Recommendation of Optimal Mitigation Actions Using Causal

Inference in LOCA Events at Nuclear Power Plants

Ji Hun Park

, Hye Seon Jo

, Ho Jun Lee

and Man Gyun Na

Department of Nuclear Engineering, Chosun University, 10 Chosundae 1-gil, Dong-gu, Gwangju, Republic of Korea

Keywords: Optimal Mitigation Actions, Causal Inference, Causal Impact, Loss of Coolant Accident,

Nuclear Power Plants.

Abstract: In nuclear power plants, ensuring safety during abnormal situations is of paramount importance. This study

focuses on the loss of coolant accident, a design basis accident, and applies the use of causal inference to

recommend optimal mitigation actions. The study utilizes data collected from the compact nuclear simulator

to analyze the effectiveness of various actions, including the activation of charging pumps and adjustments to

control valves. The results indicate that the simultaneous activation of charging pumps #2 and #3 yields the

highest cumulative absolute effect on maintaining the pressurizer water level. Additionally, keeping the

charging control valve and letdown back pressure valve fully open (100%) also contributes significantly to

managing the pressurizer water level during loss of coolant accident scenarios. These findings provide

valuable insights into improving nuclear power plant safety by guiding operators in choosing the most

effective mitigation strategies during LOCA situation.

1 INTRODUCTION

Nuclear power plants (NPPs) are electricity-

generating facilities that use nuclear fuel to produce

electricity. The use of nuclear fuel can involve the

release of radioactive materials, making safety a top

priority for NPPs.

However, accidents can occur in NPPs for various

reasons. In NPPs, accidents are categorized as

abnormal, emergency, or severe. This study focuses

on abnormal situations in NPPs. Abnormal situations

are defined as the period from when one or more

preset alarms are triggered due to the occurrence of

an abnormal event under normal conditions until the

reactor is shut down. When abnormal situations occur

in NPPs, operators diagnose the issue and take

mitigation actions based on procedures known as

abnormal operating procedures. These procedures

suggest appropriate mitigation actions but do not

specify them in detail.

For example, in a loss of coolant accident (LOCA)

situation, one of the design basis accidents for NPPs,

https://orcid.org/0000-0001-6225-5621

https://orcid.org/0000-0002-4413-5244

https://orcid.org/0009-0001-5155-9483

https://orcid.org/0000-0003-0097-3403

the general response is to activate a charging pump.

LOCA situations involve a rupture in the primary

piping of NPPs with a closed-loop configuration,

leading to the leakage of primary coolant into the

containment and a reduction in the primary coolant

inventory. The primary coolant is crucial for cooling

the heat generated by nuclear fuel, necessitating the

maintenance of an adequate coolant inventory.

Operating a charging pump, which draws water from

a separate source and injects it into the primary

system, can be an appropriate mitigation action.

However, the procedures typically do not specify how

many of the three available charging pumps should be

operational. This ambiguity provides flexibility for

operators but also places the burden of decision-

making on them, making it challenging to ensure the

optimal mitigation action for each situation.

In this study, we introduce a method that uses

causal inference to suggest mitigation actions for

operators in LOCA situations in NPPs. The causal

inference method estimates and quantifies the causal

effect of specific mitigation actions, providing a

440

Park, J., Jo, H., Lee, H. and Na, M.

Recommendation of Optimal Mitigation Actions Using Causal Inference in LOCA Events at Nuclear Power Plants.

DOI: 10.5220/0013018900003822

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 21st International Conference on Informatics in Control, Automation and Robotics (ICINCO 2024) - Volume 1, pages 440-444

ISBN: 978-989-758-717-7; ISSN: 2184-2809

quantitative evaluation of these actions. The

quantified effectiveness of these actions can inform

operators about the most likely means of mitigation.

This study is a foundational research effort on

mitigation action suggestion systems for NPPs and

focuses exclusively on LOCA situations, a design

basis accident for NPPs. In addition, a critical safety

function in LOCA situaitons is the primary coolant

inventory. The crticial safety function is a key feature

that must be maintained to mitigate an accident.

