A Nuclear Power Plant Key Parameter Anomaly Identification and Alarm System

Abstract

This invention specifically relates to a nuclear power plant critical parameter anomaly identification and alarm system, including a data warehouse and at least one set of parameter anomaly identification modules located in the data warehouse; each parameter anomaly identification module receives real-time operating data of the unit as input to identify and alarm for specified nuclear power parameters. This invention monitors the unit's operating status and can promptly and accurately identify anomalies in critical parameters, equipment, and systems.

Description

Technical Field

[0001] This invention relates to the field of nuclear power safety protection technology, and in particular to a system for identifying and alarming abnormalities in important nuclear power parameters. Background Technology

[0002] During normal operation, nuclear power plants may experience unexpected protective actions such as load shedding, shutdown, and reactor shutdown due to factors such as faulty or malfunctioning process equipment, instrumentation and control system failures, power grid failures, fires, insufficient personnel skills, and human error. Currently, the detection of abnormal situations during unit operation mainly relies on alarm information from the DCS system. Alarms are only triggered when a single measurement or a combination of multiple measurements reaches a certain threshold, lacking trend analysis. Furthermore, most typical abnormal events in the unit have different warning signs before leading to accidents, but DCS system alarms cannot fully cover these events, nor can they promptly alert operators based on the correlation of parameter fluctuations, making it difficult to attract attention in the early stages of abnormal events.

[0003] Traditional monitoring methods, based on univariate control charts, are no longer able to reflect the complex changes in industrial processes. While big data from modern industrial systems brings a wealth of information, nuclear power plant operators cannot rely on experience to make assessments of process operation based on massive amounts of data. Summary of the Invention

[0004] The purpose of this invention is to provide a nuclear power plant critical parameter anomaly identification and early warning system that monitors the unit's operating status and can identify anomalies in critical parameters, equipment, and systems in a timely and accurate manner.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A nuclear power plant critical parameter anomaly identification and alarm system includes a data warehouse and at least one set of parameter anomaly identification modules located in the data warehouse; each parameter anomaly identification module receives real-time operating data of the unit as input to identify and alarm for specified nuclear power parameters.

[0007] In this invention, each parameter anomaly identification module includes a knowledge-driven first anomaly monitoring submodule, a data-driven second anomaly monitoring submodule, and a comprehensive alarm submodule.

[0008] The first anomaly monitoring submodule constructs an anomaly identification model based on the anomaly identification rules described in the preset anomaly event tree diagram; when the first anomaly monitoring submodule determines that an anomaly exists, it determines that the data characteristics of the corresponding parameters in the received real-time operation data of the unit match the description in the anomaly event tree diagram.

[0009] The second anomaly monitoring submodule includes a normal state identification subunit and an anomaly feature identification subunit;

[0010] The normal state identification subunit uses normal data from the unit's historical operation data as training data to construct a normal state reconstruction model based on artificial intelligence technology. The normal state reconstruction model predicts the value range of the corresponding parameters based on the real-time operation data of the unit input online. When the normal state identification subunit determines that the value range of the corresponding parameters in the received real-time operation data of the unit predicted by the normal state reconstruction model exceeds the normal range, it determines that there is an anomaly.

[0011] The anomaly feature identification subunit uses anomaly data from the unit's historical operation data as training data to build an anomaly feature identification model based on artificial intelligence technology. When the anomaly feature identification subunit determines that the characteristics of the received real-time operation data of the unit match the anomaly features in the anomaly feature identification model, it determines that an anomaly exists.

[0012] The integrated alarm submodule calculates the inputs from the first anomaly monitoring submodule, the normal state identification submodule, and the anomaly feature identification submodule based on preset weights, generates corresponding integrated calculation results, and determines whether there are any anomalies in the current unit operation data based on the integrated calculation results. If there are anomalies, an anomaly alarm is triggered.

[0013] In this invention, the normal state identification subunit includes an abnormal alarm discrimination threshold unit, a normal state reconstruction model, a residual calculation unit, and an alarm unit.

