A method, system, device and storage medium for risk prediction and adaptive scheduling of a nuclear power plant supervision plan

Abstract

The application discloses a nuclear power plant supervision plan risk prediction and adaptive scheduling method, system, device and storage medium, relates to the cross technical field of nuclear power plant safety supervision and intelligent management, and comprises the following steps: constructing a hidden danger risk prediction model, outputting hidden danger probability and risk grade; calculating an adjustment coefficient, creating an inspection supervision task and grading reminding; jointly determining threshold overrun and probability, triggering early warning and emergency tasks; after abnormal description is classified through an NLP model, the abnormal description is combined with a risk prediction result and weighted, and a hidden danger list is output; based on the hidden danger list, DCS data is associated to recommend rectification schemes and time limits, and a work application is output; task progress, rectification conditions and risk prediction are summarized into a dynamic report and are regularly updated; scheduling results, hidden danger grading, rectification progress and dynamic trends are input into a risk evaluation model, and an inspection report is output; the application can realize intelligent dynamic management of the nuclear power plant supervision plan, identify equipment risks in advance, automatically optimize inspection frequency, and shorten the hidden danger response cycle.

Description

Technical Field

[0001] This invention relates to the field of cross-disciplinary technology of nuclear power plant safety supervision and intelligent management, and in particular to a method, system, equipment and storage medium for risk prediction and adaptive scheduling of nuclear power plant supervision plans. Background Technology

[0002] Safety oversight of nuclear power plants is a core aspect of ensuring nuclear safety. Its oversight targets include critical and sensitive equipment such as reactor cooling systems and radiation monitoring devices. Oversight processes must strictly adhere to nuclear safety regulations, with requirements for real-time performance, accuracy, and reliability far exceeding those in ordinary industrial sectors. However, current nuclear power plant oversight planning and management generally remains at the level of "process digitization," lacking intelligent integration and analysis of equipment operating status, historical hazard data, and real-time risk signals. This results in significant static and experience-based deficiencies in oversight plan formulation and execution, making it difficult to adapt to the complex oversight needs of nuclear power plants characterized by high safety standards, high dynamism, and multiple coupled equipment.

[0003] Existing nuclear power plant monitoring and management systems typically employ standardized functional modules from conventional office automation (OA) systems, project management software (such as Jira and Lark), or enterprise resource planning (ERP) systems. These systems are primarily implemented through manual entry of plans, manual creation of tasks, fixed-cycle checks, and post-event report filling. Furthermore, they operate independently of the nuclear power plant's DCS (Digital Control System), lacking data integration and real-time linkage. Typical drawbacks of this type of solution include: rigid scheduling, with inspection frequency entirely dependent on manual experience and unable to be dynamically adjusted based on real-time equipment operating status, easily leading to resource misallocation such as "insufficient supervision of core equipment and excessive inspection of ordinary equipment"; lack of risk prediction capability, only able to respond passively after a potential hazard occurs, unable to identify equipment degradation trends in advance; hazard classification and grading rely on manual methods, and the complex and diverse nuclear power terminology makes it difficult for conventional NLP (Natural Language Processing) technology to be accurately adapted, resulting in large classification errors and low efficiency; inability to interact with the DCS system in real time, leading to delayed emergency response and inability to obtain real-time operating parameters to support dynamic decision-making; and crude access control, with data stored in a scattered manner, resulting in information silos and difficulty in meeting the legal requirements of nuclear safety supervision for data isolation and traceability. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a method, system, device and storage medium for risk prediction and adaptive scheduling of nuclear power plant monitoring plans.

[0005] Therefore, the technical problem solved by this invention is to overcome the shortcomings of existing supervision and planning management methods, such as reliance on manual labor, rigid scheduling, lack of risk prediction, inability to link with DCS, low accuracy of hazard classification, insufficient data security, and lack of self-optimization closed loop, in the context of high safety, high sensitivity, and complex equipment in nuclear power plants. This invention provides a supervision and planning management method that integrates nuclear power-specific intelligent algorithms with real-time linkage with DCS, so as to achieve intelligent supervision of the entire process, including risk prediction, dynamic resource scheduling, real-time emergency response, automatic hazard classification, and continuous iterative optimization.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for risk prediction and adaptive scheduling of nuclear power plant monitoring plans, comprising: A hazard risk prediction model is constructed based on nuclear power plant operating parameters and historical monitoring data. The model takes real-time DCS parameters as input and outputs the hazard probability and five-level risk level. Based on the probability and risk level of potential hazards, combined with equipment level and historical pass rate, an adjustment coefficient is calculated through an adaptive scheduling algorithm to create inspection tasks and provide graded reminders. Real-time parameters are obtained from the DCS interface associated with the task, and threshold exceedances are determined in conjunction with probability to trigger early warning and emergency tasks. The descriptions of emergency-triggered anomalies or daily hazards are classified by an NLP model, weighted together with the risk prediction results, and output as a hazard list and its level. Based on the risk prediction associated with the hazard list and DCS data, combined with the recommended solutions and time limits in the case library, the work application and review conditions are output. The task progress, rectification status, DCS parameters and risk predictions are summarized into a dynamic report, and the trend is updated every hour. A risk assessment model is constructed based on nuclear safety standards, equipment risk level system and historical inspection qualification data. The model inputs planned information, risk level, scheduling results, over-limit records, hidden danger classification, rectification progress and dynamic trends, and outputs an inspection report containing risk ranking, frequency adjustment and optimization suggestions.

