Terminal equipment operation and maintenance control method, device and equipment based on retrieval enhancement

Abstract

The embodiment of the invention discloses a terminal equipment operation and maintenance control method, device and equipment based on retrieval enhancement. A specific embodiment of the method comprises the following steps: carrying out element segmentation extraction processing on operation and maintenance change information to obtain a to-be-changed terminal identifier set and change element information; performing constraint alignment processing on the to-be-changed terminal identifier set and the terminal attribute information set to obtain a target terminal information set; retrieval query information of the operation and maintenance change information is obtained; performing dual-channel fusion recall processing on the retrieval query information to obtain a candidate operation and maintenance change sample set; performing multi-layer screening enhancement processing on the candidate operation and maintenance change sample set to obtain a target operation and maintenance change sample set, and generating operation and maintenance change detection information; and generating a terminal operation and maintenance control instruction, and performing operation and maintenance control on the terminal equipment set corresponding to the target terminal information set. According to the embodiment, the change implementation time and the operation power consumption of the terminal equipment during operation and maintenance change can be reduced, and the stability of the terminal equipment is improved.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a method, apparatus, and device for operation and maintenance control of terminal devices based on retrieval enhancement. Background Technology

[0002] The retrieval-enhanced terminal device operation and maintenance control method refers to a method that utilizes knowledge retrieval technology to analyze received operation and maintenance change information, and then performs operation and maintenance control (e.g., restarting or fine-tuning terminal device configurations) on the terminal devices involved in the changes based on the analysis results. For terminal device operation and maintenance control, the typical approach is to match the operation and maintenance change information with preset rules based on a pre-defined rule engine, and output corresponding control commands when the matching results meet preset conditions, thereby executing the corresponding control operations on the terminal devices.

[0003] In practice, it has been found that when using the above methods to perform operation and maintenance control on terminal devices, the following technical problem often occurs: Due to the diverse sources and inconsistent descriptions of operation and maintenance change information, the method based on the preset rule engine is difficult to cover the combination of different terminal models, version differences, and complex scenarios. This results in low accuracy of rule matching results, which is prone to misjudgment or omission. This leads to inaccurate operation and maintenance control operations on terminal devices (e.g., restarting a terminal device that should not be restarted). This causes the terminal devices to experience unexpected abnormal operating states (e.g., service process interruption). Additional rollback operations or secondary change operations are required, which prolongs the implementation time of operation and maintenance changes on terminal devices, increases the operating power consumption of terminal devices, and reduces the stability of terminal devices.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the present disclosure concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a method, apparatus, and device for operation and maintenance control of terminal devices based on retrieval enhancement, in order to solve one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a terminal device operation and maintenance control method based on retrieval enhancement, comprising: in response to receiving operation and maintenance change information for a set of terminal devices, obtaining a set of terminal attribute information at the current time of receiving the operation and maintenance change information; performing element segmentation extraction processing on the operation and maintenance change information to obtain a set of terminal identifiers to be changed and change element information; performing constraint alignment processing on the set of terminal identifiers to be changed and the set of terminal attribute information based on the operation and maintenance change information to obtain a target terminal information set; and performing retrieval-based construction processing on the change element information and the target terminal information set to obtain a retrieval of the operation and maintenance change information. The system retrieves information; based on a pre-defined maintenance retrieval database, it performs dual-channel fusion recall processing on the retrieved information to obtain a candidate maintenance change sample set; it then performs multi-layer filtering and enhancement processing on the candidate maintenance change sample set to obtain a target maintenance change sample set; and based on the change element information, the target terminal information set, and the target maintenance change sample set, it generates maintenance change detection information; in response to the maintenance change detection information meeting pre-defined control conditions, it generates terminal maintenance control instructions based on the maintenance change detection information, and performs maintenance control on the terminal device set corresponding to the target terminal information set based on the terminal maintenance control instructions.

[0008] Secondly, some embodiments of this disclosure provide a terminal device operation and maintenance control device based on retrieval enhancement, comprising: an acquisition unit configured to acquire a set of terminal attribute information at the current time of receiving the operation and maintenance change information for a set of terminal devices in response to receiving operation and maintenance change information; an extraction unit configured to perform element segmentation extraction processing on the operation and maintenance change information to obtain a set of terminal identifiers to be changed and change element information; an alignment unit configured to perform constraint alignment processing on the set of terminal identifiers to be changed and the set of terminal attribute information based on the operation and maintenance change information to obtain a target terminal information set; and a construction unit configured to perform retrieval-based construction processing on the change element information and the target terminal information set to obtain the operation and maintenance change information. The system includes: an information retrieval and query unit; a recall unit configured to perform dual-channel fusion recall processing on the aforementioned retrieval and query information based on a preset maintenance retrieval database to obtain a candidate maintenance change sample set; an enhancement unit configured to perform multi-level filtering and enhancement processing on the aforementioned candidate maintenance change sample set to obtain a target maintenance change sample set, and to generate maintenance change detection information based on the aforementioned change element information, the aforementioned target terminal information set, and the aforementioned target maintenance change sample set; and a control unit configured to, in response to the aforementioned maintenance change detection information satisfying preset control conditions, generate a terminal maintenance control instruction based on the aforementioned maintenance change detection information, and perform maintenance control on the terminal device set corresponding to the aforementioned target terminal information set based on the aforementioned terminal maintenance control instruction.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0011] The above embodiments of this disclosure have the following beneficial effects: The terminal device operation and maintenance control method based on retrieval enhancement in some embodiments of this disclosure can reduce the implementation time and power consumption of terminal devices during operation and maintenance changes, and improve the stability of terminal devices. Specifically, the reasons for the extended implementation time, increased power consumption, and reduced stability of terminal devices during operation and maintenance changes are as follows: Due to the diverse sources and inconsistent expressions of operation and maintenance change information, the method based on the preset rule engine is difficult to cover the combination changes of different terminal models, versions, and complex scenarios, resulting in low accuracy of rule matching results and easy misjudgment or omission. This leads to inaccurate operation and maintenance control operations on terminal devices, causing unexpected abnormal operating states of terminal devices, requiring additional rollback operations or secondary change operations, thus extending the implementation time, increasing power consumption, and reducing stability of terminal devices during operation and maintenance changes. Based on this, the terminal device operation and maintenance control method based on retrieval enhancement in some embodiments of this disclosure can first, in response to receiving operation and maintenance change information for a set of terminal devices, obtain the terminal attribute information set at the current moment of receiving the aforementioned operation and maintenance change information. Here, the terminal attribute information set at the current moment is obtained only when maintenance change information is received, ensuring that the obtained terminal attribute information is the real-time status information at the current moment and avoiding deviations in subsequent constraint alignment processing due to outdated terminal attribute information. Secondly, the aforementioned maintenance change information undergoes element segmentation extraction processing to obtain the terminal identifier set to be changed and the change element information. Here, the inconsistently expressed maintenance change information is transformed into a structured terminal identifier set to be changed and change element information, reducing noise caused by differences in the format and ambiguity of the original maintenance change information, and providing standardized input for subsequent constraint alignment processing and retrieval-based construction processing. Thirdly, based on the aforementioned maintenance change information, constraint alignment processing is performed on the aforementioned terminal identifier set to be changed and the aforementioned terminal attribute information set to obtain the target terminal information set. Here, the target terminal information set can reflect the status information of the target terminal device at the current moment and within a preset time window, ensuring that the subsequently generated retrieval query information matches the maintenance change detection information and the actual status of the target terminal device, reducing the possibility of inaccurate maintenance control commands due to inconsistent terminal information. Next, the aforementioned changed element information and target terminal information set are processed using a retrieval-based construction method to obtain retrieval query information for the aforementioned operation and maintenance change information. Here, the obtained retrieval query information can simultaneously cover the semantic and keyword-level retrieval needs of operation and maintenance change information, providing accurate retrieval input for subsequent dual-channel fusion recall processing and improving the recall accuracy of the candidate operation and maintenance change sample set. Subsequently, based on a pre-set operation and maintenance retrieval database, the aforementioned retrieval query information is processed using a dual-channel fusion recall method to obtain the candidate operation and maintenance change sample set.Here, dual-channel fusion recall processing, compared to single-channel recall, reduces missed recalls and provides a more comprehensive candidate maintenance change sample set for subsequent multi-layer screening and enhancement processing. Then, multi-layer screening and enhancement processing is applied to the candidate maintenance change sample set to obtain the target maintenance change sample set. Maintenance change detection information is generated based on the aforementioned change element information, the target terminal information set, and the target maintenance change sample set. This multi-layer screening and enhancement processing filters out candidate maintenance change samples with semantic redundancy and contradictions. Finally, in response to the maintenance change detection information meeting preset control conditions, terminal maintenance control instructions are generated based on the maintenance change detection information, and maintenance control is performed on the terminal device set corresponding to the target terminal information set based on these instructions. By generating and executing terminal maintenance control instructions only when the maintenance change detection information meets preset control conditions, inaccurate maintenance control operations on terminal devices are avoided, reducing unexpected abnormal operating states of terminal devices and improving terminal device stability. Therefore, this retrieval-enhanced terminal device operation and maintenance control method can reduce the implementation time and power consumption of terminal devices during operation and maintenance changes, and improve the stability of terminal devices. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart of some embodiments of the retrieval-enhanced terminal device operation and maintenance control method according to this disclosure; Figure 2 This is a schematic diagram of the structure of some embodiments of the terminal device operation and maintenance control device based on retrieval enhancement according to the present disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1 A flowchart 100 is shown, illustrating some embodiments of the retrieval-enhanced terminal device operation and maintenance control method according to this disclosure. This retrieval-enhanced terminal device operation and maintenance control method includes the following steps: Step 101: In response to receiving maintenance change information for the set of terminal devices, obtain the set of terminal attribute information at the current time of receiving the maintenance change information.

