Journal of Radiation Research and Radiation Processing, Volume. 41, Issue 4, 040601(2023)

Consequence prediction in nuclear transport explosion accident using long short-term memory network

Lingpan RUAN1...2, Chunhua CHEN1,*, Liwei CHEN3, Fang RUAN1,2, Xiajuan LI2 and Jianye WANG1 |Show fewer author(s)
Author Affiliations
  • 1Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, China
  • 2University of Science and Technology of China, Hefei 230026, China
  • 3Hefei Normal University, Hefei 230026, China
  • show less
    Figures & Tables(15)
    Technical route
    LSTM unit ("⊙" denotes the hadamard product, "+" denotes the matrix addition)
    Stacked LSTM network
    Underlying surface model
    Horizontal distribution of Pu-239 concentration at different time (in the XY plane where Z=200 m) (color online)
    Horizontal distribution of Pu-239 concentration at 10 min (in the XY plane where Z =200 m)
    Input data format
    Nuclide concentration in area A and area B
    Comparison of prediction results in area A and area B
    Comparison of test set prediction results in area A and area B
    Comparison of loss convergence results in area A and area B
    • Table 1. Preliminary data from 780 s to 815 s in area A

      View table
      View in Article

      Table 1. Preliminary data from 780 s to 815 s in area A

      时间 / s浓度 / (Bq∙m-3)相对压强 / PaX轴风速 / (m∙s-1)Y轴风速 / (m∙s-1)Z轴风速 / (m∙s-1)
      TimeConcentrationRelative pressureWind speed-XWind speed-YWind speed-Z
      78019.187 110.3369 060.300 3331.239 8890.040 355
      78520.816 610.3372 670.300 5991.239 9670.040 489
      79022.555 580.3376 520.300 8721.240 0330.040 625
      79524.408 330.3379 430.301 1491.240 1110.040 765
      80026.379 010.3381 820.301 4291.240 1670.040 907
      80528.471 600.3384 760.301 7141.240 1780.041 052
      81030.689 790.3386 630.301 9961.240 1890.041 199
      81533.036 920.3387 990.302 2801.240 1670.041 349
    • Table 2. Model performance comparison based on grid search

      View table
      View in Article

      Table 2. Model performance comparison based on grid search

      全连接层数

      Dense layers

      平均绝对误差 / (Bq∙m-3)

      MAE

      平均绝对百分比误差 / (Bq∙m-3)

      MAPE

      LSTM层数

      LSTM layers

      24682468

      单元数=4

      Units = 4

      120.348 731.069 536.874 238.786 316.607 526.246 529.982 831.631
      223.620 719.986 629.24740.358 319.562 916.609 625.029 433.310 1
      312.3624.045 34.02915.346 510.433 43.514 13.283 412.756 7

      单元数=8

      Units = 8

      110.047 429.5057.370 412.775 68.583 124.929 66.314 810.769 6
      213.893 86.885 78.254 318.402 911.856 85.804 67.172 115.239 7
      317.251 518.29911.066 927.215 315.028 915.568 39.00223.059 8

      单元数=16

      Units = 16

      15.618 78.727 54.187 99.550 34.755 77.639 33.576 99.994 5
      23.481 45.687 85.962 61.812 92.949 74.923 53.998 81.561 6
      34.682 39.550 56.833 64.498 13.921 58.271 75.713 75.238 3

      单元数=32

      Units = 32

      110.564 55.629 79.157.836 89.2924.905 78.067 86.929 4
      27.204 43.873 19.700 56.573 25.9293.357 68.520 65.676 6
      38.94511.334 53.335 710.249 57.743 99.892 52.789 88.959 6

      单元数=64

      Units = 16

      14.863 78.404 57.964 35.635 54.080 87.281 26.771 94.728
      26.1287.228 87.5265.033 15.131 66.311 56.345 94.352 6
      36.610 410.830 39.958 93.846 15.643 39.118 48.681 92.523 4
    • Table 3. Model running time comparison based on grid search

      View table
      View in Article

      Table 3. Model running time comparison based on grid search

      全连接层数Dense layers时间 / ms Time

      LSTM层数

      LSTM layers

      2468

      单元数=4

      Units = 4

      12 8864 6456 2848 990
      23 0494 6976 2168 952
      32 9714 6636 3268 779

      单元数=8

      Units = 8

      13 0054 8136 4119 167
      23 0704 8126 5439 231
      33 05 44 8716 6179 277

      单元数=16

      Units = 16

      13 4645 6737 71816 380
      23 3965 6557 78118 768
      33 3495 6859 67017 667

      单元数=32

      Units = 32

      1162 171225 290247 738436 999
      2184 227206 015254 425438 195
      3197 060210 831263 081425 058

      单元数=64

      Units = 64

      1208 019214 643420 245427 670
      2204 623215 532421 523431 274
      3210 504215 880419 631433 215
    • Table 4. Specific values of test set prediction results and true values in area A and area B

      View table
      View in Article

      Table 4. Specific values of test set prediction results and true values in area A and area B

      区域A Area A区域B Area B

      时间 / s

      Time

      预测值 / (Bq∙m-3)

      Prediction

      CFD模拟值 / (Bq∙m-3)

      CFD simulation

      相对误差 / %

      Relative error

      时间 / s

      Time

      预测值 / (Bq∙m-3)

      Prediction

      CFD模拟值 / (Bq∙m-3)

      CFD simulation

      相对误差 / %

      Relative error

      1 620142.495140.6671.3038049.901 950.252 40.70
      1 625138.726136.5521.5938548.71549.076 10.74
      1 630134.597132.3231.7239047.487 347.875 60.81
      1 635130.188127.991.7239546.211 546.648 90.94
      1 640125.486123.5651.5540044.899 545.401 21.11
      1 645120.343119.0611.0840543.535 344.139 51.37
      1 650114.843114.4920.3141042.123 842.870 41.74
      1 655109.237109.8750.5841540.691 341.598 72.18
      1 660103.74105.2231.4142039.255 240.327 22.66
      1 66598.089100.5572.4542537.858 839.060 23.08
      1 67092.563195.88823.4743036.672 537.801 12.99
    Tools

    Get Citation

    Copy Citation Text

    Lingpan RUAN, Chunhua CHEN, Liwei CHEN, Fang RUAN, Xiajuan LI, Jianye WANG. Consequence prediction in nuclear transport explosion accident using long short-term memory network[J]. Journal of Radiation Research and Radiation Processing, 2023, 41(4): 040601

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Research Articles

    Received: Mar. 1, 2023

    Accepted: Mar. 21, 2023

    Published Online: Sep. 21, 2023

    The Author Email:

    DOI:10.11889/j.1000-3436.2023-0016

    Topics