[0007] As a preferred scheme for risk prediction and adaptive scheduling in nuclear power plant monitoring plans, wherein: The aforementioned hazard risk prediction model, constructed based on nuclear power plant operating parameters and historical monitoring data, takes real-time DCS parameters as input and outputs hazard probability and five-level risk levels, including: The system inputs basic planning information, nuclear power-specific information, algorithm support parameters, and inspection standards and regulations for nuclear power plant surveillance plans. It also links nuclear power plant equipment ledgers, historical surveillance data, and DCS interface information, encrypts and stores the data, and forms a standardized surveillance plan database. Collect historical hazard data, obtain equipment operation data from the DCS system interface, and collect environmental data. Remove abnormal parameters, convert unstructured hazard descriptions into structured feature vectors, and label the hazards. Based on historical hazard data in the monitoring plan database and historical DCS operating parameters, and using eight nuclear power-specific dimensions as input features—radiation dose, reactor coolant pressure, fuel rod temperature, core water level, equipment vibration value, primary loop operating time, cumulative radiation dose in the plant, and incremental trend of radioactive medium leakage—a hazard risk prediction model is constructed. The hazard risk prediction model is trained. After training, the model is output with the current DCS real-time operating parameters as input, and the probability of hazard occurrence and the five-level risk level of each monitored object are output.

[0008] As a preferred scheme for risk prediction and adaptive scheduling in nuclear power plant monitoring plans, wherein: The process involves calculating adjustment coefficients based on hazard probability and risk level, combined with equipment level and historical pass rate, using an adaptive scheduling algorithm to create inspection tasks and provide tiered alerts, including: Obtain the equipment risk level and historical inspection pass rate stored in the supervision plan database. The historical inspection pass rate is the ratio of the number of successful inspections to the total number of inspections within a predetermined time period. Based on the principles of prioritizing nuclear safety and optimizing resources, a calculation rule for adjustment coefficients is constructed according to the probability of hazard occurrence, equipment risk level, and historical inspection pass rate. The adjustment coefficients are positively correlated with the probability of hazard occurrence and equipment risk level, and negatively correlated with the historical inspection pass rate. Furthermore, the weight of the probability of hazard occurrence is higher than the weight of equipment risk level and historical inspection pass rate. Based on the different numerical ranges of the adjustment coefficient, the adjustment range of the inspection frequency is determined. The adjustment range includes a significant increase, a moderate increase, no change, and a decrease. When the frequency is significantly increased, the corresponding equipment will be included in the key supervision list. Based on the adjusted inspection frequency, the system automatically extracts the recorded plan information, creates inspection and supervision tasks, and associates risk prediction results, equipment DCS interface information, and historical hazard records with the inspection and supervision tasks. The alert level is determined based on the level of potential hazard risk, and the alert methods are progressively enhanced from low to high according to the alert level. The alert methods include email alerts, system pop-up alerts, and SMS alerts. SMS alerts are sent through the nuclear power plant's internal communication gateway. When the gateway is unavailable, the alerts automatically switch to a dedicated broadcast alert for the nuclear power plant. The system automatically recalculates adjustment factors periodically and dynamically adjusts the inspection frequency based on real-time DCS data, risk prediction results, and changes in historical inspection pass rates.

[0009] As a preferred scheme for risk prediction and adaptive scheduling in nuclear power plant monitoring plans, wherein: The step of obtaining real-time parameters from the DCS interface associated with the task, jointly determining threshold exceedances and probabilities, and triggering early warning and emergency tasks includes: Based on the DCS interface information associated in the task list, real-time operating parameters of the equipment are obtained through a one-way isolation gateway deployed between the nuclear power plant's management information zone and DCS control zone I. The one-way isolation gateway only allows DCS control zone I to upload parameters to the management information zone in one direction, and prohibits the management information zone from issuing any control commands to DCS control zone I. Deploy a heartbeat detection mechanism that automatically retains all DCS cache parameters during the disconnection period when the heartbeat is interrupted, and synchronously generates security audit logs and reports them to regulatory authorities. A dual-dimensional joint early warning mechanism is adopted, which combines the judgment of DCS real-time parameter threshold exceedance with the judgment of time-series risk probability output by the hidden danger risk prediction model. The early warning level is determined according to the combination of the degree of real-time parameter exceedance and the level of risk probability. After the emergency supervision task is completed, the rectification parameters will be automatically transmitted back to the DCS system and stored in the auditable log. At the same time, the rectification data will be automatically supplemented as incremental training samples for the hidden danger and risk prediction model.

[0010] As a preferred scheme for risk prediction and adaptive scheduling in nuclear power plant monitoring plans, wherein: The description of emergency-triggered anomalies or routine hazard descriptions is classified using an NLP model, weighted together with risk prediction results, and outputs a hazard list and its level, including: The abnormal descriptions generated when emergency supervision tasks are triggered or the hazard descriptions input from daily inspections are input into the fine-tuned NLP model for classification to obtain the text classification results and their corresponding preliminary hazard levels; at the same time, the time-series risk probability output by the hazard risk prediction model for the same equipment is obtained. The NLP text classification results and the hazard risk prediction results are jointly weighted and judged: if the hazard level corresponding to the NLP classification result is lower than the level corresponding to the time series risk probability output by the hazard risk prediction model, the hazard level is adjusted upward based on the hazard risk prediction result, and the hazard list and its final level are finally output.

