Power plant accident information quality assessment method and device

Abstract

The application provides a power plant accident information quality evaluation method and device, and relates to the technical field of power systems. The method comprises the following steps: obtaining alarm event data of a target alarm event to be displayed; obtaining at least one accident alarm record of a target power plant; historical processing feedback data comprises a processing operation type and an influence level in response to an accident alarm; determining an accident type code vector based on the accident type, the processing operation type and the influence level of the accident alarm record, and determining a location code vector based on the target power plant, the location code vector being used to indicate the difference between the target power plant and other power plants; determining historical event features for describing the behavior mode of the target power plant according to the accident type code vector and the location code vector; associating the historical event features with the alarm event data to generate an alarm importance score of the target alarm event, and displaying the target alarm event. The application can effectively improve the overall operation response efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, and in particular to a power plant accident information quality evaluation method and device. BACKGROUND

[0002] In the prior art, the power system control platform mainly receives accident alarm records from each power station. These records are usually in a structured or semi-structured form and contain brief information such as power plant, time, type, and basic parameter changes, which are used to trigger alarm prompts of the monitoring system and notify the operation personnel to intervene in processing. Specifically, the current processing mechanism mainly depends on preset alarm thresholds and simple classification rules, and combines artificial experience to judge the importance and urgency of the alarm. However, this kind of method generally lacks the ability to deeply evaluate the actual impact of the event, and it is difficult to distinguish between low-impact events that do not need urgent intervention and serious abnormalities that really need to be focused on, although they trigger alarms. Therefore, the prior art cannot automatically identify and highlight high-impact events that really need attention, which makes it difficult for operation and maintenance personnel to focus on key issues in a large number of low-impact alarms, and reduces the overall operation response efficiency. SUMMARY

[0003] To solve the above technical problems of the prior art, the present application provides a power plant accident information quality evaluation method and device, which solves the technical problem that it is difficult to focus on key issues in a large number of low-impact alarms due to the inability to automatically identify and highlight high-impact events that really need attention.

[0004] In a first aspect, the present application provides a power plant accident information quality evaluation method, comprising:

[0005] Obtaining alarm event data of a target alarm event to be displayed, wherein the alarm event data includes an alarm event type and a target power plant to which the target alarm event belongs;

[0006] Obtaining at least one accident alarm record of the target power plant, wherein the accident alarm record contains an accident type of an occurred accident alarm and historical processing feedback data, and the historical processing feedback data includes a processing operation type and an impact level in response to the accident alarm;

[0007] Based on the accident type, processing operation type and impact level of the accident alarm record, an accident type code vector is determined, and a location code vector is determined based on the target power plant, wherein the location code vector is used to indicate the difference between the target power plant and other power plants;

[0008] According to the accident type code vector and the location code vector of the accident alarm record, a historical event feature for describing the behavior pattern of the target power plant is determined;

[0009] correlate the historical event feature with the alarm event data to generate an alarm importance score of the target alarm event;

[0010] display the target alarm event according to a priority mapped with the alarm importance score.

[0011] In some embodiments, the determining of the accident type code vector based on the accident type, the processing operation type and the impact level of the accident alarm record, and the determining of the location code vector based on the target power plant, comprises:

[0012] encoding the accident type, the processing operation type and the impact level of the accident alarm record to obtain an accident alarm fusion feature;

[0013] converting the accident alarm fusion feature into an accident type code vector through a preset first embedding model, the accident type code vector being used to indicate the correlation between the accident type, the processing operation type and the impact level;

[0014] encoding the power plant identifier of the target power plant to obtain a power plant feature;

[0015] converting the power plant feature into a location code vector through a preset second embedding model.

[0016] In some embodiments, the accident alarm record further comprises an alarm timestamp of the accident alarm; and the determining of the historical event feature for describing the behavior pattern of the target power plant based on the accident type code vector and the location code vector of the accident alarm record, comprises:

[0017] arranging the accident alarm records occurred in the target power plant in ascending order of the alarm timestamp to obtain an accident type code sequence of the target power plant;

[0018] extracting time features from the accident type code sequence by using a pre-constructed feature extraction model to obtain a forward hidden state and a backward hidden state, and splicing the forward hidden state and the backward hidden state to obtain a context-aware feature, the context-aware feature being used for the forward and backward dependency relationship between the accident alarm records in the accident type code sequence;

[0019] fusing the location code vector and the context-aware feature to obtain a historical fusion feature of the accident alarm record;

[0020] determine a time decay weight of the accident alarm record based on an alarm timestamp of the accident alarm record, and perform weighted calculation on the historical fusion feature of each accident alarm record occurring in the target power plant and the time decay weight to obtain a historical event feature of the target power plant.

[0021] In some embodiments, the feature fusion of the position encoding vector and the context awareness feature to obtain the historical fusion feature of the accident alarm record comprises:

[0022] perform information reinforcement on the position encoding vector by using a preset first multi-layer perception machine to obtain a position importance bias vector;

[0023] splice the position importance bias vector and the context awareness feature to obtain spliced features, and process the spliced features by using a preset second multi-layer perception machine to obtain the historical fusion feature;

[0024] The determination of the time decay weight of the accident alarm record based on the alarm timestamp of the accident alarm record, and the weighted calculation on the historical fusion feature of each accident alarm record occurring in the target power plant and the time decay weight to obtain the historical event feature of the target power plant comprises:

[0025] determine a time decay weight of the accident alarm record based on an alarm timestamp of the accident alarm record, and perform weighted calculation on the historical fusion feature of each accident alarm record occurring in the target power plant and the time decay weight to obtain a historical event feature of the target power plant.

[0026] The determination of the time decay weight of the accident alarm record based on an alarm timestamp of the accident alarm record, and perform weighted calculation on the historical fusion feature of each accident alarm record occurring in the target power plant and the time decay weight to obtain a historical event feature of the target power plant.

[0027] In some embodiments, the association of the historical event feature with the alarm event data to generate the alarm importance score of the target alarm event comprises:

[0028] determine an event type feature based on an alarm event type of the target alarm event, and determine a target power plant feature based on the target power plant, and splice the event type feature and the target power plant feature into an event feature;

[0029] calculate a semantic similarity between the event feature and the historical event feature, the semantic similarity being used to indicate the relevance between the target alarm event and any accident alarm record;

[0030] The semantic similarity associated with each accident alarm record is superimposed to obtain a target similarity, and a preset third multi-layer perception machine is used to process the target similarity to obtain a context modulation factor;

[0031] Based on the context modulation factor, the event feature and the historical event feature, an alarm importance feature of the target alarm event is determined;

[0032] Based on a preset alarm importance weight matrix and a preset alarm importance bias vector, the alarm importance feature is mapped to an alarm importance score.

