Nuclear power operator supervision optimization method and system based on artificial intelligence technology

Abstract

The invention relates to the field of nuclear power industry automation, in particular to a nuclear power operator monitoring optimization method and system based on an artificial intelligence technology. The method comprises the following steps: collecting operation data of each system of the nuclear power station; constructing a deep learning prediction model by adopting a fusion architecture of a long short-term memory network and a generative adversarial network; the deep learning prediction model is used for automatically detecting abnormal events; according to an output result of the deep learning prediction model, constructing an intelligent risk assessment system, and performing comprehensive risk rating on the abnormal event; automatically generating early warning information according to the comprehensive risk level of the abnormal condition, and pushing the early warning information to a nuclear power station operator in real time to ask the operator to confirm; the deep learning prediction model is continuously updated according to a self-learning and feedback learning mechanism. According to the method, the problems of leakage, misjudgment and the like possibly occurring when the operator monitors the disk under complex working conditions and mass data are reduced, the disk monitoring efficiency of the operator of the nuclear power station is improved, and safe and stable operation of the nuclear power station is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of nuclear power industry automation, and in particular to a nuclear power operator monitoring and optimization method and system based on artificial intelligence technology. BACKGROUND

[0002] Nuclear power plants are highly complex energy facilities with high requirements for safe and stable operation. Nuclear power main control room operators assume the responsibility of operation and monitoring during work to ensure the safe and stable operation of nuclear power plants under various conditions. However, nuclear power plants contain multiple systems with numerous operating parameter limits, which can easily distract the operator's attention and increase the probability of human error to some extent.

[0003] Currently, in the operation of nuclear power plants, the combination of main control room operator monitoring and on-site operator equipment inspection is commonly used to maintain nuclear safety. This method requires both the operator and the on-site operator to have good skill levels. However, there are hundreds of alarms during the operation of nuclear power units, and the impact of different alarms on the unit varies. Some flashing alarms may be covered by other alarms, causing the operator to ignore them, which increases the risk of nuclear power plant operation.

[0004] Nuclear power plants have extremely high safety and reliability requirements, and any abnormal system parameter may impact the reactor system or other auxiliary systems. Currently, system parameter monitoring is mainly completed by nuclear power operators. During work, operators face problems such as monitoring multiple parameters, personnel fatigue, and decreased attention, which may result in human errors due to negligence, affecting the safe operation of nuclear power units.

[0005] To solve the above problems, researchers have attempted to optimize the display interface, set up redundant systems, and set up key parameter overrun alarms to improve the operator's control of the safe and stable operation of nuclear power plants. However, optimizing the display interface can only record some key parameters; setting up redundant systems increases the installation and construction cost of nuclear power; and key parameter overrun alarms indicate that the unit has certain defects, making it difficult to meet the high safety and reliability requirements of nuclear power plants. SUMMARY

[0006] The technical problem to be solved by the present application is to provide a nuclear power operator monitoring and optimization method based on artificial intelligence technology, which introduces artificial intelligence technology into the main control room of a nuclear power plant, compares and analyzes the operation data of the nuclear power plant, identifies abnormal data in advance, and lists potential operation risks, reduces the possibility of leaks and misjudgments by operators when facing complex conditions and massive data, improves the monitoring efficiency of nuclear power plant operators, and ensures the safe and stable operation of nuclear power plants.

[0007] The application provides a nuclear power operator monitoring optimization method based on artificial intelligence technology, comprising the following steps:

[0008] Step 1: collecting operation data of each system of a nuclear power plant;

[0009] Step 2: adopting a long short-term memory network and a generative adversarial network fusion architecture to build a deep learning prediction model; the deep learning prediction model is used for automatically detecting abnormal events;

[0010] Step 3: according to the output result of the deep learning prediction model, an intelligent risk assessment system is built to comprehensively risk rate the abnormal events;

[0011] Step 4: the comprehensive risk level of the abnormal condition is used to automatically generate early warning information, which is pushed to a nuclear power plant operator in real time to require him to confirm;

[0012] Step 5: the deep learning prediction model is continuously updated according to a self-learning and feedback learning mechanism.

