The machine learning model for automatic control of air compressors was implemented by Severstal Digital specialists together with specialists from the Cherepovets Metallurgical Plant. The model works in the gas-oxygen shop CherMK. The main task of compressor equipment is to compress air and supply it under pressure. Compressed air is mainly used to supply air separation units to produce oxygen, nitrogen, argon. Variable consumption of compressed air by various units leads to a constant change in compressor operating modes. Previously, the decision to put or put the compressor into operation at one time or another, as well as to reduce or increase the load, was made by the operator based on his experience. The process of redistributing the load affects the total energy consumption, so it depends on the human factor and may not be optimal.

The task of the machine learning model is to redistribute the performance of each compressor so that the amount of air does not change, and the total energy consumption decreases. The model takes into account data on air volume and electricity consumption, and also makes a recommendation on the optimal distribution of load units.

In addition, the model allows you to dynamically set technological limitations, for example, among which compressors to distribute power consumption or which consumption ranges are acceptable for each compressor.