Based on the pressurizer (PRZ) water level, which is

an indicator of the primary coolant inventory, the

appropriate action is determined (i.e., maintaining a

normal PRZ water level is a primary goal).

Data were collected using a simulator, the

Compact Nuclear Simulator (CNS). The mitigation

measures investigated included 1) the activation of

charging pumps, 2) adjustments to the charging

control valve, and 3) adjustments to the letdown Back

Pressure Valve (BPV). Based on these measures, the

following five scenarios were organized to obtain

data: 1) activation of charging pump #2, 2) activation

of charging pump #3, 3) simultaneous activation of

charging pumps #2 and #3, 4) adjusting the opening

state of the charging control valve (ranging from 10%

to 100% in 10% intervals), and 5) adjusting the

opening state of the letdown BPV (ranging from 10%

to 100% in 10% intervals).

By using the causal inference method and

providing the evaluated results to operators, this

approach is expected to significantly enhance

accident management in NPPs.

2 METHOD

Causal inference is a statistical method that aims to

identify and quantify cause-and-effect relationships

between variables. In this study, we utilize the causal

impact method (Brodersen et al., 2015), a specific

approach within the broader field of causal inference.

This method is based on Bayesian structural time-

series models and estimates the causal effect of an

intervention (e.g., mitigation actions) by comparing

data from before and after the intervention. The

causal impact method has three main components: 1)

time series modeling, 2) posterior analysis, and 3)

synthetic control and flexibility.

The causal impact method employs structural

time-series models, which include state-space

representations to account for trends, and other

temporal patterns in the data. The model comprises an

observation equation, which links observed data to

latent state variables, and a state equation, which

describes how these state variables evolve over time.

Using a Bayesian framework, the causal impact

method estimates the causal effect of an intervention

by comparing the observed data post-intervention to

a predicted counterfactual scenario based on pre-

intervention data. This comparison allows for the

quantification of the intervention's impact, including

absolute and relative effects, with uncertainty

intervals that provide insights into the confidence of

these estimates.

Additionally, the method constructs a synthetic

control group using a combination of control series

that closely match the treated unit's pre-treatment

behavior. This approach avoids rigid assumptions

about the control group and allows for the flexible

incorporation of multiple sources of variation in the

data, such as local trends and seasonality. This

flexibility is crucial for accurately capturing the

impact of interventions in complex real-world

scenarios.

3 DATA

The data were collected using the compact nuclear

simulator (CNS), developed by the Korea Atomic

Energy Research Institute. The CNS is a simulator

modeled after the Westinghouse 930 MWe 3-loop

pressurized water reactor (Park et al., 1997). This

simulator can replicate a variety of abnormal,

emergency, and normal situations, and is capable of

introducing various malfunctions.

Figure 1 illustrates the configuration of the

chemical and volume control system (CVCS), which

is responsible for maintaining the primary inventory.

Figure 1: Configuration of CVCS.

Recommendation of Optimal Mitigation Actions Using Causal Inference in LOCA Events at Nuclear Power Plants

441

In this study, data were collected for five

scenarios based on the CVCS components. The

components selected for this study are those that

operators can directly control: the charging pumps,

the charging control valve (FV122 in Figure 1), and

the letdown BPV (PV145 in Figure 1). Specifically,

charging pump #1 is always operational, so the

controllable options included charging pumps #2 and

#3. Additionally, the control valves were tested in

10% increments, ranging from 10% to 100% open.

The fully closed state (0%) was not considered, as it

is not implemented in the simulator.

The accident scenario used in this study involved

a LOCA, with the assumption that a malfunction is

introduced 30 seconds after a normal situation, and

each mitigation action is initiated at 90 seconds. The

primary variable of interest is the PRZ water level,

while the input variables include charging flow,

letdown flow, reactor vessel water level, volume

control tank outlet flow, and the open state of the

charging control valve and letdown BPV.

