Smart Cities: Forecasting Software thats a Breath of Fresh Air

Siemens has developed intelligent software that relies on artificial neural networks to accurately predict the degree of air pollution in large cities several days in advance. The software can give cities and their residents the information needed to minimize pollution peaks before they are likely to happen, thus improving the quality of life.


by Sebastian Webel

Cities have always been considered engines of industrial growth, as they offer their residents opportunities for employment and prosperity. This fact has become particularly pronounced in modern times. Indeed as of 2009, for the first time in the history of humanity, more than half of the world's population lived in urban centers. Furthermore, by 2050, 70 percent of the world's people will live in cities, almost as many people as are alive today.


But the downside of urbanization is easy to see. The explosive growth in the number of city dwellers is posing a huge challenge for urban infrastructures, which are reaching their limits in many places. For example, today more than 50 percent of the world's population has settled on less than two percent of the earth's surface area. As a result, urban centers with their traffic, industry, and energy needs already account for up to 70 percent of global greenhouse gas emissions.


Cities literally generate stuffy air. And that air is increasingly unwholesome for residents. According to an analysis published by the World Health Organization (WHO) in May 2018, 90 percent of the world's urban population breathes in air with pollutant levels that are much higher than the recommended thresholds.

Seven Million Deaths

The consequences are chilling. According to WHO data, approximately seven million people die each year from the effects of air pollution. Thus, one out of every eight deaths worldwide is a result of polluted air.


But the WHO strikes an optimistic note as well. It says that cities have the ability to greatly improve their air quality through local measures – whether it’s by means of modern and efficient solutions for smart infrastructures or through simple measures that can be implemented on short notice, such as traffic regulations and attractive incentives for pedestrians and bicyclists. Ideally, this would be done right at the places where air pollution is worst. That, however, requires knowledge of how pollutant levels change over time in specific locations.

Precise Forecasts of Air Pollution

This challenge has been taken up by Dr. Ralph Grothmann from Siemens Corporate Technology (CT). Grothmann has developed air pollution forecasting models that are based on neural networks. These models can accurately predict the degree of pollution in large cities several days in advance. "Neural networks are computer models that operate like the human brain. Through training, they learn to recognize relationships and to make predictions," says Grothmann. His models are deep neural networks, which use considerably more layers of artificial neurons than in the past. Each level is devoted to a different plane of abstraction. Because a large number of levels are interlinked, the findings are much more detailed than was the case with earlier neural networks. It sounds a bit like science fiction, but neural networks have been a proven technology at Siemens for many years and in multiple sectors. For example, they have been used to predict levels of economic activity, raw materials prices, and even the expected electricity yield from renewable energies.

London Pilot Project

During development of the forecasting system, Grothmann relied on the weather and emissions data that the city of London collects and makes available using approximately 150 sensor stations throughout its metropolitan area. "This data allowed us to train our system. Specifically, we gathered emission measurements for gases such as carbon monoxide, carbon dioxide, and nitrogen oxides. We linked the development of these emissions with the weather data from the same period of time, which included factors such as humidity, solar irradiation, cloud cover, and temperature," says Grothmann.  Recurring events such as workdays and weekends, holidays, trade shows and sports events were also programmed into the model, as these affect traffic and emissions in a variety of ways.


Based on all of the resulting data, as well as seasonal and immediate weather forecasts, the neural network had to learn how to predict the degree of air pollution. At the beginning, it did not know what effect any particular variable would have, and its forecasts therefore diverged widely from the emission levels that were actually measured. During its training process, however, which included hundreds of iterations, the program steadily reduced the difference between its forecasts and the actual levels of pollutants measured in the city’s atmosphere. It achieved this by changing the weightings of individual parameters.


"Now our system can predict the level of air pollution at 150 places in the city for every hour of the next three days with an error rate of less than ten percent," says Grothmann. "Our results also make it possible to infer what the main drivers of the predicted air pollution will be."

Data that Points to Preventive Measures

Of course, no forecast by itself can reduce a city’s air pollution. But forecasting software provides the data that’s needed to implement targeted responses. "For example, if our system predicts above-average pollution levels in certain parts of London for the next two days because of traffic, the city could temporarily raise its congestion charge, block through traffic for trucks in high-impact areas for certain hours, or make  it more appealing for people to use local public transit systems," says Grothmann.


In addition to these measures in the field of transportation or in the industrial or energy sectors, Siemens’ forecasting software could become an extended service for residents who want to avoid places and periods of time that are associated with high levels of pollution. For example, they could use an online service to find out the best place and time to do their jogging in view of the pollution forecast for the next few days – not just in London but in any city equipped with sufficient sensors.  "Our system could theoretically be extended to all cities – provided they measure their air composition," says Grothmann.


Whether as an aid to minimizing pollution or as a service for health-conscious residents, Siemens’ forecasting software offers cities a springboard for an efficient, sustainable, and smart future – one in which cities offer their residents not just opportunities for development, employment, and prosperity, but also clean, breathable air.


Sebastian Webel

Picture credits: from top: 1.picture Tang Shizeng/Xinhua News Agenc/action press

More and more cities are being sued for low air quality and must show an action plan to meet target levels. As the first city to test the new CyPT-Air tool, a software developed by Siemens, Nuremberg can present a plan to lower emissions with detailed predictions about the impact of policy and technology changes. It is called CyPT-Air because the parameter-modeling tool was adapted for air quality. Nuremberg is the first city worldwide to use CyPT-Air to measure key pollution indicators, create a package of measures to reduce pollution, and make predictions about air pollution levels over the long term. Nuremberg participated in a pilot project with Siemens that lasted roughly twelve months and resulted in three comprehensive transport scenarios that the city can use to reduce air pollution.

Subscribe to our Newsletter

Stay up to date at all times: everything you need to know about electrification, automation, and digitalization.