[0014] During normal operation, the normal state reconstruction model is invoked, and the corresponding parameter combination from the real-time operation data of the unit is input into the normal state reconstruction model. The normal state reconstruction model outputs the corresponding reconstruction value. The residual calculation unit calculates N sets of residuals between the reconstruction value and the current value of each parameter in the parameter combination. When the residual exceeds the abnormal alarm judgment threshold, the alarm unit determines that there is an abnormality.

[0015] In this invention, the normal state identification subunit extracts parameter combinations describing all modalities of the detected object from the historical operation data of the unit in the data warehouse as training data for the normal state reconstruction model.

[0016] In this invention, the various abnormal alarm discrimination thresholds in the abnormal alarm discrimination threshold unit are obtained based on the normal state reconstruction model, including the following steps:

[0017] After the normal state reconstruction model is built, the normal operation history data with corresponding parameter combinations is obtained from the data warehouse and input into the normal state reconstruction model for prediction, so as to predict the feature value based on the normal operation history data.

[0018] The difference between each feature value and the corresponding parameter value in the parameter combination is used as the abnormal alarm discrimination threshold in the abnormal alarm discrimination threshold unit.

[0019] In this invention, the alarm unit has preset alarm conditions. When the preset alarm conditions are met between the abnormal alarm judgment threshold and the residual, the normal state identification subunit determines that there is a possibility of an abnormality.

[0020] In this invention, the preset alarm conditions include the following two sets:

[0021] (1) At least one set of residuals exceeds the corresponding abnormal alarm judgment threshold;

[0022] (2) Within a specified time window, the proportion of cases exceeding the abnormal alarm judgment threshold is greater than the preset percentage.

[0023] (3) The specified time window must meet the preset length for cases that continuously exceed the abnormal alarm judgment threshold.

[0024] In this invention, the abnormal feature recognition subunit extracts the abnormal features of the detection object from the historical operation data of the unit in the data warehouse as the training data of the abnormal feature recognition model. The abnormal features of the detection object are the most relevant features under abnormal conditions of the detection object.

[0025] During normal operation, the abnormal feature identification model is invoked, and the corresponding parameter combination in the real-time operation data of the unit is input into the abnormal feature identification model. The abnormal feature identification model determines whether there is an abnormality based on the degree of abnormal feature matching. When the abnormal feature identification subunit determines that the characteristics of the real-time operation data of the unit match the abnormal features in the abnormal feature identification model, it determines that there is an abnormality.

[0026] In this invention, based on unit operation experience, the experience in judging abnormal conditions is converted into a computer-generated configuration diagram to construct an anomaly identification model, including the following steps:

[0027] (1) For different detection objects, an exception event tree diagram is drawn using VISIO diagram based on operational experience. The exception event tree diagram includes multiple branches at different levels, and each branch corresponds to an exception type.

[0028] (2) Using knowledge modeling tools, based on the abnormal event tree diagram verified by unit operation experience, develop a knowledge-driven anomaly detection configuration diagram;

[0029] (3) Construct a corresponding anomaly identification model based on the anomaly detection configuration diagram;

[0030] During normal operation, the knowledge-driven anomaly detection configuration diagram is deployed to the data warehouse. Then, the real-time operation data of the unit is input into the corresponding anomaly identification model for processing. The first anomaly monitoring submodule judges the data characteristics of the corresponding parameters in the real-time operation data of the unit based on the anomaly identification rules described in the anomaly event tree diagram, and inputs the judgment result into the comprehensive alarm submodule for processing.

[0031] In this invention, the integrated alarm submodule includes a weight adjustment unit; the weight adjustment unit presets different weights for different inputs to generate corresponding integrated calculation results;

[0032] In the weight adjustment unit, the weights input to the first anomaly monitoring submodule, the normal state identification submodule, and the anomaly feature identification submodule are adjusted according to the actual operating conditions of the unit.

[0033] In this invention, the nuclear power important parameter anomaly identification and alarm system also includes an online update module; the online update module uses updated historical operating data of the unit based on user feedback to update the data-driven normal state model and anomaly feature identification model.

