Alarm classification method and device of power system, terminal equipment and storage medium

Abstract

The invention discloses an alarm classification method and device of a power system, terminal equipment and a storage medium, and belongs to the field of alarm classification, and the method comprises the steps: inputting alarm data containing equipment operation data and an alarm description text into an alarm classification model, a first text feature used for representing equipment operation characteristics, a second text feature of a keyword feature associated with the alarm and a time sequence feature of an equipment alarm information time change trend are extracted, so that the three features are synthesized to generate an alarm type and an alarm level corresponding to the alarm data; according to the method, the equipment alarm information can be comprehensively analyzed and understood from multiple dimensions, and the limitation caused by the fact that a traditional method only depends on empirical rules is avoided, so that the accuracy and reliability of power system alarm classification can be improved, and an accurate classification result can be quickly obtained. By implementing the method and the device, the problem that the alarm data cannot be quickly and accurately classified and graded in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of alarm classification, and in particular to an alarm classification method, device, terminal equipment and storage medium for an electric power system. Background Art

[0002] In power systems, alarm information is a crucial indicator of the operational status of grid equipment. When grid equipment experiences an anomaly, it generates corresponding alarms. Accurately classifying and grading these alarms helps operations and maintenance personnel quickly locate faults and take effective measures to ensure the safe and stable operation of the grid.

[0003] However, traditional alarm information processing technology relies primarily on manual judgment and rule matching. Manual judgment requires operations personnel to analyze and process alarm information based on their own experience, while rule matching identifies alarm types and levels based on pre-set rules. This manual analysis and processing not only results in low alarm classification efficiency, but the accuracy of manual judgment and rule matching is easily affected by subjective factors and rule limitations. Differences in experience among operations personnel can lead to inconsistent judgments on the same alarm information, making it easy for missed or misjudgment cases to occur, making accurate classification and grading impossible. Consequently, existing technologies are unable to quickly and accurately classify and grade alarm data. Summary of the Invention

[0004] The embodiments of the present invention provide an alarm classification method, apparatus, terminal device and storage medium for an electric power system, which can not only automatically classify alarm data, but also analyze equipment alarm information through multi-dimensional features to obtain accurate classification results, and can effectively solve the problem in the prior art that it is impossible to quickly and accurately classify and grade alarm data.

[0005] An embodiment of the present invention provides a method for classifying alarms in a power system, comprising:

[0006] Acquire alarm data consisting of device alarm information corresponding to a plurality of moments; wherein the device alarm information includes: device operation data and alarm description text;

[0007] Inputting the alarm data into a preset alarm classification model so that the alarm classification model extracts a first text feature for characterizing an operating characteristic of the device based on the alarm data, and extracts a second text feature for characterizing a keyword feature associated with the alarm based on the alarm data;

[0008] Extracting, based on the alarm data, time series features for characterizing a temporal variation trend of the device alarm information;

[0009] An alarm type and an alarm level corresponding to the alarm data are generated according to the first text feature, the second text feature, and the time series feature; wherein the alarm level is used to characterize the severity of the alarm.

[0010] Preferably, the operating data of the equipment includes: equipment type, voltage level, environmental parameters, and operating statistics for evaluating the operating reliability of the equipment;

[0011] The alarm classification model extracts a first text feature for characterizing an operating characteristic of the device based on the alarm data, including:

[0012] The alarm classification model extracts device category features used to characterize the identity of the device based on the device type and voltage level;

[0013] Extracting environmental adaptability features for correlation between device performance and environmental factors based on the environmental parameters and the operational statistical data;

[0014] A first text feature is generated according to the device category feature and the environment adaptability feature.

[0015] Preferably, the alarm description text includes: a fault description text for characterizing the type of fault occurring in the equipment, an alarm description text for describing the urgency of the alarm, and a time description text for characterizing the duration of the alarm;

[0016] The step of extracting a second text feature for representing a keyword feature associated with the alarm based on the alarm data includes:

[0017] Segmenting the fault description text, the alarm description text, and the time description text according to a preset segmentation tool to generate a plurality of words and a plurality of phrases;

[0018] According to a preset power system fault type dictionary, a number of fault type keywords are extracted from each word and each phrase;

[0019] Extracting a number of urgency keywords from each word and each phrase according to a vocabulary table containing urgency of alarms of the power system;

[0020] Extracting time text for describing time units from each word and each phrase, and generating an alarm duration keyword based on the time text;

[0021] A second text feature is generated according to a plurality of fault type keywords, a plurality of urgency keywords, and an alarm duration keyword.

[0022] Preferably, extracting, based on the alarm data, a time series feature for characterizing a time variation trend of the device alarm information includes:

[0023] Perform sliding window segmentation on the alarm data to generate a plurality of time windows;

[0024] Perform time-attenuation weighted processing on the device alarm information in each time window to determine the timeliness characteristics of each device alarm information;

[0025] For each time window, generate the alarm frequency characteristics corresponding to the time window based on the number of alarm information of each device in the time window;

[0026] According to the alarm frequency characteristics of each time window, the frequency change rate of any two adjacent windows is determined;

[0027] Based on the frequency change rates, differential features are generated to characterize the changing trend of the alarm frequency.

[0028] A time series feature is generated according to the timeliness feature, the alarm frequency feature of each time window, and the differential feature.

