Data-Driven Incremental Learning of Takagi-Sugeno Fuzzy Models

Nowadays data-driven models become more and more an essential part in industrial systems for application tasks such as system identification and analysis, prediction, control, fault detection or simply simulation. Data driven models are mathematical models which are completely identified from data, which can be available in form of offline data sets, most commonly stored in data matrices, or in form of online measurements. Data-driven models possess the nice property that they can be built up generically in the sense that no underlying physical, chemical etc. laws about the system variables have to be known. %which are recorded with a certain frequency within an online process. Whenever measurements are recorded online with a certain frequency, usually the models should be kept up-to-date from time to time, especially when tracking highly time-variant system behaviors for online identification tasks, which requires an adaptation of some model parameters in form of incremental learning steps, as a complete rebuilding from time to time with all recorded measurements would yield a too high computational effort for a complete online training.