4 RESULT

The causal inference method was utilized to quantify

the impact of PRZ water level on the operator’s

mitigation actions. First, the causal effect of charging

pump #2 is shown in Table 1 and Figure 2. It can be

seen in Table 1 that the PRZ water level averages

37.83% with the mitigation action, compared to

20.87% without it. The absolute effect is the water

level difference between these two scenarios, while

the relative effect represents the relative difference

compared to no mitigation. As a result, the use of

charging pump #2 shows an 81.31% increase in the

PRZ level compared to no mitigation action.

Additionally, the reactor shutdown time without

mitigation action is 267 seconds compared to 487

seconds after mitigation action.

In Figure 2, “y” is the data with mitigation action,

and “predicted” is the result of predicting the data

without mitigation action. In other words, the causal

impact method performs a counterfactual analysis by

predicting data without mitigation action based on data

with mitigation action. Additionally, the second row of

Figure 2 shows the absolute effect over time, and the

third row shows the cumulative effect over time.

Second, the results for the mitigation actions using

charging pump #3 are presented in Table 2 and Figure

3. The findings indicate that charging pumps #2 and

#3 have similar causal effects, with features

consistent within the margin of error. However, it can

be seen that each charging pump #2 and #3, which

should have the same treatment effect (i.e., same

reactor shutdown time), have different predictions

(i.e., different no mitigation action). This is caused by

uncertainty in the predictions.

Table 1: Causal effect on charging pump #2.

PRZ water level

(

)

Average Cumulative

Mitigation action 37.83 15018.9

No mitigation action 20.87 8283.57

solute effect 16.97 6735.33

Relative effect 81.31% 81.31%

Reactor shutdown time 487 seconds

Figure 2: Results of estimating causal effects over time on

charging pump #2.

Table 2: Causal effect on charging pump #3.

PRZ water level

(

)

Average Cumulative

Mitigation action 37.83 15018.9

No mitigation action 20.74 8232.47

solute effect 17.09 6786.45

Relative effect 82.44% 82.44%

Reactor shutdown time 487 seconds

Figure 3: Results of estimating causal effects over time on

charging pump #3.

Third, the effects of using both charging pumps

#2 and #3 simultaneously are illustrated in Table 3

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

442

and Figure 4. The results show that the relative effect

of starting both pumps together is greater than starting

them separately. However, when charge pumps 2 and

3 are started together, the average PRZ level is lower

compared to when the charge pumps are started

individually. This is because the simultaneous start-

up of the charging pumps causes the abnormal

situation to persist longer, resulting in lower PRZ

water levels, illustrating the average fallacy.

Table 3: Causal effect on charging pumps #2 and #3.

PRZ water level

(

)

Avera

e Cumulative

Mitigation action 31.9 23477.15

No mitigation action 10.7 7871.57

solute effect 21.2 15605.58

Relative effect 198.25% 198.25%

Reactor shutdown time 826 seconds

Figure 4: Results of estimating causal effects over time on

charging pumps #2 and #3.

Fourth, the results related to the charging control

valve's opening state are shown in Figure 5. It is

observed that the effectiveness of the mitigation

actions is significantly reduced when the valve is

open below 70%. This indicates the necessity of

maintaining an opening state of 80% or more to

sustain the PRZ water level effectively.

Figure 5: Relative effect of charging control valve opening

state.

The detailed causal effect when the charging

control valve is 100% open is provided in Table 4 and

Figure 6.

Table 4: Causal effect on charging control valve 100%.

PRZ water level (%)

Average Cumulative

Miti

ation action 42.35 9061.96

No miti

ation action 32.27 6905.24

solute effect 10.08 2156.72

Relative effect 31.23% 31.23%

Reactor shutdown time 304 seconds

Figure 6: Causal effect of a 100% open charging control

valve.