[0034] In this invention, the online update module includes a manual judgment unit, an online update unit for the normal state reconstruction model, and an offline update unit for the abnormal feature recognition model;

[0035] The manual judgment unit judges the abnormal alarm status issued by the integrated alarm submodule;

[0036] If it is determined to be a false alarm, it means that there may be new normal data, and the new normal data does not exist in the unit's historical operating data. In this case, it will be transferred to the online update unit of the normal state reconfiguration model for processing.

[0037] The online update unit of the normal state reconstruction model adds new normal data to the training data of the normal state reconstruction model to obtain updated training data; subsequently, the normal state reconstruction model is retrained based on the updated training data to obtain an updated normal state reconstruction model.

[0038] And based on retraining or incremental learning, an updated normal state reconstruction model is obtained;

[0039] If it is determined to be a missed report, it means that there are features that should have been identified by the existing anomaly feature identification model, but were not actually identified. In this case, it will be transferred to the anomaly feature identification model offline update unit for processing.

[0040] The offline update unit of the anomaly feature recognition model extracts the features of missed reports and establishes an anomaly feature library. Subsequently, the anomaly feature recognition model is retrained based on the anomaly data in the anomaly feature library to obtain an updated anomaly feature recognition model.

[0041] Beneficial technical effects of the present invention:

[0042] The nuclear power critical parameter anomaly identification and alarm system of the present invention has a set of parameter anomaly identification modules for each type of critical parameter. When the unit is operating normally, the critical parameters are continuously and automatically monitored by each parameter anomaly identification module, which can provide early warning of anomalies in equipment, functions and parameters. When a single critical parameter deviates from normal operation or the status of a critical system or equipment is abnormal, the corresponding staff are alerted, reducing the workload of the operators and improving the stability of the unit. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of an embodiment of the nuclear power plant critical parameter anomaly identification and alarm system of the present invention. Detailed Implementation

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “equivalent to”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0045] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0046] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments.

[0047] See Figure 1 This embodiment provides a nuclear power important parameter anomaly identification and alarm system, including a data warehouse and at least one set of parameter anomaly identification modules located in the data warehouse; each parameter anomaly identification module receives real-time operating data of the unit as input to perform anomaly identification and alarm for specified nuclear power parameters.

[0048] In this embodiment, each parameter anomaly identification module includes a knowledge-driven first anomaly monitoring submodule, a data-driven second anomaly monitoring submodule, and a comprehensive alarm submodule.

[0049] The first anomaly monitoring submodule constructs an anomaly identification model based on the anomaly identification rules described in the anomaly event tree diagram summarized from the pre-set unit operation experience; when the first anomaly monitoring submodule determines that the data characteristics of the corresponding parameters in the received real-time unit operation data match the description in the anomaly event tree diagram, it determines that an anomaly exists;

[0050] The second anomaly monitoring submodule includes a normal state identification subunit and an anomaly feature identification subunit;

[0051] The normal state identification subunit uses normal data from the unit's historical operation data as training data to construct a normal state reconstruction model based on artificial intelligence technology. The normal state reconstruction model predicts the value range of the corresponding parameters based on the real-time operation data of the unit input online. When the normal state identification subunit determines that the value range of the corresponding parameters in the received real-time operation data of the unit predicted by the normal state reconstruction model exceeds the normal range, it determines that there is an anomaly.

[0052] The anomaly feature identification subunit uses anomaly data from the unit's historical operation data as training data to build an anomaly feature identification model based on artificial intelligence technology. When the anomaly feature identification subunit determines that the characteristics of the received real-time operation data of the unit match the anomaly features in the anomaly feature identification model, it determines that an anomaly exists.

[0053] The integrated alarm submodule calculates the inputs from the first anomaly monitoring submodule, the normal state identification submodule, and the anomaly feature identification submodule based on preset weights, generates corresponding integrated calculation results, and determines whether there are any anomalies in the current unit operation data based on the integrated calculation results. If there are anomalies, an anomaly alarm is triggered.

[0054] Under stable operating conditions, many parameters of a nuclear power plant remain stable. However, precisely because of this stability, they are easily overlooked during operational monitoring, lacking continuous dynamic monitoring. Consequently, abnormal changes in these parameters fail to be detected and identified in a timely manner until they reach an alarm threshold. The nuclear power plant critical parameter anomaly identification and alarm system of this invention employs a dual approach: a knowledge-driven first anomaly monitoring submodule and a data-driven second anomaly monitoring submodule, to identify anomalies in critical nuclear power plant parameters.

