Clustering aggregation repeated operation and maintenance order dispatching method, system, equipment and medium

Abstract

The invention provides a clustering aggregation repetitive operation and maintenance order dispatching method, system and device and a medium, and belongs to the technical field of work order processing, and the method comprises the steps: obtaining a work order, and carrying out the preprocessing; extracting work order features from the preprocessed work order; unsupervised clustering is carried out on the feature vectors, the cosine similarity of any two feature vectors is calculated, and a neighborhood radius threshold value and the minimum point number are set; identifying a core point, and attributing all points in a neighborhood of the core point to an initial cluster where the core point is located; recursively checking the neighborhood of the points in the neighborhood, and finally forming a preliminary clustering result; clustering parameters are optimized according to the evaluation result, and a final clustering result is determined; common features of the feature vectors in the cluster are analyzed, and an interpretation text is generated; and merging the same cluster work order and sending the same cluster work order to a corresponding processing party. Through cooperation of a plurality of steps, full-process automation from original operation and maintenance work order data processing to efficient order sending is realized. The accuracy and efficiency of work order processing are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of work order processing technology, specifically relating to a clustering and aggregation method, system, equipment, and medium for dispatching repetitive maintenance work orders. Background Technology

[0002] In the field of modern information technology operations and maintenance (O&M), O&M work orders are a crucial tool for ensuring the stable operation of various software and hardware systems. With the expansion and increasing complexity of enterprise information systems, the number of O&M work orders has exploded. These work orders originate from various channels, including user feedback, alarms triggered by system monitoring systems, and issues discovered during regular inspections. Traditional O&M work order dispatch systems primarily rely on manual experience or simple rule matching for work order distribution, which has the following main shortcomings: In the work order feature extraction stage, related technologies struggle to effectively extract key features that accurately distinguish different faults. Bag-of-words models or text representation methods cannot fully consider the differences in importance of words when describing faults, easily ignoring key information or being interfered with by secondary information. This results in extracted features that do not accurately reflect the essence of the fault, affecting the accuracy of subsequent clustering and fault judgment.

[0003] In terms of work order clustering, relevant clustering algorithms require pre-specifying the number of clusters. However, in actual maintenance work order scenarios, it is difficult to accurately predict the types and numbers of faults. If the pre-set number of clusters is unreasonable, it will lead to inaccurate clustering results, ineffective aggregation of similar work orders, misclassification of different work orders, and affect the efficiency of identifying and processing duplicate work orders.

[0004] In the work order dispatching process, the existing method usually dispatches work orders one by one. It cannot effectively identify and merge duplicate faulty work orders, resulting in the processor receiving a large number of duplicate work orders, increasing the workload and reducing processing efficiency. Summary of the Invention

[0005] This invention provides a clustering and aggregation method for dispatching repetitive maintenance work orders. Through the coordinated operation of multiple steps, it achieves full automation from processing original maintenance work order data to efficient work order dispatch. This effectively improves the accuracy and efficiency of work order processing; it quickly and accurately identifies and rationally dispatches repetitive work orders, shortens fault handling time, and enhances the stability and reliability of the business system.

[0006] The methods include: Step S101: Obtain the work order and perform preprocessing; Step S102: Extract work order features from the preprocessed work order, calculate the contribution score of each word to the work order fault through the attention mechanism, normalize to obtain the attention weight, and then generate the feature vector by weighted summation of the corresponding hidden vectors. Step S103: Perform unsupervised clustering on the feature vectors, calculate the cosine similarity of any two feature vectors, set the neighborhood radius threshold and the minimum number of points; identify the core point, and assign all points in the neighborhood of the core point to the initial cluster in which the core point is located; recursively check the neighborhood of the points in the neighborhood, and if it contains unassigned points and meets the density condition, then continue to include them in the current cluster and expand the neighborhood range, finally forming a preliminary clustering result containing the core point, density-connected points and recursively expanded points; Step S104: Based on the preliminary clustering results, calculate the average silhouette coefficient of all feature vectors using silhouette coefficient evaluation to measure the overall compactness and separation; optimize the clustering parameters according to the evaluation results and determine the final clustering results; Step S105: Analyze the common features of the feature vectors within the clusters based on the final clustering results, and generate explanatory text; Step S106: Based on the explanation text, merge the work orders of the same cluster and send them to the corresponding processing party.

[0007] It should be further noted that step S102 specifically includes: inputting the preprocessed work order into the LLM large-scale language model, and defining the normalized text sequence of the work order as... Where n is the number of words, and the hidden vector representation h of each word is obtained through LLM encoding. i , where h i It is a semantic vector of dimension d; The importance score e for each word is calculated using a self-attention mechanism. i Introduce inter-cluster separation constraints; if word t i In historical clustering, terms with high discriminative power are assigned an enhancement factor λ (λ>1) to their weights; conversely, terms with low discriminative power are assigned an inhibition factor μ (μ<1). ; Where, δ i For 0-1 indicator variables, if t i If it is a highly discriminative term, then δ i =1, otherwise δ i =0.

[0008] The importance scores are normalized using softmax to obtain the attention weights. ;

[0009] in, ∈(0,1), ; The hidden vectors are weighted and summed using attention weights to obtain the feature vector v of the work order. ; The feature vector v is mapped to a low-dimensional subspace of the clustering space through linear projection, and the projection formula is as follows: Where W has a dimension of db×hb, where db is the cluster space dimension and hb is the LLM hidden layer dimension.

[0010] It should be further explained that the unsupervised clustering of the feature vectors in step S103 specifically includes: Cosine similarity is used to calculate the feature vectors v of any two work orders. i and v j The similarity is calculated using the following formula:

[0011] Define distance metric ; Define the neighborhood radius threshold ε and the minimum number of points MinPts.

