Uncertainty Flashcards
What is the most common type of uncertainty?
Parameter uncertainty, which is about the “true” value of a parameter in a mathematical model.
True / False?
Probability can be said to represent “the probability that the model is correct”.
False!
No model is exactly correct. In addition, we cannot say it’s the probability that the “model is approximately correct” because all models could be called approximately correct, so probabilities wouldn’t be event exclusive.
What is Aleatory Uncertainty?
Arises through natural variability IN a system.
a) Can be quantified by measurements and statistical estimations or by experts
b) Relates to things which we either cannot (or choose not) to learn
c) The likelihood of an aleatory uncertainty cannot be changed by Bayes theorem in a Bayesian model (because there are no observations to apply Bayes theorem to!)
What is Epistemic Uncertainty?
Arises through analysts lack of knowledge ABOUT a system.
a) Can be quantified by experts but cannot be measured by statistical testing or data collection
b) Relates to things which we could learn if we were able
c) The likelihood of an epistemic uncertainty can be updated using Bayes theorem
To make the distinction between Aleatory & Epistemic we need to:
(3 points)
- Make modelling choices clear
- Provide the basis on which quantification will take place
- Demonstrate the effects of epistemic uncertainty on the output (since it’ll show how much can be controlled if the need arises).
Two other types of uncertainty are Ambiguity and Volitional uncertainty.
Describe ‘em
- Ambiguity – Lack of precision, is removed by using more careful definitions.
- Volitional Uncertainty – Whether an individual will do what he agreed to do, or not.
What are two major sources of uncertainty when building a model?
- Model Uncertainty – You might be uncertain about how well your model represents reality.
- Parameter Uncertainty – You might be unable to collect data for all parameters or do not completely trust the source.
What data can we use if specified data is not available?
Expert or Generic data. (Generic data is data that has been collected from a variety of sources).
Uncertainty Bound is the range of _______ which can be found across different application data sources.
Uncertainty Bound is the range of different failure rates which can be found across different application data sources.
The primary sources for Quantitative data for a probabilistic risk analysis are:
(5 items)
- Generic Data
- Expert Data
- Industry Wide Data
- Plant Specific Data
- Part Specific Data
Monte Carlo Simulation computes ________ and tests their effect on model predictions.
It is the repeated simulation of “____________” in a computer based model, with the intention of modelling different possible future events.
Monte Carlo Simulation computes uncertainties and tests their effect on model predictions.
It is the repeated simulation of “independent random quantities” in a computer based model, with the intention of modelling different possible future events.
Monte Carlo runs simulations ________ to apply the statistical “law of large number”.
This states that the relative frequency of events occurring in the simulation will converge to the _________.
Monte Carlo runs simulations many times to apply the statistical “law of large number”.
This states that the relative frequency of events occurring in the simulation will converge to the theoretical probability.
Monte Carlo sampling uses the computer to generate streams of “____________”.
Monte Carlo sampling uses the computer to generate streams of “independently uniformly distributed random numbers”.
Computers cannot generate ________ numbers, instead they generate ________ numbers which are numbers that are generated using a deterministic function.
The most common function is the ________ ,
Computers cannot generate true random numbers, instead they generate pseudo-random numbers which are numbers that are generated using a deterministic function.
The most common function is the Mixed Congruential Method,
What do Variance Reduction methods do?
Pseudo numbers can generate different parameter values. However, convergence issues are prominent if there is a high variance in output across different run operations – this means we would require many runs before we obtain “good” parameters.
