Alarm information pushing method and device, equipment and storage medium

Abstract

The invention relates to the technical field of information processing, in particular to an alarm information pushing method and device, equipment and a storage medium, and the method can confirm a target receiving object based on a constructed real-time feature vector through constructing a scene matching model, can avoid the redundancy problem of pushing technical alarms to non-operation and maintenance personnel, and improves the reliability of the technical alarms. Meanwhile, it is ensured that key alarms such as core system faults are directly pushed to corresponding persons in charge, and pushing pertinence and effectiveness are remarkably improved; the alarm priority is determined through the evaluation indexes, so that high-priority alarms such as payment system interruption and core database fault are pushed preferentially, the problem that the alarms are submerged by low-priority redundant information is solved, the fault response time is greatly shortened, and the operation and maintenance efficiency is improved; the congestion risk level is confirmed through the queue congestion prediction model to realize dynamic selection of the push strategy, that is, through active prediction and congestion relief, the message accumulation rate can be reduced, the alarm delivery rate can be significantly improved, and it is guaranteed that the system can still operate stably in the peak period.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to an alarm information push method, apparatus, device and storage medium. Background Technology

[0002] With the deepening of enterprise digital transformation, alarm information generated by various business systems, including transaction systems, user management systems, and log systems, is growing explosively. Unified alarm push services have become a core component in ensuring stable system operation. However, existing alarm push systems have three major shortcomings: First, the push targets are fixed; the existing alarm push service relies on robots with fixed configurations and only matches robot types through preset rules. It cannot dynamically adjust the push targets according to alarm scenarios, such as real-time interruption alarms of core transaction systems and error alarms of ordinary log systems. This leads to the redundancy of pushing technical alarms to non-operation and maintenance personnel, or the omission of key alarms of core system failures that are not pushed to the person in charge. Second, alarm information is disordered; the existing alarm push service pushes alarms according to the generation time, and high-priority alarms such as payment failures are easily overwhelmed by low-priority information such as redundant log errors, which prolongs the fault response time. Third, the response to queue congestion is passive; Kafka queues only distribute messages according to a fixed strategy and lack a congestion prediction mechanism. During peak hours, message backlog or push failures are likely to occur, affecting the alarm delivery rate.

[0003] It is evident that existing technologies still need improvement and enhancement. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the present invention aims to provide an alarm information push method that optimizes the entire process of dynamic object matching, priority sorting, and proactive congestion control to form a closed-loop alarm push mechanism, effectively improving the accuracy, orderliness, and reliability of alarm push.

[0005] The first aspect of this invention provides a method for pushing alarm information, comprising: pre-constructing and training a scene matching model to establish a mapping relationship between feature vectors and receiving objects; acquiring real-time alarm messages and real-time dynamic data, preprocessing the real-time alarm messages and the real-time dynamic data respectively to construct real-time feature vectors; inputting the real-time feature vectors into the scene matching model, and determining the target receiving object based on the output result of the scene matching model; determining an evaluation index based on the real-time alarm messages, and determining an alarm priority based on the determined evaluation index; extracting and processing the real-time alarm messages to obtain key information, and generating alarm information based on the key information and the alarm priority; acquiring real-time queue operation data, inputting the real-time queue operation data into a pre-trained queue congestion prediction model to obtain a congestion risk level, and determining a push strategy based on the congestion risk level; and pushing the alarm information to the target receiving object based on the determined push strategy.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the step of pre-constructing and training a scene matching model to establish a mapping relationship between feature vectors and receiving objects includes: constructing a model to be trained using a logistic regression algorithm; acquiring multiple historical alarm data, each of which is labeled with a correlation relationship between a feature vector and a receiving object, wherein the receiving object includes robot type and receiving group; dividing the multiple historical alarm data into a training set and a test set using a preset division ratio, and iteratively training the model to be trained based on the training set and the test set; and outputting the scene matching model when a preset iteration stopping condition is met.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the step of acquiring real-time alarm messages and real-time dynamic data, and preprocessing the real-time alarm messages and real-time dynamic data to construct a real-time feature vector, includes: acquiring real-time alarm messages and real-time dynamic data, wherein the real-time dynamic data includes real-time receiver status and real-time robot load; extracting basic fields from the real-time alarm messages, wherein the extracted basic fields include alarm source, alarm category, alarm subcategory, message type, and alarm creation time; and standardizing the basic fields and the real-time dynamic data to construct a real-time feature vector.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the step of inputting the real-time feature vector into the scene matching model and determining the target receiving object based on the output result of the scene matching model includes: inputting the real-time feature vector into the scene matching model to obtain a receiving object matching result and a matching degree; when the matching degree is greater than or equal to a preset matching threshold, determining the receiving object matching result as the target receiving object, wherein the target receiving object includes a target robot type and a target receiving group; when the matching degree is less than the preset matching threshold, using a preset receiving configuration as the target receiving object, wherein the preset receiving configuration includes a preset robot type and a preset receiving group.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the step of determining evaluation indicators based on the real-time alarm messages and determining alarm priorities based on the determined evaluation indicators includes: determining evaluation indicators based on the real-time alarm messages, wherein the evaluation indicators include alarm type, business impact scope, and urgency level; converting the evaluation indicators into standardized values ​​and calculating the information entropy of each evaluation indicator; determining the weight of each evaluation indicator based on the information entropy, and constructing a standardized indicator matrix according to the weight of each evaluation indicator and the standardized value of each evaluation indicator; determining the optimal solution and the worst solution based on the constructed standardized indicator matrix; calculating the proximity degree according to the standardized indicator matrix, the optimal solution, and the worst solution, and determining the alarm priority based on the calculated proximity degree.

