Optimal Modularization of Data-Driven Models

In the previous chapter, the modularization of data-driven models in the context of hybrid modelling was introduced. Difierent modular models were reviewed and it was observed that often there is insufficient use of hydrological knowledge in the modularization process. Therefore, in this chapter we explore alternative methods for building modular models that are based on the explicit use of hydrological knowledge (in particular, separate consideration of base- and excess flow), and also use data partitioning algorithms. The developed modelling techniques will be applied in Chapter 5 , where they will be compared with a single overall data-driven model, and with conceptual models. Application of the principle of modularization to separate modelling of processes that relate to difierent areas in space is considered here as well; it is further applied in Chapter 6 . There is also a possibility of multi-objective optimization of a model involving various modularization schemes, and this will be mentioned as well (however not explored). This chapter is to a large extent based on the methodology published by Corzo and Solomatine ( 2007a , b ) and Corzo et al. (2009a) .