Nuclear facility probability safety evaluation method based on artificial intelligence

Abstract

The invention relates to a nuclear facility probability safety evaluation method based on artificial intelligence. The method comprises the following steps: collecting data related to nuclear facility operation history; preprocessing the collected data related to the nuclear facility operation history; selecting and designing a plurality of probabilistic safety evaluation deep learning models; training a plurality of probabilistic safety evaluation deep learning models by using the preprocessed data related to the nuclear facility operation history; evaluating the plurality of trained probabilistic safety evaluation deep learning models to select an optimal deep learning model; deploying an optimal deep learning model; and applying the optimal deep learning model to an in-service or new nuclear facility, comparing model output with a result of a traditional PSA method, and determining whether to continue to perform improvement training on the optimal deep learning model according to a comparison result. According to the method disclosed by the invention, the deep learning model is introduced into a traditional nuclear facility probabilistic safety evaluation method, so that the accuracy and efficiency of nuclear facility probabilistic safety evaluation can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of probabilistic safety assessment of nuclear facilities, and specifically relates to a probabilistic safety assessment method for nuclear facilities based on artificial intelligence. Background Art

[0002] The publication of the landmark Reactor Safety Study WASH-1400 in 1975 clearly demonstrated that the probabilistic assessment method has great potential to improve the management and license application of nuclear power plants by focusing on items of safety importance. The Three Mile Island accident in the United States in 1979 confirmed the research results of WASH-1400. Since then, the U.S. Nuclear Regulatory Commission (NRC) and the nuclear energy industry have been committed to developing the mainstream probabilistic safety assessment (PSA) method for operating reactors and promoting risk-guided management based on PSA.

[0003] The PSA method is a system analysis method based on probability theory. It is one of the two safety analysis methods for nuclear facilities. It is good at analyzing the relationship and interaction between various factors. It can quantitatively evaluate the safety of nuclear power plants and other nuclear fuel cycle facilities, identify weak links in the design, construction and operation of nuclear facilities, and put forward improvement suggestions to ensure the safe operation of nuclear facilities.

[0004] The PSA method generally includes identifying initiating events and determining their frequency of occurrence, accident sequence analysis, system analysis, common cause failure analysis, personnel reliability analysis, importance analysis, uncertainty analysis and sensitivity analysis.

[0006] Introducing AI models such as deep learning into the traditional PSA method for nuclear facilities has become a general trend and will definitely improve the operational efficiency of the PSA method for nuclear facilities. Summary of the invention

[0007] In view of the defects existing in the prior art, the purpose of the present invention is to provide a probabilistic safety assessment method for nuclear facilities based on artificial intelligence. In line with the development trend of artificial intelligence, a deep learning model is introduced into the traditional probabilistic safety assessment method for nuclear facilities. This can make full and effective use of the historical operating data of nuclear facilities, improve the accuracy and efficiency of the probabilistic safety assessment of nuclear facilities, and reduce labor costs.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is: a nuclear facility probabilistic safety assessment method based on artificial intelligence, the method comprising the following steps:

[0009] S1. Collect data related to the operation history of nuclear facilities;

[0010] S2. Preprocessing the collected data related to the operation history of nuclear facilities;

[0011] S3. Select and design multiple probabilistic safety assessment deep learning models;

[0012] S4. Training multiple probabilistic safety assessment deep learning models using pre-processed data related to the nuclear facility operation history;

[0013] S5. Evaluate multiple trained probabilistic safety assessment deep learning models to select the optimal deep learning model;

[0014] S6. Deploy the optimal deep learning model on the hardware device;

[0015] S7. Use the optimal deep learning model for existing or newly built nuclear facilities, compare the output of the optimal deep learning model with the results of traditional probabilistic safety assessment methods, and determine whether to continue to improve the training of the optimal deep learning model based on the comparison results.

[0016] Furthermore, the data related to the operation history of the nuclear facility in step S1 includes various reliability parameters of basic events.

[0017] Furthermore, the data collection methods related to the operation history of the nuclear facility in step S1 include nuclear facility-specific data sources and similar data of other nuclear facilities of the same type.

[0018] Further, nuclear facility specific data sources include incident reports, maintenance reports, work orders, maintenance logs, master control logs, and operator logs.

[0019] Furthermore, step S2 includes performing statistical analysis on the reliability parameters using classical probability statistics or Bayesian statistical reasoning.