[0010] Optionally, in the fifth implementation of the first aspect of the present invention, the step of extracting and processing the real-time alarm message to obtain key information, and generating alarm information based on the key information and the alarm priority, includes: determining the message content of the real-time alarm message; when the message content only includes a text message, inputting the text message into a pre-trained BERT model to obtain key information composed of text; when the message content includes a text message and an image message, inputting the text message into a pre-trained BERT model and compressing the image message to obtain key information composed of text and a compressed image; when the message content includes a text message and a card message, inputting the text message into a pre-trained BERT model and normalizing the card message to obtain key information composed of text and normalized JSON data; and generating alarm information based on the key information and the alarm priority.

[0011] Optionally, in the sixth implementation of the first aspect of the present invention, the step of acquiring real-time queue operation data, inputting the real-time queue operation data into a pre-trained queue congestion prediction model to obtain a congestion risk level, and determining a push strategy based on the congestion risk level includes: constructing and training a queue congestion prediction model based on the LSTM algorithm; acquiring real-time queue operation data and performing outlier removal and normalization processing respectively to obtain pre-processed queue operation data; inputting the pre-processed queue operation data into the queue congestion prediction model to obtain a congestion risk level; when the congestion risk level is a low congestion risk level, the push strategy includes keeping the compression level unchanged, keeping the consumption weight unchanged, and fixing the number of threads; when the congestion risk level is a medium or high congestion risk level, the push strategy includes increasing the compression level, increasing the consumption weight corresponding to high alarm priority, and dynamically increasing the number of threads.

[0012] A second aspect of the present invention provides an alarm information push device, comprising: a training module for pre-constructing and training a scene matching model to establish a mapping relationship between feature vectors and receiving objects; a processing module for acquiring real-time alarm messages and real-time dynamic data, and preprocessing the real-time alarm messages and the real-time dynamic data respectively to construct real-time feature vectors; a matching module for inputting the real-time feature vectors into the scene matching model and determining the target receiving object based on the output result of the scene matching model; a determination module for determining an evaluation index based on the real-time alarm messages and determining an alarm priority based on the determined evaluation index; an extraction module for extracting and processing the real-time alarm messages to obtain key information, and generating alarm information based on the key information and the alarm priority; a prediction module for acquiring real-time queue operation data, inputting the real-time queue operation data into a pre-trained queue congestion prediction model to obtain a congestion risk level, and determining a push strategy based on the congestion risk level; and a push module for pushing the alarm information to the target receiving object based on the determined push strategy.

[0013] A third aspect of the present invention provides an alarm information push device, the alarm information push device comprising: a memory and at least one processor, the memory storing instructions; at least one processor calling the instructions in the memory to cause the alarm information push device to execute each step of the alarm information push method described above.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the alarm information push method described in any of the preceding claims.

[0015] In the technical solution of this invention, by constructing a scenario matching model, the target receiving object can be identified based on the constructed real-time feature vector. This avoids the redundancy problem of pushing technical alarms to non-operations and maintenance personnel, while ensuring that critical alarms such as core system failures are directly pushed to the corresponding responsible persons, significantly improving the targeting and effectiveness of the push. By determining the alarm priority through evaluation indicators, high-priority alarms, such as payment system interruptions and core database failures, are pushed first, solving the problem of being overwhelmed by low-priority redundant information, greatly shortening the fault response time, and improving operation and maintenance efficiency. By confirming the congestion risk level through a queue congestion prediction model, the push strategy can be dynamically selected. That is, by actively predicting and mitigating congestion, the message backlog rate can be reduced, the alarm delivery rate can be significantly improved, and the system can still operate stably during peak periods. Attached Figure Description