[0021] In some embodiments, the aforementioned execution entity may, in response to receiving maintenance change information for a set of terminal devices, obtain a set of terminal attribute information at the current moment of receiving the maintenance change information. The maintenance change information may be a structured data packet used to request a status change for terminal devices (e.g., switches, routers, network terminal devices, ATMs, etc.). The maintenance change information may include, but is not limited to, at least one of the following: change description text, a set of attached documents (e.g., the attached documents in the attached document set may include: technical solution documents, implementation step documents, rollback plan documents, etc.), change impact scope information (e.g., documents of upstream and downstream systems or terminals affected by the maintenance change), change constraint information, etc. The change description text may be a summary text describing the maintenance change information. The change description text may include: maintenance change type, maintenance change implementation time window, and terminal devices involved in the change, etc. The terminal attribute information in the terminal attribute information set may be information synchronized through a CMDB (Configuration Management Database) that records the current hardware and software status and business context of the terminal devices. The aforementioned terminal attribute information may include, but is not limited to, at least one of the following: terminal device identifier, attribution information, current operating status information, list of upstream terminals it depends on, list of downstream systems that depend on the aforementioned terminal device, allowed maintenance time window, physical location, etc. The aforementioned change constraint information may be information from a preset maintenance change record database. The aforementioned maintenance change record database may be a pre-defined text database that stores records of terminal status changes to be executed. The aforementioned records of terminal status changes to be executed may be text databases corresponding to historical maintenance change information, containing current terminal status change information (e.g., ATM machine number 1 will be restarted from 22:00 to 23:00 on March 15th). The aforementioned terminals to be executed may be terminal devices whose status will change within a preset time window.

[0022] Step 102: Perform element segmentation and extraction processing on the operation and maintenance change information to obtain the terminal identifier set to be changed and the change element information.

[0023] In some embodiments, the aforementioned executing entity may perform element segmentation extraction processing on the aforementioned operation and maintenance change information to obtain a set of terminal identifiers to be changed and change element information. The terminal identifiers to be changed in the aforementioned set of terminal identifiers to be changed may be identifiers of the terminal devices involved in the aforementioned operation and maintenance change information. The change element information may be structured text reflecting the aforementioned operation and maintenance change information. The change element information may include a set of change element field information and a change segment sequence. The change element field information in the aforementioned set of change element field information may be text containing key elements of the terminal device to be changed, included in the operation and maintenance change information. The key elements may include, but are not limited to, at least one of the following: terminal change implementation time window (e.g., planned start time, planned end time), terminal change type, terminal operation instructions (e.g., system commands that the terminal device actually needs to execute), pre-check items (e.g., system status indicators that must be met before change implementation), and implementation steps. The change segment information in the aforementioned change segment sequence may be text blocks in the operation and maintenance change information that retain contextual information.

[0024] In some optional implementations of certain embodiments, the above-mentioned element segmentation extraction process for the operation and maintenance change information to obtain the terminal identifier set to be changed and the change element information may include the following steps: The first step is to transcribe the set of attachment documents included in the aforementioned maintenance change information into a text set. This text set can be plain text information after removing non-text elements such as formatting, styles, and layout from the original attachment documents. In practice, the executing entity can use a text parsing tool to transcribe the set of attachment documents included in the maintenance change information into a text set. This text parsing tool can include, but is not limited to, at least one of the following: the Apache POI library and the Tesseract OCR engine.

[0025] The second step involves performing terminal entity recognition processing on the aforementioned attachment text set and the change description text included in the aforementioned operation and maintenance change information to obtain a set of terminal identifiers to be changed. In practice, the aforementioned executing entity can input the aforementioned target attachment text and the aforementioned change description text into a trained terminal identification model to obtain a set of terminal identifiers to be changed. The aforementioned target attachment text can be an attachment text in the aforementioned attachment text set that relates to the aforementioned change impact scope information. The aforementioned terminal identification model can be a model that performs named entity recognition processing on the input target attachment text and change description text to output a set of terminal identifiers to be changed. The aforementioned terminal identification model can be a model composed of a feature extraction model and a sequence decoding model concatenated. For example, the aforementioned feature extraction model can be an IDCNN (Iterated Dilated Convolutional Neural Network) model, and the aforementioned sequence decoding model can be a CRF (Conditional Random Field) model.

[0026] The third step involves identifying the element fields of the aforementioned attachment text set and change description text to obtain a set of changed element field information. In practice, the executing entity can input the aforementioned attachment text set, change description text, and preset prompt text into a trained element field extraction model to obtain the set of changed element field information. The preset prompt text can be a pre-defined structured text template used to guide the model in generating the set of changed element field information. The aforementioned element field extraction model can be a UIE (Unified Information Extraction) model.

[0027] The fourth step involves semantic segmentation of the aforementioned attachment text set and the change description text included in the aforementioned operation and maintenance change information to obtain a change segment sequence. In practice, the aforementioned executing entity can utilize the TextTiling algorithm to perform semantic segmentation of the aforementioned attachment text set and the change description text included in the aforementioned operation and maintenance change information to obtain a change segment sequence.

[0028] The fifth step is to determine the above set of changed element field information and the above sequence of changed segments as changed element information.

[0029] In addressing the technical problems mentioned above, the application scenario—where maintenance change information contains high-risk operation instructions (e.g., database maintenance change scenarios involving irreversible data changes)—often presents the following technical problem: Text segmentation methods based on semantic similarity cannot perceive the semantic boundaries of high-risk operation instructions within maintenance change information. This leads to high-risk operation instructions and their associated contextual information being split into different segments or mixed with irrelevant content within the same segment. Consequently, subsequent retrieval and detection based on the change segment sequence cannot accurately identify the risk information corresponding to the high-risk operation instructions, resulting in inaccurate generated maintenance change detection information. This leads to inaccurate maintenance control operations on terminal devices, extending the implementation time of maintenance changes, increasing power consumption, and reducing stability. Considering the following requirements for this application scenario: maintenance change information containing high-risk operation instructions, contextual integrity of high-risk instructions, and cross-segment contextual continuity of segmented information, we have decided to adopt the following solution: Optionally, the above-mentioned element segmentation and extraction processing of the above-mentioned operation and maintenance change information to obtain the terminal identifier set to be changed and the change element information, and the operation and maintenance control of the terminal device set corresponding to the above-mentioned target terminal information set based on the above-mentioned terminal identifier set to be changed and the change element information, may include the following steps: The first step is to generate an attachment text set, a set of terminal identifiers to be changed, and a set of changed element field information based on the aforementioned maintenance change information. As an example, the implementation of this step can refer to the implementation methods of steps one through three in some optional implementation methods of step 102, and will not be elaborated upon here.

[0030] The second step is to concatenate each attachment text in the aforementioned attachment text set with the change description text included in the aforementioned operation and maintenance change information to obtain the text set to be segmented. The text to be segmented in the aforementioned text set can be a continuous text formed by concatenating the attachment text and the change description text end-to-end.

[0031] The third step involves performing sentence boundary segmentation on the aforementioned text set to be segmented, resulting in a modified sentence sequence set. In practice, the execution entity can utilize the Punkt algorithm to perform sentence boundary detection on each text in the aforementioned text set to be segmented, using sentence break marks (e.g., period, question mark, exclamation mark) as segmentation boundaries to segment the text into a sequence of single-sentence texts arranged in the original order, thus obtaining the modified sentence sequence set.