[0055] The second anomaly monitoring submodule, based on data-driven principles, is further divided into a normal state identification subunit and an anomaly feature identification subunit. During operation, real-time operating data of the unit is input into the normal state identification subunit, the anomaly feature identification subunit, and the first anomaly monitoring submodule, respectively. Each module determines whether an anomaly is present based on a preset model or anomaly discrimination condition and inputs its judgment into the integrated alarm submodule. The integrated alarm submodule makes a judgment based on preset judgment logic and finally provides an anomaly identification result. This enables continuous automatic monitoring of important parameters during normal unit operation and provides early warnings when important parameters deviate from normal operation, promptly reminding operators to pay attention to parameter change trends.

[0056] In this embodiment, the normal state identification subunit includes an abnormal alarm discrimination threshold unit, a normal state reconstruction model, a residual calculation unit, and an alarm unit.

[0057] During normal operation, the normal state reconstruction model is invoked, and the corresponding parameter combination from the real-time operation data of the unit is input into the normal state reconstruction model. The normal state reconstruction model outputs the corresponding reconstruction value. The residual calculation unit calculates N sets of residuals between the reconstruction value and the current value of each parameter in the parameter combination. When the residual exceeds the abnormal alarm judgment threshold, the alarm unit determines that there is an abnormality.

[0058] The normal state reconstruction model is based on the establishment of corresponding normal samples of historical unit operation data through data analysis techniques using historical time-series data of the unit. An evaluation model is then established based on these normal samples. For different detection objects, corresponding normal state reconstruction models are constructed. During the construction of the normal state reconstruction model, the parameter combinations most relevant to the detection object are selected as the training data for that model.

[0059] To identify as many anomalies as possible in the detected object, during the training of the normal state reconstruction model, a combination of parameters that can describe all modalities of the detected object is selected as the training data. In this embodiment, the normal state identification subunit extracts a combination of parameters that describes all modalities of the detected object from the historical operation data of the unit in the data warehouse as the training data for the normal state reconstruction model.

[0060] All modes include the following combinations: (1) identification and early warning of abnormal important parameters of primary circuit, secondary circuit, turbine, peripheral, electrical, waste and ventilation and fire protection system under the power operation platform of the unit at 2%, 8%, 17%, 30%, 50%, 80% and 100%; (2) identification and early warning of abnormal important parameters during the power reduction of the unit at 2MW / min and the power increase at 0.5MW / min; (3) identification and early warning of abnormal core parameters during the up and down operation of the unit in non-power operation.

[0061] In this embodiment, the various abnormal alarm discrimination thresholds in the abnormal alarm discrimination threshold unit are obtained based on the normal state reconstruction model, including the following steps:

[0062] After the normal state reconstruction model is built, the normal operation history data with corresponding parameter combinations is obtained from the data warehouse and input into the normal state reconstruction model for prediction, so as to predict the feature values ​​based on the normal operation history data; the feature values ​​include numerical values, numerical ranges and rates of change;

[0063] The difference between each feature value and the corresponding parameter value in the parameter combination is used as the abnormal alarm discrimination threshold in the abnormal alarm discrimination threshold unit.

[0064] In this embodiment, the alarm unit has preset alarm conditions. When the preset alarm conditions are met between the abnormal alarm judgment threshold and the residual, the normal state identification subunit determines that there is a possibility of an abnormality.

[0065] In this embodiment, the preset alarm conditions include the following two sets:

[0066] (1) At least one set of residuals exceeds the corresponding abnormal alarm judgment threshold;

[0067] (2) Within a specified time window, the proportion of cases exceeding the abnormal alarm judgment threshold is greater than the preset percentage.

[0068] (3) The specified time window must meet the preset length for cases that continuously exceed the abnormal alarm judgment threshold.