[0016] Figure 1 This is a first flowchart of an alarm information push method provided in an embodiment of the present invention; Figure 2 This is a second flowchart of the alarm information push method provided in the embodiments of the present invention; Figure 3 This is a third flowchart of the alarm information push method provided in the embodiments of the present invention; Figure 4 This is a fourth flowchart of the alarm information push method provided in the embodiments of the present invention; Figure 5 This is a fifth flowchart of the alarm information push method provided in the embodiments of the present invention; Figure 6 The sixth flowchart of the alarm information push method provided in the embodiments of the present invention; Figure 7 The seventh flowchart of the alarm information push method provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the alarm information push device provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the alarm information push device provided in an embodiment of the present invention. Detailed Implementation

[0017] This invention provides a method, apparatus, device, and storage medium for pushing alarm information. In this invention, the terms "first," "second," "third," "fourth," etc. (if present)," in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the alarm information push method in this invention includes: 101. Pre-build and train a scene matching model to establish a mapping relationship between feature vectors and receiving objects; In this embodiment, the scene matching model uses the logistic regression algorithm, which is characterized by high computational efficiency and strong interpretability, and is suitable for the real-time matching requirements of alarm scenarios; the algorithm model establishes a mapping relationship between feature vectors and receiving objects, providing a basis for dynamic matching and push targets.

[0019] 102. Obtain real-time alarm messages and real-time dynamic data, and preprocess the real-time alarm messages and real-time dynamic data respectively to construct real-time feature vectors; In this embodiment, real-time alarm messages and real-time dynamic data are collected and preprocessed to form standardized real-time feature vectors, which are used as model inputs for the scene matching model.

[0020] 103. Input the real-time feature vector into the scene matching model, and determine the target receiving object based on the output of the scene matching model; In this embodiment, a scene matching model is used to analyze the real-time feature vector and output the appropriate target receiving object. The receiving object refers to the final delivery carrier and group of the alarm information, including robot types such as Lark, DingTalk, and WeChat Work, as well as receiving groups such as the operation and maintenance core group and the development special group.

[0021] 104. Determine evaluation indicators based on the real-time alarm messages, and determine alarm priorities based on the determined evaluation indicators; In this embodiment, evaluation indicators are extracted based on real-time alarm messages, and alarm priorities are divided through quantitative calculation to solve the problem of information disorder, so that high-priority alarms are processed first and avoid being overwhelmed by low-priority information, thus shortening the fault response time. The alarm priority is a level divided based on the degree of impact of the alarm on the business, the urgency, etc., and is used to distinguish the order of alarm processing.

[0022] 105. Extract and process the real-time alarm messages to obtain key information, and generate alarm information based on the key information and the alarm priority; In this embodiment, key information is extracted from real-time alarm messages and reorganized into standardized alarm information based on alarm priority, ensuring that the information is concise and effective.

[0023] 106. Obtain real-time queue operation data, input the real-time queue operation data into a pre-trained queue congestion prediction model to obtain the congestion risk level, and determine the push strategy based on the congestion risk level; In this embodiment, a queue congestion prediction model is used to determine congestion risk and adjust the push strategy to reduce message backlog and push failures, thereby improving push reliability. The real-time queue operation data includes queue length, message production rate, consumption rate, peak period identifier, etc. The push strategy refers to the combination of message distribution parameters relative to the Kafka queue, including compression level, consumption weight, number of threads, etc.

[0024] 107. Based on the determined push strategy, push the alarm information to the target receiving object; In this embodiment, alarm information is pushed to the target recipient according to the determined push strategy, thus completing the distribution loop.

[0025] This invention discloses an alarm information push method. By constructing a scenario matching model, the target recipient can be identified based on the constructed real-time feature vector, avoiding the redundancy problem of pushing technical alarms to non-operations personnel. At the same time, it ensures that critical alarms such as core system failures are directly pushed to the corresponding responsible persons, significantly improving the targeting and effectiveness of the push. The method determines the alarm priority through evaluation indicators, giving priority to high-priority alarms, such as payment system interruptions and core database failures, solving the problem of being overwhelmed by low-priority redundant information, greatly shortening the fault response time and improving operation and maintenance efficiency. The method confirms the congestion risk level through a queue congestion prediction model, enabling dynamic selection of the push strategy. That is, by actively predicting and mitigating congestion, the message backlog rate can be reduced, the alarm delivery rate can be significantly improved, and the system can still operate stably during peak periods.

[0026] Please see Figure 2 In this embodiment of the invention, the pre-construction and training of the scene matching model to establish a mapping relationship between feature vectors and receiving objects includes: 201. Construct the training model using the logistic regression algorithm; In this embodiment, a logistic regression algorithm is used to construct the model to be trained, adapting the classification mapping requirements between feature vectors and robot types, as well as between feature vectors and receiving groups.

[0027] 202. Obtain multiple historical alarm data, each of which is labeled with a feature vector and the relationship between the receiving object and the receiving object, which includes robot type and receiving group; In this embodiment, at least 6 months of historical alarm data are collected, and it is ensured that the historical alarm data covers alarm samples from different business scenarios and time periods. Each historical alarm data is clearly labeled with a feature vector (including the standardized results of basic fields and dynamic data), the corresponding robot type (such as Lark robot), and the receiving group (such as the operation and maintenance core group), forming a one-to-one correspondence between features and results, and ensuring the effectiveness of training data.