[0029] Preferably, the device type corresponds to a device critical coefficient for reflecting the importance of the device in the power grid; the time window corresponds to a time period sensitivity coefficient for reflecting the degree of influence of the time window on the alarm processing priority;

[0030] Generating an alarm type and an alarm level corresponding to the alarm data according to the first text feature, the second text feature, and the time series feature includes:

[0031] Using the device key coefficient as the first weight coefficient corresponding to the first text feature;

[0032] Using the time period sensitivity coefficient as the second weight coefficient corresponding to the second text feature;

[0033] The first text feature, the second text feature, and the time series feature are weighted and fused according to a first weight coefficient, a second weight coefficient, and a preset time series weight coefficient to generate a fused feature; wherein the preset time series weight coefficient is used to measure the influence of the time series feature on the alarm processing priority;

[0034] According to the fusion feature, the alarm type and alarm level corresponding to the alarm data are determined.

[0035] Preferably, determining the alarm type and alarm level corresponding to the alarm data based on the fusion feature includes:

[0036] According to a preset kernel function, the fused features are mapped into a preset feature space;

[0037] According to the position of the fused feature in the feature space and the position of the hyperplane, the distance between the fused feature and the hyperplane is generated;

[0038] According to the distance between the fusion feature and the hyperplane, the alarm type and alarm level corresponding to the alarm data are determined.

[0039] Preferably, the training process of the alarm classification model includes:

[0040] Taking the alarm data samples and the actual alarm type and actual alarm level corresponding to each alarm data sample as input, and the predicted alarm type and predicted alarm level of each alarm data sample as output, the alarm classification model to be trained is iteratively trained until the model converges to generate a preset alarm classification model.

[0041] Based on the above method embodiments, the present invention provides corresponding device embodiments.

[0042] An embodiment of the present invention provides an alarm classification device for a power system, comprising: an alarm data acquisition module and an alarm classification module;

[0043] The alarm data acquisition module is used to acquire alarm data consisting of device alarm information corresponding to a plurality of moments; wherein the device alarm information includes: device operation data and alarm description text;

[0044] The alarm classification module is configured to input the alarm data into a preset alarm classification model, so that the alarm classification model extracts a first text feature for characterizing an operating characteristic of the device based on the alarm data, and extracts a second text feature for characterizing a keyword feature associated with the alarm based on the alarm data;

[0045] Extracting, based on the alarm data, time series features for characterizing a temporal variation trend of the device alarm information;

[0046] An alarm type and an alarm level corresponding to the alarm data are generated according to the first text feature, the second text feature, and the time series feature; wherein the alarm level is used to characterize the severity of the alarm.

[0047] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0048] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the alarm classification method for the power system described in the above-mentioned embodiment of the invention.

[0049] Based on the above method embodiment, the present invention provides a corresponding storage medium embodiment.

[0050] Another embodiment of the present invention provides a storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the alarm classification method for the power system described in the above-mentioned embodiment of the invention.

[0051] The following beneficial effects are achieved by implementing the present invention:

[0052] Embodiments of the present invention provide a method, apparatus, terminal device, and storage medium for classifying alarms in a power system. After acquiring alarm data containing equipment operating data and alarm description text, the present invention can input the data into a preset alarm classification model for feature extraction, thereby achieving automatic classification of the alarm data. Specifically, during the processing of the alarm classification model, a first text feature used to characterize the equipment operating characteristics, a second text feature that is a keyword feature associated with the alarm, and a time series feature that represents the time-varying trend of the equipment alarm information can be simultaneously extracted. These three features are then combined to generate the alarm type and alarm level corresponding to the alarm data. Since the first text feature clarifies the operating status of the equipment at different times, the second text feature directly or indirectly reflects the keyword associated with the alarm, and the time-varying trend of the equipment alarm information is used to determine the development of equipment failures, the present invention, by comprehensively considering both text features and time series features, can comprehensively analyze and understand the equipment alarm information from multiple dimensions, avoiding the limitations of traditional methods that rely solely on empirical rules, thereby improving the accuracy and reliability of power system alarm classification. Compared with the existing technology, the present invention can not only automatically classify alarm data, but also analyze equipment alarm information through multiple dimensional features to obtain accurate classification results, thereby achieving rapid and accurate classification and grading of alarm data. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The figure is a flow chart of an alarm classification method for a power system provided in one embodiment of the present invention.

[0054] Figure 2 It is a structural diagram of multi-dimensional feature extraction and SVM model layering provided by an embodiment of the present invention.

[0055] Figure 3 The figure is a flowchart of a method for alarm classification and grading provided by one embodiment of the present invention.

[0056] Figure 4 The figure is a schematic structural diagram of an alarm classification device for a power system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0058] like Figure 1 As shown, in order to solve the problem in the prior art that alarm data cannot be quickly and accurately classified and graded, an embodiment of the present invention provides an alarm classification method for a power system, comprising:

[0059] Step S1: Acquire alarm data consisting of device alarm information corresponding to a number of moments; wherein the device alarm information includes: device operation data and alarm description text;

[0060] Illustratively, the alarm data includes alarm information corresponding to different moments, and the alarm information includes operation data about the device and an alarm description text for describing the alarm.

[0061] Step S2: inputting the alarm data into a preset alarm classification model, so that the alarm classification model extracts a first text feature for characterizing the operating characteristics of the device based on the alarm data, and extracts a second text feature for characterizing a keyword feature associated with the alarm based on the alarm data;

[0062] Extracting, based on the alarm data, time series features for characterizing a temporal variation trend of the device alarm information;

[0063] An alarm type and an alarm level corresponding to the alarm data are generated according to the first text feature, the second text feature, and the time series feature; wherein the alarm level is used to characterize the severity of the alarm.

