Operation and maintenance fault detection method and device, equipment, storage medium and program product

Abstract

The invention provides an operation and maintenance fault detection method and device, equipment, a storage medium and a program product, and relates to the field of artificial intelligence. Generating a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain; according to the monitoring data corresponding to each time point in the current period, generating a to-be-answered question for inquiring the current fault type; and reasoning according to the to-be-answered question, the prompt reasoning chain and the reasoning answer corresponding to the prompt reasoning chain to obtain a current fault type and a reasoning chain for determining the current fault type. By adopting the method provided by the embodiment of the invention, the current fault type can be determined according to the monitoring data in the current period, and the fault can be positioned.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, apparatus, device, storage medium and program product for detecting operation and maintenance faults. Background Art

[0002] With the continuous development of the internet industry and the continued expansion of internet companies, both human and machine resources have skyrocketed. Consequently, the operational and maintenance workload within internet companies has also continued to expand. Major internet companies have invested significant human and material resources in improving the efficiency and security of internet operations, aiming to increase efficiency and reduce costs. Automated operations (AO) is a method for managing and maintaining computer systems through automated tools and processes. It aims to improve O&M efficiency, reduce human error, and enhance system reliability. Automated operations have become an essential skill in modern enterprises, helping them respond more quickly to market changes, better manage resources, and reduce costs.

[0003] For the detection of operation and maintenance faults, the existing technology mainly combines deep learning and inputs historical operation and maintenance data into a convolutional neural network to train a fault prediction model. The trained fault prediction model can then be used to predict and locate operation and maintenance faults.

[0004] However, although the above method is simple to implement, it can only predict the operation and maintenance performance data at future times based on the operation and maintenance performance data at the current moment using deep learning technology. As for how abnormal operation and maintenance performance data is generated, the existing technology cannot analyze the type of operation and maintenance fault that causes abnormal operation and maintenance performance data or locate the operation and maintenance fault. Summary of the Invention

[0005] The purpose of the present invention is to provide a method, device, equipment, storage medium and program product for detecting operation and maintenance faults, which are used to solve the problem that the existing technology cannot determine the type of operation and maintenance fault that causes abnormal operation and maintenance performance data or locate the operation and maintenance fault.

[0006] To solve the above technical problems, an embodiment of the present invention provides a method for detecting operation and maintenance faults, including:

[0007] Generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and the fault type corresponding to the monitoring data;

[0008] Generate unanswered questions for inquiring about the current fault type based on the monitoring data corresponding to each time point in the current cycle;

[0009] Reasoning is performed based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain a current fault type and a reasoning chain for determining the current fault type.

[0010] Optionally, the method further includes:

[0011] Determine the warning message template that matches the current fault type;

[0012] According to the warning message template, identifying and extracting required fault information from the inference chain of the current fault type and the data corresponding to the current fault type;

[0013] Fill the fault information into the warning message template to generate a warning message.

[0014] Optionally, generating a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and a fault type corresponding to the monitoring data includes:

[0015] For each fault cycle in the historical fault cycles, a plurality of first questions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type to obtain a question set, wherein one fault cycle corresponds to one question set;

[0016] Clustering the first problem in the problem set according to a K-means clustering algorithm to obtain multiple target problems;

[0017] According to the plurality of target questions, a large language model is used to perform reasoning to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain.

[0018] Optionally, for each fault cycle in the historical fault cycles, a plurality of first questions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type to obtain a question set, including:

[0019] For each fault cycle in the historical fault cycle, multiple preconditions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type;

[0020] According to the precondition, randomly changing the monitoring data of at least one time point corresponding to the precondition to obtain a question part;

[0021] A plurality of the preconditions and the corresponding question parts are combined to generate a plurality of first questions, to obtain a question set, wherein one of the first questions includes one precondition and one corresponding question part.

[0022] Optionally, clustering the first question in the question set according to a K-means clustering algorithm to obtain multiple target questions includes:

[0023] Processing the first questions in the question set using a sentence embedding representation model to obtain a question representation for each of the first questions;

[0024] Clustering the problem representation of the first problem according to the K-means clustering algorithm to obtain multiple clusters;

[0025] For each cluster, at least one first question is selected as a target question in descending order of distances from the first questions in the cluster to the cluster center.

[0026] Optionally, the generating of a prompt inference chain for fault detection and a corresponding inference answer of the prompt inference chain by using a large language model based on the plurality of target questions includes:

[0027] Randomly combining multiple target questions to obtain multiple question combinations;

[0028] According to the zero-sample set conversion method, the large language model is used to reason the question combination to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain, wherein one question combination corresponds to one prompt reasoning chain.