[0011] As a preferred scheme for risk prediction and adaptive scheduling in nuclear power plant monitoring plans, wherein: The process, based on risk prediction associated with potential hazards and DCS data, combined with recommended solutions and time limits from the case library, outputs work application and review conditions, including: Based on the generated hazard list and its level, the risk prediction results output by the hazard risk prediction model are automatically associated with the corresponding equipment's DCS real-time data. Based on the hazard level, risk prediction results, and historical rectification case database, a rectification plan containing the resources required for rectification, the estimated completion time, and the responsible personnel is matched and recommended from the case database. Set different rectification time limits according to the hazard level from high to low, with the higher the hazard level, the shorter the rectification time limit; output a work application that includes a rectification plan and rectification time limit; After the work application is submitted, the rectification progress is tracked in real time based on the DCS equipment operating parameters, and the work application status is automatically updated according to the rectification progress. Once rectification is completed, a review task will be automatically triggered. After the review is passed, the work application will be closed, and the rectification progress will be synchronized to the dynamic reporting stage.

[0012] As a preferred scheme for risk prediction and adaptive scheduling in nuclear power plant monitoring plans, wherein: The process of summarizing task progress, rectification status, DCS parameters, and risk predictions into a dynamic report, with trends updated hourly, includes: The system automatically collects data on the progress of inspection tasks, the completion status of hazard rectification, real-time operating parameters of DCS equipment, and risk prediction changes output by the hazard risk prediction model on an hourly basis. The collected multi-source data is then aggregated to generate a dynamic status report. The dynamic status report includes a risk trend chart showing the trend of hazard risk changes in the predetermined future period based on the prediction results of the hazard risk prediction model, the rectification completion rate statistically classified by hazard level, the operation status curve of key parameters collected in real time by DCS, the details of high-risk tasks, and the statistics of resource usage.

[0013] Secondly, the present invention provides a risk prediction and adaptive scheduling system for nuclear power plant monitoring plans, comprising: The risk prediction module is used to build a hazard risk prediction model based on nuclear power plant operating parameters and historical monitoring data. It takes real-time DCS parameters as input and outputs the hazard probability and five-level risk level. The intelligent scheduling module is used to calculate adjustment coefficients based on the probability and risk level of potential hazards, combined with equipment level and historical pass rate, and to create inspection tasks and provide graded reminders. The linkage early warning module is used to obtain real-time parameters from the DCS interface associated with the task, and jointly determine the threshold exceedance with probability to trigger early warning and emergency tasks. The hazard classification module is used to classify emergency-triggered anomaly descriptions or daily hazard descriptions using an NLP model, and then weight them together with the risk prediction results to output a hazard list and its level. The rectification work order module is used to output work applications and review conditions based on the risk prediction associated with the hazard list and DCS data, combined with the recommended solutions and time limits in the case library. The dynamic reporting module is used to summarize task progress, rectification status, DCS parameters and risk predictions into a dynamic report, which is updated hourly. The comprehensive evaluation module is used to build a risk assessment model based on nuclear safety standards, equipment risk level system and historical inspection qualification data. It inputs planned information, risk level, scheduling results, over-limit records, hidden danger classification, rectification progress and dynamic trends into the risk assessment model, and outputs an inspection report containing risk ranking, frequency adjustment and optimization suggestions.

[0014] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the risk prediction and adaptive scheduling method for nuclear power plant monitoring plans.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement steps of a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans.

[0016] The beneficial effects of this invention are as follows: In the actual operation of nuclear power plant monitoring plans, this invention can effectively improve risk prevention and control and rectification response capabilities. Maintenance personnel no longer rely on experience to judge equipment status; the system can identify the deterioration trend of key parameters in advance, automatically triggering early warnings and generating inspection tasks before potential hazards develop into accidents. Many potential hazards are intercepted in their early stages, significantly reducing the impact of sudden equipment anomalies on unit operation. Simultaneously, the dynamic scheduling mechanism automatically adjusts the inspection frequency according to the actual equipment status. High-risk equipment is automatically subject to intensive monitoring during periods of parameter fluctuation, while low-risk equipment avoids ineffective and repeated inspections. The allocation of on-site monitoring resources is more precise and balanced, and the workload of manually formulating and adjusting plans is significantly reduced. In the hazard management phase, the system's automatic parsing and classification of nuclear power-related descriptive text effectively reduces the bias of manual classification. The cycle from hazard form completion to distribution is significantly shortened, allowing responsible personnel to obtain the necessary defect information more quickly. The linkage between work applications and real-time equipment data ensures that rectification plans, resource allocations, and deadlines are all based on actual data. Status changes during execution are reflected synchronously, and reviews are automatically triggered and linked to verification of rectification effectiveness, avoiding overdue or incomplete rectification due to untimely manual tracking. The automatic generation of dynamic status reports and intelligent inspection reports allows management to grasp supervision progress, risk distribution, and rectification dynamics without manually aggregating large amounts of scattered data, providing more timely and complete information for decision-making. The continuous operation of the model iteration loop and the rectification supervision loop enables the system to automatically optimize subsequent prediction and scheduling logic based on the actual results of each inspection and rectification, continuously enhancing the overall adaptability of the supervision plan over time. A four-level access control mechanism and mandatory retention of audit logs ensure that personnel in different positions can only access data within their scope of responsibility, compliance summary information required for external supervision can be extracted independently, and sensitive operating parameters are not viewed without authorization, meeting the rigid requirements of nuclear safety supervision for data security and traceability. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is an overall flowchart of a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans provided by the present invention.

[0019] Figure 2 This invention provides a logical judgment flowchart for a dual-dimensional joint early warning system based on DCS real-time threshold and hazard risk prediction probability in a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans.