[0064] For example, on a given day, 100 alarm messages were collected. After screening, 20 of them were found to be generated during equipment maintenance. These 20 messages were filtered out, resulting in 80 alarm data from non-maintenance periods. These alarm data truly reflects the equipment's operating status, and these 80 alarm data from non-maintenance periods are input into the pre-set alarm classification model.

[0065] The processing of the preset alarm classification model includes the following steps:

[0066] First, the model extracts text features from the alarm data to characterize the device's operating characteristics. For example, for an alarm message such as "A certain transformer's temperature is too high. The current temperature is 80°C, and the normal operating temperature range is 20°C-60°C," the model might extract "Transformer temperature exceeds the normal range" as the first text feature. Simultaneously, the model extracts the second text feature, which is a keyword feature associated with the alarm, such as "temperature too high."

[0067] Next, the model extracts time series features and further analyzes the temporal trends of these alarm data. For example, if a device's overcurrent alarms appear five times in the past hour, with each overcurrent value increasing, a time series feature can be generated to characterize the temporal trends of the device's alarm information.

[0068] Finally, the model generates the alarm type and level. Based on the extracted first and second text features and time series features, the model generates the corresponding alarm type and level for the alarm data. For the transformer overtemperature alarm, the model can determine that its alarm type is "equipment overheating fault." Because the temperature is significantly above the normal range, the model determines its alarm level as "critical," indicating that immediate action is required.

[0069] In a preferred embodiment, the alarm types of the present invention can be divided into: protection events, protection alarms and status changes; the alarm levels can be divided into: accidents, important alarms, general alarms, changes and notifications, and the severity of the above levels decreases step by step.

[0070] In power systems, a protection event is an event in which a protective device initiates an action due to a detected power system fault or anomaly. A protection alarm occurs when an anomaly occurs in the protection device itself or in the power system it monitors, but does not yet reach a level that triggers a protective action. A state change primarily reflects a change in the status of equipment in the power system. For example, a circuit breaker position change indicates a change in the circuit breaker's closed or open state; a switch position change indicates a change in the switch's open or closed state.

[0071] Furthermore, when the alarm level is accident, it is the most severe alarm level, indicating that a serious fault has occurred in the power system, posing a significant threat to the safe and stable operation of the system, such as system disconnection or large-scale power outages. Immediate measures must be taken to restore normal system operation. Major alarms, while not yet at the level of an accident, are still serious and may develop into an accident. Examples include failure of a key plug-in in a protective device or a serious abnormality in key equipment, requiring prompt investigation and resolution. General alarms, on the other hand, indicate general problems with the power system or equipment. While they will not cause a serious impact on the system in the short term, they still require attention and timely resolution to prevent escalation. A change in position primarily refers to a change in equipment status, which is often the result of normal operation, but may also indicate a potential equipment anomaly that requires further confirmation. Notifications, the least severe alarm, are generally informational prompts during normal system operation and have no substantial impact on system operation, such as normal self-test information from certain equipment.

[0072] Illustratively, the embodiment of the present invention can classify the alarm types according to the power grid service conditions, as shown in the following table:

[0073]

[0074] Furthermore, we can formulate alarm levels based on factors such as the urgency of the alarm, the scope of impact, and the difficulty of repair, as shown in the following table:

[0075] Level Number Level Description 1 ACCIDENT 2 Important warning 3 General Alarm 4 Conjugation 5 inform

[0076] Therefore, based on the above-mentioned alarm types and alarm levels, a preset alarm classification model can be trained to achieve classification and grading of alarm data.

[0077] Regarding step S1, in a preferred embodiment, the alarm data is generated during the non-maintenance period of the power grid. This data is closely related to the actual operating status of the power grid equipment and can directly reflect whether the equipment has faults or abnormal conditions. For example, in the power system, equipment maintenance is a planned maintenance activity. During the maintenance period, the equipment may be shut down, undergoing commissioning, or undergoing human intervention. The alarm information generated during this period is generally unrelated to actual operational failures of the equipment, but rather is caused by maintenance operations, test signals, or temporary configuration changes.

[0078] For non-maintenance alarm data, the equipment is in normal operating conditions, and the alarms it generates (such as "overcurrent protection action" and "PT line break") are strongly correlated with equipment failures or abnormal conditions, and can accurately reflect the operating characteristics of the equipment and key information in the alarm description (such as fault type and urgency).

[0079] Specifically, it can collect grid operation data, including alarm information, maintenance ticket information, switch status, etc. The collected data is cleaned and standardized, and then by parsing the maintenance ticket and switch information, the maintenance status of the alarm equipment is obtained. The corresponding alarm information is filtered out, and non-maintenance alarm data is retained.

[0080] Therefore, the embodiment of the present invention uses alarm data during non-maintenance periods to ensure that the model input is valid data that reflects the actual operating status of the equipment, thereby improving the accuracy and reliability of alarm classification and grading from the source, and avoiding misjudgment and missed judgment problems caused by the inclusion of invalid data.

[0081] Regarding step S2, in a preferred embodiment, the training process of the alarm classification model includes:

[0082] Taking the alarm data samples and the actual alarm type and actual alarm level corresponding to each alarm data sample as input, and the predicted alarm type and predicted alarm level of each alarm data sample as output, the alarm classification model to be trained is iteratively trained until the model converges to generate a preset alarm classification model.