[0032] The fourth step involves performing segmentation constraint recognition processing on the aforementioned set of modified sentence sequences to obtain a target segmentation constraint information set. The target segmentation constraint information in this set can be the positional information of characters in the modified sentence sequences that need to serve as mandatory segmentation boundaries. In practice, the executing entity can first perform high-risk instruction matching on each single sentence text of each modified sentence sequence in the aforementioned set of modified sentence sequences against a preset high-risk instruction vocabulary list to obtain a high-risk instruction sentence set and a constraint trigger word set. The preset high-risk instruction vocabulary list can be a pre-defined vocabulary list containing keywords of high-risk operation instructions in maintenance scenarios (e.g., "rm -rf", "drop table", "fdisk", "shutdown", "kill -9", etc.). The high-risk instruction sentences in the high-risk instruction sentence set can be modified sentences containing high-risk operation instructions. The constraint trigger words in the constraint trigger word set can be high-risk instruction keywords matched by the high-risk instruction sentences. Then, the position number of each high-risk instruction sentence in the aforementioned high-risk instruction sentence set within its respective modified sentence sequence is determined as the constraint sentence position index set. Finally, the tuple consisting of each constraint sentence position index and its corresponding constraint trigger word in the above constraint sentence position index set is determined as the target segmentation constraint information.

[0033] The fifth step involves semantic embedding of the modified sentence sequence set to obtain a sentence embedding vector sequence set. This sequence can be a vector sequence composed of semantic feature vectors corresponding to each individual sentence in the modified sentence sequence, arranged in the original text order. In practice, the executing entity can input each individual sentence in the modified sentence sequence set into the semantic encoding model to obtain the sentence embedding vector corresponding to that individual sentence. The semantic encoding model can be a model that performs semantic embedding on the input text and outputs the corresponding semantic feature vector. For example, the semantic encoding model could be the Sentence-BERT model.

[0034] Step 6: Based on the aforementioned target segmentation constraint information set and the aforementioned sentence embedding vector sequence set, perform semantic cohesion constraint block processing on the aforementioned changed sentence sequence set to obtain an initial changed segment sequence set. The initial changed segment sequence in the aforementioned initial changed segment sequence set can be a segment sequence obtained by merging adjacent single-sentence texts with high semantic cohesion in the sentence embedding vector sequence into continuous text blocks and forcibly severing them at high-risk instruction sentence positions. In practice, the aforementioned execution entity can perform the following block-segmentation steps for each sentence embedding vector sequence in the aforementioned sentence embedding vector sequence set: Step 1: Calculate the cosine similarity of adjacent sentence embedding vectors in the aforementioned sentence embedding vector sequence to obtain an adjacent sentence similarity sequence. Step 2: Determine the mean and standard deviation of the aforementioned adjacent sentence similarity sequence as the similarity mean and similarity standard deviation. Step 3: Determine the sum of the aforementioned similarity mean and the aforementioned similarity standard deviation as the semantic block upper bound threshold, and determine the difference between the aforementioned similarity mean and the aforementioned similarity standard deviation as the semantic block lower bound threshold. Fourth, the interval positions (e.g., the interval between two adjacent sentences) corresponding to adjacent sentences whose similarity values ​​are less than the lower bound threshold of the semantic segmentation in the above adjacent sentence similarity sequence are identified as strong semantic breakpoints, thus obtaining a set of strong semantic breakpoints. Fifth, the constraint sentence position indices corresponding to each target segmentation constraint information in the above target segmentation constraint information set are identified as the forced segmentation point set, and the union of the forced segmentation point set and the strong semantic breakpoint set is identified as the target segmentation point set. Sixth, the modified sentence sequence is segmented at each position corresponding to the target segmentation point set to obtain the initial modified segmentation sequence.

[0035] Step 7: Perform sliding window context enhancement processing on the initial changed segment sequence set to obtain the enhanced changed segment sequence set. The enhanced changed segment sequence in the enhanced changed segment sequence set can be a sequence obtained by supplementing the beginning and end of the text blocks corresponding to each initial changed segment in the initial changed segment sequence with the content of adjacent text blocks. In practice, the execution entity can perform the following enhancement steps for each text block of each initial changed segment sequence in the initial changed segment sequence set: First, determine the single-sentence text at the end of the preceding text block in the initial changed segment sequence with a preset number of sentences (e.g., 2 sentences) as the prefix context. Then, determine the single-sentence text at the beginning of the following text block in the initial changed segment sequence with a preset number of sentences (e.g., 2 sentences) as the suffix context. Next, concatenate the prefix context, the text block, and the suffix context in sequence to obtain the enhanced text block. For the first text block in the initial changed segment sequence, the prefix context is empty; for the last text block, the suffix context is empty. Finally, the enhanced text blocks are arranged in the original order to obtain the enhanced modified segment sequence.

[0036] Step 8: Determine the enhanced changed segment sequence set and the changed element field information set as the changed element information.

[0037] Step 9: Based on the aforementioned set of terminal identifiers to be changed and the information on the changed elements, perform operation and maintenance control on the set of terminal devices corresponding to the aforementioned target terminal information set. As an example, the implementation method of this step can refer to the implementation methods of steps 103-107, and will not be repeated here.

[0038] The above-described technical solution and its related content, as an inventive point of this disclosure, solves technical problem two: "reducing the implementation time and power consumption of terminal devices during maintenance changes, and improving the stability of terminal devices." Factors that lead to prolonged implementation time, increased power consumption, and reduced stability of terminal devices during maintenance changes are often as follows: Text segmentation methods based on semantic similarity cannot perceive the semantic boundaries of high-risk operation instructions in maintenance change information. This causes high-risk operation instructions and their associated context information to be split into different segments or mixed with irrelevant content in the same segment. Consequently, subsequent retrieval and detection based on the change segment sequence cannot accurately identify the risk information corresponding to high-risk operation instructions, resulting in inaccurate maintenance change detection information. This leads to inaccurate maintenance control operations performed on the terminal devices, resulting in prolonged implementation time, increased power consumption, and reduced stability. Solving these factors can reduce the implementation time and power consumption of terminal devices during maintenance changes, thereby improving their stability. To achieve this effect, this disclosure first generates an attachment text set, a set of terminal identifiers to be changed, and a set of change element field information based on the aforementioned maintenance change information. Then, each attachment text in the attachment text set is concatenated with the change description text included in the aforementioned maintenance change information to obtain a text set to be segmented. Next, sentence boundary segmentation processing is performed on the text set to be segmented to obtain a set of change sentence sequences. Here, the attachment text and change description text are concatenated and sentence boundary segmentation is performed, so that the input for subsequent segmentation processing is a sequence of change sentences at the single-sentence granularity, providing a unified granularity of basic data for subsequent segmentation constraint identification processing and semantic cohesion constraint block processing. Subsequently, segmentation constraint identification processing is performed on the aforementioned change sentence sequence set to obtain a target segmentation constraint information set. Next, semantic embedding processing is performed on the aforementioned change sentence sequence set to obtain a set of sentence embedding vector sequences. Finally, based on the aforementioned target segmentation constraint information set and the aforementioned sentence embedding vector sequence set, semantic cohesion constraint block processing is performed on the aforementioned change sentence sequence set to obtain an initial change segmentation sequence set. Here, the modified sentence sequence is segmented to ensure that the resulting initial modified segment sequence can both prevent high-risk operation instructions from being mixed with irrelevant content in the same segment and ensure that adjacent operation steps with high semantic cohesion are not split into different segments, thus improving the recall accuracy of subsequent retrieval based on the modified segment sequence. Then, the above initial modified segment sequence set is subjected to sliding window context enhancement processing to obtain the enhanced modified segment sequence set. Finally, the above enhanced modified segment sequence set and the above modified element field information set are identified as the modified element information.Here, a sliding window context enhancement is performed to ensure that each segment in the enhanced changed segment sequence maintains independent semantic integrity while possessing cross-segment contextual continuity. This reduces the loss of contextual information caused by segmentation. The enhanced changed segment sequence set and the changed element field information set are then identified as changed element information, providing semantically complete and contextually continuous input for subsequent retrieval construction and the generation of operation and maintenance change detection information. Finally, based on the aforementioned set of terminal identifiers to be changed and the changed element information, operation and maintenance control is performed on the set of terminal devices corresponding to the target terminal information set. This operation and maintenance control of the terminal device set based on semantically complete and contextually continuous changed element information reduces the likelihood of inaccurate operation and maintenance control operations on terminal devices, lowers the implementation time and power consumption of terminal devices during operation and maintenance changes, and improves the stability of terminal devices.