[0020] Figure 3 This is a flowchart of the adaptive scheduling process for inspection frequency based on the adjustment coefficient range judgment of a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans provided by the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans, including: S1: Construct a hazard risk prediction model based on nuclear power plant operating parameters and historical monitoring data. Take real-time DCS parameters as input and output the hazard probability and five-level risk level. S2: Based on the probability and risk level of potential hazards, combined with equipment level and historical pass rate, an adjustment coefficient is calculated through an adaptive scheduling algorithm to create inspection tasks and provide graded reminders. S3: Obtain real-time parameters from the DCS interface associated with the task, and jointly determine the threshold exceedance with probability to trigger early warning and emergency tasks; S4: Classify the emergency-triggered anomaly description or daily hazard description using an NLP model, and weight it together with the risk prediction results to output a hazard list and its level; S5: Based on the risk prediction associated with the hidden danger and DCS data, combined with the recommended solutions and time limits in the case library, output the work application and review conditions; S6: Summarize task progress, rectification status, DCS parameters and risk predictions into a dynamic report, and update the trend every hour; S7: Based on nuclear safety standards, equipment risk level system and historical inspection qualification data, a risk assessment model is constructed. The model inputs planned information, risk level, scheduling results, over-limit records, hidden danger classification, rectification progress and dynamic trends, and outputs an inspection report containing risk ranking, frequency adjustment and optimization suggestions.

[0023] It should be noted that through steps S1-S7, a closed-loop management scheme for the entire process of nuclear power plant supervision plan is formed, encompassing data acquisition, risk prediction, dynamic scheduling, real-time linkage, automatic hazard classification, closed-loop management of rectification, dynamic reporting and monitoring, and intelligent analysis and evaluation. This scheme is driven by nuclear power plant-specific operating parameters and historical supervision data, and supported by hazard risk prediction models, adaptive scheduling algorithms, DCS linkage judgment, and NLP hazard classification and risk assessment models. It realizes the transformation of supervision plan from passive post-event inspection to proactive pre-event control, from fixed cycle to dynamic adaptation, from manual handling to automatic linkage, and from decentralized management to collaborative closed loop, effectively improving the intelligence level of nuclear power supervision work and the nuclear safety risk prevention and control capabilities.

[0024] Example 2, refer to Figures 1-3 As one embodiment of the present invention, based on the previous embodiment, a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans is provided, including: In this embodiment, the hazard risk prediction model constructed in step S1 above based on nuclear power plant operating parameters and historical monitoring data, using real-time DCS parameters as input, outputs the hazard probability and five-level risk levels, including: Build a standardized monitoring program database, including: Enter the basic plan information for the nuclear power plant surveillance plan, including plan name, inspection type, inspection target, plan baseline cycle, responsible department and person, inspection content and nuclear safety standards; Enter nuclear power-specific information, including equipment risk level, equipment-associated DCS data interface number, and historical hazard record association number; Input algorithm support parameters, including historical hazard data from the past 5 years and threshold values ​​for nuclear safety-specific operating parameters; Enter the inspection standards and regulations, and associate them with the corresponding nuclear safety regulations and inspection standards.

[0025] After data entry is completed, the system automatically links to the nuclear power plant's equipment ledger, historical monitoring data, and DCS interface information. The data is stored in encrypted form, forming a standardized monitoring plan database.

[0026] Collect historical hazard data from the past 5 years, including hazard location, type, level, rectification cycle, and related equipment parameters; Obtain equipment operation data from the DCS system interface, including radiation dose, reactor coolant pressure, fuel rod temperature, equipment vibration value, and operating time; Environmental data was collected, including factory temperature and humidity, atmospheric pressure, totaling more than 1,500 data points.

[0027] The collected historical hazard data, equipment operation data, and environmental data are preprocessed as follows: abnormal parameters during equipment start-up and shutdown are removed; unstructured hazard descriptions are converted into structured feature vectors and labeled with hazard tags, where 0 indicates no hazard and 1 indicates a hazard; and key parameters such as radiation dose and vibration values ​​are normalized.

[0028] Based on historical hazard data and DCS historical operating parameters in the monitoring plan database, and using eight nuclear power-specific dimensions as input features—radiation dose, reactor coolant pressure, fuel rod temperature, core water level, equipment vibration value, primary loop operating time, cumulative radiation dose in the plant, and incremental trend of radioactive medium leakage—a hazard risk prediction model is constructed. The model input layer has 12 feature dimensions, including 8 nuclear power-specific dimensions and 4 basic dimensions. The basic dimensions are operating time, plant temperature and humidity, atmospheric pressure, and equipment start-up and shutdown status. The hidden layer has 3 layers with 64, 32, and 16 neurons respectively, using the ReLU activation function and a dropout coefficient of 0.2. The output layer uses the Softmax activation function to output the probability of hazard occurrence and five risk levels. The risk level determination rules are optimized in conjunction with nuclear safety regulations, with a radiation dose exceeding the threshold by 2 times considered high risk.

[0029] The preprocessed data was divided into training and test sets in a 7:3 ratio. The Adam optimizer was used with a learning rate of 0.001 for 100 iterations. Early stopping was employed to prevent overfitting; training stopped when the validation set accuracy did not improve for five consecutive iterations. The SMOTE algorithm was used to expand the high-risk hazard samples. After training, the current real-time DCS operating parameters were used as input to output the probability of hazard occurrence and the five-level risk level for each monitored object.

[0030] The system automatically calls the model at 2:00 AM every day, inputs the current DCS real-time data and historical hazard data, outputs the hazard risk prediction results for each supervised object and generates a list of high-risk equipment, and at the same time receives real-time DCS data to continuously iterate and update the prediction results.