[0012] For any work order feature vector v p ε-neighborhood , is the distance v p The set of all eigenvectors within ε; If the number of points in the neighborhood Then v p Considered the core point; All points within the ε-neighborhood of the core point are assigned to the same cluster. Starting from the core point, its neighborhood is recursively checked, and points with density connections are continuously merged into the same cluster. If the number of points in the ε-neighborhood of a certain feature vector is less than MinPts, that is, there are no enough neighbors, then it is marked as a noise point; For each sample i of the feature vector, the formula for calculating the silhouette coefficient s(i) is as follows: ; Where a(i) represents the average distance between sample i and other samples in its cluster; b(i) represents the average distance between sample i and its nearest other cluster; By calculating the average silhouette coefficient of all eigenvectors Pm represents the total number of work orders and can be used to measure the overall quality of clustering.

[0013] It should be further noted that step S103 also includes: Construct a fault type feature word mapping table, and generate a domain knowledge constraint rule base based on the fault types and corresponding high-frequency feature words marked in historical operation and maintenance work orders; Adjust ε and MinPts based on the work order's business priority and historical dispatch timeliness data: The parameter adjustment formula is: ; ; Define historical clustering quality metrics, if the current parameters are... If the clustering quality is poor, or nr > 0.3, then increase the parameter adjustment range; like If the clustering quality is poor and nr ≤ 0.2, then reduce the adjustment range. The adjusted formula is as follows:

[0014] Where k and m are adjustment coefficients.

[0015] It should be further explained that step S104 specifically includes: Define intra-cluster compactness and inter-cluster separation, and define a fault type consistency index based on the operation and maintenance scenario: statistically analyze the consistency of fault type labels of work orders within a cluster, and finally generate a comprehensive quality index. ; DB stands for Davies-Bouldin index, CH stands for Calinski-Harabasz index, and FT stands for fault type consistency score. Using the comprehensive quality index Q as the objective function, the particle swarm optimization algorithm is used to adjust the neighborhood radius ε and the minimum number of points MinPts. Each particle in the particle swarm represents a set of parameter combinations. By iteratively updating the particle's velocity and position, the search seeks the parameter combination that maximizes Q. The optimization formula is:

[0016] Among them, v i (t) is the particle velocity, x i (t) represents the parameter combination, ω represents the inertia weight, c1 and c2 are learning factors, r1 and r are random numbers, and p best,i For the individual optimal parameters, g best These are the globally optimal parameters.

[0017] It should be further explained that step S105 specifically includes: Construct a failure mode knowledge graph; Before generating explanatory text, the keyword groups of the current cluster are mapped to the fault mode knowledge graph to obtain contextual information; Generate differentiated explanations based on the needs of users at different levels; By embedding the chain of evidence for fault diagnosis into the explanatory text, and by analyzing the time series of work orders and the changing trends of system indicators within the cluster, we can better understand the process.

[0018] It should be further noted that step S105 also includes: Collect the root cause analysis results of historical clustering, and construct a root cause association graph with fault type as node and root cause similarity as edge weight; In Bayesian networks, the conditional probability parameters of root cause nodes are adjusted based on the service priority and failure frequency of work orders. The root cause hypothesis obtained from causal reasoning is cross-validated with the system log data of the work order and the user feedback description. If the log evidence conflicts with the root cause of the inference, the Bayesian network parameters are readjusted for a second inference. Based on the results of root cause reasoning, key features supporting the root cause hypothesis are extracted to generate an explanatory chain.

[0019] This application also provides a clustering and aggregation-based repetitive operation and maintenance dispatching system, the system comprising: The preprocessing module is used to obtain work orders and perform preprocessing. The feature extraction module is used to extract work order features from the preprocessed work orders. It calculates the contribution score of each word to the work order fault through the attention mechanism, normalizes it to obtain the attention weight, and then generates the feature vector by weighted summation of the corresponding hidden vectors. The clustering module is used to perform unsupervised clustering of feature vectors, calculate the cosine similarity of any two feature vectors, set the neighborhood radius threshold and the minimum number of points; identify core points, and assign all points in the neighborhood of the core point to the initial cluster in which the core point is located; recursively check the neighborhood of the points in the neighborhood, and if it contains unassigned points and meets the density condition, it continues to include them in the current cluster and expands the neighborhood range, finally forming a preliminary clustering result containing core points, density-connected points and recursively expanded points; The evaluation and optimization module, based on the preliminary clustering results, calculates the average silhouette coefficient of all feature vectors to measure the overall compactness and separation; and optimizes the clustering parameters according to the evaluation results to determine the final clustering results. The explanation generation module analyzes the common features of feature vectors within clusters based on the final clustering results and generates explanatory text. The dispatch module merges work orders belonging to the same cluster and dispatches them to the corresponding processing parties based on the explanatory text.

[0020] According to another embodiment of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the clustering and aggregation repetitive operation and maintenance dispatching method.

[0021] According to another embodiment of this application, a storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the clustering and aggregation repetitive operation and maintenance dispatching method.

[0022] As can be seen from the above technical solutions, the present invention has the following advantages: The clustering and aggregation method for repetitive maintenance dispatching provided by this invention utilizes an edit distance algorithm to correct spelling errors, ensuring the accuracy of work order text and avoiding semantic misunderstandings caused by typos. Uniform named entities ensure consistent representation of the same concepts across different work orders, facilitating accurate identification and analysis. Leveraging a large-scale language model and attention mechanism, it can accurately extract key features from work order text that differentiate between faults. The generated overall feature vector effectively highlights the differences between different faults, improving the accuracy and relevance of feature extraction.