[0013] In an embodiment of the application, in step 1, a differentiated sampling mechanism is adopted to collect operation data of each system of a nuclear power plant; for key monitoring signals, a sampling strategy with a frequency of 1Hz or higher is adopted for continuous monitoring in the whole period;

[0014] For other analog signals, a dynamic response mechanism is established to adjust the sampling speed in real time according to the actual demand of sudden conditions and the grade of sudden events;

[0015] When it is detected that the operation data exceeds a safety threshold, the collection frequency is automatically increased.

[0016] In an embodiment of the application, in step 2,

[0017] A long short-term memory network is used to establish a feature model of normal data to capture the long-term dependence relationship between data in a time sequence;

[0018] A generative adversarial network discriminates abnormal data according to the feature model, and finally forms a deep learning prediction model.

[0019] In an embodiment of the application, in step 5,

[0020] Through a back propagation algorithm, the parameters of the deep learning prediction model are dynamically optimized to reduce the error between the prediction result and the actual situation;

[0021] Real-time feedback data and historical event data are used to retrain the deep learning prediction model to dynamically adjust the parameters of the deep learning prediction model to more accurately capture the potential patterns and rules of events.

[0022] In a specific embodiment of the present application, the comprehensive risk rating Z obtained from the comprehensive risk rating in step 3 is calculated by the following formula:

[0023]

[0024] Wherein: Z represents the comprehensive risk rating, N is the total number of affected subsystems, P k represents the probability of the kth subsystem being affected, ranging from 0 to 1, I k represents the degree of influence of the kth subsystem, represented by a level of 1 to 10, with a higher value indicating greater influence.

[0025] The present application provides a nuclear power operator monitoring and optimization system based on artificial intelligence technology, comprising:

[0026] A distributed sensor network module for collecting nuclear power plant system operation data;

[0027] A deep learning prediction module for building a deep learning prediction model using a long short-term memory network and a generative adversarial network fusion architecture;

[0028] A risk assessment and reporting module for comprehensive risk rating of the output results of the deep learning prediction model and real-time pushing to the nuclear power plant operator;

[0029] A feedback and optimization module for optimizing the deep learning prediction model.

[0030] In a specific embodiment of the present application, the distributed sensor network module differentiates the collection of operation data, adopts a redundant design, and transmits data between the deep learning prediction module and the feedback and optimization module through the Profinet IRT high-precision synchronization protocol.

[0031] In a specific embodiment of the present application, the deep learning prediction module distinguishes abnormal parameters of different systems and generates corresponding prediction results and emergency response strategies according to the source and value of the abnormal parameters.

[0032] In a specific embodiment of the present application, the risk assessment and reporting module receives the results of the deep learning prediction module and divides the abnormal events into comprehensive risk ratings according to the risk assessment system;

[0033] Automatic generation of early warning information through comprehensive risk rating and real-time pushing to the nuclear power plant operator.

[0034] In a specific embodiment of the present application, the feedback and optimization module continuously optimizes the prediction model and emergency response strategy of the system through self-learning and feedback cycles;

[0035] The self-learning includes: learning the processing result of the abnormal event, continuously dynamically optimizing the model parameters through the back propagation algorithm to reduce the error between the prediction result and the actual situation;

[0036] The feedback loop includes: synchronously feeding back the real-time feedback data and the historical event data to the deep learning prediction model, and the deep learning prediction model carries out secondary training according to the data to dynamically adjust the parameters in the model.

[0037] Compared with the prior art, the nuclear power operator monitoring and optimization method and system based on artificial intelligence technology have the following beneficial effects:

[0038] (1) By integrating a high-precision distributed sensor network, all-round real-time monitoring of key parts of each subsystem of the nuclear power plant is implemented;

[0039] (2) The deep learning prediction model improves the accuracy and response speed of the early warning, and when the sensor detects an anomaly, it can quickly identify the source of the anomaly and analyze the potential risks;

[0040] (3) The real-time feedback mechanism can ensure that the nuclear power plant operator understands the relevant information in the first time. The feedback and optimization mechanism continuously improves the intelligent level of the system through self-learning, optimizes the prediction model and the emergency response strategy, and ensures the safety and stability of the nuclear power plant under different operating conditions.