[0031] In this embodiment, step S2 above, based on the probability of hidden dangers and risk level, combined with equipment level and historical pass rate, calculates adjustment coefficients through an adaptive scheduling algorithm, creates inspection tasks, and provides tiered reminders, including: Based on the principles of prioritizing nuclear safety and optimizing resources, and taking into account the probability of hazard occurrence P, equipment risk level L, and historical inspection pass rate R, the inspection frequency adjustment coefficient K is calculated and expressed as: K = α × P + β × L - γ × R L is set to 3 for Level 1, 2 for Level 2, and 1 for Level 3. R is the number of qualified inspections in the past year divided by the total number of inspections. The weighting coefficients are α=0.5, β=0.3, and γ=0.2. K ranges from 0 to 3.

[0032] like Figure 3 As shown, a first threshold, a second threshold, and a third threshold are preset, with the first threshold > the second threshold > the third threshold. In this embodiment, the first threshold is 2.0, the second threshold is 1.0, and the third threshold is 0.5. The frequency adjustment range is determined based on the range of the K value. When K≥2.0, the inspection frequency increases by 50%, from once a week to once every 4 to 5 days, and is included in the key supervision list; When 1.0≤K<2.0, the inspection frequency is increased by 20%, from once a month to once every 25 days; The reference period remains unchanged when 0.5 ≤ K < 1.0; When K < 0.5, the frequency of checks is reduced by 30%, from once a month to once every 43 days, to avoid over-checking.

[0033] Based on the adjusted inspection frequency, the system automatically extracts the entered plan information, creates inspection and supervision tasks, and associates them with risk prediction results, equipment DCS interface information, and historical hazard records.

[0034] The alerts are tiered based on the risk level of the potential hazard: low or relatively low risk alerts are sent via email; medium risk alerts are sent via email plus a system pop-up; and high or relatively high risk alerts are sent via email plus a pop-up and SMS. SMS alerts are sent through the nuclear power plant's management information regional enterprise SMS gateway via HTTP API or SMPP protocol. If the gateway is unavailable, the alert is automatically downgraded to a broadcast alert via a dedicated nuclear power plant walkie-talkie. The responsible person must log in to the intranet system to confirm receipt, thus forming a closed loop.

[0035] The system automatically recalculates the K value every week and dynamically adjusts the frequency based on real-time DCS data, risk prediction results, and changes in historical inspection pass rates.

[0036] In this embodiment, step S3 above, which involves obtaining real-time parameters based on the DCS interface associated with the task, and jointly determining threshold exceedances and probabilities to trigger early warning and emergency tasks, includes: Based on the DCS interface information associated with the task list, real-time operating parameters of the equipment are obtained through a one-way isolation gateway deployed between the nuclear power plant management information zone and DCS control zone I. The OPC UA encrypted communication protocol is adopted. The zone I DCS only allows one-way uploading of parameters to the zone II monitoring system, and prohibits the monitoring system from issuing control commands.

[0037] Deploy a heartbeat detection mechanism to automatically retain all DCS cache parameters during the disconnection period when the heartbeat is interrupted, and synchronously generate security audit logs to report to regulatory authorities.

[0038] like Figure 2As shown, a three-level early warning logic is adopted, which uses a combination of real-time thresholds from the DCS and time-series risk probability output by the hazard risk prediction model: when a single parameter slightly exceeds the limit and the hazard probability is below 0.3, only a prompt is recorded and no emergency task is triggered; when a parameter exceeds the limit and the hazard probability is greater than or equal to 0.6, the early warning level is directly upgraded, and an emergency supervision task is pushed out in advance. After the emergency supervision task is completed, the rectification parameters are automatically transmitted back to the DCS and stored in the auditable logs. At the same time, they are automatically supplemented as incremental training samples for the hazard risk prediction model, forming a cross-system closed loop of alarm-inspection-rectification-model iteration.

[0039] In this embodiment, in step S4 above, the description of emergency-triggered anomalies or daily hazard descriptions are classified using an NLP model, and then weighted together with the risk prediction results to output a hazard list and its level, including: A nuclear power classification terminology dictionary was pre-built, and after using a pre-trained intermediate model for nuclear power operation and maintenance, 1,000 texts of potential hazards from the plant were used for fine-tuning.

[0040] During fine-tuning training, the probability of masking high-risk nuclear power terminology was increased to 40%, while the probability of masking ordinary maintenance terms was 15%.

[0041] The abnormal descriptions generated when emergency supervision tasks are triggered or the hazard descriptions input during daily inspections are input into the optimized NLP model for classification, resulting in text classification results and corresponding preliminary hazard levels. At the same time, the time-series risk probability output by the hidden danger risk prediction model for the same equipment is obtained.

[0042] The NLP text classification results and the hazard risk prediction results are jointly weighted and judged: when the NLP classification considers the hazard level to be low but the hazard risk prediction model outputs a high probability of time-series risk, it means that the equipment operation data has shown a risk trend but the text description has not fully reflected the risk. The system automatically upgrades the hazard level and finally outputs the hazard list and its level.

[0043] In this embodiment, step S5 above, based on the risk prediction associated with the hazard list and DCS data, combined with the recommended solutions and time limits from the case library, outputs the work application and review conditions, including: Based on the generated hazard list and its level, the system automatically links risk prediction results with DCS real-time data. According to the hazard level, risk prediction results, and historical rectification case library, the system automatically recommends a nuclear power-specific rectification plan, specifying the personnel, equipment, and consumables required for rectification, the estimated completion time, and the responsible personnel.

