prosecuted.
© ООО "Научные технологии"

Terms Publications

Issues of the journals (Archive)

ECONOMICS AND LAW
HUMANITIES
NATURAL AND TECHNICAL SCIENCES
COGNITION

Серия - Natural and Technical Sciences