[0023] Density-based clustering algorithms do not require pre-specifying the number of clusters. They adaptively discover cluster structures based on the data distribution of work order feature vectors, automatically aggregating duplicate work orders with similar content. This effectively avoids the problem of inaccurate manually preset cluster numbers, improving the efficiency and reliability of duplicate work order identification. Silhouette coefficients are used to quantitatively evaluate clustering results, comprehensively measuring cluster quality from two key dimensions: intra-cluster compactness and inter-cluster separation, making the evaluation results more scientific and objective. Optimizing clustering parameters based on evaluation results allows for continuous adjustment of clustering effects, ensuring high-quality final clustering results. Large-scale language models automatically generate explanatory text, quickly and accurately summarizing the common features and key details of work orders in each cluster. This clearly presents information such as fault type, manifestation, and possible causes in natural language, reducing the time and workload of manual analysis and summarization. Merging work orders from the same cluster for dispatch avoids duplicate dispatch of faulty work orders, improving work order processing efficiency. Attached Figure Description

[0024] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 Flowchart of the clustering and aggregation method for dispatching repetitive maintenance tasks; Figure 2 A schematic diagram of a clustering and aggregation system for repetitive maintenance order dispatching; Figure 3 This is a schematic diagram of an electronic device. Detailed Implementation

[0026] The clustering and aggregation method for recurring maintenance work order dispatch provided by this invention automatically groups work orders with the same fault or defect together using unsupervised clustering. It also improves the accuracy of recurring work order aggregation by introducing a large language model to extract the most distinctive information from the work order descriptions. Furthermore, this invention includes a mechanism for automatically generating explanations of the clustering results to help maintenance engineers understand the basis for aggregation, enhancing the interpretability and credibility of the results.

[0027] The following describes in detail the clustering and aggregation method for repetitive maintenance dispatching involved in this application. Specific details, such as particular system structures and technologies, are presented for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details.

[0028] It should be understood that, when used in this specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0029] The terms "one embodiment" or "some embodiments" used in this application mean that one or more embodiments of this application include the specific features, structures, or characteristics described in that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this application do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0030] In embodiments of the present invention, computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (exemplarily using an Internet service provider for Internet connection).

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 The diagram shows a flowchart of a clustering and aggregation method for repetitive maintenance task dispatching in a specific embodiment. The method includes: Step S101: Obtain the work order and perform preprocessing.

[0033] In some embodiments, the original maintenance work order is obtained, and it is standardized and noise filtered. English characters in the text are uniformly converted to uppercase or lowercase. Special symbols, emoticons and other irrelevant content are removed. The date and time formats are uniformly converted according to the specified standards so that the description style of different work orders tends to be consistent.

[0034] This embodiment utilizes a dictionary or model to correct spelling errors or common typos in work order descriptions. An edit distance algorithm is used to match the nearest correct word. For a given word w, the Levenshtein distance between w and each word in the dictionary is calculated. The Levenshtein distance refers to the minimum number of edit operations required to transform one string into another, including insertion, deletion, and replacement. The word w′ that minimizes the Levenshtein distance is found, and w is replaced with w′.

[0035] This embodiment uses a unified name to represent the same entity, such as system name or module code. For example, "login service" and "login service" are unified under the same term. By setting a word frequency or inverse document frequency (IDF) threshold, noisy words (such as common stop words) and excessively rare abnormal words that do not help distinguish faults are filtered out.

[0036] The specific calculation method is as follows: calculate the document frequency df(t) of each word, where df(t) represents the number of work orders containing word t, and the inverse document frequency. If a word's df(t) is too high, meaning it appears in almost all work orders, its inverse document frequency is close to 0; or if df(t) is too low, meaning it appears only once and is not a keyword, its inverse document frequency is too high, then the word is considered noise and removed.

[0037] After preprocessing, the parts of the work order descriptions that are not related to the fault are eliminated to the greatest extent possible, thus completing the alignment of the work order description modality.

[0038] Step S102: Extract work order features from the preprocessed work orders, calculate the contribution score of each word to the work order fault through the attention mechanism, normalize to obtain the attention weight, and then generate the feature vector by weighted summation of the corresponding hidden vectors.

[0039] In some embodiments, the preprocessed work order is input into a large language model. The model encodes each word in the text sequence to generate a corresponding hidden vector, forming a hidden vector sequence. Using the model's attention mechanism, the importance score of each word in describing the work order fault is calculated. This score reflects the degree to which the word contributes to the meaning of the entire work order.

[0040] The importance scores of all words are normalized to obtain attention weights with values ​​between 0 and 1, summing to 1. A higher weight indicates a more significant ability of the word to distinguish faults. Based on the obtained attention weights, the hidden vectors corresponding to each word are weighted and summed to generate a feature vector that highlights the key information of the work order faults.

[0041] In some specific embodiments, the preprocessed work order is input into the LLM large-scale language model, and the semantic understanding ability of the pre-trained deep neural network is used to extract key feature representations that can distinguish different faults.

[0042] It can generate high-dimensional semantic vectors or extract key feature words / phrases. The preferred method is based on the attention mechanism. The steps are as follows: The normalized text sequence of the work order is Where n is the number of words, and the hidden vector representation h of each word is obtained through LLM encoding. i , where h i It is a semantic vector of dimension d.

[0043] The importance score e for each word is calculated using a self-attention mechanism. i The score reflects the contribution of words to the overall meaning of the work order. For example, it is calculated through linear transformation and activation function: e i =h i T W a h c +b a Among them, W a It is a learnable weight matrix, h c For context vectors, such as the cluster centers of similar historical work orders, b a This is a bias term.

[0044] To further bind the clustering objective, an inter-cluster separation constraint is introduced. If word t iIn historical clustering, terms with high discriminative power are assigned an enhancement factor λ (λ>1) to their weights; conversely, terms with low discriminative power are assigned an inhibition factor μ (μ<1). ; Where, δ i For 0-1 indicator variables, if t i If it is a highly discriminative term, then δ i =1, otherwise δ i =0.

[0045] The importance scores are normalized using softmax to obtain the attention weights. ;

[0046] in, ∈(0,1), The greater the weight of the word t, the better. i The more significant the ability to distinguish faults.

[0047] The hidden vectors are weighted and summed using attention weights to obtain the feature vector v of the work order. Feature vectors highlight key semantic information to represent the differences between different faults.

