Industrial Automatic Control Systems and Controllers Annotation << Back Using Bayesian Layers to Generate Probabilistic Estimates of Gas Flow Predictions Obtained from an LSTM Neural Network Abramkin S.E., Petrova A.K. An LSTM neural network can be used to identify deviations in the results of measuring gas volumes. However, it is necessary to obtain probabilistic estimates of the values predicted by the neural network, the mathematical expectation and variance of the predicted results, and, accordingly, confi dence intervals of these values, which cover the estimated parameter with a given probability. This article considers an approach in which, in order to obtain probabilistic characteristics of the results of LSTM neural network predictions, a neural network with additional Bayesian layers is synthesized, with the help of which, using the distribution of weights instead of single values of weights, it is possible to estimate the uncertainty of network forecasts. Keywords: gas transportation network, gas balance, identifi cation of deviations, LSTM neural networks, Bayesian layers. DOI: 10.25791/asu.9.2024.1528 Pp. 03-10.