ordered_model_fidelities
Specification of an hierarchy of model fidelities, ordered from low to high.
Specification
Alias: model_fidelity_sequence
Arguments: STRINGLIST
Child Keywords:
Required/Optional |
Description of Group |
Dakota Keyword |
Dakota Keyword Description |
---|---|---|---|
Optional |
Correction approaches for surrogate models |
Description
An ensemble surrogate model can manage an ordered set of model fidelities, each of which may in turn involve multiple discretization levels (in the case of a simulation model) or additional model recursions.
The ordering is assumed to be from lowest fidelity to highest
fidelity, as dictated by an accuracy versus cost trade-off.
Corresponding sequence specifications within methods (e.g.,
quadrature_order_sequence
, sparse_grid_level_sequence
,
expansion_order_sequence
, etc. within stochastic expansion methods)
should be synchronized with this model order.
Additional Discussion
Internal to the hierarchical usage of an ensemble surrogate model, only one low fidelity model instance and one high fidelity model instance are active at any given time, although various optimization and UQ algorithms can be used to traverse deep multilevel and multifidelity hierarchies by activating different model combinations and different response modes within the hierarchical model infrastructure.
Examples
model,
id_model = 'HIERARCH'
surrogate ensemble
ordered_model_fidelities = 'LF' 'MF 'HF'
correction additive zeroth_order
model,
id_model = 'LF'
simulation
interface_pointer = 'LF_DRIVER'
model,
id_model = 'MF'
simulation
interface_pointer = 'MF_DRIVER'
model,
id_model = 'HF'
simulation
interface_pointer = 'HF_DRIVER'