[0048] To ensure that the feature vector v can directly serve subsequent clustering tasks, the feature vector v is mapped to a low-dimensional subspace of the clustering space through linear projection, and the projection matrix W is obtained by contrastive learning pre-training.

[0049] The projection formula is W has dimensions of db×hb, where db is the clustering space dimension and hb is the LLM hidden layer dimension. The feature vector v condenses the key information for distinguishing different faults in the semantic space and is highly aligned with the objective function of the clustering task, providing highly discriminative input features for subsequent efficient clustering and aggregation of repetitive work orders.

[0050] The feature vector extracted in this embodiment can accurately capture the core features of work order faults. Compared with simple text representation methods, it can more effectively distinguish different types of faults, improve the accuracy and efficiency of subsequent clustering operations, and facilitate the rapid identification of duplicate fault work orders.

[0051] Step S103: Perform unsupervised clustering on the feature vectors, calculate the cosine similarity of any two feature vectors, set the neighborhood radius threshold and the minimum number of points; identify the core point, and assign all points in the neighborhood of the core point to the initial cluster in which the core point is located; recursively check the neighborhood of the points in the neighborhood, and if it contains unassigned points and meets the density condition, then continue to include them in the current cluster and expand the neighborhood range, finally forming a preliminary clustering result containing the core point, density-connected points and recursively expanded points.

[0052] In some embodiments, the cosine similarity between any two work order feature vectors is calculated, and the similarity value is mapped to a distance metric. 1 minus the similarity value indicates that the features of the two work orders are more similar.

[0053] Two key parameters are set: a neighborhood radius threshold and a minimum number of points. For each feature vector, the number of other feature vectors within its neighborhood radius is calculated. If the number reaches or exceeds the minimum number of points, the feature vector is identified as a core point, and all feature vectors within its neighborhood are assigned to the initial cluster containing that core point. For each point already assigned to a cluster, its neighborhood is recursively checked. If an unassigned point is found whose neighborhood number meets the density condition, that point is included in the current cluster, and its neighborhood is expanded from that point as a new starting point. This process is repeated until all feature vectors have been processed, forming a preliminary clustering result.

[0054] The density condition here is the criterion used in density-based clustering algorithms to determine whether a point is a core point, should be included in a cluster, and the termination condition for cluster expansion. Specifically, the density condition is whether, given a neighborhood radius threshold ε, the number of samples contained within a spherical region centered on the point and with a radius of ε reaches or exceeds a preset minimum number of points, MinPts. If this condition is met, the point is defined as a core point, and points within its neighborhood can be included in the same cluster; if not, the point is considered a noise point or a boundary point, and cluster expansion cannot be initiated.

[0055] In some specific embodiments, unsupervised clustering algorithms mainly achieve automatic discovery of duplicate work order sets and evaluation of clustering results through the following steps: Cosine similarity is used to calculate the feature vectors v of any two work orders. i and v j The similarity is calculated using the following formula:

[0056] Here, the numerator is the dot product of the two vectors, and the denominators are the product of the magnitudes of the two vectors. The similarity value ranges from 0 to 1. Based on this, a distance metric is defined. The distance value also ranges from 0 to 1. The smaller the value, the more similar the descriptions of the two work orders are.

[0057] This embodiment sets two key parameters for the DBSCAN algorithm: the neighborhood radius threshold ε and the minimum number of points MinPts.

[0058] For any work order feature vector v p Its ε-neighborhood , is the distance v pThe set of all eigenvectors within ε.

[0059] If the number of points in this neighborhood Then v p It is considered the core point.

[0060] All points within the ε-neighborhood of a core point belong to the same cluster. Then, using these points as new starting points, their neighborhoods are recursively checked, and points with density connections are continuously merged into the same cluster.

[0061] If the number of points in the ε-neighborhood of certain feature vectors is less than MinPts, meaning there are not enough neighbors, then these points are marked as noise points or isolated points. Through this process, the DBSCAN algorithm can adaptively discover clustering structures based on data distribution, automatically grouping similar duplicate work orders into one category without needing to pre-specify the number of clusters.

[0062] This embodiment calculates the silhouette coefficient to evaluate the clustering results. For each sample i, the formula for calculating the silhouette coefficient s(i) is as follows: .

[0063] Where a(i) represents the average distance between sample i and other samples in its cluster, reflecting the intra-cluster dissimilarity; b(i) represents the average distance between sample i and its nearest other cluster, i.e. the minimum inter-cluster dissimilarity.

[0064] The value of s(i) ranges from -1 to 1; a larger value indicates a higher degree of clustering rationality for sample i. The average silhouette coefficient of all feature vectors is calculated. Pm represents the total number of work orders and can be used to measure the overall quality of clustering. If... A high profile coefficient indicates good clustering density and separation, meaning that work orders with the same fault are correctly clustered, and work orders with different faults can be clearly distinguished from each other. Based on the profile coefficient evaluation results, clustering parameters can be further adjusted, such as changing the value of ε or selecting other distance metrics, to obtain the best clustering effect.

[0065] This embodiment can automatically discover clustering structures in work order data without requiring manual pre-setting of the number of clusters, and is applicable to work order data of different sizes and distributions. By aggregating similar work orders based on density judgment, it effectively identifies sets of duplicate fault work orders, improving the efficiency of operation and maintenance work order processing.

[0066] Step S104: Based on the preliminary clustering results, calculate the average silhouette coefficient of all feature vectors through silhouette coefficient evaluation to measure the overall compactness and separation; optimize the clustering parameters according to the evaluation results and determine the final clustering results.

[0067] In some embodiments, for each feature vector in the preliminary clustering results, its silhouette coefficient is calculated. The silhouette coefficient is obtained by calculating the average distance between the feature vector and other feature vectors in its cluster (i.e., intra-cluster dissimilarity) and the average distance between the feature vector and the feature vectors in the nearest other cluster (i.e., inter-cluster dissimilarity), and a value between -1 and 1 is obtained. The closer the value is to 1, the higher the rationality of the clustering of the feature vector.

