MSc' Zhengkun Jiang WUR

1
Optimization of Mobile Radioactivity Sampling
Design

• MSc. Zhengkun Jiang (WUR)
• Dr. ir. Sytze de Bruin (WUR)
• Dr. ir. Gerard B.M. Heuvelink (WUR)
• Dr. Chris J.W. Twenhöfel (RIVM)

2
Introduction
Nuclear accident
Posing severe impacts on environment and human
health
Radioactivity Monitoring Network
Providing early warning and monitoring against
nuclear accident
3
Problem definition

• More accurate information about spatial
  distribution of the radioactive contaminant near
  accident site is needed
• Static monitoring networks are usually too coarse
  to estimate local conditions.

• Proposed solution
• additional mobile measuring devices
• Key Question Where to optimally allocate
  additional mobile measuring devices?

4
Simulated accident and NPK-PUFF model
Simulated accident

• Borssele nuclear power plant
• South West wind (3-4 m / s)
• Cs-137 (Bq / m3)
• Time instant 5 hours

NPK-PUFF model

• Spatial distribution of radioactive contaminant
• Prognostic overview of radiological dose

5
Methodology

• Geostatistical model Uncertainty
• The true concentration is assumed as the sum
  of deterministic trend (NPK-PUFF prediction) and
  spatially auto-correlated stochastic residual

True concentration (x, t) NPK-PUFF prediction
(x, t) residual (x, t)
6
Methodology

• Impact factor
• ? false positive impact factor
• ? false negative impact factor
• ? ? 1 5
• Optimization Criterion
• to minimize a weighted sum of the
  expected area occupied by the two false states
  for certain action level at certain time

Correctly classified
false positive
Correctly classified
false negative
7
Methodology

• Geographical Constraints
• Openness
• Accessibility

8
Methodology Optimization procedure
preparation
Objective function calculation
Spatial simulated annealing
Semivariogram model
True concentration maps
Calculated the expected total cost
Predicted plume
New cost lt old cost
predicted concentration maps
100 Realizations of simulated residual
Candidate sampling locations
100 Interpolated residuals fields
Predicted plume
Randomly generated sampling configuration
New sampling design is accepted
New sampling design is not accepted and generates
the next design from the last accepted design
9
Results The expected total cost
10
Results initial and final sampling designs
Initial sampling design
final sampling design after 2000 spatial
simulated annealing iterations
11
Results The corresponding probabilities of
sampling design
final false positive probability map
Initial false negative probability map
12
Results The corresponding probabilities of
sampling design
Initial false positive probability map
final false positive probability map
13
Conclusions

• The optimal sampling design is achieved by
  improving data dependent objective function and
  using stochastic simulation and spatial simulated
  annealing
• Time integration is needed for dose calculation
• More elaborate residual model and simulated
  realizations is needed

14
Q A
Thank you for your attention!