[0029] Optionally, the performing reasoning based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain the current fault type and the reasoning chain for determining the current fault type includes:

[0030] Dividing and reorganizing the multiple prompt reasoning chains according to the fault types indicated by the reasoning answers corresponding to the multiple prompt reasoning chains to obtain a target reasoning chain for determining the fault type, wherein the length of the target reasoning chain meets a preset condition;

[0031] For each of the fault types, characteristic information of the fault type is obtained according to the change pattern of the monitoring data in the target reasoning chain corresponding to the fault type, wherein the characteristic information includes data on the data change stage, data change pattern, and change time point;

[0032] Reasoning is performed using a large language model according to the question to be answered, the target reasoning chain, and the feature information to obtain a current fault type and a reasoning chain for determining the current fault type.

[0033] Optionally, dividing and reorganizing the plurality of prompt reasoning chains according to the fault types indicated by the reasoning answers corresponding to the plurality of prompt reasoning chains to obtain a target reasoning chain for determining the fault type includes:

[0034] According to the fault type indicated by the reasoning answer, multiple prompt reasoning chains are merged to obtain a first prompt reasoning chain combination;

[0035] Dividing the first prompt reasoning chain combination according to the fault type to obtain a plurality of second prompt reasoning chain combinations, wherein one fault type corresponds to one second prompt reasoning chain combination;

[0036] The prompt reasoning chains in different combinations of the second prompt reasoning chains are merged to generate a target reasoning chain for determining the fault type.

[0037] An embodiment of the present invention further provides a device for detecting operation and maintenance failures, comprising:

[0038] A first generating module is configured to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and the fault type corresponding to the monitoring data;

[0039] The second generation module is used to generate questions to be answered for inquiring about the current fault type based on the monitoring data corresponding to each time point in the current cycle;

[0040] The first reasoning module is used to perform reasoning based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain the current fault type and the reasoning chain used to determine the current fault type.

[0041] An embodiment of the present invention further provides a network device comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the method for detecting operation and maintenance failures as described in any one of the above items.

[0042] An embodiment of the present invention further provides a readable storage medium, comprising: a program stored on the readable storage medium, and when the program is executed by a processor, the steps of the method for detecting operation and maintenance failure as described in any one of the above items are implemented.

[0043] An embodiment of the present invention further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the method for detecting operation and maintenance failures as described in any one of the above items.

[0044] At least one of the above technical solutions of the present invention has the following beneficial effects:

[0045] In the above scheme, first, based on the monitoring data at each time point in the historical fault cycle and the fault type corresponding to the monitoring data, a prompt reasoning chain for fault detection and the reasoning answer corresponding to the prompt reasoning chain are generated; then, based on the monitoring data corresponding to each time point in the current cycle, a question to be answered is generated to inquire about the current fault type; finally, based on the question to be answered, the prompt reasoning chain and the reasoning answer corresponding to the prompt reasoning chain, reasoning is performed to obtain the current fault type and the reasoning chain for determining the current fault type. In an embodiment of the present invention, the monitoring data of the historical fault cycle is used to generate a prompt reasoning chain for fault type detection and location of the current cycle, so that the monitoring data of different fault types in the historical fault cycle can be used to accurately detect the fault type of the current cycle and determine the time point of the fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Schematic diagram of a flow chart of a method for detecting operation and maintenance failures according to an embodiment of the present invention;

[0047] Figure 2 A graph showing changes in data transmission rate at various time points within a cycle provided by an embodiment of the present invention;

[0048] Figure 3 Schematic diagram of the structure of a device for detecting operation and maintenance failures according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] 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 and 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.

[0050] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0051] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting operation and maintenance failures, including:

[0052] Step S101: generating a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and the fault type corresponding to the monitoring data;

[0053] In step S101, monitoring data is obtained through monitoring logs, specifically by collecting monitoring logs of operation and maintenance failures in different historical periods, and then analyzing the monitoring logs of different failure periods to obtain monitoring data and corresponding failure types at each time point in each historical failure period, wherein the monitoring data includes but is not limited to data transmission rate, storage space data, performance data, and network connection data. Figure 2 As shown in FIG, it is a graph showing the data transmission rate changes at various time points within a cycle. Figure 2 Just to facilitate the explanation of the rate change law, in specific applications, the monitoring data is mainly stored in the database in the form of key-value pairs, and the difference between this application and the existing method is that the monitoring data obtained after log analysis in this scheme is stored with the monitoring data type as the key and the monitoring data of different periods in the historical fault cycle as the value.

[0054] Reasoning is performed based on the monitoring data and corresponding fault types at each time point in the historical fault cycle to generate a prompt reasoning chain for fault detection and the reasoning answer corresponding to the prompt reasoning chain, which facilitates the detection of operation and maintenance faults in the current cycle.

