“…Thus, the objective function for parameter estimation, Equation (5), is designed such that in addition to minimising residues between model prediction and experimental data, it also penalises the total number of active binary variables in the polynomial terms to avoid overfitting. In this way, the optimal hybrid model structure alongside its parameter values can be simultaneously identified via the established dynamic parameter estimation algorithm (del Rio‐Chanona et al, 2015): where and are experimental and model‐simulated value for concentration of state at time step in the k th data set (total number of data sets is ), respectively, and are the weight for each data point and the sum of binary variables, respectively.…”