[0028] 203. Divide the multiple historical alarm data into a training set and a test set using a preset division ratio, and iteratively train the model to be trained based on the training set and the test set; 204. When the preset iteration stopping condition is met, output the scene matching model; In this embodiment, the preferred ratio of training set to test set is 8:2. Preset iteration stopping conditions include: matching accuracy on the test set ≥ 90%, iteration count reaching a preset threshold (e.g., 1000 times), or the difference in model loss between two iterations ≤ 0.001. Through a standardized process of historical alarm data annotation, training set and test set splitting, and iterative verification, the trained scene matching model is ensured to have high accuracy and generalization ability, effectively avoiding mismatch of push targets caused by model bias. By establishing a mapping relationship between feature vectors and receiving objects, the scene matching model can output adaptation results based on real-time changes in the alarm scene, replacing the fixed configuration mode in existing technologies, thereby fundamentally solving the technical problem of fixed push targets.

[0029] Please see Figure 3 In this embodiment of the invention, the step of acquiring real-time alarm messages and real-time dynamic data, and preprocessing the real-time alarm messages and real-time dynamic data respectively to construct a real-time feature vector, includes: 301. Obtain real-time alarm messages and real-time dynamic data, wherein the real-time dynamic data includes the real-time receiver status and the real-time robot load; In this embodiment, real-time alarm messages and real-time dynamic data are acquired simultaneously. The real-time alarm messages are retrieved in real time through the business system interface to ensure the timeliness of the alarm information. The real-time recipient status is obtained through the open interface of the enterprise IM platform, such as the "User Online Status Query Interface" of the Lark API. The real-time robot load is collected through the Kafka monitoring interface, including the robot's message processing queue length and response latency.

[0030] 302. Extract basic fields from the real-time alarm message. The extracted basic fields include alarm source, alarm category, alarm subcategory, message type, and alarm creation time. In this embodiment, core basic fields are extracted from real-time alarm messages to serve as the basic components of the feature vector; the alarm source refers to the business system that generates the alarm, such as a transaction system, user management system, log system, etc.; the alarm category is a primary classification based on the nature of the alarm, such as availability alarms, performance alarms, configuration alarms, log alarms, etc.; the alarm sub-category is a further sub-classification under the alarm category, such as "service interruption" and "interface unavailable" under availability alarms; the message type is the presentation format of the alarm, which includes at least text and may also include images and / or cards; the alarm creation time is the specific timestamp of the alarm generation, accurate to the second.

[0031] 303. Standardize the basic fields and the real-time dynamic data respectively to construct a real-time feature vector; In this embodiment, by unifying the format and converting the values ​​of the basic fields and the real-time dynamic data, a real-time feature vector is finally constructed; specifically: Enumeration encoding is used to standardize the basic fields of text classes. For example, alarm source: core system = 3, important system = 2, general system = 1; alarm category: availability alarm = 3, performance alarm = 2, configuration alarm = 1, log alarm = 0. The basic time-related fields are standardized using a time-division method. For example, the alarm creation time of 10:30 corresponds to the time period code 6 (00:00-02:00=1, ..., 10:00-12:00=6). The receiver's status is standardized using binary encoding. When the receiver is online, the encoding is 1, and when the receiver is offline, the encoding is 0. Normalization is achieved by mapping the robot load to the [0, 1] interval. For example, if the robot queue length is 5000 (maximum queue length is 10000), the normalization value is 0.5. The constructed real-time feature vector is a fixed-dimensional numerical vector, such as [3 (alarm source), 3 (alarm category), 3 (alarm subcategory), 01 (message type), 6 (creation time), 1 (recipient status), 0.5 (robot load)]; By standardizing the processing to unify the format and value range of different types of data, model input errors caused by inconsistent data formats are effectively avoided, ensuring the accuracy of the feature vectors of the input scene matching model and providing effective input for the inference process of the scene matching model. The constructed real-time feature vectors contain both static basic fields of alarm messages and real-time dynamic data, comprehensively covering the multi-dimensional attributes of alarm scenarios and significantly improving the accuracy of scene matching.

[0032] Please see Figure 4 In this embodiment of the invention, the step of inputting the real-time feature vector into the scene matching model and determining the target receiving object based on the output of the scene matching model includes: 401. Input the real-time feature vector into the scene matching model to obtain the receiving object matching result and matching degree; In this embodiment, the real-time feature vector is input into the scene matching model to obtain the receiving object matching result and the matching degree. The receiving object matching result includes candidate robot type and candidate receiving group. The matching degree is calculated using the cosine similarity algorithm, which is the similarity between the current real-time feature vector and the historical sample feature vector in the scene matching model. The value range is [0, 1], and the closer it is to 1, the higher the scene similarity.

