Abstract
This study’s object is the process of monitoring and managing environmental risks in railroad accidents involving transportation of hazardous goods. A problem has been identified, related to the absence of a single, holistic approach to risk management during transportation, which would integrate methods of spatial-temporal forecasting with a formal assessment of uncertainty. A mathematical model has been suggested that makes it possible to process and analyze data acquired from a mobile automated air quality monitoring system (MAAQMS). The established dependences laid the foundation for the machine learning and statistical analysis model used in the operation of a simulation model (SM) of monitoring and managing environmental risks. The simulation model, unlike similar ones, has been developed in the following directions: 1) representation of data and processing of omissions; 2) construction of probabilistic risk maps taking into account uncertainty and calibration of forecasts of the state of environmental pollution at the accident site; 3) adaptation of the model in case of data variability at the accident site; 4) multi-criteria optimization of management decisions. In summary, the simulation model reported in this study provides decision-makers with the prospect of not only predicting the probability of exceeding the maximum permissible concentrations (MPC) of pollutants on the railroad infrastructure but also forming confidence risk maps. Unlike similar solutions, the constructed model is ML-oriented. In other words, the prediction of risk level is built in a spatial-temporal statement on a railroad network graph taking into account data received from MAAQMS. The adequacy of the model was confirmed by achieving the area under the ROC curve (AUC = 0.990) and the PR analysis indicator (AP = 0.940)