[0055] Step S102: generating a question to be answered for inquiring about the current fault type based on the monitoring data corresponding to each time point in the current cycle;

[0056] In step S102, the monitoring log monitored by the monitoring system is periodically obtained, and then the monitoring log is analyzed to obtain the monitoring data corresponding to each time point in the current cycle. Then, based on the monitoring data at each time point in the current cycle, questions to be answered are generated to inquire about the current fault type. Figure 2 For example, if the data transmission rates corresponding to each time point in the current cycle are as follows Figure 2 As shown, according to Figure 2 The questions to be answered are:

[0057] "When the data transmission rate at point 0 is 150 kbps, the data transmission rate at point 1 is 250 kbps, the data transmission rate at point 2 is 100 kbps, the data transmission rate at point 3 is 450 kbps, the data transmission rate at point 4 is 350 kbps, and the data transmission rate at point 5 is 300 kbps, what is the corresponding fault type?"

[0058] Step S103 : Reasoning is performed based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain the current fault type and the reasoning chain used to determine the current fault type.

[0059] In step S103, the prompt reasoning chains and reasoning answers of different fault types generated based on the monitoring data of historical fault cycles are used as prompts and input into the big model together with the questions to be answered generated based on the monitoring data of the current cycle. The big model will complete the fault type monitoring and positioning of the current cycle monitoring data based on the prompt reasoning chains and reasoning answers of different fault types, and generate the current fault reasoning chain.

[0060] In an embodiment of the present invention, first, based on the monitoring data at each time point in the historical fault cycle and the fault type corresponding to the monitoring data, a prompt reasoning chain for fault detection and the reasoning answer corresponding to the prompt reasoning chain are generated; then, based on the monitoring data corresponding to each time point in the current cycle, a question to be answered is generated for inquiring about the current fault type; finally, based on the question to be answered, the prompt reasoning chain and the reasoning answer corresponding to the prompt reasoning chain, reasoning is performed to obtain the current fault type and the reasoning chain for determining the current fault type. In an embodiment of the present invention, the monitoring data of the historical fault cycle is used to generate a prompt reasoning chain for fault type detection and location in the current cycle, so that the monitoring data of different fault types in the historical fault cycle can be used to completely and accurately detect the fault type of the current cycle, and the determination of the fault time point can be achieved.

[0061] Optionally, the method further includes:

[0062] Determine the warning message template that matches the current fault type;

[0063] According to the warning message template, identifying and extracting required fault information from the inference chain of the current fault type and the data corresponding to the current fault type;

[0064] Fill the fault information into the warning message template to generate a warning message.

[0065] In an embodiment of the present invention, after outputting the current fault type and the reasoning chain used to determine the current fault type in step S103, the server first selects a warning message template that matches the fault type from the pre-set warning message template, and then, based on the warning message template, identifies and extracts the required fault information from the reasoning chain of the current fault type and the data corresponding to the current fault type, and fills it into the warning message template to generate warning information, wherein the key information includes but is not limited to: time information of the occurrence of the fault, monitoring data at the time point of the fault, etc.

[0066] Optionally, generating a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and a fault type corresponding to the monitoring data includes:

[0067] For each fault cycle in the historical fault cycles, a plurality of first questions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type to obtain a question set, wherein one fault cycle corresponds to one question set;

[0068] Clustering the first problem in the problem set according to a K-means clustering algorithm to obtain multiple target problems;

[0069] According to the plurality of target questions, a large language model is used to perform reasoning to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain.

[0070] In this embodiment of the present invention, multiple first questions are first generated based on the monitoring data and corresponding fault types at each time point within each historical fault cycle, resulting in multiple question sets. Then, using the K-means clustering algorithm, the first questions in the question set are clustered based on their similarity, and the first question closest to the cluster center is selected as an example, or target question. Finally, based on the target question, a prompt reasoning chain for fault detection and the corresponding reasoning answer are generated. This improves the accuracy of the prompt reasoning chain.

[0071] Optionally, for each fault cycle in the historical fault cycles, a plurality of first questions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type to obtain a question set, including:

[0072] For each fault cycle in the historical fault cycle, multiple preconditions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type;

[0073] According to the precondition, randomly changing the monitoring data of at least one time point corresponding to the precondition to obtain a question part;

[0074] A plurality of the preconditions and the corresponding question parts are combined to generate a plurality of first questions, to obtain a question set, wherein one of the first questions includes one precondition and one corresponding question part.

[0075] In an embodiment of the present invention, the form of the first question and the composition of the question set are explained. The format of the first question is as follows: Question Identifier (Identity document, ID) = {precondition, question part}, wherein the question identifier is randomly generated, and the premise and question part are both described in natural language. Specifically, the premise in the question is first generated in the form of a factual statement using the monitoring data and fault type corresponding to each time point in a historical fault cycle. Then, by randomly changing the monitoring data corresponding to at least one time point in the cycle, a question part is generated to determine whether the modified monitoring data belongs to the current fault type. Finally, the premise and question part are combined to obtain the generated first question.