[0039] Step 103: Based on the operation and maintenance change information, perform constraint alignment processing on the terminal identifier set and terminal attribute information set to be changed to obtain the target terminal information set.

[0040] In some embodiments, the execution entity can perform constraint alignment processing on the set of terminal identifiers to be changed and the set of terminal attribute information based on the aforementioned operation and maintenance change information to obtain a target terminal information set. The target terminal information in the target terminal information set can be status information reflecting the target terminal device at the current time and within a preset time window (e.g., the next 48 hours). The target terminal device can be the terminal device corresponding to the terminal identifier to be changed. The target terminal information can include, but is not limited to, at least one of the following: target terminal attribute information and terminal associated status information. The target terminal attribute information can be the terminal attribute information corresponding to the target terminal device. The terminal associated status information can be identifier information reflecting whether the target terminal device can undergo a status change within the preset time window.

[0041] In addressing the technical problems mentioned above, the following technical issue arises in the application scenario: multiple operational changes overlap in their implementation time windows and involve the same terminal device (e.g., multiple operational teams submit restart and configuration distribution operations for the same batch of terminal devices within the same maintenance window). This often leads to the following third technical problem: because the terminal identifier-based matching method can only identify change conflicts where terminal device identifiers are completely identical, it cannot identify implicit conflicts at the operational semantic level between different operational changes (e.g., performing both restart and configuration distribution operations on the same terminal device within overlapping time windows). This results in conflicting operational changes not being intercepted, leading to contradictory operational control operations on the terminal device. Consequently, the implementation time of operational changes on the terminal device is prolonged, the power consumption of the terminal device increases, and the stability of the terminal device decreases. To address the following requirements for this application scenario: the ability to identify semantic conflicts across changes and the ability to perceive terminal associations under overlapping time windows, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the above-mentioned constraint alignment processing of the terminal identifier set to be changed and the terminal attribute information set to obtain the target terminal information set, and the operation and maintenance control of the terminal device set corresponding to the target terminal information set based on the target terminal information set, may include the following steps: The first step is to determine at least one terminal attribute information from the above terminal attribute information set that corresponds to the above terminal identifier set to be changed as the target terminal attribute information set.

[0042] The second step involves filling the target terminal attribute information set with slots based on the terminal change implementation time window set and terminal change type set included in the aforementioned change element information, thus obtaining the target terminal change information set. The target terminal change information in this set can be text describing the change operation that the target terminal device will perform. In practice, the executing entity can perform the following filling steps for each target terminal attribute information in the target terminal attribute information set: First, obtain the hypothesis sentence template corresponding to the terminal change type corresponding to the target terminal attribute information from a preset hypothesis sentence mapping table. The preset hypothesis sentence mapping table can be a pre-defined mapping table with the terminal change type as the index key and natural language template text containing multiple slots to be filled as the index value. For example, the hypothesis sentence template corresponding to the terminal change type "restart" could be "The terminal device {device identifier} will perform a restart operation in {time window}". Then, fill the slots in the hypothesis sentence template with the filling information corresponding to the target terminal attribute information to obtain the target terminal change information. The information to be filled in may include, but is not limited to: the terminal device identifier corresponding to the target terminal attribute information, the terminal change implementation time window corresponding to the target terminal attribute information, and the terminal change type corresponding to the target terminal attribute information.

[0043] The third step involves performing time interval overlap matching on a pre-defined maintenance change record database based on the maintenance change implementation time window included in the maintenance change information to obtain predetermined terminal status information. This predetermined terminal status information can be any pending terminal status change record in the pre-defined maintenance change record database that overlaps with the maintenance change implementation time window. In practice, the executing entity can first extract the maintenance change implementation time window from the maintenance change information to obtain the current change start time and current change end time. Then, for each pending terminal status change record in the pre-defined maintenance change record database, the planned start time and planned end time of the pending terminal status change record are extracted. Finally, at least one pending terminal status change record in the maintenance change record database whose planned start time is less than or equal to the current change end time and whose planned end time is greater than or equal to the current change start time is identified as the predetermined terminal status information.

[0044] The fourth step involves adaptive filtering of related sentences on the aforementioned predetermined terminal status information and the aforementioned target terminal change information set to obtain an initial set of operation and maintenance related sentence pairs. These initial operation and maintenance related sentence pairs can be sentence pairs composed of the pending terminal status change records in the predetermined terminal status information and the target terminal change information in the target terminal change information set. In practice, the executing entity can first pair each pending terminal status change record in the predetermined terminal status information as a premise sentence and each target terminal change information in the target terminal change information set as a hypothesis sentence to obtain a candidate sentence pair set. Then, the premise sentences and hypothesis sentences in these candidate sentence pairs are input into the aforementioned semantic encoding model to obtain a premise sentence vector set and a hypothesis sentence vector set. Next, cosine similarity is determined between each premise sentence vector in the premise sentence vector set and each hypothesis sentence vector in the hypothesis sentence vector set to obtain a semantic association score set. Finally, at least one candidate sentence pair with a semantic association score greater than or equal to a preset association threshold (e.g., 0.6) in the semantic association score set is identified as the initial set of operation and maintenance related sentence pairs.

[0045] The fifth step involves semantic feature conflict identification processing on the initial set of operation and maintenance related sentence pairs to obtain the operation and maintenance related sentence pair set and the conflict confidence set. The operation and maintenance related sentence pairs in the aforementioned set can be initial operation and maintenance related sentence pairs with operational conflicts. The conflict confidence set can be the probability value reflecting the operational conflict between the premise and hypothesis sentences in the operation and maintenance related sentence pairs. In practice, the executing entity can first input the initial set of operation and maintenance related sentence pairs into a trained text implication relation classification model to obtain a relation category probability distribution. This relation category probability distribution can include implication probability, neutral probability, and contradiction probability. The text implication relation classification model can be a classification model that performs natural language inference on the input sample pairs and outputs three relation category labels: implication, neutral, or contradiction. For example, the text implication relation classification model can be a DeBERTa (Decoding-enhanced BERT with disentangled Attention) model. Then, the contradiction probability is determined as the initial conflict confidence set. Next, at least one initial conflict confidence score in the aforementioned initial conflict confidence score set that is greater than or equal to a preset conflict detection threshold (e.g., 0.3) is determined as the conflict confidence score set. Finally, the initial operation and maintenance related sentence pair corresponding to each conflict confidence score in the aforementioned conflict confidence score set is determined as the operation and maintenance related sentence pair set.

[0046] Step 6: Based on the aforementioned conflict confidence set, perform bidirectional semantic logic verification on the aforementioned set of operation and maintenance related sentence pairs to obtain a set of terminal-related status information. The terminal-related status information in this set can be a record of a terminal status change to be executed that conflicts with the target terminal device within a preset time window. In practice, the executing entity can first perform the following verification steps for each operation and maintenance related sentence pair in the aforementioned set: First, swap the positions of the premise and hypothesis sentences in the operation and maintenance related sentence pair and re-input it to the aforementioned text implication relationship classification model to obtain the reverse contradiction probability. Then, the average of the conflict confidence of the operation and maintenance related sentence pair and the reverse contradiction probability is determined as the post-verification conflict confidence. Next, the terminal status change record to be executed corresponding to the premise sentence of at least one operation and maintenance related sentence pair in the aforementioned set whose post-verification conflict confidence is greater than or equal to a preset verification threshold (e.g., 0.5) is determined as the terminal-related status information.

[0047] Step 7: Determine the aforementioned terminal association status information set and the aforementioned target terminal attribute information set as the target terminal information set, and perform operation and maintenance control on the terminal device set corresponding to the aforementioned target terminal information set based on the aforementioned target terminal information set. As an example, the implementation method of this step can refer to the implementation method of steps 104-107, and will not be repeated here.