[0033] 402. When the matching degree is greater than or equal to a preset matching threshold, the matching result of the receiving object is determined to be the target receiving object, and the target receiving object includes the target robot type and the target receiving group; 403. When the matching degree is less than the preset matching threshold, a preset receiving configuration is used as the target receiving object. The preset receiving configuration includes a preset robot type and a preset receiving group. In this embodiment, the preset matching threshold is 0.6. When the matching degree is <0.6, a preset robot type and a preset receiving group are used as the target receiving objects: the preset robot type is the enterprise core IM platform robot, such as the default Lark robot if the enterprise mainly uses Lark, and the default DingTalk robot if the enterprise mainly uses DingTalk; the preset receiving group is a pre-built general operation and maintenance group, covering all operation and maintenance personnel to ensure that no alarm information is missed; the preset push parameters include real-time push and text format, and the text format includes core basic fields such as alarm source, alarm type, alarm priority and alarm occurrence time; the candidate receiving objects are matched through the scenario matching model, and the preset receiving configuration is set as a fallback configuration as a dual mechanism, which not only ensures accurate matching in most scenarios, but also avoids push failure when there are no similar historical samples, significantly reducing the key alarm omission rate; the existence of the fallback configuration enables the method to cope with situations where there are no historical samples, such as the launch of new business systems and sudden alarm scenarios, enhancing the robustness of the method.

[0034] Please see Figure 5 In this embodiment of the invention, determining the evaluation index based on the real-time alarm message and determining the alarm priority based on the determined evaluation index includes: 501. Determine evaluation indicators based on the real-time alarm messages, wherein the evaluation indicators include alarm type, scope of business impact, and urgency level; In this embodiment, the alarm type is derived by combining alarm major category and alarm subcategory, including: Level 3 (High Priority): Core system services are interrupted, such as the payment interface of the transaction system being unavailable; Level 2 (Medium Priority): Performance degradation of important systems, such as response latency greater than 3 seconds; Level 1 (Low Priority): General system log errors, such as abnormal logging in the test environment; The scope of business impact refers to the coverage of the alarm on the business, including: Level 3 (Large Scope): Affects more than 1,000 users, corresponding to core business; Level 2 (Medium Scope): Affects 10 to 1,000 users, corresponding to important business; Level 1 (Small Scope): Affects no more than 10 users, corresponding to general business. The urgency level refers to the urgency of the alarm needing to be handled, including: Level 3 (high urgency): real-time business interruption, which needs to be handled within 10 minutes, such as payment failure; Level 2 (medium urgency): real-time performance anomaly, which needs to be handled within 1 hour, such as interface delay; Level 1 (low urgency): non-real-time anomaly, which can be handled the next day, such as offline data synchronization failure.

[0035] 502. Convert the evaluation indicators into standardized values ​​and calculate the information entropy of each evaluation indicator; In this embodiment, the evaluation index is converted into a standardized value using a linear normalization formula. Specifically, the linear normalization formula is: Standardized value = (Original grade - Minimum grade) / (Maximum grade - Minimum grade). The steps for calculating information entropy are as follows: First, statistically analyze the value distribution of each evaluation index to obtain the probability of each value. Then, the entropy value is calculated based on the information entropy formula, which is: Where H represents information entropy, This represents the probability of taking the value of the i-th class; the smaller the entropy value, the greater the difference in the values ​​of the evaluation index, that is, the higher the discrimination, and the higher the corresponding weight.

[0036] 503. Determine the weight of each evaluation indicator based on the information entropy, and construct a standardized indicator matrix according to the weight of each evaluation indicator and the standardized value of each evaluation indicator; In this embodiment, the weight of each evaluation indicator is calculated using information entropy to quantify the distinguishing contribution of each indicator. This significantly increases the priority of critical alarms such as core business interruptions and widespread impacts, preventing them from being overwhelmed by low-priority information. The weight calculation formula is: Weight = (1 - Entropy value) / (1-Entropy value), where the smaller the entropy value, the greater the weight of the corresponding evaluation index; then, the standardized value of each evaluation index is multiplied by its weight, and all the weighted values ​​are combined to obtain the standardized index matrix. The standardized index weights comprehensively describe the priority profile of the current alarm.