[0076] by Figure 2 For example, a historical fault cycle includes monitoring data at six time points. If the data transmission rates corresponding to each time point in one of the historical fault cycles are as follows: Figure 2 As shown in the figure, the data transmission rates corresponding to each time point are 150kbps, 250kbps, 400kbps, 450kbps, 350kbps, and 300kbps, respectively. One of the first questions generated is:

[0077] "Question 1 = {Precondition: When the data transmission rate at point 0 is 150kbps, the data transmission rate at point 1 is 250kbps, the data transmission rate at point 2 is 400kbps, the data transmission rate at point 3 is 450kbps, the data transmission rate at point 4 is 350kbps, and the data transmission rate at point 5 is 300kbps, the corresponding fault type is fault type 1. Question part: When the data transmission rates corresponding to each time point are 150kbps, 250kbps, 100kbps, 450kbps, 350kbps, and 300kbps respectively, does it belong to fault type 1?"

[0078] The question set is composed of multiple first questions.

[0079] Optionally, clustering the first question in the question set according to a K-means clustering algorithm to obtain multiple target questions includes:

[0080] Processing the first questions in the question set using a sentence embedding representation model to obtain a question representation for each of the first questions;

[0081] Clustering the problem representation of the first problem according to the K-means clustering algorithm to obtain multiple clusters;

[0082] For each cluster, at least one first question is selected as a target question in descending order of distances from the first questions in the cluster to the cluster center.

[0083] In the embodiment of the present invention, the target problem is selected by clustering. The specific clustering method is as follows:

[0084] First, we use the sentence embedding representation model (Sentence Embeddings using Siamese BERT-Networks, SentenceBERT) to calculate the vector representation of each first question in the question set, and average the context vectors to form a fixed-size question representation.

[0085] The question representations are then processed using the K-means (kmeans) clustering algorithm to generate k clusters of questions. The first questions in each cluster i are sorted into a list q(i), where the first questions are sorted in ascending order of distance from the center of the corresponding cluster i. A representative example of each cluster, the target question, is then sampled. Questions closer to the cluster center are considered earlier. The target question is ultimately used to prompt the large language model to generate a reasoning chain for the corresponding question set.

[0086] Taking a problem set Q as an example, the clustering algorithm is as follows:

[0087]

[0088] It should be noted that the basis for grouping the generated first questions in the clustering processing in the example of the present invention is: clustering the questions with the same or similar distribution patterns of the premise conditions of the first questions and the same or similar distribution patterns of the modified monitoring data of the question part.

[0089] Optionally, the generating of a prompt inference chain for fault detection and a corresponding inference answer of the prompt inference chain by using a large language model based on the plurality of target questions includes:

[0090] Randomly combining multiple target questions to obtain multiple question combinations;

[0091] According to the zero-sample set conversion method, the large language model is used to reason the question combination to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain, wherein one question combination corresponds to one prompt reasoning chain.

[0092] In an embodiment of the present invention, the natural language questions input into the model can be understood based on a large language model (LLM), and corresponding answer text or pictures, etc. can be output. In an embodiment of the present invention, due to the premise conditions included in the question during the question generation stage, the first question generated has certain domain knowledge, that is, it includes the correlation between the change rules of the monitoring data and the fault type. Therefore, the pre-trained general LLM model can be used directly without fine-tuning the model based on the current domain data set. This can save design and development costs. At the same time, considering the semantic limitations of a single question, this part needs to perform semantic cross-combination on the selected target questions, that is, randomly select different numbers of different target questions for combination. Then process them through LLM. The relevant implementation is as follows:

[0093] First, all target questions need to be combined according to a random strategy. For example, if 10 target questions are selected, target question 1 can be combined with target question 2, target question 1 with target questions 2 and target question 3, and target question 2 with target questions 4, target question 6, and target question 7. By combining different target questions, the resulting combination can achieve semantic crosstalk for each target question in the combination.

[0094] Secondly, the multiple target problem combinations obtained after combination are input into LLM respectively, and the zero-shot transfer learning (Zero-shot CoT) method is used to generate multiple prompt reasoning chains for fault detection.

[0095] For example:

[0096] A set of target problems is combined as input to the LLM:

[0097] "When the data transmission rate at point 0 is 150 kbps, at point 1 is 250 kbps, at point 2 is 400 kbps, at point 3 is 450 kbps, at point 4 is 350 kbps, and at point 5 is 300 kbps, the corresponding fault type is fault type 1. Question: If the data transmission rates at each time point are 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, and 300 kbps, respectively, does this correspond to fault type 1?"

[0098] If the data transmission rate at point 0 is 150 kbps, at point 1 is 250 kbps, at point 2 is 400 kbps, at point 3 is 450 kbps, at point 4 is 350 kbps, and at point 5 is 300 kbps, the corresponding fault type is fault type 1. Question: If the data transmission rates at each time point are 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, and 100 kbps, respectively, does this correspond to fault type 1? ...

[0099] Please explain your reasoning in detail and provide your answer.

