PAO Severstal is a steel and mining company that owns the Cherepovets Metallurgical Plant, the second largest in our country. In 2019, the company produced 11,9 million tons of steel, with revenue of $8,2 billion
Business case of PAO Severstal
Task
Severstal decided to minimize the company’s losses due to erroneous forecasts for electricity consumption, as well as to eliminate unauthorized connections to the grid and theft of electricity.
Background and motivation
Metallurgical and mining companies are among the largest consumers of electricity in industry. Even with a very high share of own generation, the annual costs of enterprises for electricity amount to tens and even hundreds of millions of dollars.
Many of Severstal’s subsidiaries do not have their own power generation capacity and buy it on the wholesale market. Such companies submit bids stating how much electricity they are willing to buy on a given day and at what price. If the actual consumption differs from the declared forecast, then the consumer pays an additional tariff. Thus, due to an imperfect forecast, additional electricity costs can reach up to several million dollars a year for the company as a whole.
Solution
Severstal turned to SAP, which offered to use IoT and machine learning technologies to accurately predict energy consumption.
The solution has been deployed by Severstal’s Center for Technological Development at the Vorkutaugol mines, which do not have their own generating facilities and are the only consumer on the wholesale electricity market. The developed system regularly collects data from 2,5 thousand metering devices from all divisions of Severstal on the plans and actual values of penetration and production in all underground areas and on the active coal mine, as well as on current levels of energy consumption. The collection of values and recalculation of the model takes place on the basis of data received every hour.
implementation
Predictive analysis using machine learning technology makes it possible not only to more accurately predict future consumption, but also to highlight anomalies in electricity consumption. It was also possible to identify several characteristic patterns for abuses in this area: for example, it is known how an unauthorized connection and operation of a cryptomining farm “looks like”.
The results
The proposed solution makes it possible to significantly improve the quality of the energy consumption forecast (by 20–25% monthly) and save from $10 million annually by reducing fines, optimizing purchases, and countering electricity theft.
Plans for the future
In the future, the system can be expanded to analyze the consumption of other resources used in production: inert gases, oxygen and natural gas, various types of liquid fuels.
Subscribe and follow us on Yandex.Zen — technology, innovation, economics, education and sharing in one channel.