After the Fukushima nuclear accident in 2011, there has been increasing concern regarding severe accidents in nuclear facilities. Severe accident scenarios are difficult for operators to monitor and identify. Therefore, accurate prediction of a severe accident is important in order to manage it appr...
After the Fukushima nuclear accident in 2011, there has been increasing concern regarding severe accidents in nuclear facilities. Severe accident scenarios are difficult for operators to monitor and identify. Therefore, accurate prediction of a severe accident is important in order to manage it appropriately in the unfavorable conditions. In this study, artificial intelligence (AI) techniques, such as support vector classification (SVC), probabilistic neural network (PNN), group method of data handling (GMDH), and fuzzy neural network (FNN), were used to monitor the major transient scenarios of a severe accident caused by three different initiating events, the hot-leg loss of coolant accident (LOCA), the cold-leg LOCA, and the steam generator tube rupture in pressurized water reactors (PWRs). The SVC and PNN models were used for the event classification. The GMDH and FNN models were employed to accurately predict the important timing representing severe accident scenarios. In addition, in order to verify the proposed algorithm, data from a number of numerical simulations were required in order to train the AI techniques due to the shortage of real LOCA data. The data was acquired by performing simulations using the MAAP4 code. The prediction accuracy of the three types of initiating events was sufficiently high to predict severe accident scenarios. Therefore, the AI techniques can be applied successfully in the identification and monitoring of severe accident scenarios in real PWRs.
After the Fukushima nuclear accident in 2011, there has been increasing concern regarding severe accidents in nuclear facilities. Severe accident scenarios are difficult for operators to monitor and identify. Therefore, accurate prediction of a severe accident is important in order to manage it appropriately in the unfavorable conditions. In this study, artificial intelligence (AI) techniques, such as support vector classification (SVC), probabilistic neural network (PNN), group method of data handling (GMDH), and fuzzy neural network (FNN), were used to monitor the major transient scenarios of a severe accident caused by three different initiating events, the hot-leg loss of coolant accident (LOCA), the cold-leg LOCA, and the steam generator tube rupture in pressurized water reactors (PWRs). The SVC and PNN models were used for the event classification. The GMDH and FNN models were employed to accurately predict the important timing representing severe accident scenarios. In addition, in order to verify the proposed algorithm, data from a number of numerical simulations were required in order to train the AI techniques due to the shortage of real LOCA data. The data was acquired by performing simulations using the MAAP4 code. The prediction accuracy of the three types of initiating events was sufficiently high to predict severe accident scenarios. Therefore, the AI techniques can be applied successfully in the identification and monitoring of severe accident scenarios in real PWRs.
* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.
제안 방법
A total of 330 accident simulations were classified into three types of initiating events: hot-leg LOCA, cold-leg LOCA, and SGTR. In order to confirm the event classification using the proposed algorithm, a total of 330 simulation datasets were divided into model development data and test data.
In order to effectively manage severe accidents in nuclear power plants, many studies have examined accident management including event identification using artificial intelligence techniques. In this paper, the SVC model was designed to classify the initiating events into three types of categorized events: hot-leg LOCA, cold-leg LOCA, and SGTR. In addition, the GMDH and FNN models were used to predict severe accidents and were developed to predict the important timings representing severe accident scenarios, such as the reactor core exposure time, the time when the CET exceeds 1200 °F, and the reactor vessel failure time due to LOCA.
In this paper, the SVC models were used as non-linear pattern classifiers that categorize the initiating event representing the hot-leg LOCA, cold-leg LOCA, and SGTR using a very short time integration of some selected signals immediately after a reactor scram. The input variables to the SVC models consist of the signals acquired from the reactor coolant system, steam generators, and containment at the nuclear power plant.
In this study, the GMDH [17]-[19] and FNN models were designed to monitor the timing when the following occur: the reactor core is exposed, the core exit temperature (CET) exceeds 1200 , which is when severe accident management is normally initiated, and the reactor vessel fails. The proposed accident scenario prediction algorithms are intended to provide plant operators with valuable information, such as the core exposure time and reactor vessel failure time.