[0041] Overall, the present application significantly improves the safety and reliability of the operation of the nuclear power plant in practical application. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The flowchart shows the nuclear power operator monitoring and optimization method based on artificial intelligence technology. DETAILED DESCRIPTION

[0043] In order to further understand the present application, the embodiments of the present application are described below in conjunction with the examples, but it should be understood that these descriptions are only for further illustrating the features and advantages of the present application, and are not limitations of the present application.

[0044] The embodiments of the present application disclose a nuclear power operator monitoring and optimization method based on artificial intelligence technology, as shown in Figure 1 The method comprises the following steps:

[0045] Step 1: Collecting the operation data of each system of the nuclear power plant;

[0046] The collected data includes temperature, pressure, flow, liquid level and other parameters. By deploying high-precision real-time monitoring equipment at key points in each system of the nuclear power plant, such as the main pipe of the primary loop, the inlet and outlet of the steam generator, and the inlet of the turbine main steam valve, batch real-time collection of various operating data is realized, ensuring the timeliness and integrity of the data.

[0047] The differential sampling mechanism is used to collect operating data of each system of the nuclear power plant. For key monitoring signals, a sampling strategy with a frequency of 1 Hz or higher is used for continuous monitoring throughout the period.

[0048] For other analog signals, a dynamic response mechanism is established to adjust the sampling speed in real time according to the actual needs of sudden conditions and the level of sudden events, effectively dealing with complex fluctuations and changes during the operation of the nuclear power unit.

[0049] When the operating data such as temperature and pressure exceed the safety threshold, the collection frequency is automatically increased, and subsequent analysis is triggered simultaneously to quickly identify the root cause of the fault and ensure the safe and stable operation of the nuclear power system.

[0050] The concept of redundancy design is used to configure backup sensors for key monitoring parameters. During operation, the system monitors the sensor channels in real time. Once a channel is detected to have a fault or the collected data is abnormal, it will automatically switch to the backup channel quickly, ensuring the continuity and accuracy of the monitoring data. At the same time, the fault information is recorded in detail in the log, which includes the time and type of the fault, providing detailed evidence for subsequent equipment maintenance. To solve the problem that noise and vibration signals are easily disturbed by the environment during monitoring, the system uses advanced filtering technology. By using adaptive filtering or band-pass filtering algorithms, it can effectively filter out environmental noise interference and accurately extract and retain the characteristic frequency bands that may be related to parameter abnormalities and equipment failures, helping operators quickly discover potential equipment hazards and significantly improve the accuracy and timeliness of fault diagnosis.

[0051] In the data synchronization link, high-precision time synchronization protocols such as Profinet IRT (Isochronous Real-Time) are used to mark the time stamp of the multi-channel data collected by the sensor. With this method, the timing deviation of the readings in the transmission and processing process can be significantly reduced, ensuring the consistency of the data time dimension. On the basis of giving time stamp to multi-channel data, the key monitoring data is uniformly time-aligned, which provides high-quality and timely data support for subsequent deep learning prediction model training and prediction, helps the model to capture the internal rules between data, and improves the prediction accuracy. In the process of real-time monitoring and collection, once the network jitter or delay is detected, the buffer and queue mechanism is immediately enabled to rearrange or retransmit the data, ensuring the completeness, accuracy and reliability of the information used for real-time prediction and control.

[0052] Through the above multi-dimensional collection, redundant design and time synchronization, a perfect data support system is built, ensuring the continuity and accuracy of system parameter monitoring, and adapting to the complex and variable working conditions of nuclear power plants, laying a data foundation for real-time and accurate monitoring of system state.

[0053] Step 2: A deep learning prediction model is built using a long short-term memory network (LSTM) and a generative adversarial network (GAN) fusion architecture; the deep learning prediction model is used for automatic detection of abnormal events;

[0054] A feature model of normal data is established by the long short-term memory network to capture the long-term dependence relationship between data in the time series;

[0055] The generative adversarial network discriminates abnormal data according to the feature model, and finally forms a deep learning prediction model.