[0048] The above-mentioned technical solution and related content, as an inventive point of this disclosure, solves the third technical problem: "reducing the implementation time and power consumption of terminal devices during operation and maintenance changes, and improving the stability of terminal devices." Factors that lead to prolonged implementation time, increased power consumption, and reduced stability of terminal devices during operation and maintenance changes are often as follows: Because the matching method based on terminal identifiers can only identify change conflicts where terminal device identifiers are completely identical, it cannot identify implicit conflict relationships at the operational semantic level between different operation and maintenance changes (e.g., performing both a restart operation and a configuration distribution operation on the same terminal device within an overlapping time window). This results in operation and maintenance changes with operational conflicts not being intercepted, leading to contradictory operation and maintenance control operations being performed on the terminal devices, causing prolonged implementation time, increased power consumption, and reduced stability. Solving these factors can reduce the implementation time and power consumption of terminal devices during operation and maintenance changes, thereby improving the stability of terminal devices. To achieve this effect, this disclosure first determines at least one terminal attribute information corresponding to the terminal identifier set to be changed in the above-mentioned terminal attribute information set as the target terminal attribute information set. Secondly, based on the terminal change implementation time window set and terminal change type set included in the aforementioned change element information, slot filling processing is performed on the aforementioned target terminal attribute information set to obtain the target terminal change information set. Thirdly, based on the operation and maintenance change implementation time window included in the operation and maintenance change information, time interval overlap matching is performed on the preset operation and maintenance change record database to obtain the predetermined terminal status information. Next, adaptive filtering of related sentences is performed on the aforementioned predetermined terminal status information and the aforementioned target terminal change information set to obtain an initial set of operation and maintenance related sentence pairs. Here, performing time interval overlap matching on the preset operation and maintenance change record database based on the operation and maintenance change implementation time window, filtering out terminal status change records that overlap in the time window, and then performing adaptive filtering of related sentences can reduce the computational overhead of subsequent conflict identification processing, while also reducing noise interference introduced by irrelevant sentences. Then, semantic feature conflict identification processing is performed on the aforementioned initial set of operation and maintenance related sentence pairs to obtain the operation and maintenance related sentence pair set and conflict confidence set. Afterwards, based on the aforementioned conflict confidence set, bidirectional semantic logic verification processing is performed on the aforementioned set of operation and maintenance related sentence pairs to obtain the terminal related status information. Here, bidirectional semantic logic verification is performed on the set of operation and maintenance related sentences. Bidirectional verification can reduce the misjudgment caused by unidirectional recognition, improve the reliability of conflict recognition results, and enable the obtained terminal association status information to accurately reflect the operational conflict relationship between the current operation and maintenance change for the same target terminal device and other operation and maintenance changes to be executed.Finally, the aforementioned terminal association status information and the aforementioned target terminal attribute information set are determined as the target terminal information set. Based on this target terminal information set, operation and maintenance control is performed on the terminal device set corresponding to the target terminal information set. Here, an accurate target terminal information set allows for the simultaneous perception of the target terminal device's own attributes and operational conflicts with other operation and maintenance changes when generating operation and maintenance change detection information. This avoids executing contradictory operation and maintenance control operations on terminal devices with operational conflicts, reduces the likelihood of inaccurate operation and maintenance control operations on terminal devices, lowers the implementation time and power consumption of terminal devices during operation and maintenance changes, and improves the stability of terminal devices.

[0049] Step 104: Perform retrieval-based construction processing on the changed element information and target terminal information set to obtain retrieval query information for operation and maintenance change information.

[0050] In some embodiments, the executing entity may perform retrieval-based construction processing on the aforementioned changed element information and the aforementioned target terminal information set to obtain retrieval query information for the aforementioned operation and maintenance change information. The aforementioned retrieval query information may be information extracted from the aforementioned changed element information and the aforementioned target terminal information set, used for subsequent construction of retrieval conditions, and reflecting the semantic or lexical features of the aforementioned operation and maintenance change information.

[0051] In some optional implementations of certain embodiments, the above-described retrieval-based construction process of the modified element information and the target terminal information set to obtain the retrieval query information of the operation and maintenance change information may include the following steps: The first step involves targeted extraction of the aforementioned changed element information and target terminal information set to obtain a set of retrieval feature items. These retrieval feature items can be structured key-value pairs reflecting the association between the target terminal information set and the key elements included in the changed element information. For example, the retrieval feature items could be structured key-value pairs with the key "operation instruction" and the value "restart database". In practice, the executing entity can input the changed element information and target terminal information set into a retrieval feature targeted extraction model to obtain the retrieval feature item set. This model can perform feature encoding, feature fusion, and decoding on the input changed element information and target terminal information set, outputting the retrieval feature item set. This model can be a concatenated structure consisting of a targeted encoding layer, a semantic purification layer, and a sequence decoding output layer. The targeted encoding layer can be a model that performs character mapping and positional encoding on the input changed element information and target terminal information set, outputting a targeted query matrix, a key matrix reflecting the changed element information, and a value matrix. The aforementioned directional query matrix can be a matrix composed of query vectors encoded from the target terminal information set. For example, the aforementioned directional encoding layer can be BERT (Bidirectional Encoder Representations from Transformers). The aforementioned semantic purification layer can be a model that performs cross-attention feature fusion on the obtained directional query matrix, key matrix, and value matrix to output a sequence of latent state vectors. The latent state vectors in the aforementioned sequence of latent state vectors can represent the deep semantic feature vectors of each word included in the aforementioned changed element information under the constraints of the target terminal context. For example, the aforementioned semantic purification layer can be a cross-attention network layer of the Transformer decoder. The aforementioned sequence decoding output layer can be a model that performs sequence labeling on the obtained sequence of latent state vectors to output a set of retrieved feature terms. For example, the aforementioned sequence decoding output layer can be a BiLSTM-CRF (Bidirectional Long Short-Term Memory with a Conditional Random Field layer) model.

[0052] The second step involves generating a semantic query information set based on the aforementioned retrieval feature item information set and the change segment sequence included in the aforementioned change element information. This semantic query information can be text reflecting the overall semantic characteristics of the aforementioned operational change information, used to generate query vectors for subsequent knowledge retrieval. In practice, the executing entity can first identify the retrieval feature item information in the aforementioned retrieval feature item information set as the key representing the change type, and determine it as the target retrieval feature item information. Then, the target retrieval feature item information is matched with a preset syntax template mapping table using key-value addressing to obtain the target syntax template. The preset syntax template mapping table can be a pre-defined hash table with change type as the index key and template text as the mapping value. The template text can be a pre-written template string containing multiple slots to be filled (e.g., curly braces). For example, the template text could be "Execute {operation action} operation on {target device} node". The target syntax template can be the template text in the preset syntax template mapping table that matches the current change type. Next, the aforementioned retrieval feature item information set is filled into the target syntax template to obtain the feature-guided sentence. Subsequently, the Punkt algorithm is used to segment the aforementioned modified segmented sequence into sentences, resulting in a single-sentence text sequence. Each sentence in this sequence can be a character fragment with independent and complete semantics, defined by sentence breakpoints. Next, the TextRank algorithm is used to perform extractive text summarization on the single-sentence text sequence, yielding the target single-sentence text sequence. This target single-sentence text sequence can be a sequence of single-sentence texts retained after filtering out invalid sentences (e.g., redundant descriptive content, background information). Finally, the aforementioned feature-guided sentence is placed at the beginning and concatenated with each target single-sentence text in the target single-sentence text sequence to obtain a semantic query information set.

[0053] The third step involves weighting the aforementioned set of search feature information to obtain the target set of search feature information, which serves as the keyword query information set. The keyword query information in this set can be text reflecting high-discrimination lexical features of the aforementioned maintenance change information. For example, the text with high-discrimination lexical features can be terms that appear frequently in the aforementioned maintenance change information. In practice, the executing entity can first use TF-IDF (Term Frequency-Inverse Document Frequency) to generate a set of search feature weight values ​​based on the number of times the search feature information appears in the aforementioned change segment sequence and the number of attachment texts in the aforementioned attachment text set. The search feature weight values ​​in this set can reflect the importance of the search feature in the aforementioned maintenance change information. Then, at least one search feature weight value in the set that is greater than or equal to a preset weight threshold is determined as the target set of search feature weight values. The preset weight threshold can be a pre-set value used to determine whether the search feature weight value is a target search feature weight value. For example, the preset weight threshold mentioned above could be 0.15. Finally, at least one retrieval feature information corresponding to the above target retrieval feature weight value set is determined as the target retrieval feature information set.

