Toolkit for description and management of event risks

Abstract

The work is based on the proposition that the probabilistic-statistical methods of risk assessment in modern engineering no longer meet the needs of users in various fields of human activity. The paper shows the capabilities of neural networks as a mechanism for predicting risks from undesirable events in various fields of application. A three-synapse model of a direct propagation perceptron has been developed to solve the problem of automobile transportation of flammable liquids, in particular, gasoline, which is very relevant for the current state of the issue in Ukraine. A software product has been created using the Python language and Keras applications that allow to operate with neural networks of this kind. The results of «training» of the three-synapse model for solving similar problems in the field of transport are shown. Preliminary weight coefficients for each input action are systematized. The example shows the results of training a neural network, in particular, the resulting values of the weight coefficients are calculated. The operation of such a network is shown. A mechanism for predicting the risks associated with accidents in some of the most relevant conditions of road transportation has been methodically developed. The result of the model operation is the idea that during the specified time period of gasoline transportation in a tank truck of a given configuration with a specific volume of transportation and a given state of roads, the main risk, oriented to a set of possible event factors, is associated with the accumulation of electrostatic charges as a result of system swaying on the way, followed by explosion, for example, when liquid is drained at the destination. Moreover, such a risk is unambiguous and does not depend on the state of other system parameters. The results of the analysis are confirmed by practical data, namely, the risks associated with the accumulation of static electricity have neutralization mechanisms even before their accumulation. The dependence of the volume of accumulation of electrostatic charges on the duration of liquid transportation is shown