[0044] The rectification timeframes are set according to the following rules: major hazards no more than 24 hours, significant hazards no more than 72 hours, and general hazards no more than 7 days. A work application is then generated. After the work application is submitted, the system captures the DCS equipment operating parameters in real time and automatically updates the application status based on the rectification progress. Upon completion of rectification, the system automatically triggers a review task. Once the review is passed, the process is closed, and the rectification progress is simultaneously synchronized to the dynamic status report, achieving full-process linkage between inspection, hazard identification, rectification, monitoring, and review.

[0045] In this embodiment, step S6 above summarizes the task progress, rectification status, DCS parameters, and risk predictions into a dynamic report, with hourly updates including: The system automatically captures the progress of inspection tasks, the status of hazard rectification, changes in DCS equipment operating parameters and risk predictions every hour, and updates the status report in real time.

[0046] The report includes a risk trend chart showing the trend of hazard risk changes over the next 7 days based on the prediction results of the hazard risk prediction model; statistics on the completion rate of rectification categorized by hazard level; equipment operation status curves showing changes in key parameters collected by DCS in real time; high-risk task reminders; and statistics on resource usage.

[0047] Management uses reports to keep abreast of the progress of oversight work and changes in nuclear safety risks, and to promptly coordinate and resolve difficulties encountered during implementation.

[0048] In this embodiment, step S7 above constructs a risk assessment model based on nuclear safety standards, equipment risk level system, and historical inspection qualification data. The model inputs planned information, risk level, scheduling results, over-limit records, hazard classification, rectification progress, and dynamic trends, and outputs an inspection report including risk ranking, frequency adjustment, and optimization suggestions. Based on nuclear safety regulations and standards, equipment risk level system, and historical inspection compliance data, a risk assessment model is constructed. The model incorporates original plans and equipment ledger information, risk probability and level distribution, task scheduling results and frequency adjustment coefficients, DCS overrun records and emergency task execution status, hazard classification and grading results, work applications and rectification progress, and dynamic trend summary data. It calculates quantitative indicators such as the coverage rate of Level 1 equipment supervision and the probability of high-risk missed inspections. Outputs include a risk prediction analysis with a ranking of hazard occurrence probabilities for each supervised object, a list of high-risk equipment, and risk prevention and control recommendations; an inspection frequency adjustment analysis with the basis for frequency adjustment and effect analysis; statistics on high-frequency hazard types and causes based on NLP classification results; an evaluation of the rectification effect by combining real-time DCS data to assess the equipment operating status after rectification; and supervision plan optimization recommendations with quantitative regulatory compliance optimization suggestions.

[0049] The report uses a risk heat map combined with the layout of the nuclear power plant to accurately mark high-risk areas, a trend line chart to show the trend of risk changes, and a pie chart to show the proportion of each type of hazard for visualization. After generation, it is automatically pushed to the corresponding management and supports exporting an encrypted version.

[0050] This embodiment also includes tiered encryption access control: based on the four types of entities in nuclear safety regulations—system administrators, power plant supervisors, management, and external national regulatory authorities—differentiated data access scopes are set. Administrators can view all supervisory data, model parameters, DCS interface information, and encrypted data, and have system management configuration permissions. Supervisors can only view the tasks they are responsible for, hazard data, and corresponding DCS parameters. Management can view risk prediction results, inspection reports, optimization suggestions, and summary data, but cannot view sensitive interface information. External regulatory authorities can only view compliance-related data such as inspection completion rate and hazard rectification rate, but cannot view equipment operation secrets and model parameters. All data is stored using AES-256 encryption. Core and radiation-related sensitive data are subject to double encryption: AES-256 storage encryption plus transmission OPC UA session encryption, and are bound to the equipment's three-dimensional coordinates and nuclear safety thresholds. A full operation audit log for at least 10 years is mandatory. Each sensitive data access log is associated with the operator, DCS equipment number, and access time period, meeting the legal requirements for nuclear safety regulatory traceability.

[0051] This embodiment also includes three closed-loop feedback mechanisms: the structured data of inspection, rectification, and review are automatically written back to the encrypted DCS database, triggering automatic reconstruction of time-series features without manual input; newly added hidden danger samples automatically complete feature preprocessing and incremental training, and the model automatically updates the prediction weights monthly or based on incremental samples; non-compliant rectification automatically triggers K-value recalculation and task re-issuance, continuously verifying the rectification effect through real-time DCS parameters, and automatically re-issuing the scheduling task when non-compliant, linking the adaptive K-value scheduling algorithm to increase the inspection frequency until the rectification is qualified and the loop is closed.

[0052] Example 3 illustrates a method for risk prediction and adaptive scheduling of a nuclear power plant monitoring plan. It should be noted that the technical solution of a nuclear power plant monitoring plan risk prediction and adaptive scheduling system and the technical solution of the aforementioned method for risk prediction and adaptive scheduling of a nuclear power plant monitoring plan belong to the same concept. Details not described in detail in the technical solution of the nuclear power plant monitoring plan risk prediction and adaptive scheduling system in this embodiment can be found in the description of the aforementioned method for risk prediction and adaptive scheduling of a nuclear power plant monitoring plan.