[0037] 504. Based on the constructed standardized index matrix, determine the optimal and worst solutions; 505. Calculate the proximity score based on the standardized index matrix, the optimal solution, and the worst solution, and determine the alarm priority based on the calculated proximity score; In this embodiment, the optimal solution is the maximum value after weighting all evaluation indicators, and the worst solution is the minimum value after weighting all evaluation indicators. The formula for calculating proximity is: Proximity = Euclidean distance between the current alarm and the optimal solution / (Euclidean distance between the current alarm and the optimal solution + Euclidean distance between the current alarm and the worst solution). The closer the proximity is to 1, the higher the alarm priority. Specifically, proximity of 0.9-1.0 corresponds to level 1 (highest priority), proximity of 0.7-0.9 corresponds to level 2, proximity of 0.5-0.7 corresponds to level 3, proximity of 0.3-0.5 corresponds to level 4, and proximity of 0.0-0.3 corresponds to level 5 (lowest priority). By using evaluation index quantification, entropy weighting, and TOPSIS to calculate proximity, the subjective bias of human judgment can be avoided, ensuring the fairness and consistency of alarm priority classification. After the alarm priority is clear, maintenance personnel can process alarms in order of priority. The average processing time for high-priority alarms of levels 1-2 is shortened to within 10 minutes, and the fault location time is significantly reduced.

[0038] Please see Figure 6 In this embodiment of the invention, the step of extracting and processing the real-time alarm message to obtain key information, and generating alarm information based on the key information and the alarm priority, includes: 601. Determine the message content of the real-time alarm message; 602. When the message content only includes text messages, the text messages are input into the pre-trained BERT model to obtain key information composed of text. In this embodiment, a Chinese pre-trained BERT model, such as bert-base-chinese, is selected. The model parameters are set as follows: hidden layer dimension 768, attention head number 12, sequence length 128, to ensure the accuracy of semantic parsing. The extracted key information composed of text includes at least the alarm object (such as the transaction system), the anomaly type (such as payment failure), and the occurrence time, so that operation and maintenance personnel can quickly obtain key information and improve information acquisition efficiency.

[0039] 603. When the message content includes text messages and image messages, the text messages are input into the pre-trained BERT model, and the image messages are compressed to obtain key information composed of text and compressed images; In this embodiment, a hybrid compression algorithm combining LZ4 and WebP is used to compress image messages. Specifically, based on the interface limitations of mainstream IM platforms (such as Lark, which supports a maximum of 1MB and a resolution of 1280×720), the original images are compressed to a file size ≤1MB and a resolution ≤1280×720, with a compression ratio controlled between 1:2 and 1:3. Through compression, the file size is significantly reduced while maintaining clarity, thereby reducing network transmission pressure and queue storage pressure, and shortening push latency.

[0040] 604. When the message content includes text messages and card messages, the text messages are input into the pre-trained BERT model, and the card messages are normalized to obtain key information composed of text and normalized JSON data; In this embodiment, for the card message scenario, the original card's JSON data is first read, then the field mapping relationship of the target platform is parsed, then the field names and structural hierarchy are automatically converted, and finally the format compatibility is verified. Through standardization processing, it is ensured that key information can adapt to the card template specifications of the target robot platform. For example, DingTalk cards require the field names to be named "cardTitle" and "cardContent", while Lark requires "title" and "content".

[0041] 605. Based on the key information and the alarm priority, generate alarm information; In this embodiment, when the message content only includes text messages, the generated alarm information is formatted as: [Priority] Alarm Object - Anomaly Type (Occurrence Time); when the message content includes both text and image messages, the generated alarm information includes formatted text and a compressed image file; when the message content includes both text and card messages, the generated alarm information includes formatted text and standardized JSON data. By adopting a standardized information format, the presentation of alarm information is unified and standardized, avoiding misunderstandings caused by format confusion.

[0042] Please see Figure 7 In this embodiment of the invention, the step of acquiring real-time queue operation data, inputting the real-time queue operation data into a pre-trained queue congestion prediction model to obtain a congestion risk level, and determining a push strategy based on the congestion risk level includes: 701. Construct and train a queue congestion prediction model based on the LSTM algorithm; In this embodiment, the queue congestion prediction model adopts a 3-layer LSTM network structure, which consists of an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence. The input dimension of the queue congestion prediction model is 4, and the corresponding input information includes queue length, production rate, consumption rate, and peak indicator. The number of neurons in the first hidden layer is 64, the number of neurons in the second hidden layer is 32, and the number of neurons in the third hidden layer is 16. The time step of the queue congestion prediction model is set to 5, that is, it predicts the future congestion risk based on data from 5 consecutive historical time windows, and can predict the congestion risk in the next 5 minutes, replacing the passive processing mode of the existing technology, realizing the advance adjustment of the push strategy, and reducing the queue congestion rate. The output dimension of the queue congestion prediction model is 1, and the corresponding output result is the congestion risk probability.

[0043] 702. Obtain real-time queue operation data and perform outlier removal and normalization processing respectively to obtain preprocessed queue operation data; In this embodiment, the acquired real-time queue operation data includes queue length (number of messages in the current queue), message production rate (messages / minute), message consumption rate (messages / minute), and peak period identifier (1 indicates peak period 8:00-22:00, 0 indicates off-peak period); the collection granularity is set to a time window of 5 minutes to ensure the temporal continuity of the data; after acquiring the real-time queue operation data, the following methods are used: Outliers are detected in principle and filled with the average of the two adjacent windows before and after the time window to complete the outlier detection and filling process. Then, the min-max normalization algorithm is used to normalize the real-time queue running data after removing outliers to obtain the preprocessed queue running data.