[0068] Calculate the average silhouette coefficient of all feature vectors to evaluate the compactness and separation of the entire clustering result. If the average silhouette coefficient is low, it indicates poor clustering effect. Adjust clustering parameters such as neighborhood radius threshold and minimum number of points, and repeat the clustering operation until a higher average silhouette coefficient is obtained to determine the final clustering result.

[0069] By evaluating the indicators and optimizing the parameters, the quality of clustering results can be objectively assessed, avoiding clustering bias caused by improper parameter settings. This ensures that the final clustering results have good quality.

[0070] Step S105: Analyze the common features of the feature vectors within the clusters based on the final clustering results, and generate explanatory text.

[0071] In some embodiments, for each cluster in the final clustering result, all feature vectors within the cluster are analyzed to extract common key features, such as the faulty components, error phenomena, and error codes involved. The extracted common features are input into a large language model, which analyzes and integrates these features to generate detailed explanatory text in natural language, clearly describing the fault type, fault manifestation, and possible causes reflected in the work orders within that cluster.

[0072] It should be noted that after clustering is completed, this invention automatically generates explanatory text for each cluster through the interpretation module, explaining the reasons why work orders within that cluster are classified into one category.

[0073] Specifically, a large-scale language model is used to analyze the work order descriptions within clusters, summarizing their common characteristics and key details. For example, the model can extract common information such as problem symptoms, error codes, and occurrence environments corresponding to the cluster, and generate explanatory sentences in natural language, such as "All work orders in this cluster report database connection timeout errors, and the errors occurred after the system upgrade of version XYZ." This explanatory text clearly points out the basis for clustering and the essential characteristics of the fault, helping engineers understand why these work orders are considered to be the same problem. Finally, work orders are dispatched based on the clustering results: the system can assign the work order set of each cluster as a merged work order task to the corresponding processing team or engineer, thereby avoiding duplicate dispatch. For example, if a cluster indicates that several user reports are related to the "login timeout" problem, only one merged task order needs to be generated and dispatched to the engineer responsible for the login module, instead of dispatching each duplicate work order individually. With the help of automatically generated explanations, engineers can quickly understand that the problem has occurred multiple times and its common symptoms when they receive merged work orders, greatly improving the efficiency of problem response and resolution.

[0074] As can be seen, the generated explanatory text can clearly present the core characteristics and root causes of the fault, helping maintenance personnel to quickly understand the clustering results, reduce the time spent repeatedly reading and analyzing work order content, and improve the efficiency of fault diagnosis and handling.

[0075] Step S106: Based on the explanation text, merge the work orders of the same cluster and send them to the corresponding processing party.

[0076] In some embodiments, the fault type and processing requirements corresponding to each cluster are clarified based on the explanatory text generated in step S105. All work orders in the same cluster are merged into a task unit, and the merged work order task is assigned to the corresponding processing team or engineer according to the pre-defined correspondence between the processing party and the fault type. During the work order assignment process, an automatically generated explanatory text is attached, enabling the processing personnel receiving the work order to quickly understand the main information and common characteristics of the fault. This avoids duplicate assignment of work orders with duplicate faults, reducing the workload and information redundancy of the processing party. By merging work orders and attaching explanatory text, processing personnel can quickly locate the fault, understand the whole picture of the fault, speed up the fault handling process, and improve the overall operation and maintenance efficiency.

[0077] The method in this embodiment achieves automatic aggregation and interpretation of duplicate maintenance work orders through the above steps. Without manual intervention, it intelligently merges a large number of work orders describing the same fault into a single order, thereby reducing the workload of manual screening and dispatching. Simultaneously, by introducing the semantic understanding and generation capabilities of a large language model, it overcomes the limitations of traditional methods in handling semantic differences and lacking interpretation.

[0078] In one embodiment of the present invention, based on step S103, the following will provide a possible embodiment and its specific implementation will be described in a non-limiting manner. Step S103 specifically includes: Step S1031: Construct a fault type feature word mapping table. Based on the fault types and corresponding high-frequency feature words annotated in historical maintenance work orders, generate a domain knowledge constraint rule base.

[0079] Before calculating the similarity of feature vectors, the feature vectors are first filtered for knowledge: if the work order feature vector contains non-fault-related words, its weight in the similarity calculation is reduced; if it contains strong fault-related words, its weight is increased, ensuring that the similarity calculation focuses on the core fault features.

[0080] Step S1032: Adjust ε and MinPts based on the work order's business priority and historical dispatch timeliness data: For high-priority work orders, reduce ε and decrease MinPts to ensure that a small number of high-value work orders can also be clustered; For low-priority work orders, increase ε and improve MinPts to avoid invalid clustering.

[0081] The parameter adjustment formula is: ; The priority coefficient ranges from 0 to 1, with 1 for the production environment and 0.3 for the testing environment.

[0082] Define historical clustering quality metrics, such as the average silhouette coefficient. The noise point ratio (nr) serves as a feedback factor for parameter adjustment. If the current parameters... If the clustering quality is poor, or nr>0.3, meaning there is too much noise, then increase the parameter adjustment range.

[0083] like If the clustering quality is poor and nr≤0.2, which means the clustering quality is good, then reduce the adjustment range.

[0084] The adjusted formula is as follows:

[0085] Where k and m are adjustment coefficients. Optionally, k = 0.5, m = 0.3. .

[0086] To prevent parameters from deviating excessively from a reasonable range, upper and lower limit constraints are set: ε min =0.1, where ε is the minimum neighborhood radius. max =1.0, where is the maximum neighborhood radius.

[0087] MinPts min =2, MinPtsmax =20. If the calculated ε or MinPts exceeds the boundary, then the boundary value is taken.