[0083] Illustratively, after the training process is complete, the resulting alarm classification model has learned the characteristics and patterns of a large number of alarm data samples and is capable of classifying and grading new alarm data. In practical applications, new alarm data can be fed into this pre-set alarm classification model, which will then output the corresponding predicted alarm type and level, helping operations and maintenance personnel to promptly and accurately address alarms in the power system.

[0084] In a preferred embodiment, the operating data of the device includes: device type, voltage level, environmental parameters, and operating statistics for evaluating the operating reliability of the device;

[0085] Then, the alarm classification model extracts a first text feature for characterizing the operating characteristics of the device based on the alarm data, specifically including:

[0086] The alarm classification model extracts device category features used to characterize the identity of the device based on the device type and voltage level;

[0087] Extracting environmental adaptability features for correlation between device performance and environmental factors based on the environmental parameters and the operational statistical data;

[0088] A first text feature is generated according to the device category feature and the environment adaptability feature.

[0089] Schematically, device type determines its function and role in the power grid. Different types of devices have different operating characteristics and failure modes. Voltage level, on the other hand, reflects the device's hierarchy and importance within the power grid. High-voltage devices typically transmit greater power and have a greater impact on grid stability. Combining these two characteristics provides a device with an identity, helping to accurately identify and differentiate different devices. The resulting device category signature can then be used to clarify a device's identity and role within the power system.

[0090] Furthermore, environmental parameters can include temperature, humidity, and air pressure, while operational statistics include historical equipment failure counts, operating hours, and maintenance records. By analyzing this data, we can understand the equipment's operational reliability and performance under different environmental conditions. For example, if a device frequently fails in a high-temperature environment, the environmental adaptability characteristics will reflect that the device has poor adaptability to high-temperature environments. Conversely, if the device can operate stably under various environmental conditions, it indicates good environmental adaptability. Therefore, we can derive environmental adaptability characteristics that reflect the correlation between equipment performance and environmental factors.

[0091] The embodiment of the present invention combines the device category feature and the environmental adaptability feature to generate a first text feature, which can more comprehensively and accurately describe the operating characteristics of the device. When classifying and grading alarms, the model can more accurately determine the nature and severity of the alarm based on the first text feature. For example, for the same overcurrent alarm, if it occurs on a high-voltage transformer and the ambient temperature of the transformer is chronically high, the model can determine that it may be more serious than the same alarm occurring on a low-voltage device, thereby more accurately classifying and grading.

[0092] By combining multi-dimensional features, the alarm classification model can learn richer equipment operation information, improving the model's adaptability to different equipment and different operation scenarios.

[0093] In a preferred embodiment, the alarm description text includes: a fault description text for characterizing the type of fault occurring in the device, an alarm description text for describing the urgency of the alarm, and a time description text for characterizing the duration of the alarm;

[0094] Then, the second text feature for characterizing the keyword feature associated with the alarm is extracted from the alarm data, specifically including:

[0095] Segmenting the fault description text, the alarm description text, and the time description text according to a preset segmentation tool to generate a plurality of words and a plurality of phrases;

[0096] According to a preset power system fault type dictionary, a number of fault type keywords are extracted from each word and each phrase;

[0097] Extracting a number of urgency keywords from each word and each phrase according to a vocabulary table containing urgency of alarms of the power system;

[0098] Extracting time text for describing time units from each word and each phrase, and generating an alarm duration keyword based on the time text;

[0099] A second text feature is generated according to a plurality of fault type keywords, a plurality of urgency keywords, and an alarm duration keyword.

[0100] For example, fault type keywords are words extracted from the fault description text that directly reflect the specific type of equipment failure. In power systems, different faults trigger different alarms, such as "short circuit," "open circuit," "overcurrent," and "overvoltage." These words can clearly identify the nature of the equipment problem.

[0101] Urgency keywords are words representing the severity of an alarm, filtered from the alarm description text. The urgency of power system alarms varies greatly, and words like "urgent," "serious," and "normal" clearly reflect the impact of an alarm on system operations. Extracting keywords based on a table of urgency terms for power system alarms provides a unified standard for determining urgency.

[0102] The alarm duration keyword, derived from the time description text, records the duration of the alarm occurrence and duration. In power systems, alarm duration is a crucial factor in measuring the impact of a fault. Short-lived and long-lasting alarms have different impacts on equipment and systems. Extracting time units (such as "seconds," "minutes," and "hours") from the text and generating corresponding alarm duration keywords can intuitively reflect the time dimension of the alarm.

[0103] Integrating the keywords of fault type, urgency and alarm duration to generate the second text feature can provide richer and more detailed information for the alarm classification model, so as to achieve more accurate classification and grading of alarms and output results that are more in line with the actual situation.

[0104] In a preferred embodiment, extracting, based on the alarm data, a time series feature for characterizing a time variation trend of the device alarm information includes:

[0105] Perform sliding window segmentation on the alarm data to generate a plurality of time windows;

[0106] Perform time-attenuation weighted processing on the device alarm information in each time window to determine the timeliness characteristics of each device alarm information;

[0107] For each time window, generate the alarm frequency characteristics corresponding to the time window based on the number of alarm information of each device in the time window;

[0108] According to the alarm frequency characteristics of each time window, the frequency change rate of any two adjacent windows is determined;

[0109] Based on the frequency change rates, differential features are generated to characterize the changing trend of the alarm frequency.