[0044] 703. Input the preprocessed queue operation data into the queue congestion prediction model to obtain the congestion risk level; In this embodiment, when the predicted congestion probability is ≤0.6, it corresponds to a low congestion risk level, and when the predicted congestion probability is >0.6, it corresponds to a medium-high congestion risk level.

[0045] 704. When the congestion risk level is low congestion risk level, the push strategy includes keeping the compression level unchanged, keeping the consumption weight unchanged, and keeping the number of threads fixed. In this embodiment, when the congestion risk level is low, the compression level remains at the default level 3, the consumption weight remains unchanged, that is, all alarm weights = 1.0, and the number of threads is fixed at 8 consumption threads to avoid resource waste caused by over-configuration.

[0046] 705. When the congestion risk level is medium or high, the push strategy includes increasing the compression level, increasing the consumption weight corresponding to high alarm priority, and dynamically increasing the number of threads. In this embodiment, when the congestion risk level is low, the compression level is maintained at level 5, i.e., the compression ratio is increased by 30%-40%; the priority of high alarms is increased, with the weight of level 1-2 alarms adjusted to 1.5 and the weight of level 3-5 alarms adjusted to 0.8; the number of threads is dynamically increased to 12-16, specifically 16 for high congestion risk levels and 12 for medium congestion risk levels; in addition, 1-2 temporary queues can be created to distribute level 3-5 alarms; by dynamically adjusting the compression level, consumption weight, and number of threads, the message transmission volume is reduced, key alarms are prioritized, queue processing capacity is improved, and push failure rate is reduced.

[0047] The alarm information push method in the embodiments of the present invention has been described above. The alarm information push device in the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 8One embodiment of the alarm information push device in this invention includes: Training module 801 is used to pre-build and train a scene matching model to establish a mapping relationship between feature vectors and receiving objects; The processing module 802 is used to acquire real-time alarm messages and real-time dynamic data, and preprocess the real-time alarm messages and real-time dynamic data respectively to construct real-time feature vectors; Matching module 803 is used to input the real-time feature vector into the scene matching model and determine the target receiving object based on the output of the scene matching model; The determination module 804 is used to determine the evaluation index based on the real-time alarm message, and to determine the alarm priority based on the determined evaluation index. The extraction module 805 is used to extract and process the real-time alarm message to obtain key information, and generate alarm information based on the key information and the alarm priority. Prediction module 806 is used to acquire real-time queue operation data, input the real-time queue operation data into a pre-trained queue congestion prediction model to obtain the congestion risk level, and determine the push strategy based on the congestion risk level. The push module 807 is used to push the alarm information to the target receiving object based on the determined push strategy.

[0048] The above combination Figure 8 The alarm information push device in the embodiments of the present invention will be described in detail from the perspective of modular functional entities. The alarm information push device in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0049] Figure 9 This is a schematic diagram of an alarm information push device 900 provided in an embodiment of the present invention. The alarm information push device 900 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 910 (e.g., one or more processors) and a memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) for storing application programs 933 or data 932. The memory 920 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the alarm information push device 900. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute a series of instruction operations on the alarm information push device 900 to implement the steps of the alarm information push method provided in the above-described method embodiments.

[0050] The alarm information push device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 9 The alarm information push device structure shown does not constitute a limitation on the alarm information push device. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0051] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of an alarm information push method.

[0052] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0053] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0054] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for pushing alarm information, characterized in that, include: A scene matching model is pre-built and trained to establish a mapping relationship between feature vectors and receiving objects; Acquire real-time alarm messages and real-time dynamic data, and preprocess the real-time alarm messages and real-time dynamic data respectively to construct real-time feature vectors; The real-time feature vector is input into the scene matching model, and the target receiving object is determined based on the output of the scene matching model. Evaluation indicators are determined based on the real-time alarm messages, and alarm priorities are determined based on the determined evaluation indicators. The real-time alarm messages are extracted and processed to obtain key information. Based on the key information and the alarm priority, alarm information is generated. Acquire real-time queue operation data, input the real-time queue operation data into a pre-trained queue congestion prediction model to obtain the congestion risk level, and determine the push strategy based on the congestion risk level; Based on the determined push strategy, the alarm information is pushed to the target receiving object.