[0033] In a second aspect, the application provides a power plant accident information quality evaluation device, comprising:

[0034] An event data acquisition module is configured to acquire alarm event data of a target alarm event to be displayed, wherein the alarm event data comprises an alarm event type and a target power plant to which the target alarm event belongs;

[0035] A historical data acquisition module is configured to acquire at least one accident alarm record of the target power plant, wherein the accident alarm record comprises an accident type of an occurred accident alarm and historical processing feedback data, and the historical processing feedback data comprises a processing operation type and an influence level in response to the accident alarm;

[0036] An encoding module is configured to determine an accident type encoding vector based on the accident type, the processing operation type and the influence level of the accident alarm record, and determine a location encoding vector based on the target power plant, wherein the location encoding vector is used to indicate the difference between the target power plant and other power plants;

[0037] A historical event feature extraction module is configured to determine a historical event feature used to describe the behavior mode of the target power plant according to the accident type encoding vector and the location encoding vector of the accident alarm record;

[0038] An alarm importance evaluation module is configured to associate the historical event feature with the alarm event data to generate an alarm importance score of the target alarm event;

[0039] A display module is configured to display the target alarm event according to a priority mapped with the alarm importance score.

[0040] The power plant accident information quality evaluation method and device provided by the application, by the accident type, processing operation type and influence level in the accident alarm record, determine the accident type code vector, which integrates the accident type and historical processing feedback, and by the target power plant to which the target alarm event belongs, determine the location code vector, taking into account the differences between the target power plant and other power plants, and then by the accident type code vector and the location code vector, determine the historical event characteristics of the target power plant as the historical experience benchmark for evaluating the target alarm event, improve the reliability of evaluating the target alarm event, by associating the target alarm event with the historical event characteristics, generate the alarm importance score of the target alarm event, and thus display according to the alarm importance score, improve the identification ability of real high-impact events, and improve the operation response efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A flowchart of a power plant accident information quality evaluation method provided by the application;

[0042] Figure 2 A structural diagram of a power plant accident information quality evaluation device provided by the application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0044] As Figure 1 shown, the application provides a power plant accident information quality evaluation method, which comprises steps S101 to S106.

[0045] S101, obtain alarm event data of a target alarm event to be displayed, the alarm event data comprising an alarm event type and a target power plant to which the target alarm event belongs.

[0046] In the application, the target alarm event refers to an alarm to be disposed triggered by real-time monitoring data of a power plant, which reflects abnormality of equipment of the power plant or existing safety risks, for example, abnormality of a boiler water level, abnormality of a transformer temperature, etc. The alarm event type comprises but is not limited to a device fault type, a parameter out-of-limit type, an environmental abnormality type, etc. The target power plant represents a specific power plant where the target alarm event occurs.

[0047] S102, acquire at least one accident alarm record of the target power plant, the accident alarm record comprising an accident type of an occurred accident alarm and historical processing feedback data, the historical processing feedback data comprising a processing operation type and an impact level in response to the accident alarm.

[0048] In the present application, the accident alarm record represents the archived data of an accident event occurred in the target power plant and has been disposed. The accident type represents the category of the accident alarm occurred in the power plant historically, such as equipment short circuit, bearing temperature exceeding, valve sticking, power grid frequency fluctuation, etc., to distinguish different nature faults or alarms. It should be noted that the accident type of the accident alarm record extracted in the present application corresponds to the alarm event type of the target alarm event. For example, the target alarm event is transformer temperature overrun, and the record of "transformer temperature related accident" occurred in the power plant needs to be collected. Specifically, the historical processing feedback data represents the record of the historical accident disposal process and disposal result. The processing operation type represents the operation type of the disposal action actually performed for the accident alarm, for example, emergency shutdown for maintenance, remote parameter adjustment, and no abnormality after on-site troubleshooting. The impact level represents the impact degree actually caused by the historical accident alarm, i.e., the actual consequence of the historical accident alarm.

[0049] Therefore, the present application introduces the accident alarm record occurred in the target power plant, combines the historical disposal feedback of the power plant itself, makes the evaluation of the target alarm event fit the actual operation situation of the power plant, improves the accuracy of the event evaluation, realizes the differentiation of the alarm event with high actual impact degree, and can effectively improve the operation efficiency.

[0050] S103, determine an accident type code vector based on the accident type, the processing operation type, and the impact level of the accident alarm record, and determine a location code vector based on the target power plant, the location code vector being used to indicate the difference between the target power plant and other power plants.

[0051] In the present application, the accident type, the processing operation type, and the impact level are encoded and converted into a vector to obtain an accident type code vector, which reflects the quantitative characteristics (i.e., data fusion characteristics) of the accident type, the disposal method, and the impact degree of the accident alarm, i.e., reflects the properties of the accident itself. In addition, the difference characteristics (such as unit type, installed capacity, geographical location, historical accident frequency, etc.) of the power plant are converted into a vector through the power plant identifier of the target power plant to obtain a location code vector. The location code vector represents the difference between the target power plant and other power plants, i.e., the difference between different power plants in physical characteristics, operation scale, region, etc., and the location code vector reflects the characteristics of the power plant scene properties.

[0052] In some embodiments, the determining, based on the accident type, the handling operation type, and the impact level of the accident alarm record, an accident type code vector and determining, based on the target power plant, a location code vector comprises:

[0053] encoding the accident type, the handling operation type, and the impact level of the accident alarm record to obtain an accident alarm fusion feature;

[0054] converting the accident alarm fusion feature into an accident type code vector through a preset first embedding model, the accident type code vector being used to indicate the correlation between the accident type, the handling operation type, and the impact level;

[0055] encoding the power plant identifier of the target power plant to obtain a power plant feature;

[0056] converting the power plant feature into a location code vector through a preset second embedding model.

[0057] In the embodiment, the accident type, the handling operation type, and the impact level of the accident alarm record are converted into numerical forms to obtain an accident alarm fusion feature. The encoding mode can be one-hot encoding. Then, the first embedding model (Embedding) is used to convert the accident alarm fusion feature into an accident type code vector representing a continuous vector, which integrates the accident itself and historical handling feedback to improve the evaluation accuracy.

[0058] In addition, the power plant identifier of the target power plant is encoded to obtain a power plant feature, and the encoding mode can be one-hot encoding. The power plant feature is used to identify the identity information of the power plant. The power plant identifier is used to uniquely distinguish the symbols or strings of different power plants, such as “power plant A01” and “hydropower plant 05”. Then, the second embedding model (Embedding) is used to convert the one-hot encoding of the power plant identifier into a continuous vector to obtain a location code vector, which reflects the physical characteristics, operation scale, and regional differences of the power plant, and is used to distinguish the characteristics of different power plants.

[0059] For example, the accident type code vector is calculated by the following formula:

[0060] ;

[0061] wherein, represents the accident type code vector of the i th accident alarm record, represents the accident type of the i th accident alarm record, represents the handling operation type of the i th accident alarm record, represents the impact level of the i th accident alarm record.

[0062] The position coding vector is calculated by the following formula:

[0063] ;

[0064] wherein, represents the position coding vector of the i th accident alarm record, represents the power plant to which the i th accident alarm record belongs.