[0069] The preset alarm conditions are set in this way to minimize false alarms. Due to data fluctuations, business knowledge is incorporated into the alarm unit to reduce the false alarm rate. For example, business knowledge can confirm that no more than one-third of the data within one minute is fluctuating data. The alarm unit only needs to trigger an alarm when more than one-third of the data within one minute has a residual exceeding the anomaly alarm threshold. Furthermore, users can modify the anomaly alarm threshold based on experience. By leveraging the experience of operators, fixed upper and lower limits and rate of change thresholds for various parameters can be established, thereby achieving automatic monitoring of operating parameters. This shifts the focus from human error detection to problem identification, assisting operators in discovering unit anomalies and improving work efficiency and unit safety.

[0070] The anomaly feature identification model is based on data analysis techniques used to establish corresponding anomaly samples of historical unit operation data. An evaluation model is then built based on these anomaly samples. For different detection targets, corresponding anomaly feature identification models are constructed. During the construction of the anomaly feature identification model, the parameter combinations most relevant to the detection target are selected as the training data for that model.

[0071] The corresponding anomaly feature settings vary depending on the object being detected. In this embodiment, the anomaly feature recognition subunit extracts the anomaly features of the object being detected from the historical operating data of the unit in the data warehouse as training data for the anomaly feature recognition model. The anomaly features of the object being detected are the most relevant features under abnormal conditions, such as numerical values, numerical ranges, or rates of change.

[0072] During normal operation, the abnormal feature identification model is invoked, and the corresponding parameter combination in the real-time operation data of the unit is input into the abnormal feature identification model. The abnormal feature identification model determines whether there is an abnormality based on the degree of abnormal feature matching. When the abnormal feature identification subunit determines that the characteristics of the real-time operation data of the unit match the abnormal features in the abnormal feature identification model, it determines that there is an abnormality.

[0073] The anomaly identification model is built based on knowledge and experience; for different detection objects, corresponding anomaly feature identification models are constructed separately. In this embodiment, based on the unit's operating experience, the experience in judging abnormal conditions is converted into a computer-run configuration diagram to construct the anomaly identification model, including the following steps:

[0074] (1) For different detection objects, an exception event tree diagram is drawn using VISIO diagram based on operational experience. The exception event tree diagram includes multiple branches at different levels, and each branch corresponds to an exception type.

[0075] (2) Using knowledge modeling tools, based on the abnormal event tree diagram verified by unit operation experience, develop a knowledge-driven anomaly detection configuration diagram;

[0076] (3) Construct a corresponding anomaly identification model based on the anomaly detection configuration diagram;

[0077] During normal operation, the knowledge-driven anomaly detection configuration diagram is deployed to the data warehouse. Then, the real-time operation data of the unit is input into the corresponding anomaly identification model for processing. The first anomaly monitoring submodule judges the data characteristics of the corresponding parameters in the real-time operation data of the unit based on the anomaly identification rules described in the anomaly event tree diagram, and inputs the judgment result into the comprehensive alarm submodule for processing.

[0078] In this embodiment, the integrated alarm submodule includes a weight adjustment unit; the weight adjustment unit presets different weights for different inputs to generate corresponding integrated calculation results;

[0079] In the weight adjustment unit, the weights of the first anomaly monitoring submodule, the normal state identification submodule, and the anomaly feature identification submodule are adjusted according to the actual operation of the unit. For example, if in the actual operation of the unit, it is found that the anomaly feature identification model is more suitable for anomaly identification of a certain type of parameter, the normal state reconstruction model is second best, and the anomaly identification model is even worse, then for that type of parameter, a higher weight is set for the anomaly feature identification model in the integrated alarm submodule.

[0080] In this embodiment, the object of detection is an important parameter of nuclear power.

[0081] All the models mentioned above are detection models built based on historical operating data of the unit, and their accuracy is constrained by the data. Since historical operating data cannot cover all normal or abnormal states, anomaly detection models developed based on historical operating data will experience false alarms or missed alarms when encountering new normal or abnormal states. For example, as the performance of equipment such as fans and pumps decreases, or the efficiency of components such as heat exchangers decreases, differences arise between real-time operating data and historical operating data. For instance, reduced electrical power and different trends in heater condensate flow rates compared to historical data will lead to increased false alarm rates if previous models are used for detection. Therefore, an online update process is added; historical operating data is used for verification, and adjustments are made based on the new data.