[0110] A time series feature is generated according to the timeliness feature, the alarm frequency feature of each time window, and the differential feature.

[0111] Illustratively, an embodiment of the present invention can perform time-attenuation weighted processing on the device alarm information of each time window to obtain a timeliness feature, which can highlight the importance of recent alarms, because recent alarms can better reflect the current actual operating status of the device, thereby determining whether the alarm level needs to be increased.

[0112] By generating alarm frequency characteristics based on the number of alarms generated by each device within a time window, you can intuitively understand the frequency of device alarms and determine the stability of device operation. If a device has a high alarm frequency over a period of time, it indicates that its operating status is unstable and there may be potential fault hazards, so the alarm level should be increased.

[0113] Furthermore, the present invention generates differential features based on the frequency change rates of adjacent windows to characterize the changing trends in alarm frequency. These differential features clearly indicate whether the alarm frequency is increasing, decreasing, or remaining stable. If the differential features indicate an upward trend in alarm frequency, this indicates a potential worsening of the equipment failure. Conversely, a downward trend may indicate that the failure is mitigating or that the equipment has been effectively maintained. Based on these differential features, it can be used to determine whether a critical alarm level should be marked.

[0114] Therefore, the embodiment of the present invention fully considers the time correlation and mutual influence between alarms, and generates differential features by analyzing the frequency change rate of adjacent windows and other operations, thereby mining the time series characteristics between alarm information, thereby achieving a more accurate assessment of alarm types and alarm levels.

[0115] In a preferred embodiment, the device type corresponds to a device critical coefficient for reflecting the importance of the device in the power grid; the time window corresponds to a time period sensitivity coefficient for reflecting the degree of influence of the time window on the alarm processing priority;

[0116] Generating an alarm type and an alarm level corresponding to the alarm data according to the first text feature, the second text feature, and the time series feature includes:

[0117] Using the device key coefficient as the first weight coefficient corresponding to the first text feature;

[0118] Using the time period sensitivity coefficient as the second weight coefficient corresponding to the second text feature;

[0119] The first text feature, the second text feature, and the time series feature are weighted and fused according to a first weight coefficient, a second weight coefficient, and a preset time series weight coefficient to generate a fused feature; wherein the preset time series weight coefficient is used to measure the influence of the time series feature on the alarm processing priority;

[0120] Based on the fusion features, the alarm type and alarm level corresponding to the alarm data are determined.

[0121] According to a preset kernel function, the fused features are mapped into a preset feature space;

[0122] According to the position of the fused feature in the feature space and the position of the hyperplane, the distance between the fused feature and the hyperplane is generated;

[0123] According to the distance between the fusion feature and the hyperplane, the alarm type and alarm level corresponding to the alarm data are determined.

[0124] Indicatively, the device critical coefficient is used as the first weight coefficient to reflect the impact of the importance of different devices in the power grid on alarm processing. It is understandable that in the power system, the failure of certain key equipment (such as main transformers, busbars, etc.) may have a significant impact on the safe and stable operation of the power grid, while the impact of the failure of some secondary equipment is relatively small. For example, the device critical coefficient corresponding to the main transformer is higher. When the main transformer has an alarm, the weight of its first text feature will be larger. In this way, when generating the fusion feature, the alarm will receive more attention, thereby timely and accurately classifying and grading the alarms of important equipment.

[0125] Furthermore, the present invention can use the time period sensitivity coefficient as the second weight coefficient, reflecting the impact of different time windows on the alarm processing priority. It is understandable that certain time periods (such as peak power consumption periods, important activity power supply periods, etc.) have higher requirements for the stability of the power grid, and the alarms that appear at this time need to be handled more quickly and carefully. For example, during peak power consumption periods, the time period sensitivity coefficient is higher, and the weight of the second text feature increases, making the alarm information related to the time period more prominent in the fusion feature, which can more accurately reflect the urgency of the alarm in the time period, thereby reasonably adjusting the alarm level of the alarm.

[0126] This embodiment of the present invention combines multi-dimensional information, such as the device's operating characteristics, the alarm's keyword features, and the temporal trends of alarm information, by weightedly fusing the first and second text features with temporal features. Because different types of features can describe the alarm situation from different perspectives, the weighted fusion of features can more comprehensively and accurately depict the essential characteristics of the alarm, avoiding the limitations of a single feature and improving the accuracy of alarm classification and grading.

[0127] Furthermore, the preset kernel function of the present invention can map the fusion features into a preset feature space, which can effectively handle the nonlinear relationships in the alarm data. In actual power systems, there are generally complex nonlinear relationships between the features of the alarm data, and traditional linear classification methods have difficulty accurately classifying and grading them. However, the embodiment of the present invention maps the data into a high-dimensional feature space through a kernel function, making the data that was originally linearly inseparable in the low-dimensional space linearly separable in the high-dimensional space. The hyperplane is then used for classification, further improving the accuracy of the classification.

[0128] This method determines the alarm type and level based on the distance between the fused feature's position in feature space and a hyperplane, providing a quantitative basis for alarm classification and grading. The farther away from the hyperplane, the greater the likelihood that the alarm belongs to a certain category, and the higher the confidence level; the closer to the hyperplane, the greater the uncertainty in the classification. This quantitative approach makes alarm classification and grading more objective and accurate, avoiding errors caused by subjective judgment, and also facilitates the evaluation and adjustment of classification results.