[0054] The fourth step involves filtering the target terminal information set based on the aforementioned retrieval feature item information set to obtain a terminal constraint information set. The terminal constraint information in this set can be a tuple reflecting the category attributes (e.g., device type, implementation environment, change type, etc.) of the target terminal device and its corresponding change operation. This tuple can include a constraint dimension identifier and a constraint dimension value. The constraint dimension identifier can be the name text of the category attribute. The constraint dimension value can be the specific category label to which the target terminal device or change operation belongs under the constraint dimension identifier. For example, the constraint dimension identifier might be "system level," and the constraint dimension value might be "P1." In practice, the executing entity can first perform the following constraint candidate information determination steps: First, input the key text corresponding to each retrieval feature item in the aforementioned retrieval feature item information set into the aforementioned semantic encoding model to obtain a feature key vector set. Second, using a cosine similarity algorithm, perform similarity determination processing on the semantic vectors of each constraint dimension included in the preset constraint field dimension table and each feature key vector in the aforementioned feature key vector set to obtain a first similarity matrix. The aforementioned preset constraint field dimension table can be a pre-defined table containing multiple constraint dimension records. Each constraint dimension record can include a constraint dimension identifier and a constraint dimension semantic vector. The constraint dimension semantic vector can be a numerical representation of the constraint dimension identifier. The elements of the aforementioned first similarity matrix can characterize the semantic matching degree between the key of the retrieved feature item information and the constraint dimension identifier. Third, the maximum value of each row of the aforementioned first similarity matrix is ​​determined as the first field matching degree value set. Fourth, the constraint dimension identifier and retrieved feature item information corresponding to at least one first field matching degree value in the first field matching degree value set whose value is greater than or equal to a preset matching threshold (e.g., 0.75) are determined as the first constraint candidate information set. Then, the text information corresponding to each target terminal information in the aforementioned target terminal information set is used as the key text input in the aforementioned constraint candidate information determination step, and the aforementioned constraint candidate information determination step is executed again to obtain the second constraint candidate information set. Next, at least one first constraint candidate and at least one second constraint candidate with the same constraint dimension identifier in the aforementioned first constraint candidate information set and the aforementioned second constraint candidate information set are determined as the first terminal constraint information set. Next, the first set of constraint candidate information after removing the target constraint candidate information and the aforementioned second set of constraint candidate information are determined as the second terminal constraint information set. The aforementioned target constraint candidate information can be constraint candidate information from the aforementioned first and second sets of constraint candidate information that has already been used to determine the first terminal constraint information set. Finally, the aforementioned first and second sets of terminal constraint information are determined as the terminal constraint information set.

[0055] The fifth step is to determine the above keyword query information set, the above semantic query information set, and the above terminal constraint information set as the retrieval query information.

[0056] Step 105: Based on the preset operation and maintenance retrieval database, perform dual-channel fusion recall processing on the retrieval query information to obtain a candidate operation and maintenance change sample set.

[0057] In some embodiments, the aforementioned execution entity can perform dual-channel fusion recall processing on the aforementioned retrieval query information based on a preset operation and maintenance retrieval database to obtain a candidate operation and maintenance change sample set. The candidate operation and maintenance change samples in the aforementioned candidate operation and maintenance change sample set can be information recalled from the aforementioned operation and maintenance retrieval database that has semantic or lexical association with the aforementioned operation and maintenance change information. The aforementioned operation and maintenance retrieval database can be a hybrid retrieval database that integrates vector retrieval capabilities and keyword retrieval capabilities. For example, the aforementioned operation and maintenance retrieval database can be a Milvus vector database.

[0058] In some optional implementations of certain embodiments, the above-mentioned operation and maintenance retrieval database is obtained through the following steps: The first step is to perform text segmentation on the acquired set of operation and maintenance change cases to obtain a set of segmented change case information. The operation and maintenance change cases in this set can be historical text records related to the execution of operation and maintenance changes on terminal devices. For example, these operation and maintenance change cases could be fault review reports and operation and maintenance manuals for historical operation and maintenance change failures. The segmented change case information in the set of segmented change case information can be text blocks with relatively independent semantics within the operation and maintenance change cases. In practice, the executing entity can first use a preset metadata extraction rule set to extract metadata fields from each operation and maintenance change case in the set, obtaining a case metadata dataset. The metadata extraction rules in the set of metadata extraction rules can be pre-defined regular expression matching rules used to extract various types of case metadata from operation and maintenance change cases. For example, the metadata extraction rules could be rules for extracting the case occurrence time by matching regular expressions in the format "YYYY-MM-DD". The case metadata in the case metadata set can be structured field information recording the source attributes of the operation and maintenance change cases. For example, the aforementioned case metadata may include, but is not limited to: the ID of the operation and maintenance change case, the time of occurrence of the operation and maintenance change case, location identifier, etc. (e.g., the storage address identifier in the aforementioned operation and maintenance retrieval database). Then, the TextTiling algorithm is used to semantically segment each operation and maintenance change case in the aforementioned operation and maintenance change case set to obtain an initial change case segment information set. Afterwards, the segment information of each initial change case in the aforementioned initial change case segment information set is concatenated with its corresponding case metadata to obtain the change case segment information set.

[0059] In the second step, an inverted index is constructed for the above-mentioned segmented information set of change cases to obtain a keyword inverted index. Among them, the above-mentioned keyword inverted index can be an index structure with the word segments included in the segmented information set of change cases as index keys and the position identifiers of the segmented information of change cases containing the term as index values. In practice, the above-mentioned execution entity can first use a word segmentation tool to perform word segmentation on each segmented information of change cases in the above-mentioned segmented information set of change cases to obtain a set of segmented semantic unit sequences. The above-mentioned word segmentation tool can be the Jieba word segmentation tool. Then, using a preset stop word list, the segmented semantic units in each segmented semantic unit sequence in the above-mentioned set of segmented semantic unit sequences that match the preset stop word list are removed to obtain a target set of segmented semantic unit sequences. The above-mentioned preset stop word list can be a pre-set word list containing common words that do not have substantial retrieval significance (for example, "de", "le", "zai", etc.). Finally, each target segmented semantic unit included in the above-mentioned target set of segmented semantic unit sequences is used as a key, and its corresponding position identifier is used as a value, and input into an initial inverted index to obtain a keyword inverted index. The above-mentioned initial inverted index can be a hash table structure with an empty list as the value.

[0060] In the third step, the above-mentioned segmented information set of change cases is input into the above-mentioned semantic encoding model to obtain a set of change case segmented vectors. Among them, the change case segmented vectors in the above-mentioned set of change case segmented vectors can be numerical vectors representing the semantic features of the corresponding segmented information of change cases.

[0061] In the fourth step, an approximate nearest neighbor index is constructed for the above-mentioned set of change case segmented vectors to obtain a semantic vector index. Among them, the above-mentioned semantic vector index can be a multi-layer directed graph structure with each change case segmented vector in the above-mentioned set of change case segmented vectors as graph nodes and the neighbor relationship between nodes as connection edges. In practice, the above-mentioned execution entity can use the HNSW (Hierarchical Navigable Small World) algorithm to perform multi-layer neighbor graph construction processing on the above-mentioned set of change case segmented vectors to obtain a semantic vector index.

[0062] In the fifth step, an association mapping process is performed on the above-mentioned segmented information set of change cases, the above-mentioned keyword inverted index, and the above-mentioned semantic vector index to obtain an operation and maintenance retrieval database. In practice, the above-mentioned execution entity can use the above-mentioned segmented information set of change cases and the corresponding set of change case segmented vectors as data records, and store them together with the above-mentioned keyword inverted index and the above-mentioned semantic vector index in a preset hybrid retrieval database to obtain an operation and maintenance retrieval database.

[0063] Optionally, the above-mentioned dual-channel fusion recall processing of the search query information based on the preset operation and maintenance retrieval database to obtain a candidate operation and maintenance change sample set may include the following steps: The first step is to perform Boolean compilation on the terminal constraint information set included in the above-mentioned retrieval query information to obtain the retrieval filter expression. This retrieval filter expression can be a Boolean logical expression used to narrow the scope of data records in the above-mentioned operation and maintenance retrieval database (e.g., change case segment information and its corresponding change case segment vector). For example, the retrieval filter expression could be (Device Type = ATM) AND (Change Type = Restart). In practice, the execution entity can first use the constraint dimension identifier included in each terminal constraint information set as the field name and the constraint dimension value as the matching value to generate an equivalent predicate expression set (e.g., "Device Type = ATM"). Then, the various equivalent predicate expressions in the above-mentioned equivalent predicate expression set are combined using a logical AND operator to obtain the retrieval filter expression.

[0064] The second step involves vectorizing the semantic query information included in the aforementioned retrieval query information to obtain a semantic query vector set. The semantic query vectors in this set represent the semantic features of the semantic query information. In practice, the executing entity can input the semantic query information set into the semantic encoding model to obtain the semantic query vector set.