[0088] Noisy points in the initial clustering labels are processed in the following way: Extract the feature vectors of noise points and compare them with the historical noise point database; If similar noise points are matched, they are classified into a noise cluster to be confirmed and a manual review process is triggered. If no match is found, the keywords in its feature vector are analyzed, and it is attempted to be associated with the nearest cluster. If the similarity threshold within the cluster is met after association, it is formally included in the cluster; otherwise, the noise label is maintained.

[0089] In one embodiment of the present invention, based on step S104, the following will provide a possible embodiment and its specific implementation will be described in a non-limiting manner. Step S104 specifically includes: Step S1041: In addition to the profile coefficient, define intra-cluster compactness and inter-cluster separation. Combined with the operation and maintenance scenario, define the fault type consistency index: statistically analyze the consistency of fault type labels of work orders within the cluster, and finally generate a comprehensive quality index. ; DB stands for Davies-Bouldin index, CH stands for Calinski-Harabasz index, and FT stands for Fault Type Consistency Score. These scores can measure the tightness, separation, and business relevance of clusters.

[0090] Step S1042: Using the comprehensive quality index Q as the objective function, the particle swarm optimization algorithm is used to adjust the neighborhood radius ε and the minimum number of points MinPts.

[0091] Each particle in the particle swarm represents a set of parameter combinations (εi, MinPtsi). By iteratively updating the particle's velocity and position, the search seeks the parameter combination that maximizes Q. The optimization formula is:

[0092] Among them, v i (t) is the particle velocity, x i (t) represents the parameter combination, ω represents the inertia weight, c1 and c2 are learning factors, r1 and r are random numbers, and p best,i For the individual optimal parameters, g best These are the globally optimal parameters.

[0093] Step S1043: To avoid clustering bias caused by changes in work order distribution over time, select work order data from the past 3 days, perform clustering, and calculate the comprehensive quality index Q. If the difference between the current clustering result and the mean Q of the historical data exceeds the threshold, it is marked as an unstable cluster, triggering the manual review process.

[0094] Step S1044: Generate dispatch strategy labels based on the cluster's comprehensive quality index Q and business characteristics.

[0095] In this embodiment, from an evaluation perspective, the construction of the comprehensive quality index Q encompasses the cluster density, separation, and fault type consistency, accurately reflecting the quality of the clustering results and avoiding misjudgments due to incomplete evaluation. The particle swarm optimization algorithm is used to automatically search for optimal clustering parameters, more efficiently finding the parameter combination that achieves the best clustering quality, while improving the accuracy and stability of clustering and adapting to work order data with different distributions.

[0096] To address data variation issues, unstable clustering caused by changes in work order distribution is promptly identified through comparison with historical data and threshold judgment. This triggers a manual review process, effectively avoiding clustering bias caused by data changes and ensuring the reliability of clustering results. In work order dispatching decisions, dispatching strategy labels are generated based on the comprehensive quality indicator Q and business characteristics. This allows the dispatching process to closely integrate the quality of clustering results with actual business needs, optimizing work order allocation strategies and improving work order processing efficiency.

[0097] In one embodiment of the present invention, based on step S152, the following will provide a possible embodiment and its specific implementation will be described in a non-limiting manner. Step S105 specifically includes: Step S1051: Construct a fault mode knowledge graph. Based on historical clustering results and expert annotations, structure entities and relationships such as fault types, symptoms, affected components, and associated error codes to form a knowledge network for the operation and maintenance domain.

[0098] Before generating explanatory text, the keyword groups of the current cluster are mapped to the fault mode knowledge graph to obtain more comprehensive contextual information.

[0099] Step S1052: Generate differentiated explanations for user needs at different levels: For front-line operations and maintenance personnel, generate technical explanations containing detailed repair suggestions.

[0100] Step S1053: Embed the chain of evidence for fault diagnosis into the explanatory text by analyzing the time series of work orders and the changing trends of system indicators within the cluster. For example: at time T1, a user reports a login failure; at time T2, the system log shows that the authentication service has restarted abnormally; at time T3, the database connection pool monitoring shows that the number of connections has reached the limit, suggesting that the cause is unreasonable database configuration parameters.

[0101] Step S1054: Generate a credibility score for the explanatory text by calculating multi-dimensional indicators such as semantic consistency of work orders within the cluster, logical coherence of the evidence chain, and matching degree with the fault mode knowledge graph.

[0102] For example, if more than 80% of the work orders within a cluster mention "500 error," and the knowledge graph clearly associates this error code with "database connection timeout," then the credibility score is 0.85. This score is used to guide subsequent work order dispatch decisions, and explanations with credibility scores below the threshold trigger a secondary verification process.

[0103] This embodiment constructs a fault mode knowledge graph, making full use of historical experience and domain knowledge to improve the depth and accuracy of explanatory text, enabling maintenance personnel to obtain fault information; it generates differentiated explanations for different user needs, improving the relevance and effectiveness of information transmission, helping front-line maintenance personnel to quickly locate problems and implement repairs; it provides a reliable basis for accurate work order dispatch and efficient processing, thereby improving overall maintenance efficiency and the scientific nature of fault handling.

[0104] It should be noted that step S1052 specifically includes: Step S10521: Collect the root cause analysis results of historical clustering, and construct a root cause association graph with fault type as nodes and root cause similarity as edge weights. For the multimodal feature vectors of the current cluster, match similar root cause nodes through a graph search algorithm to obtain the root cause patterns and related influencing factors of similar faults in the past, providing prior knowledge for causal inference.

[0105] Step S10522: Define the weighted causal inference mechanism. In the Bayesian network, adjust the conditional probability parameters of the root cause node based on the service priority and fault occurrence frequency of the work order. For high-priority or high-frequency faults, increase the weight coefficient of known key root causes; for low-frequency faults, decrease the default root cause weight and increase the probability of exploring rare factors.

[0106] Step S10523: Cross-validate the root cause hypothesis obtained from causal reasoning with the system log data of the work order (such as error stack information) and user feedback description.

[0107] The textual implication model is used to determine whether the user description supports the root cause hypothesis. If the log evidence conflicts with the inferred root cause, the Bayesian network parameters are readjusted for secondary inference.

