Resumen
Digitalization and intellectualization as part of the mass introduction of cyber-physical systems "Industry 4.0" has become a real trend in which cyber technologies provide automated and automatic control, greater efficiency, and increased security. At the same time, the integration of such technologies into critical infrastructure facilities is often subject to cyber threats, and as a result, cyber-attacks, violating not only the confidentiality and integrity of data, but also accessibility, for example, using DDoS attacks, which indicates the imperfection of most data filtering methods. attacks at various levels of the OSI model. This leads to the fact that many organizations whose physical and computing resources have access to the external global Internet network face the inaccessibility of their own services, which leads to the inability to provide the necessary data and services for both their own employees and customers, which leads to represents the financial loss of the company from equipment downtime. To minimize losses from this problem, it is proposed to use a spike (impulse) neural network to filter attacks by unauthorized external traffic (DDoS).The main features of the proposed neural network are both high speed and quality (due to constant learning on big data) of self-learning, and quick response to DDoS attacks (including those that are unknown), as well as structural dependence (the number of neurons and layers of the impulse neural network) from the physical (computing) resources of the server/cluster. A modified method of nested mathematical models of self-learning (unsupervised learning) of a pulsed neural network is proposed, which is based on the standard ANN training method with error backpropagation (gradient descent), which allows the pulsed neural network to quickly and efficiently learn in order to filter attacks by external unauthorized traffic