Time series and data analysis, prediction models

A time series is the term used to describe observed data points that occur continuously at regular intervals. If one wants to analyze these data, for example to make a prediction about the future course, the basic approach is to find a model that describes the past data as well as possible.

In time series analysis, we humans often develop an intuitive feeling for the underlying observation and can quickly answer whether there are any anomalies or how the time series will develop further.

For an artificial intelligence (AI), however, this is not always an easy task. In addition to the selection of the mathematical model, it is also true that any AI is only as good as the data provided. To create meaningful predictive models, for example, questions have to be answered such as how domain knowledge and expert knowledge or trends and seasonalities can be integrated into the model. The Fraunhofer IOSB is researching these questions in several projects.

Fields of application

The business unit develops forecasting solutions for energy suppliers so that they can optimally coordinate energy generation and consumption. The software solution EMS-EDM PROPHET® is a powerful time series forecasting module. In addition, we offer customer-specific individual projects and a comprehensive range of seminars to become an “Energy Data Analyst”.

Machine learning for industry: The Fraunhofer-Gesellschaft's Machine Learning Research Center is investigating the advantages of deep learning methods compared to conventional linear methods and how prior knowledge can be integrated into these methods.

The project W-Net 4.0 has among other things a focus on the question of transfer learning: to what extent can “general” models for the prediction of drinking water consumption be trained and adapted for specific consumer groups, such as industrial enterprises or schools.