[0082] In this embodiment, the nuclear power important parameter anomaly identification and alarm system also includes an online update module; the online update module quickly updates the data-driven normal state model and anomaly feature identification model based on user feedback and updated historical unit operation data.

[0083] In this embodiment, the online update module includes a manual judgment unit, an online update unit for the normal state reconstruction model, and an offline update unit for the abnormal feature recognition model.

[0084] The manual judgment unit judges the abnormal alarm status issued by the integrated alarm submodule;

[0085] If it is determined to be a false alarm, it means that there may be new normal data, and the new normal data does not exist in the unit's historical operating data. In this case, it will be transferred to the online update unit of the normal state reconfiguration model for processing.

[0086] The online update unit of the normal state reconstruction model adds new normal data to the training data of the normal state reconstruction model to obtain updated training data; subsequently, the normal state reconstruction model is retrained based on the updated training data to obtain an updated normal state reconstruction model.

[0087] And based on retraining or incremental learning, an updated normal state reconstruction model is obtained;

[0088] If it is determined to be a missed report, it means that there are features that should have been identified by the existing anomaly feature identification model, but were not actually identified. In this case, it will be transferred to the anomaly feature identification model offline update unit for processing.

[0089] The offline update unit of the anomaly feature recognition model extracts the features of missed reports and establishes an anomaly feature library. Subsequently, the anomaly feature recognition model is retrained based on the anomaly data in the anomaly feature library to obtain an updated anomaly feature recognition model.

[0090] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A nuclear power plant critical parameter anomaly identification and alarm system, characterized in that, It includes a data warehouse and at least one set of parameter anomaly identification modules located in the data warehouse; each parameter anomaly identification module receives real-time operating data of the unit as input to identify and alarm for specified nuclear power parameters; Each parameter anomaly identification module includes a knowledge-driven first anomaly monitoring submodule, a data-driven second anomaly monitoring submodule, and a comprehensive alarm submodule; The first anomaly monitoring submodule constructs an anomaly identification model based on the anomaly identification rules described in the preset anomaly event tree diagram; when the first anomaly monitoring submodule determines that an anomaly exists, it determines that the data characteristics of the corresponding parameters in the received real-time operation data of the unit match the description in the anomaly event tree diagram. The second anomaly monitoring submodule includes a normal state identification subunit and an anomaly feature identification subunit; The normal state identification subunit uses normal data from the unit's historical operation data as training data to build a normal state reconstruction model based on artificial intelligence technology; the normal state reconstruction model predicts the value range of the corresponding parameters based on the real-time operation data of the unit input online; The normal state identification subunit determines that there is an anomaly when the value range of the corresponding parameter in the received real-time operation data of the unit, predicted by the normal state reconstruction model, exceeds the normal range. The anomaly feature identification subunit uses anomaly data from the unit's historical operating data as training data to build an anomaly feature identification model based on artificial intelligence technology. When the anomaly feature identification subunit determines that an anomaly exists, it will find that the characteristics of the received real-time operation data of the unit match the anomaly features in the anomaly feature identification model. The integrated alarm submodule calculates the inputs from the first anomaly monitoring submodule, the normal state identification submodule, and the anomaly feature identification submodule based on preset weights, generates corresponding integrated calculation results, and determines whether there are any anomalies in the current unit operation data based on the integrated calculation results. If there are anomalies, an anomaly alarm is triggered.

2. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 1, characterized in that, The normal state identification subunit extracts parameter combinations describing all modalities of the detected object from the historical operation data of the unit in the data warehouse as training data for the normal state reconstruction model.

3. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 1, characterized in that, The normal state identification subunit includes an abnormal alarm discrimination threshold unit, a normal state reconstruction model, a residual calculation unit, and an alarm unit; During normal operation, the normal state reconfiguration model is invoked, and the corresponding parameter combinations from the unit's real-time operating data are input into the normal state reconfiguration model. The normal state reconfiguration model then outputs the corresponding reconfiguration value. The residual calculation unit calculates N sets of residuals between the reconfiguration value and the current values ​​of each parameter in the parameter combination. The alarm unit determines that an anomaly exists when the residual exceeds the anomaly alarm discrimination threshold.

4. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 3, characterized in that, The abnormal alarm discrimination thresholds in the abnormal alarm discrimination threshold unit are obtained based on the normal state reconstruction model, including the following steps: After the normal state reconstruction model is built, the normal operation history data with corresponding parameter combinations is obtained from the data warehouse and input into the normal state reconstruction model for prediction, so as to predict the feature value based on the normal operation history data. The difference between each feature value and the corresponding parameter value in the parameter combination is used as the abnormal alarm discrimination threshold in the abnormal alarm discrimination threshold unit.

5. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 3, characterized in that, The alarm unit has preset alarm conditions. When the abnormal alarm judgment threshold and the residual meet the preset alarm conditions, the normal state identification subunit judges that there is a possibility of an abnormality.

6. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 5, characterized in that, The preset alarm conditions include the following two sets: (1) At least one set of residuals exceeds the corresponding abnormal alarm judgment threshold; (2) Within a specified time window, the percentage of cases exceeding the abnormal alarm judgment threshold is greater than the preset percentage; (3) The specified time window must meet the preset length for cases that continuously exceed the abnormal alarm judgment threshold.

7. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 1, characterized in that, The anomaly feature identification subunit extracts the anomaly features of the detected objects from the historical operation data of the units in the data warehouse as training data for the anomaly feature identification model; During normal operation, the abnormal feature identification model is invoked, and the corresponding parameter combination in the real-time operation data of the unit is input into the abnormal feature identification model. The abnormal feature identification model determines whether there is an abnormality based on the degree of abnormal feature matching. When the abnormal feature identification subunit determines that the characteristics of the real-time operation data of the unit match the abnormal features in the abnormal feature identification model, it determines that there is an abnormality.

8. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 1, characterized in that, Based on unit operation experience, the experience in judging abnormal conditions is converted into a computer-generated configuration diagram to construct an anomaly identification model, including the following steps: (1) For different detection objects, an exception event tree diagram is drawn using VISIO diagram based on operational experience. The exception event tree diagram includes multiple branches at different levels, and each branch corresponds to an exception type. (2) Using knowledge modeling tools, based on the abnormal event tree diagram verified by unit operation experience, develop a knowledge-driven anomaly detection configuration diagram; (3) Construct the corresponding anomaly identification model based on the anomaly detection configuration diagram.

9. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 1, characterized in that, The integrated alarm submodule includes a weight adjustment unit; the weight adjustment unit presets different weights for different inputs to generate corresponding integrated calculation results; In the weight adjustment unit, the weights input to the first anomaly monitoring submodule, the normal state identification submodule, and the anomaly feature identification submodule are adjusted according to the actual operating conditions of the unit.

10. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 1, characterized in that, The nuclear power plant critical parameter anomaly identification and alarm system also includes an online update module; the online update module uses updated historical unit operating data based on user feedback to update the data-driven normal state model and anomaly feature identification model.

11. The nuclear power plant critical parameter anomaly identification and alarm system according to claim 10, characterized in that, The online update module includes a manual judgment unit, an online update unit for the normal state reconstruction model, and an offline update unit for the abnormal feature recognition model; the manual judgment unit judges the abnormal alarm status issued by the comprehensive alarm submodule. If it is determined to be a false alarm, it means that there may be new normal data, and the new normal data does not exist in the unit's historical operating data. In this case, it will be transferred to the online update unit of the normal state reconfiguration model for processing. The online update unit of the normal state reconstruction model adds new normal data to the training data of the normal state reconstruction model, and retrains the normal state reconstruction model based on the updated training data to obtain an updated normal state reconstruction model. If it is determined to be a missed report, it means that there are features that should have been identified by the existing anomaly feature identification model, but were not actually identified. In this case, it will be transferred to the anomaly feature identification model offline update unit for processing. The offline update unit of the anomaly feature recognition model extracts the features of missed reports and establishes an anomaly feature library. Based on the anomaly data in the anomaly feature library, the anomaly feature recognition model is retrained to obtain an updated anomaly feature recognition model.

Images