[0129] Schematically, the position of the hyperplane can be adaptively determined based on training data, and the classification boundaries can be automatically adjusted based on the distribution of different alarm data. In practical applications, the operation of the power system is constantly changing, and the distribution of alarm data will also change accordingly. Through the hyperplane-based classification method, the model of the embodiment of the present invention can dynamically adjust the position of the hyperplane based on new alarm data, thereby adapting to different situations and ensuring the accuracy and stability of alarm classification and grading.

[0130] In a preferred embodiment, the alarm classification model of the present invention combines multi-dimensional feature enhancement technology and dynamically weighted support vector machine (SVM), and can analyze and process alarm data from multiple dimensions.

[0131] like Figure 2As shown in the structural diagram of multi-dimensional feature extraction and SVM model layering, it can be seen that after the original alarm text data is first input into the text feature extraction sublayer in the alarm classification model, the "CT secondary open circuit" can be converted into "current transformer secondary circuit break" according to the "Relay Protection Standard Terminology Library" built into the text feature extraction sublayer to achieve terminology standardization.

[0132] In addition, the BiGram word segmenter generates two-word phrases through a sliding window. For example, "protection action" and "reclosing failure" can be combined into "protection action_reclosing failure", capturing complex alarm vocabulary.

[0133] Furthermore, the domain-enhanced TF-IDF method can be used to adjust the inverse document frequency calculation and improve the weight of professional vocabulary. The formula is: TF-IDF'(t) = TF(t)*log[N / (DF(t)+α)] (α=0.5), thereby highlighting key words or phrases.

[0134] For the time series frequency analysis sublayer, it can extract time series features used to characterize the time change trend of equipment alarm information. The corresponding processing flow is: original time series, sliding window segmentation, time attenuation weighting, frequency statistics and differential feature generation.

[0135] The original time series data can be segmented into sliding windows, and the window size is adaptively adjusted based on "max(12h, average interval between similar alarms × 3)". An exponential decay weighter (weight(t) = λ^(Δt / τ), λ = 0.85, τ = 1 hour) is used to perform time-decay weighted processing on the alarm information in each window to highlight the importance of recent alarms.

[0136] The alarm frequency in each window is counted, and the frequency change rate of adjacent windows is calculated (ΔFreq = (current window value - previous window value) / previous window value), thereby obtaining differential features to quantify the changing trend of the alarm frequency.

[0137] Furthermore, the feature enhancement and fusion sublayer can automatically generate high-order interaction features through a cross-feature generator. For example, by multiplying the text keyword weight with the equipment failure history rate, the high-order interaction feature F_cross = (text keyword weight) × (equipment failure history rate) can be obtained. The Box-Cox transformation (y = (x^λ-1) / λ, λ = 0.2) can also be used to perform nonlinear transformations on long-tail distribution features to optimize data distribution.

[0138] Finally, the processed text features, time series features and cross features are spliced ​​together, and the spliced ​​features are normalized and output to obtain comprehensive mixed features.

[0139] Furthermore, the text features, time series features and cross features output by the hybrid feature extraction layer can be combined into a feature vector in the format of X = [text features, time series features, cross features] to obtain a fusion feature, and the fusion feature is input into the dynamic weighted SVM classification layer to perform a distance comparison with the hyperplane, thereby classifying the alarm type and alarm.

[0140] In the support vector machine (SVM), it includes the kernel function mapping layer, the weight adaptive adjustment layer and the classification decision layer, which are:

[0141] Kernel function mapping layer: This embodiment of the present invention combines the Gaussian kernel function (RBF kernel), the polynomial kernel function, and the power field custom kernel function to obtain a hybrid kernel function, which is specifically in the following form:

[0142] K(x_i,x_j)=α*K_rbf+β*K_poly+γ*K_custom;

[0143] Among them, x_i, x_j are different samples in the sample data, and the mixed kernel function can calculate the similarity between two samples in high-dimensional space.

[0144] K_rbf=exp(-γ||x_i-x_j||2), which is a Gaussian kernel function used to measure the local similarity between samples, γ=0.1.

[0145] K_poly = (x_i·x_j+c)^d, which is a polynomial kernel function that can capture high-order polynomial relationships between sample features; d = 3, c = 1.

[0146] K_custom is a similarity function in the power field, which combines the calculation results of device topology association, time correlation and alarm semantic similarity.

[0147] Device topology association degree: topo_sim = A[x_i.device_id][x_j.device_id], which is calculated using the grid topology matrix A. This function reflects the calculation of the degree of association between two devices in the grid topology structure.

[0148] Time correlation: time_sim = exp(-|t_i-t_j| / 3600), where t_i and t_j are the times corresponding to the two samples respectively. This function reflects the similarity of the samples in the time dimension. The shorter the time interval, the closer the value of time_sim is to 1.

[0149] Alarm semantic similarity: text_sim = cosine_similarity(TF-IDF(x_i), TF-IDF(x_j)), obtained by calculating the cosine similarity of TF-IDF(x_i) and TF-IDF(x_j). This function measures the semantic similarity between two alarm texts.