[0065] Therefore, the embodiment realizes the conversion of the power plant difference into a calculable vector difference. In the power plant accident information quality assessment, the importance of the alarm of the same accident type in different power plants may be different (for example, the fan failure is more important in the wind power-based power plant than in the thermal power-based power plant), and the difference (such as the equipment characteristics and the difference in the operating environment) of different power plants is reflected through the position coding vector, thereby improving the accuracy of the current alarm event assessment.

[0066] S104, determining a historical event feature for describing the behavior mode of the target power plant according to the accident type coding vector and the position coding vector of the accident alarm record.

[0067] In the present application, the accident type coding vector and the position coding vector are subjected to feature extraction to obtain the historical event feature of the target power plant, which reflects the experience law of the accident frequency, the disposal habit, and the influence consequence of the power plant, that is, the behavior mode of the power plant. The behavior mode represents the correlation mode of the event-response-consequence formed in the disposal of the power plant, and embodies the equipment characteristics and the operation and maintenance habit of the power plant.

[0068] Therefore, the embodiment converts the accident alarm record into the behavior mode by determining the historical event feature for describing the behavior mode of the target power plant, uses the behavior mode and the law of the power plant to assess the importance of the current alarm event, and improves the reliability of the importance assessment.

[0069] S105, associating the historical event feature with the alarm event data to generate an alarm importance score of the target alarm event.

[0070] In the present application, the historical event feature is matched with the alarm event data, and the corresponding alarm importance score is generated based on the matching result, so as to realize the improvement of the objectivity and the operability of the assessment through the association of the historical law.

[0071] S106, displaying the target alarm event according to the priority mapped with the alarm importance score.

[0072] It should be noted that the corresponding rule of configuring the alarm importance score and the priority is configured so that the target alarm event is displayed differently according to the mapped priority. The priority represents the classification of the emergency order of the current alarm based on the alarm importance score, and directly guides the response priority. Therefore, the application differentiates the display of the alarm event through the priority, identifies the alarm event that is urgent and important (i.e., high impact), and improves the handling efficiency.

[0073] In some embodiments, the accident alarm record further comprises an alarm timestamp of the accident alarm; and the historical event feature used to describe the behavior pattern of the target power plant is determined according to the accident type code vector and the position code vector of the accident alarm record, including:

[0074] The accident alarm records occurring in the target power plant are arranged in ascending order of the alarm timestamp to obtain an accident type code sequence of the target power plant;

[0075] The accident type code sequence is subjected to time feature extraction by using a pre-constructed feature extraction model to obtain a forward hidden state and a backward hidden state, and the forward hidden state and the backward hidden state are spliced to obtain a context-aware feature, which is used for the forward and backward dependency relationship between each accident alarm record in the accident type code sequence;

[0076] The position code vector and the context-aware feature are subjected to feature fusion to obtain a historical fusion feature of the accident alarm record;

[0077] The time decay weight of the accident alarm record is determined based on the alarm timestamp of the accident alarm record, and the historical fusion feature and the time decay weight of each accident alarm record occurring in the target power plant are subjected to weighted calculation to obtain the historical event feature of the target power plant.

[0078] In the embodiment, the alarm timestamp represents the time record when the accident alarm occurs, and then the accident alarm records are sorted in the order of time from the past to the present. The accident type code sequence table represents a one-dimensional ordered array composed of the accident type code vectors of the target power plant in chronological order. Then, the feature extraction model refers to a bidirectional recurrent neural network (BiRNN) and its variants (such as BiLSTM, BiGRU) for processing time series data, aiming to capture the forward and backward dependencies of the elements in the accident type code sequence. Then, the forward hidden state is obtained by using the feature extraction model, specifically, the feature vector output by the feature extraction model at each time step when traversing the sequence in the direction of “time from early to late”, which contains the information of all accidents before this time step. At the same time, the backward hidden state is obtained by using the feature extraction model, specifically, the feature vector output by the feature extraction model at each time step when traversing the sequence in the direction of “time from late to early”, which contains the information of all accidents after this time step. Thus, by splicing the forward hidden state and the backward hidden state, the context-aware feature is obtained, which reflects the bidirectional dependency between accident alarms, for example, after the event of high boiler pressure alarm, the event of safety valve action alarm often occurs, and at the same time, the event of safety valve action alarm also reversely verifies the authenticity of the event of high boiler pressure alarm.

[0079] Further, by feature fusion of the position encoding vector and the context-aware feature, the position-enhanced context-aware feature is obtained, which realizes the fusion of the individualized attributes of the target power plant (i.e. power plant difference) and the time correlation logic of the target power plant accident (i.e. accident time correlation), and improves the adaptability of the feature to the power plant scene.

[0080] Secondly, according to the difference between the alarm timestamp of each accident alarm record and the occurrence time of the current target alarm event, the time decay weight of the accident alarm record is calculated by a preset decay function (such as exponential decay, linear decay). Exemplarily, the time decay weight indicates that the closer the accident, the higher the weight; the farther the accident, the lower the weight.

[0081] Further, the historical fusion features (i.e. position-enhanced context-aware features) of all accident alarm records of the target power plant are respectively summed with their respective time decay weights to obtain the historical event features representing the overall accident behavior pattern of the target power plant.

[0082] Therefore, the embodiment constructs an accident type coding sequence, extracts context-aware features from the accident type coding sequence, fuses the location coding vector and the context-aware features to obtain historical fusion features of the accident alarm record, determines a time decay weight according to an alarm timestamp of the accident alarm record, and obtains historical event features of the target power plant by the historical fusion features and the time decay weight, thereby strengthening the timeliness of the features, making the historical fusion features fit the current state of the target power plant, improving the accuracy of the historical event features, and providing the accuracy of the alarm event evaluation.

[0083] In some embodiments, the fusing the location coding vector and the context-aware features to obtain the historical fusion features of the accident alarm record comprises:

[0084] performing information strengthening on the location coding vector by using a preset first multi-layer perception machine to obtain a location importance bias vector;

[0085] splicing the location importance bias vector and the context-aware features to obtain spliced features, and processing the spliced features by using a preset second multi-layer perception machine to obtain the historical fusion features.

[0086] In the embodiment, the first multi-layer perception machine is used to strengthen key information related to accident evaluation in the location coding to obtain the location importance bias vector. The location importance bias vector can highlight power plant difference information that has an impact on accident importance score, and weaken irrelevant differences (such as non-functional differences of power plant names and geographical positions). Then, the location importance bias vector and the context-aware features are spliced to obtain spliced features, and the integration between power plant differences and time-dependent features is realized. The second multi-layer perception machine is used to process the spliced features to extract the correlation between the features, and the historical fusion features reflect the integrated power plant difference information and historical accident time-dependent information.

[0087] For example, the location importance bias vector is determined by the following formula:

[0088] ;

[0089] wherein, represents the location importance bias vector of the i th accident alarm record.

[0090] The historical fusion features are determined by the following formula:

[0091] ;

[0092] wherein, represents the historical fusion features of the i th accident alarm record, represents a context-aware feature, k represents the number of accident alarm records, represents a concatenation operation.