2. The alarm information push method according to claim 1, characterized in that, The pre-built and trained scene matching model, used to establish a mapping relationship between feature vectors and receiving objects, includes: The logistic regression algorithm is used to construct the model to be trained; Multiple historical alarm data are acquired, and each historical alarm data is labeled with a feature vector and the relationship between the receiving object and the receiving object, which includes robot type and receiving group; The historical alarm data are divided into training set and test set using a preset division ratio, and the model to be trained is iteratively trained based on the training set and the test set. When the preset iteration stopping condition is met, the scene matching model is output.

3. The alarm information push method according to claim 1, characterized in that, The process of acquiring real-time alarm messages and real-time dynamic data, and preprocessing the real-time alarm messages and real-time dynamic data to construct real-time feature vectors, includes: Acquire real-time alarm messages and real-time dynamic data, including real-time receiver status and real-time robot load; The basic fields are extracted from the real-time alarm messages. The extracted basic fields include alarm source, alarm category, alarm subcategory, message type, and alarm creation time. The basic fields and the real-time dynamic data are standardized to construct real-time feature vectors.

4. The alarm information push method according to claim 1, characterized in that, The step of inputting the real-time feature vector into the scene matching model and determining the target receiving object based on the output of the scene matching model includes: The real-time feature vector is input into the scene matching model to obtain the receiving object matching result and matching degree; When the matching degree is greater than or equal to the preset matching threshold, the matching result of the receiving object is determined to be the target receiving object, and the target receiving object includes the target robot type and the target receiving group; When the matching degree is less than the preset matching threshold, the preset receiving configuration is used as the target receiving object. The preset receiving configuration includes a preset robot type and a preset receiving group.

5. The alarm information push method according to claim 1, characterized in that, The step of determining evaluation indicators based on the real-time alarm messages and determining alarm priorities based on the determined evaluation indicators includes: Evaluation indicators are determined based on the real-time alarm messages, including alarm type, scope of business impact, and urgency level. The evaluation indicators are converted into standardized values, and the information entropy of each evaluation indicator is calculated. The weight of each evaluation index is determined based on the information entropy, and a standardized index matrix is ​​constructed based on the weight of each evaluation index and the standardized value of each evaluation index. Based on the constructed standardized index matrix, the optimal and worst solutions are determined. The proximity score is calculated based on the standardized index matrix, the optimal solution, and the worst solution, and the alarm priority is determined based on the calculated proximity score.

6. The alarm information push method according to claim 1, characterized in that, The step of extracting and processing the real-time alarm message to obtain key information, and generating alarm information based on the key information and the alarm priority, includes: Determine the message content of the real-time alarm message; When the message content consists only of text messages, the text messages are input into the pre-trained BERT model to obtain key information composed of text. When the message content includes both text and image messages, the text message is input into the pre-trained BERT model, and the image message is compressed to obtain key information composed of text and compressed images. When the message content includes text messages and card messages, the text messages are input into the pre-trained BERT model, and the card messages are normalized to obtain key information composed of text and normalized JSON data. Based on the key information and the alarm priority, alarm information is generated.

7. The alarm information push method according to claim 1, characterized in that, The process of acquiring real-time queue operation data, inputting the real-time queue operation data into a pre-trained queue congestion prediction model to obtain a congestion risk level, and determining a push strategy based on the congestion risk level includes: A queue congestion prediction model was constructed and trained based on the LSTM algorithm; Real-time queue operation data is acquired and outlier removal and normalization are performed to obtain preprocessed queue operation data. The preprocessed queue operation data is input into the queue congestion prediction model to obtain the congestion risk level; When the congestion risk level is low, the push strategy includes keeping the compression level unchanged, keeping the consumption weight unchanged, and keeping the number of threads fixed. When the congestion risk level is medium or high, the push strategy includes increasing the compression level, increasing the consumption weight corresponding to high alarm priority, and dynamically increasing the number of threads.

8. An alarm information push device, characterized in that, include: The training module is used to pre-build and train the scene matching model to establish a mapping relationship between feature vectors and receiving objects; The processing module is used to acquire real-time alarm messages and real-time dynamic data, and to preprocess the real-time alarm messages and real-time dynamic data respectively to construct real-time feature vectors; The matching module is used to input the real-time feature vector into the scene matching model and determine the target receiving object based on the output of the scene matching model; The determination module is used to determine evaluation indicators based on the real-time alarm messages, and to determine alarm priorities based on the determined evaluation indicators. The extraction module is used to extract and process the real-time alarm messages to obtain key information, and generate alarm information based on the key information and the alarm priority. The prediction module is used to acquire real-time queue operation data, input the real-time queue operation data into a pre-trained queue congestion prediction model to obtain the congestion risk level, and determine the push strategy based on the congestion risk level. The push module is used to push the alarm information to the target receiving object based on the determined push strategy.

9. An alarm information push device, characterized in that, The alarm information push device includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the alarm information push device to perform the steps of the alarm information push method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instruction is executed by the processor, it implements each step of the alarm information push method as described in any one of claims 1-7.

Images