[0100] An example of a prompt inference chain output by LLM is as follows, which includes the inference process:

[0101] “First, when the data transmission rate is 150kbps, 250kbps, 400kbps, 450kbps, 350kbps, and 300kbps, the corresponding fault type is fault type 1;

[0102] Secondly, the variation pattern of the number of transmissions at data rates of 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, and 300 kbps is different from that of the data rates of 150 kbps, 250 kbps, 400 kbps, 450 kbps, 350 kbps, and 300 kbps corresponding to fault type 1;

[0103] Finally, when the data transmission rates are 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, and 300 kbps, the corresponding fault types do not belong to fault type 1.

[0104] First, when the data transmission rate is 150kbps, 250kbps, 400kbps, 450kbps, 350kbps, and 300kbps, the corresponding fault type is fault type 1;

[0105] Secondly, the variation pattern of the number of transmissions at data rates of 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, and 100 kbps is different from that of the data rates of 150 kbps, 250 kbps, 400 kbps, 450 kbps, 350 kbps, and 300 kbps corresponding to fault type 1;

[0106] Finally, when the data transmission rates are 150kbps, 250kbps, 100kbps, 450kbps, 350kbps, and 100kbps, the corresponding fault types do not belong to fault type 1.

[0107] It should be noted that the regular distribution of data transmission rates in the above examples includes time information.

[0108] Finally, the reasoning answers are extracted from multiple prompt reasoning chains, and the reasoning answers are merged corresponding to the prompt reasoning chains to obtain the answer set of each prompt reasoning chain.

[0109] Optionally, the performing reasoning based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain the current fault type and the reasoning chain for determining the current fault type includes:

[0110] Dividing and reorganizing the multiple prompt reasoning chains according to the fault types indicated by the reasoning answers corresponding to the multiple prompt reasoning chains to obtain a target reasoning chain for determining the fault type, wherein the length of the target reasoning chain meets a preset condition;

[0111] For each of the fault types, characteristic information of the fault type is obtained according to the change pattern of the monitoring data in the target reasoning chain corresponding to the fault type, wherein the characteristic information includes data on the data change stage, data change pattern, and change time point;

[0112] Reasoning is performed using a large language model according to the question to be answered, the target reasoning chain, and the feature information to obtain a current fault type and a reasoning chain for determining the current fault type.

[0113] In the embodiment of the present invention, first, to improve the efficiency of operation and maintenance fault detection in the current cycle, before inputting into the large language model, multiple prompt reasoning chains are divided and reorganized to obtain a target reasoning chain with the shortest reasoning process for determining the fault type; then, feature information of each fault type is extracted. The feature information includes but is not limited to data on the data change stage, data change pattern, and change time point. The specific method is as follows:

[0114] Extract the data information included in each reasoning corresponding to each fault type, divide it according to the time information included in the data information, determine the data within a continuous time as a data change stage, and use the data change law as the data change stage feature of the fault type at this stage, and simultaneously determine the overall change law of the data in adjacent continuous times, as well as the data of the change time points of adjacent data change stages. Figure 2 For example, Figure 2The corresponding characteristic information is: the first time point to the fourth time point is the first data change stage, the fourth time point to the sixth time point is the second data change stage, the data change rule in the first data change stage is rising, and the data change rule in the second data change stage is falling. The change time point between the first data change stage and the second data change stage is the fourth time point, and the corresponding data is 450kbps.

[0115] Finally, the prompt is obtained by combining the unanswered question of the current cycle, the characteristic information corresponding to the fault type obtained based on the reasoning chain, and the prompt reasoning chain generated based on the historical fault cycle. The combined prompt is then input into the large language model to obtain the output result of fault location according to the stage. The output result includes the fault type of the current cycle and the specific reasoning chain.

[0116] The following example illustrates the logic for determining the relevant fault type based on the Generative Pre-trained Transformer (GPT) model. The corresponding code can be implemented using Python programming.

[0117] #Hypothetical GPT model calling function

[0118] def generate_fault type(prompt):

[0119] #Actually call the GPT model to generate the code for the fault type

[0120] return "According to the input monitoring data, the fault type is:..." #Here returns the fault inference process and fault type text generated by the model

[0121] #User input data

[0122] question_data="When the data transmission rate at point 0 is 150 kbps, the data transmission rate at point 1 is 250 kbps, the data transmission rate at point 2 is 100 kbps, the data transmission rate at point 3 is 450 kbps, the data transmission rate at point 4 is 350 kbps, and the data transmission rate at point 5 is 300 kbps, what is the corresponding fault type?" #In specific applications, the input data is real-time monitoring data.

