This work presents a modeling framework that combines first principle models with Gaussian process model and aims to account for model discrepancy due to the effect of fouling on the system from previous cycle of operation.
It is argued that mixtures of experts are often insightful for describing model discrepancy and ways in which the computer model can be improved.
In the Kennedy and O'Hagan framework for Bayesian calibration of physics models, selection of an appropriate prior form for the model discrepancy function is a challenging issue due to the lack of physics knowledge regarding model inadequacy.
We found (Fig. 8B) that the relative discrepancy between the AQLV model and data is larger (i.e., the data/model viscoelastic ratio is lower) at small initial elongations (this is not necessarily true for the absolute model discrepancy, as the forces at play are higher at large lengths).
Such checks reveal discrepancies: for example, models indicate that rainfall over the major land masses of Indonesia is lower than that over the neighbouring oceans though the reverse is true (Qian 2008) and that the flow of water out of the Amazon River should be half what is observed (e.g., Hagemann et al. 2011).
Although model 2 had the lowest discrepancies for fasts up to 50 days, between 50 and 200 days it was more discrepant than model 3. On average, model 3 provided the least discrepant predictions over the widest duration of fasts.
For instance, we compared the EC numbers obtained with the EFICAz [ 52] and PRIAM [ 53] tools to the EC numbers of the reactions linked to a unique gene in our model (287 reactions) and found discrepancies for 30 and 27 reactions respectively.
Although there is good agreement between experiment and model for solutions XNO3<0.5, there are discrepancies for both solution composition and relative humidity for nitrate-rich solutions near the eutonic.
As mentioned in the section for Model I, discrepancies between theory and simulation are generally found to decrease with increasing K M ; for the case of competitive inhibition, the effective K M of the reaction is significantly larger than that of the enzyme (see Eq. (11)), which explains the increased agreement between theory and simulations for Model III compared to the previous two models.
Additional putative functional PPO gene models with discrepancies were identified for some genomes, but were excluded from this.
