CN-121981453-A - Intelligent service work order processing method and system based on multi-mode data fusion and dynamic self-adaptive scheduling

CN121981453ACN 121981453 ACN121981453 ACN 121981453ACN-121981453-A

Abstract

The invention provides an intelligent service work order processing method and system based on multi-mode data fusion and dynamic self-adaptive scheduling, which are characterized in that through collecting work order texts, user operation logs, equipment monitoring data, contract databases and historical work order records, word segmentation denoising and entity identification are carried out on the texts, and key events and index sequences are extracted from the logs; the method comprises the steps of extracting semantic features and time sequence features by using text branches and log branches, dynamically weighting and fusing a main class and subclasses of a work order through a attention mechanism, identifying the intention of the work order based on a rule template and semantic similarity, calculating and dynamically adjusting priority by combining client level, SLA residual time, real-time load and waiting time, dispatching to a target team according to a nearby and load balancing strategy, starting cross-regional backup and synchronizing the collaboration state through a message queue if necessary, collecting solution time and evaluation data, updating a model according to the period, and supporting quick adaptation of few samples to new service scenes.

Inventors

HU XU
LI MING
WANG YANPING
LIU YILI
LIU YUNFENG

Assignees

紫光云技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (10)

1. An intelligent service work order processing method based on multi-mode data fusion and dynamic self-adaptive scheduling is characterized by comprising the following steps: Acquiring a work order processing request, and acquiring work order text, a user operation log, equipment monitoring data, contract database information and historical work order records associated with the work order processing request from a data source, wherein the contract database information at least comprises client grade information and clause information related to service grade agreement SLA; performing word segmentation and denoising on the work order text and performing entity recognition; extracting key operation events from the user operation log to obtain an operation event sequence and/or a behavior index obtained by statistics of the operation event sequence; Obtaining semantic features through a text feature extraction model based on the work order text, obtaining time sequence features through a time sequence feature extraction model based on the user operation log and/or the equipment monitoring data, and dynamically weighting and fusing the semantic features and the time sequence features through an attention mechanism to obtain fusion features; outputting classification results of the main class and the sub class of the worksheet based on the fusion characteristics; Performing intention recognition on the work order text based on the rule template and the semantic similarity calculation to obtain one or more candidate intentions and confidence degrees thereof; Determining a client grade and calculating SLA residual time based on the contract database information, calculating a work order priority grade according to the business category to which the work order belongs, the real-time system load and the work order waiting time, and dynamically adjusting the priority grade according to a time sensitive factor related to the work order waiting time; Generating a scheduling decision and distributing a work order or subtasks thereof to a target processing team for execution according to the adjusted priority score, the real-time availability of the processing resources and the cross-department or cross-region load state; and collecting work order processing result data and forming feedback data for updating the classification and scheduling strategies.
2. The method of claim 1, wherein the text feature extraction model is an encoder based on a pre-trained language model for outputting upper and lower Wen Yuyi vectors of a work order text, the time sequence feature extraction model is a cyclic neural network or a long and short term memory network for outputting time sequence pattern vectors of device monitoring data, and the attention mechanism is used for assigning weights to semantic features and time sequence features so as to adapt importance differences of text information and log information in different scenes.
3. The method of claim 1, wherein the intent recognition comprises: obtaining a first candidate intention set based on rule template matching; Encoding the work order text to obtain a text semantic vector, and encoding a preset intention description text and/or a historical work order text to obtain a corresponding intention semantic vector; Calculating semantic similarity between the text semantic vector and each intention semantic vector, screening and/or selecting first K corresponding intentions from high to low according to a similarity threshold value to obtain a second candidate intention set; Combining and de-duplicating the first candidate intention set and the second candidate intention set, and outputting the candidate intention and the confidence level thereof.
4. The method of claim 1, wherein the worksheet priority score comprises a base priority score and a time-sensitive factor-based adjusted target priority score; The basic priority score is obtained by weighting the work order type or business category score, the client grade score, the SLA residual time score and the real-time system load score according to corresponding weights; And the time sensitive factor is determined according to the work order waiting time length and a preset threshold value and is used for carrying out additive or multiplicative adjustment on the basic priority grade so as to obtain the target priority grade.
5. The method of claim 1, wherein the time sensitive factor is such that when the work order waiting time exceeds a predetermined threshold, the work order priority score is incremented by a predetermined rate of rise over a time period to reduce the probability of long-term retention of the low priority work order.
6. The method of claim 1, wherein the scheduling decision comprises: Determining a target service region based on a region entity obtained by the identification of the work order text entity and/or a fault resource region identification determined based on equipment monitoring data; selecting a processing team in the same area as the target service region from candidate processing teams meeting processing qualification as a local team; when a plurality of local teams exist, selecting a processing team with the minimum network delay and/or the lowest real-time load; when the real-time load of the local team exceeds a preset threshold value or the existence of processing overrun risk in the residual time of the SLA is predicted, at least one trans-regional backup team is automatically activated to enter a standby or take over state, and a work order or subtasks thereof are distributed to the trans-regional backup team.
7. The method of claim 1, wherein the assigning the work order or the subtasks thereof to the target processing team for execution comprises splitting the same work order into at least two parallel subtasks and assigning the parallel subtasks to different team for execution, and synchronizing the work order status, the subtask progress and the collaboration information through a message queue to achieve cross-team parallel collaboration processing.
8. The method of claim 1, wherein the feedback data includes at least a work order solution time, a user score, and a processor rating, wherein the classification model and the scheduling parameters are incrementally updated at predetermined periods based on the feedback data, and wherein historical knowledge is maintained during model updates to reduce risk of forgetting.
9. The method of claim 1, wherein when a new business scenario or a new job ticket category occurs, a small sample study is performed based on a small number of annotation samples to quickly generate classification labels and parameter configurations for the corresponding category and incorporate the new category into a subsequent closed loop update flow.
10. The intelligent service work order processing system based on multi-mode data fusion and dynamic self-adaptive scheduling is characterized by comprising a multi-mode data acquisition and preprocessing module, a closed-loop optimization and increment learning module, a dynamic priority and self-adaptive scheduling module, a closed-loop optimization and increment learning module and a small sample quick adaptation module, wherein the multi-mode data acquisition and preprocessing module is used for acquiring work order texts, user operation logs, equipment monitoring data, contract database information and historical work order records, performing text preprocessing, entity recognition and log key event extraction, the intelligent classification and intention recognition module is used for respectively extracting work order text semantic features and equipment monitoring data time sequence features, performing feature fusion based on an attention mechanism to output work order classification results, and performing intention recognition based on rule templates and semantic similarity, the dynamic priority and self-adaptive scheduling module is used for calculating and dynamically adjusting priority scores based on client grades, SLA residual time, real-time system loads and waiting time, and generating scheduling decisions by combining cross-department or cross-region resource loads, and the closed-loop optimization and increment learning module is used for collecting processing results and user evaluation to form feedback data, and performing periodic increment updating and small sample quick adaptation on a new increment scene according to the classification model and scheduling strategies.