[0065] The third step involves performing multi-channel semantic recall processing on the semantic query vector set and the maintenance retrieval database based on the aforementioned retrieval filtering expression, resulting in a semantic recall sample set. The semantic recall samples in this set can be information from the maintenance retrieval database that exhibits semantic similarity to the semantic query information. In practice, the executing entity can perform Boolean filtering on the data records in the maintenance retrieval database based on the aforementioned retrieval filtering expression, obtaining a filtered subset of data records. This Boolean filtering can exclude records that do not satisfy the aforementioned retrieval filtering expression. Then, for each semantic query vector in the semantic query vector set, an approximate nearest neighbor search is performed within the filtered subset of data records using the semantic vector index of the maintenance retrieval database, resulting in a semantic recall sample set and a semantic similarity value set. The semantic similarity values ​​included in the semantic similarity value set characterize the closeness in semantic space between the semantic query vector and the change case segment vectors in the filtered subset of data records.

[0066] The fourth step involves merging and deduplicating the aforementioned semantic recall sample sets to obtain a semantic candidate sample set. The semantic candidate samples in this set can be the semantic recall samples retained after filtering out redundant samples with duplicate location identifiers from the semantic recall sample sets. In practice, the executing entity can first identify at least two semantic recall samples with the same location identifier for each semantic recall sample in the aforementioned semantic recall sample set as the target semantic recall sample set. Then, the target semantic recall sample with the highest corresponding semantic similarity value in each target semantic recall sample set is identified as a semantic candidate sample, thus obtaining the semantic candidate sample set.

[0067] Fifth, based on the aforementioned retrieval filtering expression, lexical channel recall processing is performed on the keyword query information set included in the aforementioned retrieval query information and the aforementioned operation and maintenance retrieval database to obtain a lexical candidate sample set. The lexical candidate samples in the aforementioned lexical candidate sample set can be information from the aforementioned filtered data record subset that has a lexical matching relationship with the keyword query information. In practice, the aforementioned execution entity can first use the BM25 (Best Matching 25) algorithm to perform word frequency matching calculations on the inverted indexes of keywords in each keyword query information set and each change case segment information in the filtered data record subset corresponding to the aforementioned retrieval filtering expression, to obtain a lexical candidate sample set and a lexical similarity value set. The lexical similarity values ​​in the aforementioned lexical similarity value set can characterize the degree of matching between the keyword query information and the change case segment information at the lexical level.

[0068] The sixth step involves ranking and fusing the aforementioned semantic candidate sample set and lexical candidate sample set to obtain a fused candidate sample set. The fused candidate samples in this set can be information from the aforementioned operations and maintenance retrieval database that has both semantic and lexical associations with the aforementioned operations and maintenance change information. In practice, the executing entity can utilize the RRF (Reciprocal Rank Fusion) algorithm to rank and fuse the semantic and lexical candidate sample sets, obtaining a fused candidate sample set and a fused relevance value set. The fused relevance values ​​in this set characterize the overall relevance of the fused candidate samples after considering both semantic similarity and lexical matching.

[0069] Step 7: Perform time decay matching processing on the aforementioned fusion candidate sample set and the obtained current time value to obtain a time decay degree value set. The time decay degree values ​​in this set characterize the timeliness of the operation and maintenance change case corresponding to the fusion candidate sample relative to the current time. In practice, the executing entity can first obtain the current system time as the current time value. Then, for each fusion candidate sample in the aforementioned fusion candidate sample set, perform the following generation steps: extract the case occurrence time from the case metadata corresponding to the fusion candidate sample. Next, determine the time difference (e.g., in days) between the current time value and the case occurrence time. Then, using an exponential decay function, generate the time decay degree value corresponding to the fusion candidate sample based on the time difference.

[0070] Step 8: Based on the aforementioned time decay value set, the aforementioned fusion candidate sample set is rearranged and filtered to obtain a candidate operation and maintenance change sample set. In practice, the aforementioned execution entity can first perform weighted fusion of each time decay value in the aforementioned time decay value set and its corresponding fusion relevance value to obtain a target relevance value set. The target relevance value in the aforementioned target relevance value set can characterize the retrieval matching degree of the fusion candidate samples after considering both relevance and timeliness. Then, each fusion candidate sample in the aforementioned fusion candidate sample set is arranged in descending order according to the size of the aforementioned target relevance value set to obtain a fusion candidate sample sequence. Next, at least one fusion candidate sample in the aforementioned fusion candidate sample sequence that is located before a preset rearrangement threshold position (e.g., 50) and whose corresponding target relevance value is greater than or equal to a preset filtering threshold (e.g., 0.3) is determined as the candidate operation and maintenance change sample set.

[0071] Step 106: Perform multi-level screening and enhancement processing on the candidate operation and maintenance change sample set to obtain the target operation and maintenance change sample set, and generate operation and maintenance change detection information based on the change element information, the target terminal information set and the target operation and maintenance change sample set.

[0072] In some embodiments, the execution entity may perform multi-level screening and enhancement processing on the candidate operation and maintenance change sample set to obtain a target operation and maintenance change sample set, and generate operation and maintenance change detection information based on the change element information, the target terminal information set, and the target operation and maintenance change sample set. The target operation and maintenance change sample in the target operation and maintenance change sample set may be a candidate operation and maintenance change sample that has a semantic association with the operation and maintenance change information and does not have semantic contradictions with other candidate operation and maintenance change samples in the candidate operation and maintenance change sample set. The operation and maintenance change detection information may be information reflecting the semantic and logical association features between the change element information and the target operation and maintenance change sample set. The operation and maintenance change detection information may include, but is not limited to, at least one of the following: change risk confidence, operation and maintenance automation adaptation value, change analysis text, and change control information. The change risk confidence may be a probability value reflecting whether the implementation of the operation and maintenance change information will cause abnormal conditions (e.g., service process interruption or crash) in the terminal device. The operation and maintenance automation adaptation value may be a characterizing the degree of adaptation of the implementation scheme of the operation and maintenance change information that can be automatically executed. The aforementioned change analysis text can be natural language text reflecting the main risk points (e.g., high-risk instructions, missing steps) in the aforementioned operation and maintenance change detection information, as well as the target operation and maintenance change samples associated with the main risk points. The aforementioned change control information can be a standardized set of parameters reflecting the status changes made to terminal devices by the aforementioned operation and maintenance change information.

[0073] In some optional implementations of certain embodiments, the above-mentioned multi-level screening and enhancement processing of the candidate operation and maintenance change sample set to obtain the target operation and maintenance change sample set may include the following steps: The first step is to perform context completion processing on the aforementioned candidate operation and maintenance change sample set to obtain the first operation and maintenance change sample set. The first operation and maintenance change sample in the first operation and maintenance change sample set can be a change case segment information that expands upon the candidate operation and maintenance change samples by adding adjacent contextual information. In practice, the executing entity can perform the following concatenation steps for each candidate operation and maintenance change sample in the candidate operation and maintenance change sample set: First, according to a preset context window size (e.g., one segment before and one after), retrieve at least one change case segment information adjacent to the candidate operation and maintenance change sample in the same operation and maintenance change case from the aforementioned operation and maintenance retrieval database, and determine it as the context segment information set. Second, concatenate the aforementioned context segment information set and the aforementioned candidate operation and maintenance change sample to obtain the first operation and maintenance change sample.

[0074] The second step involves performing similarity clustering deduplication on the first maintenance change sample set to obtain the second maintenance change sample set. The second maintenance change samples in the second maintenance change sample set can be the first maintenance change samples retained after filtering out semantically redundant samples from the first maintenance change sample set. In practice, the executing entity can first input each first maintenance change sample from the first maintenance change sample set into the semantic encoding model to obtain a first maintenance change sample vector set. Then, pairwise cosine similarity is determined for each first maintenance change sample vector in the first maintenance change sample vector set to obtain a sample similarity matrix. Next, two first maintenance change samples with a similarity value greater than or equal to a preset deduplication threshold (e.g., 0.9) in the sample similarity matrix are identified as duplicate sample pairs, and the first maintenance change samples with lower fusion relevance values ​​in the duplicate sample pairs are marked as redundant samples. Finally, the first maintenance change samples retained after removing all redundant samples from the first maintenance change sample set are determined as the second maintenance change sample set.

[0075] The third step involves processing the second set of maintenance and operation change samples for conflict identification, resulting in a third set of maintenance and operation change samples. The third set of maintenance and operation change samples can be the second set of maintenance and operation change samples retained after filtering out those with semantically contradictory relationships from the second set. In practice, the executing entity can first pair each second set of maintenance and operation change samples in the second set to obtain a set of sample pairs. Then, the set of sample pairs is input into the textual implication relationship classification model to obtain a set of relationship category labels. Next, at least one second set of maintenance and operation change samples with an earlier occurrence time among the sample pairs in the set of relationship category labels that have contradictory relationship category labels is identified as the conflict sample set. Finally, the second set of maintenance and operation change samples after removing the conflict sample set is identified as the third set of maintenance and operation change samples.