The aim of this study is to develop and verify monitoring techniques for severe accidents in pressurized water reactors (PWRs) using AI methods, such as support vector classification (SVC), probabilistic neural network (PNN), group method of data handling (GMDH), and fuzzy neural networks (FNNs). The SVC and PNN models are used to classify the initiating events, and the GMDH and FNN models are used to monitor the severe accidents.
The data for a total of 330 accident simulations were performed using the MAAP4 code to acquire the data. The following 15 simulated sensor signals acquired from these simulations were used: core exit temperature, containment pressure and temperature, pressurizer pressure and water level, sump water level, collapsed water level, broken side SG pressure and temperature, broken side SG water level, unbroken side SG pressure and temperature, unbroken side SG water level, refueling water storage tank water level, and containment mole fraction of H2. The containment pressure and temperature are the values measured at the central position of the containment that is located between the operating deck and the polar crane, which is known as the upper compartment below the dome.
(10) for the SVC model was 90 sec, which means that the SVC model employs time-integrated signals at 90 sec intervals immediately after a reactor scram. The integrating time span was selected using several numerical simulations of the proposed algorithm in order to minimize the classification error. The two SVC models were trained to categorize the hot-leg LOCA, cold-leg LOCA, and SGTR as (1, 1), (1, -1), and (-1, -1), respectively, as shown in Table 1.
In order to confirm the event classification using the proposed algorithm, a total of 330 simulation datasets were divided into model development data and test data. The model development data was used to develop the proposed algorithm and the test data was used to test it independently. Therefore, a total of 300 simulation datasets were used to develop the proposed SVC classification algorithm, which consisted of 100 hot-leg LOCAs, 100 cold-leg LOCAs, and 100 SGTRs.
The simulation results demonstrated that the proposed SVC could accurately classify numerous initiating events into three types of categorized events: hot-leg LOCA, cold-leg LOCA, and SGTR. In addition, the proposed GMDH and FNN models could predict within approximately 30% RMS error the important timings representing severe accident scenarios, such as the reactor core exposure time, the time when the CET exceeds 1200 °F, and the reactor vessel failure time due to LOCA.
(10) depends on the types of input signals and initiating events. The time span was determined using the correlation between the related timing (the output of the SVR and FNN models) and the integrated input signals (Table 2). The correlation coefficient matrices between the output and input signals were calculated every 10 seconds of the integration time span from 30 seconds to 90 seconds, and the integration time span with maximum correlation degree was chosen.
The model development data was used to develop the proposed algorithm and the test data was used to test it independently. Therefore, a total of 300 simulation datasets were used to develop the proposed SVC classification algorithm, which consisted of 100 hot-leg LOCAs, 100 cold-leg LOCAs, and 100 SGTRs. The remaining 30 test simulation datasets consisted of 10 hot-leg LOCAs, 10 cold-leg LOCAs, and 10 SGTRs.
This study examined the effectiveness of the proposed accidentscenario-prediction algorithm using the GMDH and FNN models in predicting the timing when the following occurred: the reactor core was exposed, the CET exceeded 1200 °F, and the reactor vessel failed.
대상 데이터
Using the same methods as the event classification, a total of 330 simulation datasets were divided into model development data and test data. A total of 300 simulation datasets were used to develop the GMDH and FNN models, which consisted of 100 hot-leg LOCAs, 100 cold-leg LOCAs, and 100 SGTRs. The remaining 30 test simulation datasets consisted of 10 hot-leg LOCAs, 10 cold-leg LOCAs, and 10 SGTRs.
The remaining 30 test simulation datasets consisted of 10 hot-leg LOCAs, 10 cold-leg LOCAs, and 10 SGTRs. Each 100 development data consisted of 80 training data and 20 verification data. Figure 6 shows the RMS error averaged for nine cases (three different break locations each using three different event timings) using the FNN model.
A total of 330 accident simulations were classified into three types of initiating events: hot-leg LOCA, cold-leg LOCA, and SGTR. In order to confirm the event classification using the proposed algorithm, a total of 330 simulation datasets were divided into model development data and test data. The model development data was used to develop the proposed algorithm and the test data was used to test it independently.
