Dr. Hans Georg Zimmermann - The Science of Prediction
Before people make decisions they need good forecasts. Energy suppliers have to know how much electricity a wind farm will generate so that they can bridge a period of calm winds using other power generation systems. Railway companies need forecasts of the electricity consumed by their trains so that they can better manage their own power plants. Pharmaceutical companies must be able to calculate how to compensate for the fluctuating quality of healing herbs so that medications always have the same concentration of active ingredients. Companies also have to take many different factors into account when selecting a new production site. Among other things, they need to have some idea of how salaries, transportation costs, and raw material prices will develop. Siemens researchers are developing customized software for all of these needs. Their most important tool is the Software Development Environment for Neural Networks (German acronym: SENN).
Hans Georg Zimmermann is Siemens’ leading expert in this field. His task is to develop software that works like the nerve cells (neurons) in the human brain, learning from examples and using this knowledge as a basis for making decisions. “Neural networks enable us to gain information from past events to use in the future,” explains Zimmermann, who studied mathematics at Bonn University, where he mainly examined the behavior of dynamic systems. In 1987 he completed his doctoral dissertation on game theory, in which participants – stockbrokers, for example – are viewed as players in a big game. “I was fascinated by the boom in artificial neural networks, because it allowed me to contribute my knowledge of dynamic systems.”
Zimmermann was in the right place at the right time when he joined Siemens in 1987. At the time, CT created a working group to find out if neural networks could provide Siemens with interesting results. The work initially focused on image and voice recognition and robot control. Today neural networks from Siemens can also be found in steel rolling mills, sewage treatment plants, and washing machines, where they enable the systems to operate more efficiently and conserve resources.
From the very start, Zimmermann was also convinced that neural networks could be used to develop new methods for the quantitative analysis of economic phenomena such as trends in electricity prices. “I wanted to find out what mathematical tools could be used to help neural networks learn the essential facts from a set of data,” he says.
In the case of electricity prices, the data involves dependence on the markets for coal, oil, natural gas, and other resources. Market participants’ aims play a key role. An energy company, for example, wants to be able to buy electricity as cheaply as possible even five years down the road. Researchers therefore build a dynamic model to simulate price development, including data on the company’s past strategies and forecasts about general conditions in five years. To date, Zimmermann has laid the mathematical groundwork for over 60 industrial applications and registered 22 patents to protect associated software system architectures.