[0053] This embodiment also provides a risk prediction and adaptive scheduling system for nuclear power plant monitoring plans, including: The risk prediction module is used to build a hazard risk prediction model based on nuclear power plant operating parameters and historical monitoring data. It takes real-time DCS parameters as input and outputs the hazard probability and five-level risk level. The intelligent scheduling module is used to calculate adjustment coefficients based on the probability and risk level of potential hazards, combined with equipment level and historical pass rate, and to create inspection tasks and provide graded reminders. The linkage early warning module is used to obtain real-time parameters from the DCS interface associated with the task, and jointly determine the threshold exceedance with probability to trigger early warning and emergency tasks. The hazard classification module is used to classify emergency-triggered anomaly descriptions or daily hazard descriptions using an NLP model, and then weight them together with the risk prediction results to output a hazard list and its level. The rectification work order module is used to output work applications and review conditions based on the risk prediction associated with the hazard list and DCS data, combined with the recommended solutions and time limits in the case library. The dynamic reporting module is used to summarize task progress, rectification status, DCS parameters and risk predictions into a dynamic report, which is updated hourly. The comprehensive evaluation module is used to build a risk assessment model based on nuclear safety standards, equipment risk level system and historical inspection qualification data. It inputs planned information, risk level, scheduling results, over-limit records, hidden danger classification, rectification progress and dynamic trends into the risk assessment model, and outputs an inspection report containing risk ranking, frequency adjustment and optimization suggestions.

[0054] This embodiment also provides a computer device applicable to a risk prediction and adaptive scheduling method for a nuclear power plant monitoring plan, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans, as proposed in the above embodiments.

[0055] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a risk prediction and adaptive scheduling method for nuclear power plant monitoring plans as proposed in the above embodiments.

[0056] The storage medium proposed in this embodiment and the risk prediction and adaptive scheduling method for nuclear power plant monitoring plans proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0057] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A risk prediction and adaptive scheduling method for nuclear power plant monitoring plans, characterized in that, include: A hazard risk prediction model is constructed based on nuclear power plant operating parameters and historical monitoring data. The model takes real-time DCS parameters as input and outputs the hazard probability and five-level risk level. Based on the probability and risk level of potential hazards, combined with equipment level and historical pass rate, an adjustment coefficient is calculated through an adaptive scheduling algorithm to create inspection tasks and provide graded reminders. Real-time parameters are obtained from the DCS interface associated with the task, and threshold exceedances are determined in conjunction with probability to trigger early warning and emergency tasks. The descriptions of emergency-triggered anomalies or daily hazards are classified by an NLP model, weighted together with the risk prediction results, and output as a hazard list and its level. Based on the risk prediction associated with the hazard list and DCS data, combined with the recommended solutions and time limits in the case library, the work application and review conditions are output. The task progress, rectification status, DCS parameters and risk predictions are summarized into a dynamic report, and the trend is updated every hour. A risk assessment model is constructed based on nuclear safety standards, equipment risk level system and historical inspection qualification data. The model inputs planned information, risk level, scheduling results, over-limit records, hidden danger classification, rectification progress and dynamic trends, and outputs an inspection report containing risk ranking, frequency adjustment and optimization suggestions.

2. The risk prediction and adaptive scheduling method for nuclear power plant monitoring plans as described in claim 1, characterized in that, The aforementioned hazard risk prediction model, constructed based on nuclear power plant operating parameters and historical monitoring data, takes real-time DCS parameters as input and outputs hazard probability and five-level risk levels, including: The system inputs basic planning information, nuclear power-specific information, algorithm support parameters, and inspection standards and regulations for nuclear power plant surveillance plans. It also links nuclear power plant equipment ledgers, historical surveillance data, and DCS interface information, encrypts and stores the data, and forms a standardized surveillance plan database. Collect historical hazard data, obtain equipment operation data from the DCS system interface, and collect environmental data. Remove abnormal parameters, convert unstructured hazard descriptions into structured feature vectors, and label the hazards. Based on historical hazard data and DCS historical operating parameters in the monitoring plan database, and using eight nuclear power-specific dimensions as input features, namely radiation dose, reactor coolant pressure, fuel rod temperature, core water level, equipment vibration value, primary loop operating time, cumulative radiation dose in the plant, and incremental trend of radioactive medium leakage, a hazard risk prediction model is constructed. The hazard risk prediction model is trained. After training, the model is output with the current DCS real-time operating parameters as input, and the probability of hazard occurrence and the five-level risk level of each monitored object are output.

3. The risk prediction and adaptive scheduling method for nuclear power plant monitoring plans as described in claim 2, characterized in that, The process involves calculating adjustment coefficients based on hazard probability and risk level, combined with equipment level and historical pass rate, using an adaptive scheduling algorithm to create inspection tasks and provide tiered alerts, including: Obtain the equipment risk level and historical inspection pass rate stored in the supervision plan database. The historical inspection pass rate is the ratio of the number of successful inspections to the total number of inspections within a predetermined time period. Based on the principles of prioritizing nuclear safety and optimizing resources, a calculation rule for adjustment coefficients is constructed according to the probability of hazard occurrence, equipment risk level, and historical inspection pass rate. The adjustment coefficients are positively correlated with the probability of hazard occurrence and equipment risk level, and negatively correlated with the historical inspection pass rate. Furthermore, the weight of the probability of hazard occurrence is higher than the weight of equipment risk level and historical inspection pass rate. Based on the different numerical ranges of the adjustment coefficient, the adjustment range of the inspection frequency is determined. The adjustment range includes a significant increase, a moderate increase, no change, and a decrease. When the frequency is significantly increased, the corresponding equipment will be included in the key supervision list. Based on the adjusted inspection frequency, the system automatically extracts the entered plan information, creates inspection and supervision tasks, and associates risk prediction results, equipment DCS interface information, and historical hazard records with the inspection and supervision tasks. The alert level is determined based on the level of potential hazard risk, and the alert methods are progressively enhanced from low to high according to the alert level. The alert methods include email alerts, system pop-up alerts, and SMS alerts. SMS alerts are sent through the nuclear power plant's internal communication gateway. When the gateway is unavailable, the alerts automatically switch to a dedicated broadcast alert for the nuclear power plant. The system automatically recalculates adjustment factors periodically and dynamically adjusts the inspection frequency based on real-time DCS data, risk prediction results, and changes in historical inspection pass rates.