Finally, the impact of the letdown BPV’s opening

state is depicted in Figure 7. Unlike the charging

control valve, there isn't a proportional relationship

between the opening state and effectiveness, but a

significant mitigation effect is observed at 100% open.

This is due to the fact that the charging control valve

exhibits a proportional increase in charging flow in

accordance with the opening state of the valve,

whereas the letdown BPV demonstrates a variation in

letdown flow as a consequence of the pressure

difference. A 100% opening of the letdown BPV

indicates that the pressure at the front and back are

equal, resulting in a letdown flow of 0. Consequently,

only when the letdown BPV is fully open is the

letdown flow 0, which yields a superior treatment

effect compared to other conditions.

Figure 7: Relative effect of letdown BPV opening state.

Recommendation of Optimal Mitigation Actions Using Causal Inference in LOCA Events at Nuclear Power Plants

443

Specifically, the causal effect of a fully open

letdown BPV is shown in Table 5 and Figure 8.

Table 5: Causal effect on letdown BPV 100%.

PRZ water level (%)

Average Cumulative

Miti

ation action 34.28 20262.02

No miti

ation action 20.5 12114.5

solute effect 13.79 8147.52

Relative effect 67.25% 67.25%

Reactor shutdown time 681 seconds

Figure 8: Causal effect of a 100% open letdown BPV.

In conclusion, the most effective mitigation action

for maintaining the PRZ water level involves the

simultaneous activation of charging pumps #2 and #3,

as evidenced by the cumulative absolute effect.

Additionally, the recommended actions based on the

results are: 1) simultaneous start-up of charging

pumps #2 and #3, 2) keeping the charging control

valve fully open at 100%, and 3) keeping the letdown

BPV fully open at 100%.

5 CONCLUSIONS

In accident situations at NPPs, operators perform

mitigation actions based on established procedures.

However, these procedures often lack specificity and

leave critical decisions to the operators. This study

aims to recommend optimal mitigation actions for

LOCA situations in NPPs.

We explored the application of causal inference to

evaluate and recommend optimal mitigation actions

during LOCA situations. By analyzing the effects of

various mitigation actions on the PRZ water level, we

identified the most effective strategies for managing

abnormal conditions.

The study utilized data collected from simulations

involving different combinations of charging pumps

and control valve settings. The results consistently

showed that the simultaneous activation of charging

pumps #2 and #3 led to the most significant

improvement in maintaining the PRZ water level,

evidenced by the highest cumulative absolute effect.

Additionally, keeping the charging control valve and

the letdown BPV fully open (100%) was found to be

particularly effective.

The findings suggest that adopting these specific

mitigation strategies can substantially enhance

reactor safety during LOCA events. By providing

operators with clear and quantifiable

recommendations, this approach helps ensure that the

most effective actions are taken promptly, reducing

the risk of reactor damage and improving overall

safety protocols in NPPs. This study lays the

groundwork for developing more detailed and

specific guidelines for emergency response,

potentially leading to better-prepared operators and

safer nuclear plant operations.

ACKNOWLEDGEMENTS

This work was supported by the Korea Institute of

Energy Technology Evaluation and Planning

(KETEP) grant funded by the Korea government

(MOTIE) (20224B10100130, Development of

operational state simulator for operating nuclear

power plant and commercialization technology for

artificial intelligence decision-making support system

to prevent human error in accident operation) and the

National Research Council of Science & Technology

(NST) grant by the Korea government (MSIT) (No.

GTL24031-000).

REFERENCES

Brodersen, K. H., Gallusser, F., Koehler, J., Remy, N., &

Scott, S. L. (2015). Inferring causal impact using

Bayesian structural time-series models.

Park, J. C., Kwon, K. C., Sim, B. S., Kim, J. T., Lee, D. Y.,

Kim, C. H., ... & Yang, K. N. (1997). Performance and

equipments upgrade of compact nuclear simulator (No.

KAERI/RR--1794/97). Korea Atomic Energy Research

Institute.

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

444