[0093] Therefore, the embodiment realizes highlighting the power plant difference information that has an impact on the accident importance score and weakening irrelevant differences by determining the position importance bias vector, avoids irrelevant information in the power plant identification interfering with accident assessment, and focuses on the difference features (such as device configuration and operation and maintenance capability) related to accident handling and impact level. At the same time, it is ensured that the historical accident features of different power plants can reflect their own accident occurrence rules when being fused. By determining the historical fusion features, highlighting the power plant differentiation and time-dependent features, the accident recurrence risk of the power plant can be accurately reflected, the effective association is strengthened, and the accuracy of the alarm event assessment is improved.

[0094] In some embodiments, the time decay weight of the accident alarm record is determined based on the alarm timestamp of the accident alarm record, and the historical fusion features of each accident alarm record occurring in the target power plant and the time decay weight are weighted and calculated to obtain the historical event features of the target power plant, including:

[0095] The time decay weight of the accident alarm record is determined based on the alarm timestamp of the accident alarm record, the occurrence timestamp of the target alarm event, and a preset time decay rate. The time decay weight is used to indicate the influence degree of the accident alarm record with a time difference value less than a preset time difference threshold on the target alarm event. The time difference value is determined by the alarm timestamp of the accident alarm record and the occurrence timestamp.

[0096] The time decay weight and the historical fusion features of any accident alarm record occurring in the target power plant are calculated to obtain the historical event features of the target power plant.

[0097] In the embodiment, the time decay weight of the accident alarm record is determined by the alarm timestamp of the accident alarm record, the occurrence timestamp of the target alarm event, and the time decay rate, and the reference value of the historical accident to the current alarm event is considered.

[0098] For example, the time decay weight is determined by the following formula:

[0099] ;

[0100] wherein, represents the time decay weight of the i th accident alarm record, represents the time decay rate, represents the alarm timestamp of the i th accident alarm record, represents the occurrence timestamp of the target alarm event, a serial number representing a target alarm event.

[0101] The historical event feature is determined by the following formula:

[0102]

[0103] wherein, represents the historical event feature of the target power plant, and k represents the number of accident alarm records.

[0104] Therefore, the embodiment obtains the historical event feature of the target power plant through weighted calculation of the historical fusion feature and the time decay weight, ensures that the recent and highly correlated historical accident information accounts for a higher proportion in the final feature, and better reflects the potential risk of the current alarm. At the same time, the accident occurrence regularity of the target power plant is quantified, historical experience and behavior regularity are provided for the evaluation of the current alarm event, and the accuracy of the score is improved.

[0105] Therefore, in the present application, the accident type coding sequence of the target power plant is constructed, the accident alarm records of the power plant are arranged in ascending order of timestamp, the context-aware feature is generated by using the bidirectional long short-term memory network to deeply mine the context dependence relationship of the accident type coding sequence, the time sequence evolution law between events is captured, the position importance bias vector is used to perform nonlinear mapping on the position coding vector, the position-enhanced context feature is formed to fuse the spatial context information. At the same time, the embodiment dynamically weights the historical accident alarms through the time decay weight, so that the recent events obtain a higher weight, and finally the historical event feature is obtained through weighted aggregation. In this way, the embodiment can accurately evaluate the alarm quality because the disposal value of the power plant alarm is highly dependent on time sensitivity, that is, for the same power plant, the same type of event occurring recently is more likely to reflect the current operation hidden danger. The embodiment automatically suppresses the interference of old events through the time decay weight, strengthens the contribution of key historical segments, so that the historical event feature can not only reflect the cumulative effect of accidents at the same position, but also distinguish the operation characteristics of different power plants, thereby filtering out inert alarms that trigger alarms but actually have decreasing impact, and significantly improving the recognition ability of real high-impact events.

[0106] In some embodiments, the association of the historical event feature with the alarm event data generates an alarm importance score of the target alarm event, including:

[0107] Based on the alarm event type of the target alarm event, an event type feature is determined, and a target power plant feature is determined based on the target power plant, and the event type feature and the target power plant feature are spliced into an event feature;

[0108] ​Calculate the semantic similarity between the event features and the historical event features, whereby the semantic similarity is used to indicate the correlation between the target alarm event and any accident alarm record;

[0109] The semantic similarity associated with each accident alarm record is superimposed to calculate the target similarity, and the target similarity is processed by a preset third multilayer perceptron to obtain the context modulation factor.

[0110] Based on the context modulation factor, the event characteristics, and the historical event characteristics, the alarm importance characteristics of the target alarm event are determined;

[0111] Based on a preset alarm importance weight matrix and a preset alarm importance bias vector, the alarm importance features are mapped to alarm importance scores.

[0112] For example, event characteristics can be determined using the following formula:

[0113] ;

[0114] in, Indicates the first The event characteristics of an alarm event (i.e., the target alarm event). Indicates the first Event type characteristics of each alarm event Indicates the first The characteristics of the target power plant for each alarm event.

[0115] The importance characteristics of alarms are determined using the following formula:

[0116] ;

[0117] in, Indicates the first The importance characteristics of each alarm event. This represents the context modulation factor.

[0118] The alarm importance score is determined using the following formula:

[0119] ;

[0120] in, Indicates the first Alarm importance score for each alarm event This represents the alarm importance weight matrix. This represents the alarm importance bias vector.

[0121] In the embodiment, by constructing the alarm importance feature, the alarm event type and the location encoding vector of the current alarm event are fused into the event feature, and the semantic similarity calculation is performed with the historical event feature to capture the matching degree of the event and the historical behavior pattern of the power plant. Then, by context modulation gain, the similarity accumulation result is nonlinearly mapped to generate a context modulation factor. The context modulation factor is used for the fusion of the historical event feature and the current event feature, which improves the accuracy of feature fusion. Thus, the context modulation factor is multiplied by the event feature to add the historical event feature to form the alarm importance feature.

[0122] It should be noted that, since the actual impact of the power plant alarm is often implied in the historical experience of the historical disposal feedback, when a certain type of alarm frequently occurs but the historical disposal is marked as low impact, the semantic similarity will be significantly higher, and the context modulation factor will automatically suppress the feature weight of such event to avoid misjudgment of inertial alarm as high risk. Conversely, if the event is highly deviated from the historical high-impact pattern, the similarity is reduced, and the context modulation factor strengthens the feature expression to ensure that the real anomaly obtains a higher score. Therefore, by the context modulation factor, the event feature and the historical event feature, the alarm importance feature of the target alarm event is determined, which can accurately distinguish the alarm value. Secondly, by quantifying the historical experience by semantic similarity, the feature is adaptively modulated by dynamic gain, which not only avoids isolated judgment of a single event, but also overcomes the noise interference of historical data, so that the alarm importance feature can truly reflect the actual value of the event in the power plant operation process, thereby filtering low-impact alarms and significantly improving the discrimination accuracy and decision reliability of the evaluation result.