4. The risk prediction and adaptive scheduling method for nuclear power plant monitoring plans as described in claim 3, characterized in that, The step of obtaining real-time parameters from the DCS interface associated with the task, jointly determining threshold exceedances and probabilities, and triggering early warning and emergency tasks includes: Based on the DCS interface information associated in the task list, real-time operating parameters of the equipment are obtained through a one-way isolation gateway deployed between the nuclear power plant's management information zone and DCS control zone I. The one-way isolation gateway only allows DCS control zone I to upload parameters to the management information zone in one direction, and prohibits the management information zone from issuing any control commands to DCS control zone I. Deploy a heartbeat detection mechanism that automatically retains all DCS cache parameters during the disconnection period when the heartbeat is interrupted, and synchronously generates security audit logs and reports them to regulatory authorities. A dual-dimensional joint early warning mechanism is adopted, which combines the judgment of DCS real-time parameter threshold exceedance with the judgment of time-series risk probability output by the hidden danger risk prediction model. The early warning level is determined according to the combination of the degree of real-time parameter exceedance and the level of risk probability. After the emergency supervision task is completed, the rectification parameters will be automatically transmitted back to the DCS system and stored in the auditable log. At the same time, the rectification data will be automatically supplemented as incremental training samples for the hidden danger and risk prediction model.

5. The risk prediction and adaptive scheduling method for nuclear power plant monitoring plans as described in claim 4, characterized in that, The description of emergency-triggered anomalies or routine hazard descriptions is classified using an NLP model, weighted together with risk prediction results, and outputs a hazard list and its level, including: The abnormal descriptions generated when emergency supervision tasks are triggered or the hazard descriptions input from daily inspections are input into the fine-tuned NLP model for classification to obtain text classification results and their corresponding preliminary hazard levels; at the same time, the time-series risk probability output by the hazard risk prediction model for the same equipment is obtained. The NLP text classification results and the hazard risk prediction results are jointly weighted and judged: if the hazard level corresponding to the NLP classification result is lower than the level corresponding to the time series risk probability output by the hazard risk prediction model, the hazard level is adjusted upward based on the hazard risk prediction result, and the hazard list and its final level are finally output.

6. The risk prediction and adaptive scheduling method for nuclear power plant monitoring plans as described in claim 5, characterized in that, The process, based on risk prediction associated with potential hazards and DCS data, combined with recommended solutions and time limits from the case library, outputs work application and review conditions, including: Based on the generated hazard list and its level, the risk prediction results output by the hazard risk prediction model are automatically associated with the corresponding equipment's DCS real-time data. Based on the hazard level, risk prediction results, and historical rectification case database, a rectification plan containing the resources required for rectification, the estimated completion time, and the responsible personnel is matched and recommended from the case database. Set different rectification time limits according to the hazard level from high to low, with the higher the hazard level, the shorter the rectification time limit; output a work application that includes a rectification plan and rectification time limit; After the work application is submitted, the rectification progress is tracked in real time based on the DCS equipment operating parameters, and the work application status is automatically updated according to the rectification progress. Once rectification is completed, a review task will be automatically triggered. After the review is passed, the work application will be closed, and the rectification progress will be synchronized to the dynamic reporting stage.

7. The risk prediction and adaptive scheduling method for nuclear power plant monitoring plans as described in claim 6, characterized in that, The process of summarizing task progress, rectification status, DCS parameters, and risk predictions into a dynamic report, with trends updated hourly, includes: The system automatically collects data on the progress of inspection tasks, the completion status of hazard rectification, real-time operating parameters of DCS equipment, and risk prediction changes output by the hazard risk prediction model on an hourly basis. The collected multi-source data is then aggregated to generate a dynamic status report. The dynamic status report includes a risk trend chart showing the trend of hazard risk changes in the predetermined future period based on the prediction results of the hazard risk prediction model, the rectification completion rate statistically classified by hazard level, the operation status curve of key parameters collected in real time by DCS, the details of high-risk tasks, and the statistics of resource usage.

8. A risk prediction and adaptive scheduling system for nuclear power plant monitoring programs, using the method described in any one of claims 1-7, characterized in that, include: The risk prediction module is used to build a hazard risk prediction model based on nuclear power plant operating parameters and historical monitoring data. It takes real-time DCS parameters as input and outputs the hazard probability and five-level risk level. The intelligent scheduling module is used to calculate adjustment coefficients based on the probability and risk level of potential hazards, combined with equipment level and historical pass rate, and to create inspection tasks and provide graded reminders. The linkage early warning module is used to obtain real-time parameters from the DCS interface associated with the task, and jointly determine the threshold exceedance with probability to trigger early warning and emergency tasks. The hazard classification module is used to classify emergency-triggered anomaly descriptions or daily hazard descriptions using an NLP model, and then weight them together with the risk prediction results to output a hazard list and its level. The rectification work order module is used to output work applications and review conditions based on the risk prediction associated with the hazard list and DCS data, combined with the recommended solutions and time limits in the case library. The dynamic reporting module is used to summarize task progress, rectification status, DCS parameters and risk predictions into a dynamic report, which is updated hourly. The comprehensive evaluation module is used to build a risk assessment model based on nuclear safety standards, equipment risk level system and historical inspection qualification data. It inputs planned information, risk level, scheduling results, over-limit records, hidden danger classification, rectification progress and dynamic trends into the risk assessment model, and outputs an inspection report containing risk ranking, frequency adjustment and optimization suggestions.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images