[0150] Furthermore, the embodiment of the present invention also introduces a weight adaptive adjustment layer, that is, the sample weight w can be adjusted through a dynamic weight calculation module according to the real-time nature of the alarm and the system status. The formula is:

[0151] w = w_base × (1 + α * equipment key coefficient + β * time period sensitivity coefficient);

[0152] Indicatively, in terms of equipment key coefficients, w_base is a preset benchmark parameter, which can be 0.3 for transformers, 0.2 for buses, and 0 for others; in terms of time period sensitivity coefficients, it can be 0.25 during peak hours and 0 during normal hours, α=0.6, β=0.4 (trainable parameters), which allows the model to flexibly adjust the attention paid to different alarms according to actual conditions.

[0153] Finally, based on the dynamic weights and the output of the hybrid kernel function, the classification decision function f(x) = sign(Σ(α_i*y_i K(x_i,x))+b) can be calculated, where α_i is the support vector coefficient, y_i is the category label, K is the hybrid kernel function, and b is the bias term.

[0154] By mapping the output value of the classification decision function to the confidence score, the type and level of the alarm can be determined according to the threshold standards of accident class (confidence>0.9), important class (0.7<confidence≤0.9), general class (0.5<confidence≤0.7), displacement and notification class (confidence≤0.5), and complete the alarm classification and grading output.

[0155] Illustratively, this embodiment of the present invention combines a hybrid kernel function consisting of a Gaussian kernel, a polynomial kernel, and a custom kernel function for the power sector to measure the similarity of samples in high-dimensional space from different perspectives. The Gaussian kernel measures local similarity, the polynomial kernel captures high-order polynomial relationships, and the custom kernel function integrates device topology, temporal, and semantic similarity, fully accounting for the specialized characteristics of power systems. This enables the model to effectively process complex power alarm data and improve classification accuracy.

[0156] By adjusting sample weights based on alarm real-time performance and system status, and taking into account equipment criticality and time-of-day sensitivity, the model prioritizes alarms of varying importance and at different times. For example, alarms from critical equipment like transformers and during peak hours receive higher weights, meeting the actual needs of power system operations and maintenance, improving the model's adaptability and decision-making rationality.

[0157] A classification decision function is calculated based on dynamic weights and hybrid kernel function outputs, mapped to confidence scores, and the alarm type and level are determined according to threshold standards. This quantitative classification and grading approach makes alarm processing more objective and accurate, providing operations and maintenance personnel with a clear basis for decision-making, facilitating rapid response and handling of alarms of varying severity, and ensuring safe and stable power grid operation.

[0158] like Figure 3 As shown, in a preferred embodiment, the method flow of alarm classification and grading of the present invention may also include the following process:

[0159] Through the data collection module, the power grid operation data is collected in real time, including alarms sent by protection devices, protection device maintenance ticket information, switch status information, etc., to build a protection alarm information database.

[0160] The collected protection alarm information is preprocessed through the preprocessing module, including filtering and maintenance information processing, to ensure the accuracy and consistency of the data.

[0161] Through the feature extraction module, key features are extracted from the preprocessed protection alarm information, such as equipment voltage level, equipment type, alarm type, alarm description keywords, etc.

[0162] Through the intelligent classification and grading module, based on machine learning algorithms (such as support vector machines, random forests, neural networks, etc.), a protection alarm information classification and grading model is constructed, the alarm information is automatically classified according to the extracted features, and the classified alarm information is graded and evaluated.

[0163] Through the result display module, the classified and graded protection alarm information is displayed to dispatchers and operation and maintenance personnel in a visual manner, making it easy to quickly understand the power grid operation status and alarm information and make timely and effective responses.

[0164] like Figure 4 As shown, based on the above-mentioned embodiments of the alarm classification method of various power systems, the present invention provides corresponding device embodiments;

[0165] An embodiment of the present invention provides an alarm classification device for a power system, comprising: an alarm data acquisition module and an alarm classification module;

[0166] The alarm data acquisition module is used to acquire alarm data consisting of device alarm information corresponding to a plurality of moments; wherein the device alarm information includes: device operation data and alarm description text;

[0167] The alarm classification module is configured to input the alarm data into a preset alarm classification model, so that the alarm classification model extracts a first text feature for characterizing an operating characteristic of the device based on the alarm data, and extracts a second text feature for characterizing a keyword feature associated with the alarm based on the alarm data;

[0168] Extracting, based on the alarm data, time series features for characterizing a temporal variation trend of the device alarm information;

[0169] An alarm type and an alarm level corresponding to the alarm data are generated according to the first text feature, the second text feature, and the time series feature; wherein the alarm level is used to characterize the severity of the alarm.

[0170] It should be noted that the device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, and may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without paying any creative effort.

[0171] Those skilled in the art can clearly understand that, for the sake of convenience and brevity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0172] Based on the above-mentioned embodiments of the alarm classification method for various power systems, the present invention provides corresponding embodiments of terminal equipment items.

[0173] An embodiment of the present invention provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an alarm classification method for an electric power system as described in any method embodiment of the present invention.

[0174] The terminal device may be a computing terminal device such as a desktop computer, a notebook computer, a palmtop computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0175] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0176] The memory can be used to store the computer program, and the processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created based on the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device or other volatile solid-state storage device.

[0177] Based on the above-mentioned embodiments of the alarm classification method for various power systems, the present invention provides corresponding embodiments of storage media items.

[0178] An embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute an alarm classification method for an electric power system as described in any method embodiment of the present invention.