[0020] Furthermore, the output of the probabilistic safety assessment deep learning model in step S3 includes accident sequence analysis, system analysis, common cause failure analysis, personnel reliability analysis, importance analysis, uncertainty analysis and sensitivity analysis.

[0021] Furthermore, in step S4, batch regularization technology is used to train multiple probabilistic safety assessment deep learning models.

[0022] Furthermore, in step S4, the prediction results of multiple probabilistic safety assessment deep learning models are combined as the prediction result of the probabilistic safety assessment deep learning model.

[0023] Further, step S6 includes using different reasoning frameworks to perform reasoning conversion on the optimal deep learning model.

[0024] Furthermore, the input of the optimal deep learning model in step S7 is the reliability parameters of basic events specific to the nuclear facility.

[0025] The beneficial technical effect of the present invention is that the artificial intelligence-based probabilistic safety assessment method for nuclear facilities disclosed by the present invention conforms to the development trend of artificial intelligence, introduces a deep learning model into the traditional probabilistic safety assessment method for nuclear facilities, and can fully and effectively utilize the historical operating data of nuclear facilities, thereby improving the accuracy and efficiency of the probabilistic safety assessment of nuclear facilities and reducing labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a probabilistic safety assessment method for nuclear facilities based on artificial intelligence, shown in Example 1 of the present invention. DETAILED DESCRIPTION

[0027] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0028] Embodiment 1

[0029] like Figure 1 As shown, an embodiment of the present invention provides a nuclear facility probabilistic safety assessment method based on artificial intelligence, and the method comprises the following steps:

[0030] S1. Collect data related to the operating history of nuclear facilities.

[0031] In the PSA analysis process, the reliability model of basic events is indispensable, and parameter estimates need to be provided to determine the reliability parameters of various basic events involved in the PSA analysis model. The reliability parameters of various basic events include: the frequency of initiating events, equipment unavailability, and the probability of human error, etc. Equipment unavailability includes unavailability caused by hardware and unavailability caused by factors such as testing and maintenance.

[0032] Collect data related to the operation history of nuclear facilities, including various reliability parameters of basic events.

[0033] The means of collecting data related to the operating history of nuclear facilities include nuclear facility-specific data sources: event reports, maintenance reports, work orders, maintenance logs, master control logs, and operator logs; and also include similar data from other nuclear facilities of the same type.

[0034] S2. Preprocess the collected data related to the operation history of nuclear facilities.

[0035] Reliability parameters include both discrete random variables and continuous random variables. Discrete random variables obey binomial distribution, Poisson distribution, hypergeometric distribution and other distribution types; continuous random variables obey uniform distribution, normal distribution, exponential distribution, Weibull distribution and other distribution types. Statistical analysis of reliability parameters can be performed by classical probability statistics or Bayesian statistical reasoning. Then, the collected data related to the operation history of nuclear facilities are preprocessed, and the data are converted into a consistent format through data cleaning (removing duplicate data, removing outliers, filling missing values), aggregation and change, and considering issues such as data normalization, data segmentation, and data labeling.

[0036] S3. Select and design multiple probabilistic safety assessment deep learning models.

[0037] Select and design a deep learning model based on the PSA tasks of nuclear facilities and the characteristics of data related to the operation history of nuclear facilities. First, select a suitable network structure. Common network structures include convolutional neural networks, recurrent neural networks, self-attention models, etc. In general, the first two can be selected. Second, select the appropriate number of network layers and width. Because PSA itself has a large amount of calculation, it is necessary to select the number of network layers and width based on the computing power available, and make a balance between duration and accuracy.

[0038] The unique design of the model output layer can directly produce important quantitative analysis results such as accident sequence analysis, system analysis, common cause failure analysis, personnel reliability analysis, importance analysis, uncertainty analysis and sensitivity analysis.

[0039] S4. Training multiple probabilistic safety assessment deep learning models using pre-processed data related to the nuclear facility operation history;

[0040] Stop training the model when the model performance stops improving.

[0041] Use a pre-trained model as the initial model to speed up the training process.

[0042] The pre-trained model is applied to various types of probabilistic safety assessment of nuclear facilities to enable its application to a wider range.

[0043] Batch regularization techniques can help train more stable and accurate models.

[0044] Combine the prediction results of multiple models to obtain more accurate prediction results.