[0123] In some embodiments, the method further comprises:

[0124] obtaining the operating condition and the topology structure of the target power plant, wherein the operating condition includes the overall operating state of the power plant and the device operating state and the device associated state of the plurality of devices in the target power plant, the associated state is used to indicate the concurrent alarm event and the associated device data of the associated device accompanied by the target alarm event, and the overall operating state of the power plant includes the regional power grid load level and the scheduling requirement of the region where the target power plant is located, and the topology structure is used to indicate the connection relationship between the devices;

[0125] determining a first impact degree based on the operating condition and the topology structure, the first impact degree being used to indicate the correlation between the target alarm event, the processing device and the associated device;

[0126] matching the accident alarm record and the impact level related to the alarm event type of the target alarm event;

[0127] determine a second influence degree based on the device running state, the device association state, and the power plant overall running state, the second influence degree being used to indicate an influence degree of the running condition on the target alarm event;

[0128] determine a real-time influence factor based on the first influence degree and the second influence degree, and determine a historical correlation coefficient according to the real-time influence factor and the matched influence level;

[0129] correct a priority mapped with the alarm importance score according to the first influence degree, the second influence degree, the influence level, and the historical correlation coefficient, to obtain an execution priority, and indicate processing of the target alarm event according to the execution priority.

[0130] In the embodiment, the running condition represents a current dynamic running state set of the power plant and the device, and the embodiment reflects the current risk environment through the real-time condition. The topology represents the physical connection and functional dependence relationship between the devices, and is used to determine whether a device failure will affect other devices or the entire system. For example, the overall running state of the power plant can be the load rate (full load / half load / low valley) of the power plant, whether it is in the start-stop stage, whether it is executing a power protection task, and the like. The regional power grid load level represents the overall electricity demand (such as summer peak load and deep night low valley load) of the region where the target power plant is located, and reflects the power supply pressure of the power plant, that is, the alarm processing needs to be more urgent during the peak period to avoid affecting the regional power supply. The device running state can be the real-time parameters (such as temperature, pressure, current, vibration value, and whether it is supercritical threshold) of the device and the associated device involved in the target alarm event. The device association state can be a concurrent alarm event (indicating a possible chain failure) occurring simultaneously with the target alarm event, for example, the “boiler water level abnormality” event is accompanied by the “low pressure of the feed water pump” event, and the real-time data of the associated device (such as the flow and speed of the feed water pump).

[0131] Then, the correlation between the target alarm event, the processing device, and the associated device is calculated through the association relationship between the devices, the concurrent alarm relationship in the running condition, and the physical connection relationship in the topology, and the first influence degree is determined, that is, the device importance, the connection tightness, and the number of concurrent alarms. Further, the risk degree of the running condition on the target alarm event is calculated through the device running state in the running condition, the overall state of the power plant, and the power grid state, that is, the direct influence of the real-time running condition on the target alarm risk, and the risk change caused by the difference in the running condition is realized.

[0132] Further, the real-time influence factor represents quantification of the device association risk and the operation condition risk of the current scene. By quantifying the correlation strength between the real-time influence factor and the influence level, a historical correlation coefficient is determined to reflect the reference value of the influence level in the history to the current real-time scene. Thus, the priority is corrected in combination with the first influence degree, the second influence degree, the influence level and the historical correlation coefficient to obtain an execution priority, so as to process the current alarm event.

[0133] Therefore, the embodiment combines the historical alarm experience with the real-time operation environment to avoid the situation that only relying on the historical experience cannot cope with the real-time dynamic scene, adaptively executes the scene, improves the alarm processing efficiency, and meets the actual needs of power plant operation and maintenance.

[0134] In some embodiments, the first influence degree is determined based on the operation condition and the topology structure, including:

[0135] The event influence degree of the target alarm event is determined based on the topology structure and the operation condition, wherein the event influence degree includes an event influence range and an event influence time, the event influence range is used to indicate at least one device affected by the target alarm event, the number of devices and the first device correlation between devices, and the event influence time is used to indicate the time length between the occurrence time stamp of the target alarm event and the time when the target alarm event is completed.

[0136] The device influence degree of the processing device is determined based on the topology structure, the device operation state and the device association state, wherein the device influence degree includes a device influence time of the processing device, a device processing operation difficulty, a device importance degree and a second device correlation between the processing devices, and the device influence time is used to indicate the time length between the time when the processing device is abnormal due to the influence of the target alarm event and the time when it returns to normal.

[0137] The concurrent influence degree of the concurrent alarm event is determined based on the topology structure and the device association state, wherein the concurrent influence degree includes the number of events of the concurrent alarm event, a concurrent event processing difficulty, a concurrent event importance degree, an event correlation degree between the concurrent alarm events and the association device data of the corresponding association devices, the concurrent event processing difficulty is used to indicate the association device operation difficulty of the association devices involved in processing the concurrent alarm event, and the concurrent event importance degree is used to indicate the importance of the concurrent alarm event compared to the target power plant and the target alarm event.

[0138] The first influence degree is obtained based on the event influence degree, the device influence degree and the concurrent influence degree.

[0139] In the embodiment, the event influence degree represents a quantitative description of the overall influence scale of the target alarm event, reflecting how large the influence range is and how long the influence lasts, to determine whether the alarm needs emergency response. Among them, the device of the target alarm event is located based on the topology structure, the upstream and downstream device levels along the topology link are traced, and the affected device level and the number of devices are determined. The device level represents the device level divided according to the importance of the topology (such as first-level devices are core production devices, second-level devices are auxiliary devices, and third-level devices are auxiliary facilities), or the device level according to the functional causal relationship of the device. The first device correlation represents the functional dependence strength between the affected processing devices, identifying the global influence of the event level. The event influence time represents the total time from the occurrence of the target alarm event to the complete processing, directly reflecting the interference period of the alarm on the continuous operation of the power plant. Therefore, by determining the event influence degree, the influence quantification of the device structure, device level, device number, device correlation and the like is considered.

[0140] Then, the device influence time represents the time from the abnormality (triggered by the alarm) to the recovery of the processing device; the device processing operation difficulty represents the technical complexity and operation risk required for disposal processing device, which can be determined by the complexity of the processing device, the operation process and the historical operation success rate; the device importance is determined by the non-replaceability of the processing device in the production link of the power plant based on the topology structure; and the second device correlation represents the cooperative dependence strength between multiple processing devices, and high correlation means that the operation needs to be strictly synchronized, otherwise it is easy to cause new risks. Therefore, by determining the device influence degree, the importance of the processing device is determined, and the potential problems of multi-device cooperation are identified, so as to reduce the risk in the disposal process.

[0141] Further, the concurrent influence degree represents the chain failure risk caused by the concurrent alarm which occurs at the same time as the target alarm. The concurrent alarm event represents other alarms overlapping in time with the target alarm event, which may be the result of the target alarm event (such as abnormal boiler water level causing low water pump pressure) or the same source (such as power grid fluctuation causing voltage abnormality and current fluctuation). The concurrent event processing difficulty represents the overall operation complexity required for disposing all concurrent alarms. The concurrent event importance can be determined by combining the topology structure to determine the device level to which the concurrent alarm belongs, and comparing the importance of the current alarm event. The event correlation degree represents the system belonging consistency and causal correlation between the concurrent alarms and between the concurrent alarms and the current target alarm event, to determine whether it is a chain failure. Therefore, by determining the concurrent influence degree, the pressure of quantitative multi-alarm cooperative disposal is realized, and the chain failure risk caused by the current alarm event is improved.

