The article addresses current issues of ensuring safety in industrial areas, using Krasnoyarsk – a major Siberian center with a high concentration of chemically hazardous facilities – as a case study. Particular attention is paid to risks associated with the use of chlorine at water treatment plants, which are classified as hazardous production facilities in accordance with Federal Law No. 116-FZ. An innovative methodology for predicting chemical contamination zone parameters, based on a cascade architecture of deep neural networks and serving as an alternative to traditional approaches under SP 165.1325800.2014, is proposed. The network architecture includes preprocessing modules, a feature encoder, an attention mechanism, and regression heads that process 14 input parameters to calculate the equivalent quantity of the substance, contamination zone depths, and exposure duration for both primary and secondary clouds. The model was trained on 50,000 scenarios using the Monte Carlo method, demonstrating high accuracy (deviations <1%) compared to the "TOXI+Risk" software suite. Validation results for a chlorine leak scenario of 10 tons confirm the model's physical consistency and practical value for emergency response management, damage minimization, and sustainable regional development. The methodology based on the cascade neural network represents a scientifically grounded, accurate, and ultra-fast alternative to regulatory methods. Its integration into the civil defense risk management system of industrialized territories (exemplified by Krasnoyarsk) enables a fundamentally new level of operational efficiency in assessing accident consequences, planning evacuation measures, and minimizing potential damage.
Grebnev et al. (Mon,) studied this question.