[0045] S5. Evaluate multiple trained probabilistic safety assessment deep learning models to select the optimal deep learning model;

[0046] Evaluate the model using some evaluation indicators and compare the performance of different models to select the best model. Adjust model parameters to obtain more accurate prediction results. Randomly search for hyperparameters without trying all possible hyperparameter combinations.

[0047] The adjusted model parameters include the model's learning rate, regularization coefficient, batch size and other parameters.

[0048] S6. Deploy the optimal deep learning model on the hardware device.

[0049] Taking the PSA task as the final requirement, the model is converted for reasoning using different reasoning frameworks and deployed on appropriate hardware devices.

[0050] S7. Use the optimal deep learning model for existing or newly built nuclear facilities, compare the output of the optimal deep learning model with the results of traditional probabilistic safety assessment methods, and determine whether to continue to improve the training of the optimal deep learning model based on the comparison results.

[0051] The deployed model is applied to existing or newly built nuclear facilities, and the specific reliability parameters of the facility are used to obtain important quantitative analysis results such as PSA accident sequence analysis, system analysis, common cause failure analysis, personnel reliability analysis, importance analysis, uncertainty analysis and sensitivity analysis.

[0052] The results of the AI ​​method will be compared with the traditional PSA results from various major analysis dimensions. Not only does the PSA project team need to be deeply involved in this step, but industry practitioners and field experts are also required to discuss and review it together, and determine whether to continue to improve the model training based on the comparison results.

[0053] It can be seen from the above embodiments that the artificial intelligence-based probabilistic safety assessment method for nuclear facilities disclosed in the present invention conforms to the development trend of artificial intelligence. In the traditional probabilistic safety assessment method for nuclear facilities, a deep learning model is introduced, which can fully and effectively utilize the operating historical data of nuclear facilities, improve the accuracy and efficiency of the probabilistic safety assessment of nuclear facilities, reduce labor costs, and improve the operating benefits of the PSA method for nuclear facilities.

[0054] The method described in the present invention is not limited to the embodiments described in the specific implementation manner. Those skilled in the art may derive other implementation manners based on the technical solution of the present invention, which also fall within the technical innovation scope of the present invention.

Claims

1. A probabilistic safety assessment method for nuclear facilities based on artificial intelligence, the method comprising the following steps: S1. Collect data related to the operation history of nuclear facilities; S2. Preprocessing the collected data related to the operation history of nuclear facilities; S3. Select and design multiple probabilistic safety assessment deep learning models; S4. Training multiple probabilistic safety assessment deep learning models using pre-processed data related to the nuclear facility operation history; S5. Evaluate multiple trained probabilistic safety assessment deep learning models to select the optimal deep learning model; S6. Deploy the optimal deep learning model on the hardware device; S7. Use the optimal deep learning model for existing or newly built nuclear facilities, compare the output of the optimal deep learning model with the results of traditional probabilistic safety assessment methods, and determine whether to continue to improve the training of the optimal deep learning model based on the comparison results.

2. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that: The data related to the operation history of the nuclear facility in step S1 include various reliability parameters of basic events.

3. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that: The data collection methods related to the operation history of the nuclear facility in step S1 include nuclear facility-specific data sources and similar data of other nuclear facilities of the same type.

4. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 3, characterized in that: Nuclear facility specific data sources include incident reports, maintenance reports, work orders, maintenance logs, master control logs, and operator logs.

5. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that: Step S2 includes performing statistical analysis on the reliability parameters using classical probability statistics or Bayesian statistical reasoning.

6. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that: The output of the probabilistic safety assessment deep learning model in step S3 includes accident sequence analysis, system analysis, common cause failure analysis, personnel reliability analysis, importance analysis, uncertainty analysis and sensitivity analysis.

7. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that : In step S4, batch regularization technology is used to train multiple probabilistic safety evaluation deep learning models.

8. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that: In step S4, the prediction results of multiple probabilistic safety assessment deep learning models are combined as the prediction result of the probabilistic safety assessment deep learning model.

9. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that: Step S6 includes using different reasoning frameworks to perform reasoning conversion on the optimal deep learning model.

10. The artificial intelligence-based nuclear facility probabilistic safety assessment method according to claim 1, characterized in that: The input of the optimal deep learning model in step S7 is the reliability parameters of the basic events specific to the nuclear facility.

Images