Description

Intelligent service work order processing method and system based on multi-mode data fusion and dynamic self-adaptive scheduling Technical Field The invention belongs to the field of information technology service management and intelligent operation and maintenance, and particularly relates to an intelligent service work order processing method and system based on multi-mode data fusion and dynamic self-adaptive scheduling. Background In the enterprise IT service and cloud service operation and maintenance scenarios, service worksheets are typically used to carry matters such as fault repair, configuration consultation, project delivery, after-sales support, and the like. The existing work order processing system generally adopts keyword matching of a predefined rule base to classify work orders, customer service personnel carry out manual assignment according to titles or contents, priorities are statically set according to fixed rules, meanwhile, the work orders are often linearly circulated according to fixed processes of submission, classification, distribution, processing and closing, and the work order system is lack of deep association with data such as user behavior data, equipment monitoring logs, contract databases and the like, and decisions mainly depend on work order text information. The method is easy to generate the following problems in practical application, namely firstly, keyword matching is easy to be influenced by expression diversity, so that classification accuracy is insufficient, composite work orders related to multiple matters are difficult to identify, secondly, manual assignment depends on experience, processing efficiency is limited by manpower, response delay is easy to occur in peak period, and complex work orders can be repeatedly transferred to cause cycle extension, thirdly, static priority rules are difficult to be dynamically adjusted by combining real-time system loads and emergencies, and non-emergency matters occupy key fault processing resources possibly occur, fourthly, processing team loads are unbalanced, cross-region coordination capability is insufficient, reasonable distribution according to geographic positions and resource states is difficult to be performed, fifthly, a system lacks continuous optimization capability, a manual adjustment rule base is relied, and service expansion or new service scenes are difficult to adapt. Disclosure of Invention In view of the above, the present invention is directed to an intelligent service work order processing method and system based on multi-mode data fusion and dynamic adaptive scheduling, so as to solve at least one problem in the background art. In order to achieve the above purpose, the technical scheme of the invention is realized as follows: An intelligent service work order processing method based on multi-mode data fusion and dynamic self-adaptive scheduling comprises the following steps: Acquiring a work order processing request, and acquiring work order text, a user operation log, equipment monitoring data, contract database information and historical work order records associated with the work order processing request from a data source, wherein the contract database information at least comprises client grade information and clause information related to service grade agreement SLA; performing word segmentation and denoising on the work order text and performing entity recognition; extracting key operation events from the user operation log to obtain an operation event sequence and/or a behavior index obtained by statistics of the operation event sequence; Obtaining semantic features through a text feature extraction model based on the work order text, obtaining time sequence features through a time sequence feature extraction model based on the user operation log and/or the equipment monitoring data, and dynamically weighting and fusing the semantic features and the time sequence features through an attention mechanism to obtain fusion features; outputting classification results of the main class and the sub class of the worksheet based on the fusion characteristics; Performing intention recognition on the work order text based on the rule template and the semantic similarity calculation to obtain one or more candidate intentions and confidence degrees thereof; Determining a client grade and calculating SLA residual time based on the contract database information, calculating a work order priority grade according to the business category to which the work order belongs, the real-time system load and the work order waiting time, and dynamically adjusting the priority grade according to a time sensitive factor related to the work order waiting time; Generating a scheduling decision and distributing a work order or subtasks thereof to a target processing team for execution according to the adjusted priority score, the real-time availability of the processing resources and the cross-department or cross-regi