Therefore, a total of 300 simulation datasets were used to develop the proposed SVC classification algorithm, which consisted of 100 hot-leg LOCAs, 100 cold-leg LOCAs, and 100 SGTRs. The remaining 30 test simulation datasets consisted of 10 hot-leg LOCAs, 10 cold-leg LOCAs, and 10 SGTRs.
This study examined the effectiveness of the proposed accidentscenario-prediction algorithm using the GMDH and FNN models in predicting the timing when the following occurred: the reactor core was exposed, the CET exceeded 1200 °F, and the reactor vessel failed. Using the same methods as the event classification, a total of 330 simulation datasets were divided into model development data and test data. A total of 300 simulation datasets were used to develop the GMDH and FNN models, which consisted of 100 hot-leg LOCAs, 100 cold-leg LOCAs, and 100 SGTRs.
이론/모형
Normally, fi(x1, …, xm) is a polynomial in the input variables, but it can be any function if it can appropriately describe the output of the fuzzy inference system within the fuzzy region specified by the rule antecedent. In this paper, the symmetric Gaussian membership function was used. The output of an arbitrary i-th rule, fi, consists of the first-order polynomial of inputs, as given in Eq.
The GMDH algorithm uses high-order polynomials in their Kolmogorov-Gabor form [20]-[23]. The KolmogorovGabor form (called Ivakhnenko polynomials) can be expressed as follows:
4. The coefficient parameters were decided using a normal least squares method, and the variables of the elements were calculated. The threshold value at each generation determines if the outputs of the elements in a generation are acceptable.
The potential of a data point is defined as being higher when it is surrounded by a greater amount of neighboring data. The input/output data positioned in the cluster centers were used to train the GMDH model.
In addition, the GMDH and FNN models were used to predict severe accidents and were developed to predict the important timings representing severe accident scenarios, such as the reactor core exposure time, the time when the CET exceeds 1200 °F, and the reactor vessel failure time due to LOCA. The proposed AI techniques were applied and verified using the data acquired using the MAAP4 code. In addition, more informative data obtained from an SC scheme were used to train the models.
The proposed algorithm was verified using the simulation data from the MAAP4 code [14] for the advanced power reactor 1400 (APR1400), which is an advanced PWR that was developed by the Korea Hydro & Nuclear Power Company (KHNP).
성능/효과
In addition, the proposed GMDH and FNN models could predict within approximately 30% RMS error the important timings representing severe accident scenarios, such as the reactor core exposure time, the time when the CET exceeds 1200 °F, and the reactor vessel failure time due to LOCA.
The time span was determined using the correlation between the related timing (the output of the SVR and FNN models) and the integrated input signals (Table 2). The correlation coefficient matrices between the output and input signals were calculated every 10 seconds of the integration time span from 30 seconds to 90 seconds, and the integration time span with maximum correlation degree was chosen. This means that the GMDH model can predict these important scenario timings using the initial short time integration of the measured input signals for 30 to 90 seconds immediately after a reactor scram.
Note that the GMDH and FNN models used the same data and conditions. The prediction errors were 11.8743%, averaged from the nine different applications of the GMDH model (refer to Table 3), and 12.6450% for that of the FNN model. By comparing Tables 3 and 4, the GMDH model is slightly superior to the FNN model in severe accident scenarios.
참고문헌 (27)
Akio Gofuku, Hidekazu Yoshikawa, Shunsuke Hayashi, Kenji Shimizu, Jiro Wakabayashi, "Diagnostic techniques of a small-break loss-of-coolant accident at a pressurized water reactor plant," Nucl. Technol., vol. 81, pp. 313-332, 1988.
S. W. Cheon and S. H. Chang, "Application of neural networks to a connectionist expert system for transient identification in nuclear power plants," Nucl. Technol., vol. 102, no. 2, pp. 177-191, May 1993.
Y. Bartal, J. Lin, and R. E. Uhrig, "Nuclear power plant transient diagnostics using artificial neural networks that allow "don't-know" classifications," Nucl. Technol., vol. 110, no. 3, pp. 436-449, June 1995.
K. Nabeshima, T. Suzudo, T. Ohno, K. Kudo, "Nuclear reactor monitoring with the combination of neural network and expert system," Mathematics and Computers in Simulation, vol. 60, pp. 233-244, 2002.
