Adaptive calibration of fuzzy models for early warning of emergency situations

This paper examines the problem of improving the quality and effectiveness of emergency forecasting in rapidly changing and uncertain conditions. The relevance of the study is determined by the increasing frequency and complexity of emergency situations, as well as the heterogeneity, noisy nature, and incompleteness of monitoring data, which limit the applicability of traditional methods and fuzzy early warning models that do not take into account changes in the structure of input flows and real risk criteria. The goal of the study is to improve the accuracy and robustness of emergency forecasting by developing an adaptive method for calibrating fuzzy early warning models for emergency situations. The paper also proposes a mechanism for dynamically updating the parameters of membership functions and rule weights, ensuring stable and efficient model behavior in the face of structural shifts, noise, and missing observations. А software product was developed, and a computational experiment was conducted on various emergency scenarios, including changes in the intensity of input signals and abrupt shifts in data distribution. The obtained results confirm an increase in the quality of forecasting and a reduction in the level of fuzzy uncertainty compared to the baseline model without adaptation, while the root mean square error decreased by 26-40 %, and the level of uncertainty by 18-27 %, which indicates the practical applicability of the proposed approach in early warning and real-time monitoring systems.