[0142] Therefore, the first influence degree is obtained by the event influence degree, the device influence degree and the concurrent influence degree in the embodiment, the correlation and influence degree between the global influence of the event, the risk of the processing device and the risk of the concurrent chain are reflected, and it is ensured that the priority correction is more consistent with the actual device operation and the associated logic of the power plant, and the accuracy of the execution priority of the subsequent current alarm event is improved.

[0143] In some embodiments, the device operation state includes a working condition state and a load rate; and the second influence degree is determined based on the device operation state, the device association state and the overall operation state of the power plant, including:

[0144] When it is detected that the working condition state is not in a specified critical working condition, and the load rate is within a preset normal load rate range, a device risk coefficient is determined based on the device operation state, wherein the critical working condition is a peak load stage or a start-stop stage; or, when it is detected that the working condition state is in the critical working condition, and the load rate exceeds the normal load rate range, the device risk coefficient is increased based on the device operation state; or, when it is detected that the working condition state is in a shutdown state or a standby state, the device risk coefficient is decreased based on the device operation state;

[0145] When it is detected that the number of concurrent alarm events is less than a preset first number threshold, and the event correlation degree is less than a preset correlation degree threshold, a concurrent correlation coefficient is determined based on the device association state; or, when it is detected that the number of concurrent alarm events is greater than the first number threshold, and the event correlation degree is greater than the correlation degree threshold, the concurrent correlation coefficient is increased based on the device association state; or, when it is detected that the number of concurrent alarm events is greater than a preset second number threshold, or the event correlation degree reaches a preset cross-system alarm condition, the increased concurrent correlation coefficient is increased based on the device association state, and the second number threshold is greater than the first number threshold;

[0146] When it is detected that the regional power grid load level is at a flat load level or a valley load level, a power grid constraint coefficient is determined based on the overall operation state of the power plant, and the power grid constraint coefficient is used to indicate the emergency degree of processing the target alarm event; or, when it is detected that the regional power grid load level is at a peak load level, the power grid constraint coefficient is increased based on the overall operation state of the power plant;

[0147] The second influence degree is determined based on the device risk coefficient, the concurrent correlation coefficient and the power grid constraint coefficient.

[0148] In the embodiment, when in non-critical working condition and normal load, the device risk coefficient can be calculated based on the current parameters of the device (e.g. temperature, vibration value deviating from the normal threshold). Then, when in critical working condition (i.e. the device is in peak load stage or start-stop stage (large fluctuation of device parameters), such as summer peak power consumption, the device is in full load operation) and over normal load, the device is in a high risk scenario, because the fault tolerance rate of the device is low in the critical working condition, and abnormal load is easy to cause failure, so the device risk coefficient is increased. In addition, when in shutdown (has exited operation) / standby state (does not bear actual load), the device risk coefficient is decreased, because the device has little impact on production at this time, and the risk is controllable. Therefore, by increasing / decreasing the device risk coefficient, real-time risk is reflected, and invalid disposal resource consumption is reduced.

[0149] In addition, if the number of concurrent alarms is less than the first number threshold (e.g. less than 3), and the event correlation degree is less than the correlation degree threshold (i.e. the lower the value, the weaker the correlation with the target alarm or other concurrent alarms), it is considered to be a low chain risk scenario, and the concurrent correlation coefficient can be calculated based on the average importance of the concurrent alarms. Then, if the number of concurrent alarms is greater than the first number threshold (e.g. greater than 3), and the event correlation degree is greater than the correlation degree threshold, it is considered to be a medium chain risk scenario, and the correlation coefficient is increased based on the basic coefficient, because multiple and correlated concurrent alarms may indicate a potential failure of the same system. In addition, if the number of concurrent alarms is greater than the second number threshold (e.g. greater than 5, and the second threshold is greater than the first threshold), or the event correlation degree meets the cross-system alarm condition (e.g. concurrent alarms involve the water supply system and the electrical system, breaking the single system boundary), it is considered to be a high chain risk scenario, and the correlation coefficient is further increased based on the medium risk. Therefore, by determining the concurrent correlation coefficient, the overall risk assessment is raised, and the resources are focused on the real chain risk, and the invalid concurrent interference is reduced.

[0150] In addition, if the regional power grid load level is in a flat section (i.e. stable power demand, such as 10 am-3 pm) or a low valley (i.e. low power demand, such as 0-5 am), the basic power grid constraint coefficient is calculated based on whether the power plant currently undertakes the power protection task. If the regional power grid load level is in a peak (i.e. power demand surges, such as 18-22 pm in summer), the power grid constraint coefficient is increased based on the basic coefficient (e.g. 0.3→0.5 after increasing), because the power plant needs to prioritize power supply during the peak period, and any alarm that may cause output to decrease needs to be disposed urgently (to avoid regional power failure). Therefore, by determining the power grid constraint coefficient, the embodiment ensures that the alarm priority is consistent with the power grid power supply target, and avoids scheduling errors caused by the disconnection between internal evaluation of the power plant and the demand of the power grid.

[0151] Therefore, the embodiment determines the second influence degree based on the device risk coefficient, the concurrent correlation coefficient and the power grid constraint coefficient, realizes the quantification of the correlation between the real-time operation condition and the current target alarm event. The second influence degree changes in real time with the condition, so that the alarm priority can respond to the condition change in real time, avoid the problem that the priority corresponding to the alarm importance score cannot adapt to the dynamic scene, realize the conversion of the influence of the real-time condition on the alarm from the experience judgment to the result driven by the fusion of the real-time condition and the historical experience, can accurately reflect the real risk level under the current condition, and can cooperate with the first influence degree (device correlation risk) to provide complete real-time risk basis for the priority correction of the historical experience and the real-time scene, and ensure that the alarm disposal decision is more suitable for the dynamic actuality of the power plant operation.

[0152] In some embodiments, the target alarm event is processed according to the execution priority indication, including:

[0153] Obtaining personnel scheduling data, the personnel scheduling data including personnel working state and processing capacity for the alarm event type;

[0154] Obtaining historical processing data of processing operation related to the alarm event type in the accident alarm record, the historical processing data including historical success rate and historical processing efficiency;

[0155] Based on the operation condition, the topology structure and the historical processing data, determining an event prediction processing time length, and determining an efficiency difference value between the event prediction processing time length and the historical processing efficiency;

[0156] Based on the personnel scheduling data, the historical processing data, the efficiency difference value and the execution priority, scheduling personnel to process the target alarm event.