M. G. Na, S. M. Lee, S. H. Shin, D. W. Jung, S. P. Kim, J. H. Jeong, and B. C. Lee, "Prediction of major transient scenarios for severe accidents of nuclear power plants," IEEE Trans. Nucl. Sci., vol. 51, no. 2, pp. 313-321, April 2004.
Antonio C.A. Mol, et al., "Neural and genetic-based approaches to nuclear transient identification including 'don't know' response," Progress in Nuclear Energy, vol. 48, pp. 268-282, 2006.
T.V. Santosh, et al., "Diagnostic system for identification of accident scenarios in nuclear power plants using artificial neural networks," Reliability Engineering and System Safety, vol. 94, pp. 759-762, 2009.
M. G. Na, W. S. Park, and D. H. Lim, "Detection and diagnostics of loss of coolant accidents using support vector machines," IEEE Trans. Nucl. Sci., vol. 55, no. 1, pp. 628-636, Feb. 2008.
S. H. Lee, Y. G. No, M. G. Na, K.-I. Ahn and S.-Y. Park, "Diagnostics of loss of coolant accidents using SVC and GMDH models," IEEE Trans. Nucl. Sci., vol. 58, no. 1, pp. 267-276, Feb. 2011.
I.-Y.Seo, B.-N. Ha, S.-W. Lee, C.-H. Shin, and S.-J. Kim, "Principal components based support vector regression model for on-line instrument calibration monitoring in NPPs," Nucl. Eng. Technol., vol. 42, no. 2, pp. 219-230, Apr. 2010.
E. Zio and R. Bazzo, "Optimization of the test intervals of a nuclear safety system by genetic algorithms, solution clustering and fuzzy preference assignment," Nucl. Eng. Technol., vol. 42, no. 4, pp. 414-425, Aug. 2010.
R. E. Henry et al., MAAP4 - Modular Accident Analysis Program for LWR Power Plants, User's Manual. Burr Ridge, IL: Fauske, vol. 1-4, 1990.
B.-S. Yang, W.-W. Hwang, M.-H. Ko, and S.-J. Lee, "Cavitation detection of butterfly valve using support vector machines," J. Sound Vibr., vol. 287, nos. 1-2, pp. 25-43, Oct. 2005.
A. G. Ivakhnenko, "The group method of data handling; a rival of method of stochastic approximation," Soviet Automatic Control, vol. 1, no. 3, pp. 43-55, 1968.
M. C. Acock and Y. A. Pachepsky, "Estimating missing weather data for agricultural simulations using group method of data handling," J. Applied Meteorology, vol. 39, no. 7, pp. 1176-1184, 2000.
T. Kondo, A. S. Pandya, "GMDH-type neural network algorithm with sigmoid function," Intl. J. Knowledge-Based Engineering Systems, vol. 7, no. 4, pp. 198-205, 2003.
S. J. Farlow, Self-Organizing Methods in Modeling: GMDH Type Algorithms. Marcel Dekker, New York, 1984.
C. R. Hild, "Development of The Group Method of Data Handling With Information-based Model Evaluation Criteria: A New Approach to Statistical Modeling," Ph.D. Dissertation, Univ. Tennessee, Knoxville, 1998.
P. B. Ferreira and B. R. Upadhyaya, Incipient Fault Detection and Isolation of Sensors and Field Devices, Nuclear Engineering Dept., Univ. Tennessee, Knoxville, UTNE/BRU/99-02, December 1999.
A. G. Ivakhnenko, "Polynomial theory of complex systems", IEEE Trans. Syst. Man & Cybern, SMC-1, pp. 364-378, 1971
Jang, J.-S. R., "ANFIS: Adaptive-network-based fuzzy inference systems," IEEE Transactions on Systems, Man, and Cybernetics, vol. 23, pp. 665-685, 1993.
Mamdani, E.H. and Assilian, S., "An experiment in linguistic synthesis with a fuzzy logic controller," Int. J. Man-Machine Studies, vol. 7, pp. 1-13, 1975.
Takagi, T. and Sugeno, M., "Fuzzy identification of systems and its applications to modeling and control," IEEE Trans. System, Man, Cybern., vol. 1, pp. 116-132, 1985.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.