[0108] Step S10524: Based on the root cause reasoning results, extract key features supporting the root cause hypothesis and generate an explanation chain. For example, present error code 503, insufficient server resources, and the unavailability of the payment interface in a structured way to help engineers quickly understand the fault logic.

[0109] In this embodiment, by constructing and searching a root cause correlation graph, historical root cause knowledge is combined with current fault analysis. This avoids repetitive reasoning and improves the efficiency of root cause analysis. It enables rapid and accurate root cause localization for high-frequency, high-priority faults, while maintaining an exploratory approach for low-frequency faults to avoid missed diagnoses, thus enhancing the generalization ability of reasoning. The generation of structured explanation chains transforms abstract root cause reasoning into intuitive logical chains, helping engineers quickly understand the nature and scope of the fault and shortening fault diagnosis time.

[0110] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0111] The following are embodiments of the clustering and aggregation repetitive operation and maintenance dispatch system provided in this disclosure. This system and the clustering and aggregation repetitive operation and maintenance dispatch methods in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the clustering and aggregation repetitive operation and maintenance dispatch system, please refer to the embodiments of the above clustering and aggregation repetitive operation and maintenance dispatch methods.

[0112] like Figure 2 As shown, the system includes: The preprocessing module 201 is used to obtain work orders and perform preprocessing. The feature extraction module 202 is used to extract work order features from the preprocessed work order, calculate the contribution score of each word to the work order fault through the attention mechanism, normalize to obtain the attention weight, and then generate the feature vector by weighted summation of the corresponding hidden vectors. Clustering module 203 is used to perform unsupervised clustering on feature vectors, calculate the cosine similarity of any two feature vectors, set the neighborhood radius threshold and the minimum number of points; identify core points, assign all points in the neighborhood of the core point to the initial cluster in which the core point is located; recursively check the neighborhood of the points in the neighborhood, if it contains unassigned points and meets the density condition, then continue to include them in the current cluster and expand the neighborhood range, and finally form a preliminary clustering result containing core points, density-connected points and recursively expanded points; The evaluation and optimization module 204, based on the preliminary clustering results, calculates the average silhouette coefficient of all feature vectors to measure the overall compactness and separation; optimizes the clustering parameters according to the evaluation results, and determines the final clustering results; The explanation generation module 205 analyzes the common features of the feature vectors within the clusters based on the final clustering results and generates explanatory text; The dispatch module 206 merges work orders belonging to the same cluster and dispatches them to the corresponding processing parties based on the explanatory text.

[0113] like Figure 3As shown, this application also provides an electronic device, including a display module 103, a memory 102, a processor 101, and a computer program stored in the memory and executable on the processor 101. When the processor 101 executes the program, it implements the steps of a clustering and aggregation method for repetitive maintenance dispatching.

[0114] In embodiments of the present invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments described and / or claimed herein.

[0115] In this embodiment, processor 101 may be implemented using at least one of an application-specific integrated circuit, a programmable logic device, a field-programmable gate array, a processor, a controller, a microcontroller, a microprocessor, or an electronic unit designed to perform the functions described herein. In some cases, such an implementation may be implemented within a controller. For software implementation, implementations such as processes or functions may be implemented with separate software modules that allow the performance of at least one function or operation. Software code may be implemented by a software application (or program) written in any suitable programming language, and the software code may be stored in memory and executed by the controller.

[0116] The display module 103 is used to display information input by the user or information provided to the user. The display module 103 may include a display panel, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like.

[0117] The memory 102 can be used to store software programs and various data. The memory 102 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0118] This application also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the clustering and aggregation repetitive operation and maintenance dispatching method.

[0119] The storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0120] In a storage medium, a readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0121] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A clustering and aggregation method for repetitive operation and maintenance dispatching, characterized in that, The methods include: Step S101: Obtain the work order and perform preprocessing; Step S102: Extract work order features from the preprocessed work order, calculate the contribution score of each word to the work order fault through the attention mechanism, normalize to obtain the attention weight, and then generate the feature vector by weighted summation of the corresponding hidden vectors. Step S103: Perform unsupervised clustering on the feature vectors, calculate the cosine similarity of any two feature vectors, set the neighborhood radius threshold and the minimum number of points; identify the core point, and assign all points in the neighborhood of the core point to the initial cluster in which the core point is located; recursively check the neighborhood of the points in the neighborhood, and if it contains unassigned points and meets the density condition, then continue to include them in the current cluster and expand the neighborhood range, finally forming a preliminary clustering result containing the core point, density-connected points and recursively expanded points; Step S104: Based on the preliminary clustering results, calculate the average silhouette coefficient of all feature vectors to measure the overall compactness and separation by using silhouette coefficient evaluation; optimize the clustering parameters according to the evaluation results and determine the final clustering results; Step S105: Analyze the common features of the feature vectors within the clusters based on the final clustering results, and generate explanatory text; Step S106: Based on the explanation text, merge the work orders of the same cluster and send them to the corresponding processing party.

2. The clustering and aggregation method for repetitive maintenance dispatching according to claim 1, characterized in that, Step S102 specifically includes: inputting the preprocessed work order into the LLM large-scale language model, and defining the normalized text sequence of the work order as... Where n is the number of words, and the hidden vector representation h of each word is obtained through LLM encoding. i , where h i It is a semantic vector of dimension d; The importance score e for each word is calculated using a self-attention mechanism. i Introduce inter-cluster separation constraints; if word t i In historical clustering, terms with high discriminative power are assigned an enhancement factor λ (λ>1) to their weights; conversely, terms with low discriminative power are assigned an inhibition factor μ. ; Where, δ i For 0-1 indicator variables, if t i If it is a highly discriminative term, then δ i =1, otherwise δ i =0; The importance scores are normalized using softmax to obtain the attention weights. ; in, ∈(0,1), ; The hidden vectors are weighted and summed using attention weights to obtain the feature vector v of the work order. ; The feature vector v is mapped to a low-dimensional subspace of the clustering space through linear projection, and the projection formula is as follows: Where W has a dimension of db×hb, where db is the cluster space dimension and hb is the LLM hidden layer dimension.

