Modeling of neural network monitoring agent to predict traffic spikes and agent training

The article completes the study of solutions to the problem of creating distributed QoS monitoring control in IoT and IIoT telecommunications. A multi-agent architecture is used. For the proposed structure of the LSTM network, approaches to the process of its training (configuring the network configuration) are considered, based on the representation of this process in the form of a Markov decision-making process, which made it possible to solve the problem of determining the training times of the network within one epoch, as well as its retraining using a variant of the dynamic programming method – the iteration algorithm by values, which In the future, it allows you to proceed to the synthesis of an iterative algorithm for dynamically configuring the parameters of the developed LSTM network.