[0179] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. The computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0180] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for classifying alarms in a power system, characterized in that: include: Acquire alarm data consisting of device alarm information corresponding to a plurality of moments; wherein the device alarm information includes: device operation data and alarm description text; Inputting the alarm data into a preset alarm classification model so that the alarm classification model extracts a first text feature for characterizing an operating characteristic of the device based on the alarm data, and extracts a second text feature for characterizing a keyword feature associated with the alarm based on the alarm data; Extracting, based on the alarm data, time series features for characterizing a temporal variation trend of the device alarm information; An alarm type and an alarm level corresponding to the alarm data are generated according to the first text feature, the second text feature, and the time series feature; wherein the alarm level is used to characterize the severity of the alarm.

2. The method for classifying alarms in a power system according to claim 1, wherein: The operating data of the equipment, including: equipment type, voltage level, environmental parameters, and operating statistics used to evaluate the operating reliability of the equipment; The alarm classification model extracts a first text feature for characterizing an operating characteristic of the device based on the alarm data, including: The alarm classification model extracts device category features used to characterize the identity of the device based on the device type and voltage level; Extracting environmental adaptability features for correlation between device performance and environmental factors based on the environmental parameters and the operational statistical data; A first text feature is generated according to the device category feature and the environment adaptability feature.

3. The method for classifying alarms in a power system according to claim 2, wherein: The alarm description text includes: a fault description text for characterizing the type of fault occurring in the device, an alarm description text for describing the urgency of the alarm, and a time description text for characterizing the duration of the alarm; The step of extracting a second text feature for representing a keyword feature associated with the alarm based on the alarm data includes: Segmenting the fault description text, the alarm description text, and the time description text according to a preset segmentation tool to generate a plurality of words and a plurality of phrases; According to a preset power system fault type dictionary, a number of fault type keywords are extracted from each word and each phrase; Extracting a number of urgency keywords from each word and each phrase according to a vocabulary table containing urgency of alarms of the power system; Extracting time text for describing time units from each word and each phrase, and generating an alarm duration keyword based on the time text; A second text feature is generated according to a plurality of fault type keywords, a plurality of urgency keywords, and an alarm duration keyword.

4. The method for classifying alarms in a power system according to claim 3, wherein: Extracting, based on the alarm data, a time series feature for characterizing a time variation trend of the device alarm information includes: Perform sliding window segmentation on the alarm data to generate a plurality of time windows; Perform time-attenuation weighted processing on the device alarm information in each time window to determine the timeliness characteristics of each device alarm information; For each time window, generate the alarm frequency characteristics corresponding to the time window based on the number of alarm information of each device in the time window; According to the alarm frequency characteristics of each time window, the frequency change rate of any two adjacent windows is determined; Based on the frequency change rates, differential features are generated to characterize the changing trend of the alarm frequency; A time series feature is generated according to the timeliness feature, the alarm frequency feature of each time window, and the differential feature.

5. The method for classifying alarms in a power system according to claim 4, wherein: The device type corresponds to a device critical coefficient for reflecting the importance of the device in the power grid; the time window corresponds to a time period sensitivity coefficient for reflecting the degree of influence of the time window on the alarm processing priority; Generating an alarm type and an alarm level corresponding to the alarm data according to the first text feature, the second text feature, and the time series feature includes: Using the device key coefficient as the first weight coefficient corresponding to the first text feature; Using the time period sensitivity coefficient as the second weight coefficient corresponding to the second text feature; The first text feature, the second text feature, and the time series feature are weighted and fused according to a first weight coefficient, a second weight coefficient, and a preset time series weight coefficient to generate a fused feature; wherein the preset time series weight coefficient is used to measure the influence of the time series feature on the alarm processing priority; According to the fusion feature, the alarm type and alarm level corresponding to the alarm data are determined.

6. The method for classifying alarms in a power system according to claim 5, wherein: Determining the alarm type and alarm level corresponding to the alarm data based on the fusion feature includes: Mapping the fused features into a preset feature space according to a preset kernel function; According to the position of the fused feature in the feature space and the position of the hyperplane, the distance between the fused feature and the hyperplane is generated; According to the distance between the fusion feature and the hyperplane, the alarm type and alarm level corresponding to the alarm data are determined.

7. The method for classifying alarms in a power system according to claim 6, wherein: The training process of the alarm classification model includes: Taking the alarm data samples and the actual alarm type and actual alarm level corresponding to each alarm data sample as input, and the predicted alarm type and predicted alarm level of each alarm data sample as output, the alarm classification model to be trained is iteratively trained until the model converges to generate a preset alarm classification model.

8. An alarm classification device for a power system, characterized in that: include: Alarm data acquisition module and alarm classification module; The alarm data acquisition module is used to acquire alarm data consisting of device alarm information corresponding to a plurality of moments; wherein the device alarm information includes: device operation data and alarm description text; The alarm classification module is configured to input the alarm data into a preset alarm classification model, so that the alarm classification model extracts a first text feature for characterizing an operating characteristic of the device based on the alarm data, and extracts a second text feature for characterizing a keyword feature associated with the alarm based on the alarm data; Extracting, based on the alarm data, time series features for characterizing a temporal variation trend of the device alarm information; An alarm type and an alarm level corresponding to the alarm data are generated according to the first text feature, the second text feature, and the time series feature; wherein the alarm level is used to characterize the severity of the alarm.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for classifying alarms of a power system as claimed in any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is executed, the device where the storage medium is located is controlled to execute the alarm classification method for a power system according to any one of claims 1 to 7.

Images