[0056] The deep learning prediction model can take advantage of fusion, simultaneously analyze real-time data and historical data, analyze and predict the causes of abnormal data and the possible impact range. Among them, the feature model of normal data is established by LSTM to capture the long-term dependence relationship between data, and the discriminator in GAN discriminates abnormal data according to the feature model constructed by LSTM. This model architecture can distinguish abnormal parameters of different systems, and generate corresponding prediction results and emergency response strategies according to the source and value of abnormal parameters.

[0057] The deep learning model used in the present application integrates the advantages of LSTM and GAN. LSTM is used to capture the long-term dependence relationship in the time series, and its basic formula is as follows:

[0058] i t =σ(W i ·x t +Ui ·h t-1 +b i (1)

[0059] o t =σ(W o ·x t +U o ·h t-1 +b o (2)

[0060] f t =σ(W f ·x t +U f ·h t-1 +b f (3)

[0061]

[0062] h t =o t *tanh(C t (6)

[0063] Among them, i t For the input gate, o t For output gate, f t For the Gate of Oblivion, C t In cellular state, h t Indicates the hidden state, x t This is the current input. σ is the sigmoid function, W i W o W f and W c For the input weights, U i U o U f and U c To be in a hidden state, b i b o b f and b c This is a bias term.

[0064] GANs are used for adversarial training on data, and their optimization objective function is:

[0065]

[0066] Where G is the generator, D is the discriminator, and p data (x) is the distribution of the real data x, p z (z) is the distribution of noise z, and IE represents the true expectation.

[0067] By combining LSTM and GAN, we can accurately characterize the normal state of the system and effectively detect abnormal states, thereby improving the accuracy and reliability of predictions.

[0068] Step 3: Based on the output of the deep learning prediction model, construct an intelligent risk assessment system to conduct a comprehensive risk rating of abnormal events;

[0069] The overall risk level Z obtained from the comprehensive risk rating is calculated using the following formula:

[0070]

[0071] Where: Z represents the overall risk level, N is the total number of affected subsystems, and P... k I represents the probability that the k-th subsystem is affected, with a value ranging from 0 to 1. k This indicates the degree of impact on the k-th subsystem, expressed in levels from 1 to 10, with higher values ​​indicating greater impact.

[0072] Step 4: Automatically generate early warning information based on the comprehensive risk level of the abnormal situation and push it to the nuclear power plant operator in real time for confirmation;

[0073] Step 5: The deep learning prediction model is continuously updated based on self-learning and feedback learning mechanisms.

[0074] By leveraging the self-learning mechanism of end-to-end intelligent systems, the results of event processing are analyzed, thereby optimizing prediction results and emergency response strategies.

[0075] The learning mechanism updates model parameters based on the following formula:

[0076]

[0077] Where L is the loss function, η is the learning rate, and θ is the model parameters. The backpropagation algorithm continuously and dynamically optimizes the model parameters to reduce the error between the predicted results and the actual situation, thereby improving the model's prediction accuracy.

[0078] By using real-time feedback data and historical event data to retrain the deep learning prediction model, the parameters of the deep learning prediction model are dynamically adjusted to more accurately capture the potential patterns and rules of events.

[0079] Specifically:

[0080] The system uses a feedback mechanism to transmit various operational data, such as pipeline pressure, tank liquid level, and fluid flow rate, to the deep learning prediction model in real time. The transmitted data includes information from both normal and abnormal states, enabling the deep learning prediction model to understand the system's operating status under different conditions. Furthermore, the system stores historical event data and transmits it to the deep learning prediction model to facilitate trend analysis and anomaly identification. By integrating historical and real-time data, the deep learning prediction model can identify potential risks and provide corresponding solutions.

[0081] After each exception is handled, a detailed report is automatically generated, recording the source of the exception, the handling method, and the result, and is fed back to the operators. The report content is used to further analyze and optimize the next emergency response plan, continuously improving the overall performance of the system.

