(340d) Data-Driven Nonlinear System Identification of Covid-19 Pandemic Behaviour

The outbreak of the COVID-19 pandemic causing pneumonia has attracted worldwide attention. In order to both observe the dynamic behaviour of this disease and to take the right measures for optimization and forecasting purpose, a reliable and sustainable model represents a crucial part. However, the proper understanding of the dynamic behaviour is not ensured a priori, such that a variety of different models leads to different results and uncertainties. Thus, data-driven nonlinear modelling becomes more important. In general, the basic challenges in data-driven modelling consist of unknown model structure, unknown functional candidates to represent the appropriate behaviour, providing a reduced model order with an acceptable computation effort, and handling the effect of noise on the overall performance. Therefore, this investigation examines a data-driven nonlinear system identification method using sparse regression with approximated L1-regularization to represent the dynamic behaviour of COVID-19 by determining the appropriate nonlinear functional candidates with its loadings based on the states of the SEIR-model assumptions. The proposed method was able to obtain an accurate model with tight confidence intervals that reflected the original data well. The successful computation of the estimates is related to an appropriate selection of the nonlinear candidates through Feature selection and a good setting of the hyperparameter in the regularization.