Data-based Modelling

Data-based models are considered black-box models which are quite in contrast with the physics-based models where the entire input-output relations of the model are established primarily based on the experimental/simulation data using statistics, probability, and network architecture.  This aids in avoiding the complexities of physics-based models, intensive calibration, and also provides better accuracy in most cases. Data-based models are computationally faster when compared to complex physics-based models which is extremely important in real-time control, and diagnostics applications, especially where the complex mechanisms cannot be modelled using simple governing equations

Data-based models tend to have a quicker turnaround in the preliminary phase as multiple iterations are computed straight away without significantly affecting the development time. Considering the fuel cells and lithium-ion battery commercialization timeline, it is essential to develop such data-based models to accelerate the commercialization of clean energy technologies by reaping the benefits of data acquisition, storage, and analytic resources available in today's digital age. An illustration of the data-based model using a feed-forward neural network for fuel cell and the lithium-ion battery is mentioned below

Schematic of the data-based model to predict the fuel cell voltage based on operating conditions (Multi Input Single Output)

Schematic of the data-based model to predict the fuel cell voltage based on CFD data (Multi Input Multi Output)

Schematic of the artificial neural network (ANN) architecture with extended Kalman filter (EKF) to predict the thermal behavior of lithium batteries.