[0082] A specific embodiment of the present invention discloses a nuclear power plant operator monitoring optimization system based on artificial intelligence technology, comprising:

[0083] Distributed sensor network module, used to collect operational data from various systems in nuclear power plants;

[0084] The distributed sensor network module collects operational data in a differentiated manner, adopts a redundant design, and transmits data with the deep learning prediction module and the feedback and optimization module through the Profinet IRT high-precision synchronization protocol.

[0085] The collected data includes parameters such as temperature, pressure, flow rate, and liquid level. By deploying high-precision real-time monitoring equipment at key locations in various systems of the nuclear power plant, such as the primary loop main pipeline, steam generator inlet and outlet, and turbine main steam valve inlet, various types of operational data can be collected in batches in real time, ensuring the timeliness and completeness of the data.

[0086] A differentiated sampling mechanism is used to collect operational data from various systems of the nuclear power plant. For key monitoring signals, a sampling strategy with a frequency of 1Hz or higher is used for continuous monitoring throughout the day.

[0087] For other analog signals, a dynamic response mechanism is established to adjust the sampling rate in real time according to the actual needs of the emergency and the level of the emergency, so as to effectively cope with the complex fluctuations and changes in the operation of the nuclear power unit.

[0088] When operating data such as temperature and pressure are detected to exceed safety thresholds, the acquisition frequency is automatically increased, and subsequent analysis is triggered simultaneously to quickly carry out emergency response, identify the root cause of the fault in the shortest possible time, and ensure the safe and stable operation of the nuclear power system.

[0089] The deep learning prediction module, a core component of the system's intelligent analysis, is used to build a deep learning prediction model using a fusion architecture of Long Short-Term Memory (LSTM) and Generative Adversarial Networks (GANs). This model fully leverages the advantages of this fusion, simultaneously processing real-time and historical data. It can not only accurately identify abnormal data but also deeply analyze the causes of anomalies and predict the potential scope of impact. LSTM is primarily responsible for building a feature model of normal data, establishing a dynamic model of data changes by mining long-term dependencies between data points. The discriminator in the GAN then uses the feature model built by LSTM to systematically identify each point in the input data, accurately locating abnormal data. This unique architectural design enables the model to distinguish different abnormal parameters of the system and automatically generate targeted prediction reports and emergency response plans based on the attribute characteristics of these abnormal parameters.

[0090] The risk assessment and reporting module is used to perform comprehensive risk rating on the output results of the deep learning prediction model and push them to the nuclear power plant operators in real time. It integrates fuzzy neural network and rule engine technology and relies on the output results of the deep learning prediction model. It is the core control unit of the system.

[0091] The risk assessment and reporting module receives the results from the deep learning prediction module and classifies the abnormal events into comprehensive risk levels according to the risk assessment system.

[0092] Early warning information is automatically generated based on the comprehensive risk level and pushed to nuclear power plant operators in real time.

[0093] The feedback and optimization module is used to optimize deep learning prediction models.

[0094] The feedback and optimization module continuously optimizes the system's prediction model and emergency response strategy through self-learning and feedback loops, thereby improving the overall performance of the system and its ability to adapt to different working conditions.

[0095] Self-learning includes: learning from the results of handling abnormal events and continuously optimizing the model parameters dynamically through backpropagation algorithm to reduce the error between the prediction results and the actual situation;

[0096] The feedback loop includes: synchronously feeding real-time feedback data and historical event data back to the deep learning prediction model, and the deep learning prediction model conducts secondary training based on this data to dynamically adjust the parameters within the model.

[0097] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0098] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing nuclear power plant operator monitoring based on artificial intelligence technology, characterized in that, Includes the following steps: Step 1: Collect operational data from various systems of the nuclear power plant; Step 2: A deep learning prediction model is built using a fusion architecture of Long Short-Term Memory Network and Generative Adversarial Network; the deep learning prediction model is used to automatically detect abnormal events; Step 3: Based on the output of the deep learning prediction model, construct an intelligent risk assessment system to conduct a comprehensive risk rating of abnormal events; Step 4: Automatically generate early warning information based on the comprehensive risk level of the abnormal situation and push it to the nuclear power plant operator in real time for confirmation; Step 5: The deep learning prediction model is continuously updated based on self-learning and feedback learning mechanisms.