[0123] CoT_data = "When the data transmission rate is 150kbps, 250kbps, 400kbps, 450kbps, 350kbps, or 300kbps, the corresponding fault type is fault type 1;

[0124] The variation pattern of the number of transmissions at data transmission rates of 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, and 300 kbps is different from the data transmission rates of 150 kbps, 250 kbps, 400 kbps, 450 kbps, 350 kbps, and 300 kbps corresponding to fault type 1;

[0125] When the data transmission rate is 150kbps, 250kbps, 100kbps, 450kbps, 350kbps, or 300kbps, the corresponding fault type does not belong to fault type 1. #In specific applications, this data is automatically generated based on historical periodic data.

[0126] Features_data="The data change phase characteristics for fault type 1 are: the first data change phase shows an increasing pattern, the second data change phase shows a decreasing pattern, and the rate of change at adjacent time points in the first and second data change phases is 450 kbps."# In specific applications, this data is automatically generated feature information based on the fault type inference chain.

[0127] #Splicing prompt

[0128] prompt=f"{question_data}{CoT_data}{Features_data}"

[0129] #Call the model to generate fault type answers

[0130] answer = generate_fault type (prompt)

[0131] #Print answer

[0132] print(answer).

[0133] Optionally, dividing and reorganizing the plurality of prompt reasoning chains according to the fault types indicated by the reasoning answers corresponding to the plurality of prompt reasoning chains to obtain a target reasoning chain for determining the fault type includes:

[0134] According to the fault type indicated by the reasoning answer, multiple prompt reasoning chains are merged to obtain a first prompt reasoning chain combination;

[0135] Dividing the first prompt reasoning chain combination according to the fault type to obtain a plurality of second prompt reasoning chain combinations, wherein one fault type corresponds to one second prompt reasoning chain combination;

[0136] The prompt reasoning chains in different combinations of the second prompt reasoning chains are merged to generate a target reasoning chain for determining the fault type.

[0137] In this embodiment of the present invention, in order to improve the efficiency of operation and maintenance fault detection in the current cycle, before inputting the large language model, multiple prompt reasoning chains are divided and reorganized to obtain the target reasoning chain with the shortest reasoning process for determining the fault type. The specific method is as follows:

[0138] In the first step, according to the fault type indicated by the reasoning answer, multiple prompt reasoning chains indicating the same fault type are merged to obtain a first prompt reasoning chain combination;

[0139] In the second step, the first prompt reasoning chain combination in the first step is divided according to the fault type to obtain multiple second prompt reasoning chain combinations, where one fault type corresponds to one second prompt reasoning chain combination;

[0140] In the third step, the prompt reasoning chains in the different combinations of the second prompt reasoning chains are merged, with the goal of determining the shortest reasoning chain for the monitoring data fault type. This ensures that the intermediate process between the reasoning process corresponding to the question and the reasoning process corresponding to the answer in the reasoning chain is the shortest, thereby generating a target reasoning chain for determining the fault type. An example of a target reasoning chain is as follows:

[0141] “Fault types include fault type 1, fault type 2, fault type 3, and fault type 4;

[0142] When the data transmission rate is 150kbps, 250kbps, 400kbps, 450kbps, 350kbps, or 300kbps, the corresponding fault type is fault type 1.

[0143] The variation pattern of the number of transmissions at data transmission rates of 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, and 300 kbps is different from the data transmission rates of 150 kbps, 250 kbps, 400 kbps, 450 kbps, 350 kbps, and 300 kbps corresponding to fault type 1;

[0144] When the data transmission rate is 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, or 300 kbps, the corresponding fault type does not belong to fault type 1.

[0145] …

[0146] When the data transmission rate is 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, or 300 kbps, the corresponding fault type does not belong to fault type 2.

[0147] …

[0148] When the data transmission rate is 150 kbps, 250 kbps, 100 kbps, 450 kbps, 350 kbps, or 300 kbps, the corresponding fault type does not belong to fault type 3.

[0149] Therefore, when the data transmission rate is 150kbps, 250kbps, 100kbps, 450kbps, 350kbps, and 300kbps, the corresponding fault type belongs to fault type 4.

[0150] In the above example, when the known fault types include fault type 1, fault type 2, fault type 3, and fault type 4, the reasoning chain determines that when the data transmission rates are 150kbps, 250kbps, 100kbps, 450kbps, 350kbps, and 300kbps, the corresponding fault types do not belong to fault type 1, fault type 2, and fault type 3. Therefore, it is determined that when the data transmission rates are 150kbps, 250kbps, 100kbps, 450kbps, 350kbps, and 300kbps, the corresponding fault type belongs to fault type 4. Only four steps are used to determine the fault type, meeting the goal of the shortest reasoning chain for the monitoring data fault type, and there is no repeated judgment and reasoning process.

[0151] like Figure 3 As shown, an embodiment of the present invention further provides an operation and maintenance fault detection device, comprising:

[0152] The first generating module 301 is configured to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and the fault type corresponding to the monitoring data;

[0153] The second generating module 302 is used to generate a question to be answered for inquiring about the current fault type based on the monitoring data corresponding to each time point in the current cycle;

[0154] The first reasoning module 303 is configured to perform reasoning based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain the current fault type and the reasoning chain used to determine the current fault type.

