HERON Economic Uncertainty Quantification as a new Valued Param

[GabrielSoto-INL]

HERON has recently added uncertainty quantification capabilities for economic parameters within merged pull request #331. Users may now associate a distribution with certain cash flow parameters to simulate uncertainty in cost or other economic inputs. For example, a user may want to capture the uncertainty of capital expenditures of an advanced nuclear power plant because public data is hard to find or is not available yet. This new addition allows users to account for that uncertainty.

Refresher on HERON: HERON’s standard workflow (the bi-level, RAVEN-runs-RAVEN stochastic optimization) already provides uncertainty quantification when users run technoeconomic simulations with trained ARMA or TSA reduced order models (ROMs): market uncertainty. The outer RAVEN level determines values for variable component capacities:

• In opt mode via stochastic gradient descent (Bayesian Optimization is coming soon) for component capacities with <opt_bounds>
• In sweep mode via iterating over a grid of values for component capacities with <sweep_values>

Values from each “step” in the outer are held constant in the inner level, where decisions over resource dispatch are optimized for different market scenarios. Each scenario/realization takes a time series sampled from the included ARMA or TSA ROM using a Monte Carlo sampler. In this way, HERON can incorporate uncertainty over market pricing, hourly load demand, wind, and solar availability, etc.

What’s new: Users can now add uncertainty to cash flow parameters alongside ARMA/TSA ROM runs. This is particularly useful for components such as advanced nuclear reactors which do not have settled, publicly available cost numbers for capital expenditures (CAPEX), operations and maintenance costs, etc., and may instead have a certain range or distribution. The uncertainty is applied as a distribution for a cash flow parameter (e.g., the reference price or cost). In the HERON standard workflow, when the Monte Carlo sampler generates a synthetic history from the ARMA/TSA ROM, it will also sample from the provided distribution(s) a new cost value or cash flow parameter for each realization. HERON may now produce results such as the expected value of net present value (NPV) given uncertainty in market conditions and uncertainty in component costs.

How to use it: Users must include an <uncertainty> node below the desired cash flow parameter; this uncertainty node is a new ValuedParam (e.g. “fixed_value”, “opt_bounds”, “ARMA”, “Function”). Within the <uncertainty> node, users must include a distribution XML node—with the same syntax as the ones in RAVEN (e.g., “Normal”, “Uniform”, “Beta”, “Weibull”)—to represent the uncertainty in the parameter. Below is an example of adding uncertainty to the CAPEX cost/reference price cash flow – it is assumed to be a normal distribution with mean 10,000 and standard deviation of 2 (this is low just for testing purposes).

Important Notes/Caveats:

• Economic Uncertainty can only be specified for the following cash flow parameters:
  o <reference_price> (sometimes referred to as ‘alpha’)
  o <driver>
  o <reference_driver>
  o <scaling_factor_x>
• A full list of distributions can be found in the RAVEN User Manual (Section 9)
• Economic Uncertainty is not currently available when using a “StaticHistory” (i.e., using <CSV> instead of an ARMA/TSA ROM)