Scientists from HSE University in Perm Receive First Artificial Intelligence Patent
Aleksey Kychkin, a researcher at the Laboratory for Interdisciplinary Empirical Studies (LINES), and Oleg Gorshkov, a research assistant at LINES, received a patent for a system that predicts the spatial distribution of harmful substances in atmospheric air using an artificial intelligence unit. The invention can be used for integrated planning and notification of the risks of atmospheric air pollution by harmful substances. The work was carried out under a grant from the HSE University Artificial Intelligence Centre. The HSE News Service spoke with Aleksey Kychkin about the benefits of the new system, its application, and how it came to life.
We worked on the invention in 2021–2022. It took some more time to complete the application and receive feedback from the Federal Service for Intellectual Property. Our development is part of a larger project on identifying emission sources carried out at the HSE University Artificial Intelligence Centre.
We developed our solution consistently, based on the results of pre-project studies and analysis of production needs. We also had working meetings with our industrial partners, environmentalists from enterprises, representatives of city environmental monitoring services, scientists and specialists, and developers of equipment and software systems.
Many industries use specialised dispersion models to analyse air pollution measurements. However, such models can only be effective in a static regime, while atmospheric processes are dynamic. To correctly predict changes in the concentrations of harmful substances, large computing resources and a lot of calculations are required.
In the patent, we have proposed introducing an AI unit with auxiliary units to support its functioning in a cluster of computing appliances into the eco-monitoring system.
The new system includes units for collecting, storing, and processing data, making it possible to increase the efficiency of calculating the concentrations of harmful substances and obtain dynamic forecasts. The AI unit reveals trends, seasonality, anomalies, and patterns in the dynamics of emissions, which makes the forecast more resistant to changes in the work patterns of enterprises.
The technology enables the model to continuously learn and adapt to changes, which allows us to take into account the dynamics of atmospheric processes. This new scientific and technical solution complements existing systems for forecasting harmful substance emissions. We have filed a patent, published several papers in scientific journals, and created a prototype after verification.
Emission predictions were tested for a range of pollutants, including PM2.5 dust particles, at more than 20 real-world control points in Moscow, Novosibirsk, Perm, and Chelyabinsk.
The training itself takes about 30 minutes, while each new forecast is calculated instantly—within a second. In the vast majority of cases, the constructed forecasts turned out to be more accurate than forecasts based on existing dispersion models; the increase in accuracy was 6–40%, depending on the day of the week and season. We have seen that our data does not contradict such models, but works in dynamics, and this is our main advantage.
The forecasting system for the distribution of harmful substances in atmospheric air based on artificial intelligence technologies can be integrated with other existing eco-monitoring systems for large industrial enterprises in the oil and gas, chemical, pulp and paper, metallurgical and mining industries, as well as at infrastructure facilities and fuel and energy companies. The system will be of interest to specialists at environmental monitoring services, supervisory authorities, and public organisations. The information obtained using the monitoring system may be of interest to residents of cities and settlements located near enterprises.
Modern environmental monitoring of atmospheric air for industrial enterprises is not only a tool for limiting emissions, but also a key to increasing production efficiency. To do this, tools for predicting the distribution of harmful substances really need to be improved to the level of working in dynamics—this allows for the optimisation of technological processes and the detection of accidents (such as the leakage or ignition of hazardous gases, pipeline breaks, etc) in a timely manner. Taking into account changing weather conditions, our AI forecasting system makes it possible to form production plans which consider the environmental criterion.
It is also important for the state that the ecological situation in cities is safe, especially in cases when enterprises are part of the urban environment.
The main task of our project at the Artificial Intelligence Centre is to dynamically identify emission sources and predict their distribution. We started with the second point, ie, the development of a forecasting system. Source identification is a more complex task that requires a lot of predictions. Based on the obtained picture of emission distribution, we must obtain an inverse picture of the dispersion and indicate from which source the emission was made. This task, of course, is being solved today, but, as a rule, manually and after documenting emissions—in the format of an observation history analysis.
If enterprises have a technology that allows them to dynamically determine not only the risks of exceedances at control points, but also the release source, it will improve their productivity and environmental safety. Today, there is a demand for such products from enterprises.