[0155] Optionally, the device further comprises:

[0156] A first determination module is used to determine a warning message template that matches the current fault type;

[0157] A first processing module is configured to identify and extract required fault information from the inference chain of the current fault type and the data corresponding to the current fault type according to the warning message template;

[0158] The third generating module is used to fill the fault information into the warning message template to generate a warning message.

[0159] Optionally, the first generating module 301 includes:

[0160] a first generating unit configured to generate, for each fault cycle in a historical fault cycle, a plurality of first questions according to monitoring data at each time point in the fault cycle and a corresponding fault type, to obtain a question set, wherein one fault cycle corresponds to one question set;

[0161] a first clustering unit, configured to cluster the first questions in the question set according to a K-means clustering algorithm to obtain a plurality of target questions;

[0162] The first reasoning unit is configured to perform reasoning based on the plurality of target questions using a large language model to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain.

[0163] Optionally, the first generating unit includes:

[0164] A second generating unit is configured to generate, for each fault cycle in the historical fault cycles, a plurality of preconditions according to monitoring data at each time point in the fault cycle and the corresponding fault type;

[0165] a third generating unit, configured to randomly change the monitoring data of at least one time point corresponding to the precondition according to the precondition, to obtain a question part;

[0166] The fourth generating unit is configured to combine the plurality of the preconditions with the corresponding question parts to generate a plurality of first questions, thereby obtaining a question set, wherein one of the first questions includes one precondition and one corresponding question part.

[0167] Optionally, the first clustering unit includes:

[0168] A first processing unit is configured to process the first questions in the question set using a sentence embedding representation model to obtain a question representation of each of the first questions;

[0169] a first clustering unit, configured to cluster the problem representation of the first problem according to a K-means clustering algorithm to obtain a plurality of clusters;

[0170] The first selection unit is configured to select, for each cluster, at least one of the first questions as a target question in descending order of distances from the first questions in the cluster to the cluster center.

[0171] Optionally, the first reasoning unit includes:

[0172] A first combining unit is used to randomly combine a plurality of the target questions to obtain a plurality of question combinations;

[0173] The second reasoning unit is used to use the large language model to reason on the question combination according to the zero-sample set conversion method, generate a prompt reasoning chain for fault detection and an inference answer corresponding to the prompt reasoning chain, wherein one question combination corresponds to one prompt reasoning chain.

[0174] Optionally, the first reasoning module 303 includes:

[0175] a second processing unit, configured to divide and reorganize the plurality of prompt reasoning chains according to the fault types indicated by the reasoning answers corresponding to the plurality of prompt reasoning chains, to obtain a target reasoning chain for determining the fault type, wherein the length of the target reasoning chain satisfies a preset condition;

[0176] a third processing unit, configured to obtain, for each of the fault types, characteristic information of the fault type based on a change pattern of the monitoring data in the target inference chain corresponding to the fault type, the characteristic information including data on a data change stage, a data change pattern, and a change time point;

[0177] The third reasoning unit is configured to perform reasoning using a large language model based on the question to be answered, the target reasoning chain, and the feature information to obtain a current fault type and a reasoning chain for determining the current fault type.

[0178] Optionally, the second processing unit includes:

[0179] a first merging unit, configured to merge the plurality of prompt reasoning chains according to the fault type indicated by the reasoning answer to obtain a first prompt reasoning chain combination;

[0180] a first dividing unit, configured to divide the first prompt reasoning chain combination according to the fault type to obtain a plurality of second prompt reasoning chain combinations, wherein one fault type corresponds to one second prompt reasoning chain combination;

[0181] The second merging unit is configured to merge the prompt reasoning chains in different combinations of the second prompt reasoning chains to generate a target reasoning chain for determining the fault type.

[0182] It should be noted that the embodiment of the device is a device corresponding to the embodiment of the above method, and all implementation methods in the embodiment of the above method are applicable to the embodiment of the device and can achieve the same technical effect.

[0183] An embodiment of the present invention also provides a network device, comprising: a processor, a memory, and a program stored in the memory and runnable on the processor. When the program is executed by the processor, it implements the method for detecting operation and maintenance failures as described in any of the above items and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0184] An embodiment of the present invention further provides a computer-readable storage medium, comprising: a program stored on the computer-readable storage medium, wherein the program, when executed by a processor, implements the steps of the method for detecting operation and maintenance failures as described in any of the above items, and can achieve the same technical effects. To avoid repetition, the details are not described here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0185] An embodiment of the present invention also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the steps of the operation and maintenance failure detection method as described in any of the above items are implemented, and the same technical effects can be achieved. To avoid repetition, they will not be described here.