3. The clustering and aggregation method for repetitive maintenance dispatching according to claim 1, characterized in that, Step S103, which involves unsupervised clustering of the feature vectors, specifically includes: Cosine similarity is used to calculate the feature vectors v of any two work orders. i and v j The similarity is calculated using the following formula: Define distance metric ; Define the neighborhood radius threshold ε and the minimum number of points MinPts; For any work order feature vector v p ε-neighborhood , is the distance v p The set of all eigenvectors within ε; If the number of points in this neighborhood Then v p Considered the core point; All points within the ε-neighborhood of the core point are assigned to the same cluster. Starting from the core point, its neighborhood is recursively checked, and points with density connections are continuously merged into the same cluster. If the number of points in the ε-neighborhood of a certain feature vector is less than MinPts, that is, there are no enough neighbors, then it is marked as a noise point; For each sample i of the feature vector, the formula for calculating the silhouette coefficient s(i) is as follows: ; Where a(i) represents the average distance between sample i and other samples in its cluster; b(i) represents the average distance between sample i and its nearest other cluster; By calculating the average silhouette coefficient of all eigenvectors Pm represents the total number of work orders and can be used to measure the overall quality of clustering.

4. The clustering and aggregation method for repetitive maintenance dispatching according to claim 3, characterized in that, Step S103 also includes: Construct a fault type feature word mapping table, and generate a domain knowledge constraint rule base based on the fault types and corresponding high-frequency feature words marked in historical operation and maintenance work orders; Adjust ε and MinPts based on the work order's business priority and historical dispatch timeliness data: The parameter adjustment formula is: ; ； Define historical clustering quality metrics, if the current parameters are... If the clustering quality is poor, or nr > 0.3, then increase the parameter adjustment range; like If the clustering quality is poor and nr ≤ 0.2, then reduce the adjustment range. The formula is adjusted as follows: Where k and m are adjustment coefficients.

5. The clustering and aggregation method for repetitive maintenance dispatching according to claim 1, characterized in that, Step S104 specifically includes: Define intra-cluster compactness and inter-cluster separation, and define a fault type consistency index based on the operation and maintenance scenario: statistically analyze the consistency of fault type labels of work orders within a cluster, and finally generate a comprehensive quality index. ; DB stands for Davies-Bouldin index, CH stands for Calinski-Harabasz index, and FT stands for fault type consistency score. Using the comprehensive quality index Q as the objective function, the particle swarm optimization algorithm is used to adjust the neighborhood radius ε and the minimum number of points MinPts. Each particle in the particle swarm represents a set of parameter combinations. By iteratively updating the particle's velocity and position, the search seeks the parameter combination that maximizes Q. The optimization formula is: Among them, v i (t) is the particle velocity, x i (t) represents the parameter combination, ω represents the inertia weight, c1 and c2 are learning factors, r1 and r are random numbers, and p best,i For the individual optimal parameters, g best These are the globally optimal parameters.

6. The clustering and aggregation method for repetitive maintenance dispatching according to claim 1, characterized in that, Step S105 specifically includes: Construct a failure mode knowledge graph; Before generating explanatory text, the keyword groups of the current cluster are mapped to the fault mode knowledge graph to obtain contextual information; Generate differentiated explanations based on the needs of users at different levels; By embedding the chain of evidence for fault diagnosis into the explanatory text, and by analyzing the time series of work orders and the changing trends of system indicators within the cluster, we can better understand the process.

7. The clustering and aggregation method for repetitive maintenance dispatching according to claim 6, characterized in that, Step S105 also includes: Collect the root cause analysis results of historical clustering, and construct a root cause association graph with fault type as node and root cause similarity as edge weight; In Bayesian networks, the conditional probability parameters of root cause nodes are adjusted based on the service priority and failure frequency of work orders. The root cause hypothesis obtained from causal reasoning is cross-validated with the system log data of the work order and the user feedback description. If the log evidence conflicts with the root cause of the inference, the Bayesian network parameters are readjusted for a second inference. Based on the results of root cause reasoning, key features supporting the root cause hypothesis are extracted to generate an explanatory chain.

8. A clustering and aggregation-based repetitive operation and maintenance dispatch system, characterized in that, The system is used to implement the clustering and aggregation repetitive operation and maintenance dispatching method as described in any one of claims 1 to 7; The system includes: The preprocessing module is used to obtain work orders and perform preprocessing. The feature extraction module is used to extract work order features from the preprocessed work orders. It calculates the contribution score of each word to the work order fault through the attention mechanism, normalizes it to obtain the attention weight, and then generates the feature vector by weighted summation of the corresponding hidden vectors. The clustering module is used to perform unsupervised clustering of feature vectors, calculate the cosine similarity of any two feature vectors, set the neighborhood radius threshold and the minimum number of points; identify core points, and assign all points in the neighborhood of the core point to the initial cluster in which the core point is located; recursively check the neighborhood of the points in the neighborhood, and if it contains unassigned points and meets the density condition, it continues to include them in the current cluster and expands the neighborhood range, finally forming a preliminary clustering result containing core points, density-connected points and recursively expanded points; The evaluation and optimization module, based on the preliminary clustering results, calculates the average silhouette coefficient of all feature vectors to measure the overall compactness and separation; and optimizes the clustering parameters according to the evaluation results to determine the final clustering results. The explanation generation module analyzes the common features of feature vectors within clusters based on the final clustering results and generates explanatory text. The dispatch module merges work orders belonging to the same cluster and dispatches them to the corresponding processing parties based on the explanatory text.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the clustering aggregation repetitive operation and maintenance dispatching method as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the clustering and aggregation repetitive operation and maintenance dispatching method as described in any one of claims 1 to 7.

Images