Tapioca M. Chiwewe is a research scientist at IBM Research in Johannesburg, South Africa. He and his colleagues are expanding the company's machine learning technology to predict air quality. Image source: IBM Research.
In more and more places, machine learning is changing the traditional physical model of air quality prediction. The latest development is in South Africa. In the recently established IBM Research Laboratory, computer engineer Tapiwa M. Chiwewe used IBM's air quality prediction software for local requirements and added new features. This study is an extension of what is known as the "Green Horizon" project. IBM researchers began working with Chinese government researchers and officials two years ago.
At the 14th International Conference on Industrial Informatics held in France last month, Chiwewe presented the first results of South African laboratories, including the prediction of ground-level ozone. "You can do a lot of physics research to understand how ozone comes to different places." He said, "But what we do is collect a lot of data and train machines on this data, and they can predict the local ozone level, and There is no need to understand what principle ozone operates in the atmosphere."
Like China, a large part of South Africa’s energy also relies on coal power, and it is also trapped in toxic air filled with particulate matter. Although Chiwewe said that he and his colleagues in South Africa can reuse the air prediction tool developed in some of China, they must be debugged for local characteristics. For example, South Africa has a long mining history. Mining has left a lot of tailings plants exposed to the air. Strong winds often blow tiny particles. In the downwind residential areas, the air quality drops. Chiwewe said he hopes to develop a tool that can identify signs of wind and give early warning to nearby residents.
Johannesburg also lacks the intensive air quality monitoring stations in Beijing: There are reports that there are 35 monitoring stations in Beijing, compared with only 8 in Johannesburg. To compensate for this shortcoming, IBM's system design also includes data from two other, lower-cost sensors that may only include one or two types of monitoring (eg, particulates), rather than the full set of monitoring stations covered by the primary monitoring station. Gas and Particles - Chiwewe said that Beijing may have about 1,000 smaller sensors. Therefore, his team must debug the "professor" stage of the machine learning system and use less data more creatively. Before getting more ground data, they were studying an intermediate approach: so-called "virtual monitoring stations" that could use remote sensing platforms such as satellites.
All of these studies can help guide the government. The current government officials are providing IBM researchers with data from public monitoring stations. In return, they will receive the forecast results for free. As predictions become more mature, government officials here or elsewhere can use these results to order heavily polluting power plants to reduce production during periods of severe pollution or other hazy weather and to help existing laws perform better. Long-term predictions can help the government plan road layout and differentiation, thereby reducing emissions, or at least reducing the health impact of emissions.
Via IEEE Spectrum
Langrui Energy (Shenzhen) Co.,Ltd , https://www.langruibattery.com