[0157] In the embodiment, the personnel working state includes a basic state, a load condition and a physical position. The basic state is used to judge whether the personnel can respond immediately, i.e. on duty / temporarily off duty / rest. The load condition represents the number of tasks being processed and the task progress. The physical position represents the current power plant area where the personnel is located, and the response time of the personnel to the alarm device is judged (e.g. personnel close to the scene can arrive faster). The processing capacity can be skill matching degree and qualification certification, i.e. the processing experience of the personnel for the current alarm event type.

[0158] Then, a proportion of a number of successful processing times of the same type of the current alarm event in total processing times in historical processing feedback data is obtained, that is, a historical success rate. The historical processing efficiency represents an average processing speed of the same type of alarm event in history, including an average processing time, that is, an average time from alarm response to device recovery.

[0159] Therefore, the embodiment performs personnel scheduling by personnel scheduling data, historical experience, efficiency difference value and execution priority, guarantees timeliness and accuracy of alarm disposal, reduces power plant operation risk and improves alarm processing efficiency.

[0160] In a second aspect, the application provides a power plant accident information quality evaluation device, which comprises:

[0161] An event data acquisition module 201 is configured to acquire alarm event data of a target alarm event to be displayed, wherein the alarm event data comprises an alarm event type and a target power plant to which the target alarm event belongs.

[0162] A historical data acquisition module 202 is configured to acquire at least one accident alarm record of the target power plant, wherein the accident alarm record comprises an accident type of an occurred accident alarm and historical processing feedback data, and the historical processing feedback data comprises a processing operation type and an influence level in response to the accident alarm.

[0163] An encoding module 203 is configured to determine an accident type encoding vector based on the accident type, the processing operation type and the influence level of the accident alarm record, and determine a location encoding vector based on the target power plant, wherein the location encoding vector is used to indicate the difference between the target power plant and other power plants.

[0164] A historical event feature extraction module 204 is configured to determine a historical event feature used to describe a behavior mode of the target power plant according to the accident type encoding vector and the location encoding vector of the accident alarm record.

[0165] An alarm importance evaluation module 205 is configured to associate the historical event feature with the alarm event data to generate an alarm importance score of the target alarm event.

[0166] A display module 206 is configured to display the target alarm event according to a priority mapped with the alarm importance score.

[0167] The system of the embodiments of the present application can execute the method provided by the embodiments of the present application, the implementation principles are similar, and the actions performed by each module in the system of the embodiments of the present application correspond to the steps in the method of the embodiments of the present application. For the detailed function description of each module of the system, refer to the description in the corresponding method shown in the foregoing, which will not be described here again.

[0168] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the quality of power plant accident information, characterized in that, include: Obtain the alarm event data of the target alarm event to be displayed, the alarm event data including the alarm event type and the target power plant to which the target alarm event belongs; Acquire at least one accident alarm record of the target power plant, the accident alarm record containing the accident type of the accident alarm and historical processing feedback data, the historical processing feedback data including the processing operation type and impact level in response to the accident alarm; Based on the accident type, handling operation type, and impact level of the accident alarm record, an accident type coding vector is determined, and a location coding vector is determined based on the target power plant. The location coding vector is used to indicate the differences between the target power plant and other power plants. Based on the accident type encoding vector and location encoding vector of the accident alarm record, historical event features describing the behavior pattern of the target power plant are determined. The accident alarm record also includes the alarm timestamp of the accident alarm, including: arranging the accident alarm records occurring at the target power plant in ascending order of the alarm timestamp to obtain the accident type encoding sequence of the target power plant; extracting time features from the accident type encoding sequence using a pre-built feature extraction model to obtain a forward hidden state and a backward hidden state, and concatenating the forward hidden state and the backward hidden state to obtain context-aware features. The context-aware features are used to determine the sequential dependencies between each accident alarm record in the accident type encoding sequence; and connecting the location encoding vector with the above... The following describes the use of perceived features to perform feature fusion, obtaining the historical fusion features of the accident alarm records; based on the alarm timestamp of the accident alarm records, the time decay weight of the accident alarm records is determined, and the historical fusion features of each accident alarm record occurring at the target power plant and the time decay weight are weighted and calculated to obtain the historical event features of the target power plant. Specifically, based on the alarm timestamp of the accident alarm records, the occurrence timestamp of the target alarm event, and a preset time decay rate, the time decay weight of the accident alarm records is determined. The time decay weight is used to indicate the degree of influence of accident alarm records with a time difference less than a preset time difference threshold on the target alarm event. The time difference is determined by the alarm timestamp of the accident alarm records and the occurrence timestamp. The historical event characteristics are correlated with the alarm event data to generate an alarm importance score for the target alarm event; The target alarm events are displayed based on the priority mapped to the alarm importance score.

2. The power plant accident information quality assessment method according to claim 1, characterized in that, The process of determining an accident type coding vector based on the accident type, handling operation type, and impact level of the accident alarm record, and determining a location coding vector based on the target power plant, includes: The accident type, processing operation type, and impact level of the accident alarm records are encoded to obtain accident alarm fusion features; The accident alarm fusion features are converted into an accident type encoding vector by a preset first embedding model. The accident type encoding vector is used to indicate the correlation between the accident type, the processing operation type and the impact level. The power plant identifier of the target power plant is encoded to obtain the power plant characteristics; The power plant features are converted into location encoding vectors using a pre-defined second embedding model.

3. The power plant accident information quality assessment method according to claim 2, characterized in that, The step of fusing the location encoding vector with the context-aware features to obtain the historical fusion features of the accident alarm record includes: The location encoding vector is enhanced using a preset first multilayer perceptron to obtain a location importance bias vector; The positional importance bias vector and the context-aware features are concatenated to obtain concatenated features, and the concatenated features are processed using a preset second multilayer perceptron to obtain historical fusion features.

4. The power plant accident information quality assessment method according to claim 3, characterized in that, The step of associating the historical event features with the alarm event data to generate an alarm importance score for the target alarm event includes: Based on the alarm event type of the target alarm event, determine the event type feature, and based on the target power plant, determine the target power plant feature, and then combine the event type feature and the target power plant feature to form the event feature; Calculate the semantic similarity between the event features and the historical event features, whereby the semantic similarity is used to indicate the correlation between the target alarm event and any accident alarm record; The semantic similarity associated with each accident alarm record is superimposed to calculate the target similarity, and the target similarity is processed by a preset third multilayer perceptron to obtain the context modulation factor. Based on the context modulation factor, the event characteristics, and the historical event characteristics, the alarm importance characteristics of the target alarm event are determined; Based on a preset alarm importance weight matrix and a preset alarm importance bias vector, the alarm importance features are mapped to alarm importance scores.