2. The nuclear power plant operator monitoring optimization method based on artificial intelligence technology according to claim 1, characterized in that, In step 1, a differentiated sampling mechanism is used to collect operating data of various systems in the nuclear power plant. For key monitoring signals, a sampling strategy with a frequency of 1Hz or higher is used for continuous monitoring throughout the day. For other analog signals, a dynamic response mechanism is established to adjust the sampling rate in real time according to the actual needs of the emergency and the level of the emergency. The collection frequency is automatically increased when the detected running data exceeds the safety threshold.

3. The nuclear power plant operator monitoring optimization method based on artificial intelligence technology according to claim 1, characterized in that, In step 2, By establishing a feature model of normal data through long short-term memory networks, long-term dependencies between data in time series can be captured. Generative adversarial networks (GANs) identify anomalous data based on the feature model and ultimately form a deep learning prediction model.

4. The nuclear power plant operator monitoring optimization method based on artificial intelligence technology according to claim 3, characterized in that, In step 5 By using the backpropagation algorithm, the parameters of the deep learning prediction model are continuously and dynamically optimized to reduce the error between the prediction results and the actual situation. By using real-time feedback data and historical event data to retrain the deep learning prediction model, the parameters of the deep learning prediction model are dynamically adjusted to more accurately capture the potential patterns and rules of events.

5. The nuclear power plant operator monitoring optimization method based on artificial intelligence technology according to claim 1, characterized in that, In step 3, the comprehensive risk level Z obtained from the comprehensive risk rating is calculated using the following formula: Where: Z represents the overall risk level, N is the total number of affected subsystems, and P... k I represents the probability that the k-th subsystem is affected, with a value ranging from 0 to 1. k This indicates the degree of impact on the k-th subsystem, expressed in levels from 1 to 10, with higher values ​​indicating greater impact.

6. A nuclear power plant operator monitoring and optimization system based on artificial intelligence technology, characterized in that, include: Distributed sensor network module, used to collect operational data from various systems in nuclear power plants; The deep learning prediction module is used to build deep learning prediction models using a fusion architecture of long short-term memory networks and generative adversarial networks. The risk assessment and reporting module is used to perform a comprehensive risk rating on the output of the deep learning prediction model and push it to the nuclear power plant operator in real time. The feedback and optimization module is used to optimize deep learning prediction models.

7. The nuclear power plant operator monitoring and optimization system based on artificial intelligence technology according to claim 6, characterized in that, The distributed sensor network module collects operational data in a differentiated manner, adopts a redundant design, and transmits data with the deep learning prediction module and the feedback and optimization module through the Profinet IRT high-precision synchronization protocol.

8. The nuclear power plant operator monitoring and optimization system based on artificial intelligence technology according to claim 6, characterized in that, The deep learning prediction module distinguishes abnormal parameters of different systems and generates corresponding prediction results and emergency response strategies based on the source and value of the abnormal parameters.

9. The nuclear power plant operator monitoring and optimization system based on artificial intelligence technology according to claim 6, characterized in that, The risk assessment and reporting module receives the results from the deep learning prediction module and classifies the abnormal events into comprehensive risk levels according to the risk assessment system. Early warning information is automatically generated based on the comprehensive risk level and pushed to nuclear power plant operators in real time.

10. The nuclear power plant operator monitoring and optimization system based on artificial intelligence technology according to claim 6, characterized in that, The feedback and optimization module continuously optimizes the system's prediction model and emergency response strategy through self-learning and feedback loops. Self-learning includes: learning from the results of abnormal event handling and continuously optimizing model parameters through backpropagation to reduce the error between prediction results and actual situation; feedback loop includes: synchronously feeding real-time feedback data and historical event data back to the deep learning prediction model, and conducting secondary training based on this data to dynamically adjust the parameters within the model.

Images