[0186] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0187] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for detecting operation and maintenance failures, characterized in that: include: Generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and the fault type corresponding to the monitoring data; Generate unanswered questions for inquiring about the current fault type based on the monitoring data corresponding to each time point in the current cycle; Reasoning is performed based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain a current fault type and a reasoning chain for determining the current fault type.

2. The method for detecting operation and maintenance failure according to claim 1, characterized in that: The method further comprises: Determine the warning message template that matches the current fault type; According to the warning message template, identifying and extracting required fault information from the inference chain of the current fault type and the data corresponding to the current fault type; Fill the fault information into the warning message template to generate a warning message.

3. The method for detecting operation and maintenance failure according to claim 1, characterized in that: The generating of a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and a fault type corresponding to the monitoring data includes: For each fault cycle in the historical fault cycles, a plurality of first questions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type to obtain a question set, wherein one fault cycle corresponds to one question set; Clustering the first problem in the problem set according to a K-means clustering algorithm to obtain multiple target problems; According to the plurality of target questions, a large language model is used to perform reasoning to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain.

4. The method for detecting operation and maintenance failure according to claim 3, characterized in that: For each fault cycle in the historical fault cycle, a plurality of first questions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type to obtain a question set, including: For each fault cycle in the historical fault cycle, multiple preconditions are generated according to the monitoring data at each time point in the fault cycle and the corresponding fault type; According to the precondition, randomly changing the monitoring data of at least one time point corresponding to the precondition to obtain a question part; A plurality of the preconditions and the corresponding question parts are combined to generate a plurality of first questions, to obtain a question set, wherein one of the first questions includes one precondition and one corresponding question part.

5. The method for detecting operation and maintenance failure according to claim 3, characterized in that: The step of clustering the first problem in the problem set according to the K-means clustering algorithm to obtain multiple target problems includes: Processing the first questions in the question set using a sentence embedding representation model to obtain a question representation for each of the first questions; Clustering the problem representation of the first problem according to the K-means clustering algorithm to obtain multiple clusters; For each cluster, at least one first question is selected as a target question in descending order of distances from the first questions in the cluster to the cluster center.

6. The method for detecting operation and maintenance failure according to claim 3, characterized in that: The method of performing reasoning based on the plurality of target questions using a large language model to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain includes: Randomly combining multiple target questions to obtain multiple question combinations; According to the zero-sample set conversion method, the large language model is used to reason the question combination to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain, wherein one question combination corresponds to one prompt reasoning chain.

7. The method for detecting operation and maintenance failure according to claim 1, characterized in that: The performing reasoning based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain the current fault type and the reasoning chain for determining the current fault type includes: Dividing and reorganizing the multiple prompt reasoning chains according to the fault types indicated by the reasoning answers corresponding to the multiple prompt reasoning chains to obtain a target reasoning chain for determining the fault type, wherein the length of the target reasoning chain meets a preset condition; For each of the fault types, characteristic information of the fault type is obtained according to the change pattern of the monitoring data in the target reasoning chain corresponding to the fault type, wherein the characteristic information includes data on the data change stage, data change pattern, and change time point; Reasoning is performed using a large language model according to the question to be answered, the target reasoning chain, and the feature information to obtain a current fault type and a reasoning chain for determining the current fault type.

8. The method for detecting operation and maintenance failure according to claim 7, characterized in that: The method of dividing and reorganizing the plurality of prompt reasoning chains according to the fault types indicated by the reasoning answers corresponding to the plurality of prompt reasoning chains to obtain a target reasoning chain for determining the fault type includes: According to the fault type indicated by the reasoning answer, multiple prompt reasoning chains are merged to obtain a first prompt reasoning chain combination; Dividing the first prompt reasoning chain combination according to the fault type to obtain a plurality of second prompt reasoning chain combinations, wherein one fault type corresponds to one second prompt reasoning chain combination; The prompt reasoning chains in different combinations of the second prompt reasoning chains are merged to generate a target reasoning chain for determining the fault type.

9. A device for detecting operation and maintenance failures, characterized in that: include: A first generating module is configured to generate a prompt reasoning chain for fault detection and a reasoning answer corresponding to the prompt reasoning chain based on monitoring data at each time point in a historical fault cycle and the fault type corresponding to the monitoring data; The second generation module is used to generate questions to be answered for inquiring about the current fault type based on the monitoring data corresponding to each time point in the current cycle; The first reasoning module is used to perform reasoning based on the question to be answered, the prompt reasoning chain, and the reasoning answer corresponding to the prompt reasoning chain to obtain the current fault type and the reasoning chain used to determine the current fault type.

10. A network device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the method for detecting operation and maintenance failures according to any one of claims 1 to 8.

11. A readable storage medium, characterized in that: include: The readable storage medium stores a program, and when the program is executed by the processor, the steps of the method for detecting operation and maintenance failures according to any one of claims 1 to 8 are implemented.

12. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the method for detecting operation and maintenance failures as described in any one of claims 1 to 8.

Images