5. The method for assessing the quality of power plant accident information according to any one of claims 1 to 4, characterized in that, The method further includes: The current operating conditions and topology of the target power plant are obtained. The operating conditions include the overall operating status of the power plant and the operating status and associated status of multiple devices in the target power plant. The associated status is used to indicate the associated device data of concurrent alarm events that occur at the same time as the target alarm event and their associated devices. The overall operating status of the power plant includes the regional power grid load level and dispatch requirements of the area where the target power plant is located. The topology is used to indicate the connection relationship between the devices. Based on the operating conditions and the topology, a first degree of influence is determined. The first degree of influence is used to indicate the correlation between the target alarm event, the processing device, and the associated device. The processing device is used to indicate the device involved in processing the target alarm event. Match the incident alarm records and their impact levels related to the alarm event type of the target alarm event; Based on the equipment operating status, the equipment association status, and the overall operating status of the power plant, a second degree of influence is determined, which is used to indicate the degree of influence of the operating condition on the target alarm event. Based on the first degree of influence and the second degree of influence, a real-time influence factor is determined, and based on the real-time influence factor and the matched influence level, a historical correlation coefficient is determined; Based on the first degree of impact, the second degree of impact, the impact level, and the historical correlation coefficient, the priority mapped to the alarm importance score is adjusted to obtain the execution priority, and the target alarm event is processed according to the execution priority indication.

6. The power plant accident information quality assessment method according to claim 5, characterized in that, The determination of the first degree of influence based on the operating conditions and the topology includes: Based on the topology and the operating conditions, the event impact degree of the target alarm event is determined, wherein the event impact degree includes the event impact range and the event impact time. The event impact range is used to indicate at least one level of equipment affected by the target alarm event, the number of equipment, and the first equipment correlation between equipment. The event impact time is used to indicate the duration between the occurrence timestamp of the target alarm event and the completion time of the target alarm event. Based on the topology, the device operating status, and the device association status, the device impact degree of the processing device is determined. The device impact degree includes the device impact time of the processing device, the difficulty of the device processing operation, the importance of the device, and the second device correlation between the processing devices. The device impact time is used to indicate the duration between the moment when the processing device becomes abnormal due to the target alarm event and the moment when it returns to normal. Based on the topology and the device association status, the concurrent impact of the concurrent alarm events is determined. The concurrent impact includes the number of concurrent alarm events, the difficulty of handling concurrent events, the importance of concurrent events, the correlation between concurrent alarm events, and the associated device data of the corresponding associated devices. The difficulty of handling concurrent events is used to indicate the difficulty of operating the associated devices involved in handling the concurrent alarm events. The importance of concurrent events is used to indicate the importance of the concurrent alarm events compared to the target power plant and the target alarm events. Based on the degree of impact of the event, the degree of impact of the device, and the degree of impact of concurrency, a first degree of impact is obtained.

7. The power plant accident information quality assessment method according to claim 6, characterized in that, The equipment operating status includes operating condition and load rate; determining the second degree of influence based on the equipment operating status, the associated equipment status, and the overall operating status of the power plant includes: When the operating condition is detected to be not in the specified critical operating condition and the load rate is within the preset normal load rate range, the equipment risk coefficient is determined based on the equipment operating status, wherein the critical operating condition is a peak load phase or a start-up / shutdown phase; or, when the operating condition is detected to be in the critical operating condition and the load rate exceeds the normal load rate range, the equipment risk coefficient is increased based on the equipment operating status; or, when the operating condition is detected to be in a shutdown or standby state, the equipment risk coefficient is decreased based on the equipment operating status. When the number of concurrent alarm events detected is less than a preset first threshold and the event correlation is less than a preset correlation threshold, a concurrent correlation coefficient is determined based on the device correlation status; or, when the number of concurrent alarm events detected is greater than the first threshold and the event correlation is greater than the correlation threshold, the concurrent correlation coefficient is increased based on the device correlation status; or, when the number of concurrent alarm events detected is greater than a preset second threshold, or the event correlation reaches a preset cross-system alarm condition, the increased concurrent correlation coefficient is added based on the device correlation status, where the second threshold is greater than the first threshold. When the load level of the regional power grid is detected to be at a flat load level or a low load level, a power grid constraint coefficient is determined based on the overall operating status of the power plant. The power grid constraint coefficient is used to indicate the urgency of handling the target alarm event; or, when the load level of the regional power grid is detected to be at a peak load level, the power grid constraint coefficient is increased based on the overall operating status of the power plant. The second degree of influence is determined based on the equipment risk coefficient, the concurrent correlation coefficient, and the power grid constraint coefficient.

8. The power plant accident information quality assessment method according to claim 7, characterized in that, The process of handling the target alarm event according to the execution priority indication includes: Acquire personnel scheduling data, which includes personnel work status and the ability to handle the alarm event types; Obtain historical processing data of the processing operations related to the alarm event type in the accident alarm record, the historical processing data including historical success rate and historical processing efficiency; Based on the operating conditions, the topology, and the historical processing data, the event prediction processing time is determined, and the efficiency difference between the event prediction processing time and the historical processing efficiency is determined. Personnel scheduling is performed based on the personnel scheduling data, the historical processing data, the efficiency difference, and the execution priority to handle the target alarm event.

9. A power plant accident information quality assessment device, characterized in that, include: The event data acquisition module is used to acquire the alarm event data of the target alarm event to be displayed. The alarm event data includes the alarm event type and the target power plant to which the target alarm event belongs. The historical data acquisition module is used to acquire at least one accident alarm record of the target power plant. The accident alarm record includes the accident type of the accident alarm and historical processing feedback data. The historical processing feedback data includes the processing operation type and impact level in response to the accident alarm. The coding module is used to determine an accident type coding vector based on the accident type, handling operation type and impact level of the accident alarm record, and to determine a location coding vector based on the target power plant. The location coding vector is used to indicate the difference between the target power plant and other power plants. The historical event feature extraction module is used to determine historical event features describing the behavior patterns of the target power plant based on the accident type encoding vector and location encoding vector of the accident alarm record. The accident alarm record also includes the alarm timestamp of the accident alarm. The module includes: arranging the accident alarm records occurring at the target power plant in ascending order of the alarm timestamp to obtain the accident type encoding sequence of the target power plant; extracting time features from the accident type encoding sequence using a pre-built feature extraction model to obtain a forward hidden state and a backward hidden state; concatenating the forward hidden state and the backward hidden state to obtain context-aware features. The context-aware features are used to establish the sequential dependencies between each accident alarm record in the accident type encoding sequence; and then... The code vector is fused with the context-aware features to obtain the historical fusion features of the accident alarm records. Based on the alarm timestamp of the accident alarm records, the time decay weight of the accident alarm records is determined, and the historical fusion features of each accident alarm record occurring at the target power plant and the time decay weight are weighted and calculated to obtain the historical event features of the target power plant. The time decay weight of the accident alarm records is determined based on the alarm timestamp of the accident alarm records, the occurrence timestamp of the target alarm event, and a preset time decay rate. The time decay weight is used to indicate the degree of influence of accident alarm records with a time difference less than a preset time difference threshold on the target alarm event. The time difference is determined by the alarm timestamp of the accident alarm records and the occurrence timestamp. The alarm importance assessment module is used to associate the historical event characteristics with the alarm event data to generate an alarm importance score for the target alarm event; The display module is used to display the target alarm events according to the priority mapped to the alarm importance score.

Images