In cases where little data exists from real-life applications, i.e. few measurements, newly installed sensors, limited resources, etc., synthetic data can be generated to ensure a good foundation for training machine learning models or other data-driven methods. To succeed at this, a simulation environment must be developed where expert knowledge (e.g. models, domain knowledge, sensor noise) must be utilized to ensure realistic and representative data generation. This can be seen as a type of data augmentation. Deep learning methods in particular require large amounts of data to be trained properly
Synthesizing data for machine learning and other data-driven methods.