Some approaches to building models of chemical processes based on machine learning

Building data-driven models is a requirement of the current state-of-the-art in computing and data science. Currently, every industrial facility accumulates a large amount of data in 1-3 years. Data-driven models do not require additional identification as they already incorporate all the features of equipment and technological processes. Complex chemical processes are characterized by a delayed response to changes in input parameters, and LSTM neural network is an effective tool for solving such problems. This paper demonstrates the influence of the activation function on the LSTM network constructed to predict the volume of nitric acid in a chemical manufacturing plant.