[0076] The fourth step involves performing cross-coding rearrangement on the third maintenance change sample set based on the aforementioned retrieval query information to obtain the target maintenance change sample set. In practice, the executing entity can first concatenate each third maintenance change sample in the third maintenance change sample set with the semantic query information included in the aforementioned retrieval query information to obtain a concatenated sample information set. Then, the concatenated sample information set is input into the trained cross-coding rearrangement model to obtain a rearranged relevance value set. The rearranged relevance values ​​in the rearranged relevance value set can characterize the semantic relevance between the third maintenance change sample and the aforementioned maintenance change information. The cross-coding rearrangement model can be a model that performs joint attention encoding on the input concatenated sample information set to obtain the rearranged relevance values. For example, the cross-coding rearrangement model can be a Cross-Encoder model. Then, the third maintenance change samples in the aforementioned third maintenance change sample set are arranged in descending order according to the rearranged relevance values ​​to obtain the third maintenance change sample sequence. Finally, the third maintenance change sample ranked first by a preset coding rearrangement threshold (e.g., 5) is determined as the target maintenance change sample set.

[0077] In addressing the technical problems mentioned above, the application scenario—complex change scenarios involving multiple steps and long chains of operations related to maintenance change information—often presents the following technical problem: Because text-matching-based maintenance change detection methods assess maintenance change information as a whole, they fail to perceive the execution dependencies between various operation steps and the attribution relationships between each operation step and the target terminal device. This results in the generated maintenance change detection information failing to accurately reflect the risks caused by local operation steps in the long chain of operations and the constraint conflicts between operation steps and terminal devices. Consequently, inaccurate maintenance control operations are performed on the terminal devices, leading to prolonged implementation time, increased power consumption, and reduced stability during maintenance changes. To address the following requirements for this application scenario: the ability to model execution dependencies between operation steps and the ability to locate local risks in multi-step operation chains, we have decided to adopt the following solution: Optionally, the process of generating maintenance change detection information based on the aforementioned change element information, the aforementioned target terminal information set, and the aforementioned target maintenance change sample set, and then performing maintenance control on the terminal device set corresponding to the aforementioned target terminal information set based on the aforementioned maintenance change detection information, may include the following steps: The first step is to perform semantic enhancement processing on the aforementioned target maintenance change sample set to obtain a sample enhancement information set. The sample enhancement information in this set can be text that adds semantic information about the operational behavior of the target maintenance change sample compared to the target maintenance change sample. This sample enhancement information can include, but is not limited to, at least one of the following: target maintenance change sample, operation triple set, and operation timing label sequence. The operation triple can be a structured information unit consisting of an operation subject (e.g., automation script), an operation action (e.g., restart, upgrade, rollback), and an operation object (e.g., target terminal device identifier or service name). The operation timing labels in the operation timing label sequence can be category labels reflecting the execution stage of the operation action in the operation triple in the implementation steps of this maintenance change (e.g., pre-step, core step, post-verification). In practice, the execution subject can perform the following enhancement steps on each target maintenance change sample in the aforementioned target maintenance change sample set: First, input the aforementioned target maintenance change sample into the trained semantic role labeling model to obtain a semantic role labeling information set. The semantic role labeling information in the aforementioned semantic role labeling information set can be a semantic framework centered on predicates, recording the role relationships between predicates and their associated semantic components. This semantic role labeling information can include: predicate V (e.g., a verb expressing an operational action), agent role A0 (e.g., the subject performing the operation), and patient role A1 (e.g., the object being operated on). The aforementioned semantic role labeling model can be a model that identifies and classifies predicates and their arguments in the input target operation and maintenance change sample, outputting a semantic role labeling information set. For example, the aforementioned semantic role labeling model can be a BERT-based SRL (Semantic Role Labeling) model. The second step involves determining the predicate V, agent role A0, and patient role A1 included in each semantic role labeling information set as the operation action, operation subject, and operation object, respectively. Furthermore, the operation action, operation subject, and operation object corresponding to the same semantic role labeling information are determined as operation triples, resulting in a set of operation triples. The third step involves using a pre-defined operation timing rule base to perform timing category matching on each operation triplet in the aforementioned operation triplet set, resulting in an operation timing tag sequence. This pre-defined operation timing rule base can be a pre-set mapping table using operation action keywords as index keys and operation timing tags as index values ​​(e.g., backup corresponds to a pre-processing step, restart corresponds to a core process, and verification corresponds to a post-processing check). The fourth step involves identifying the aforementioned target maintenance change sample, the aforementioned operation triplet set, and the aforementioned operation timing tag sequence as sample enhancement information.

[0078] The second step involves performing multi-dimensional dynamic association encoding on the aforementioned sample enhancement information set and the aforementioned changed element information to obtain a change semantic association vector set. The change semantic association vectors in this set can represent the degree of association between the sample enhancement information and the changed element information in terms of semantic dimension and operational structure (e.g., the combination relationship between the operating subject, operating action, and operating object). In practice, the executing subject can first input the changed element field information set and the changed segment sequence, including the changed element information, into the semantic encoding model to obtain an element field vector set and a changed segment vector set. Then, the element field vector set and the changed segment vector set are averaged and pooled to obtain a field summary vector and a segment context vector. The field summary vector can represent the overall semantic features of the text corresponding to each changed element field information in the changed element field information set. The segment context vector can represent the overall semantic features of the text corresponding to each changed segment in the changed segment sequence. Next, for each sample enhancement information in the above sample enhancement information set, the following weighted summation steps are performed: First, the operation subject, operation action, and operation object of each operation triplet included in the above sample enhancement information are concatenated to obtain a set of operation description phrases. Second, the above operation description phrase set and the target operation and maintenance change samples included in the above sample enhancement information are input into the above semantic encoding model to obtain a set of triplet vectors and a sample semantic vector. Third, the preset temporal stage weight mapping table and the operation temporal label sequence included in the above sample enhancement information are matched and mapped to obtain a set of operation weight values. The operation weight values ​​in the above operation weight value set can characterize the importance of the operation of the category corresponding to the operation temporal label in the change implementation steps (e.g., core steps correspond to a weight of 0.6, pre-steps correspond to a weight of 0.3, and post-verification corresponds to a weight of 0.1). The above preset temporal stage weight mapping table can be a pre-defined mapping table with the category of the operation temporal label as the index key and the operation weight value as the index value. Fourth, the above operation weight value set is used as weight values ​​to perform weighted pooling on the above triplet vector set to obtain an operation structure vector. The aforementioned operation structure vector can represent the overall semantic features of each operation triplet in the aforementioned operation triplet set. Fifth step: Multiply and subtract the aforementioned sample semantic vector element-wise from the aforementioned segmented context vector to obtain the first association vector and the second association vector. The first association vector can represent the degree of matching between the sample semantic vector and the segmented context vector in each semantic dimension. The second association vector can represent the degree of difference between the sample semantic vector and the segmented context vector in each semantic dimension. Sixth step: Concatenate the aforementioned first association vector and the second association vector to obtain the semantic dimension association vector. Seventh step: Multiply and subtract the aforementioned operation structure vector element-wise from the aforementioned field summary vector, and concatenate the two to obtain the operation structure dimension association vector.The aforementioned operational structure dimension association vector can characterize the matching degree and difference features between the operational structure vector and the field summary vector in each dimension. In the eighth step, the aforementioned sample semantic vector, segmented context vector, operational structure vector, and field summary vector are concatenated to obtain the semantic operation concatenation vector. This semantic operation concatenation vector is then input into the trained dynamic weight generation model to obtain the semantic operation weight vector. The aforementioned dynamic weight generation model can be a model that performs nonlinear feature transformation processing on the input semantic operation concatenation vector and outputs the semantic operation weight vector. This dynamic weight generation model can be a single-hidden-layer FNN (Feedforward Neural Network). The aforementioned semantic operation weight vector can characterize the degree of information quality difference between the current sample enhancement information and the changed element information in both the semantic and operational structure dimensions. In the ninth step, the two components of the aforementioned semantic operation weight vector are scalar-weighted summed with the aforementioned semantic dimension association vector and operational structure dimension association vector, respectively, to obtain the changed semantic association vector.

[0079] The third step involves performing multi-level feature decoding on the aforementioned change semantic association vector set to obtain the change risk confidence score. This change risk confidence score can reflect the probability of anomalies occurring after the implementation of the aforementioned operational changes. In practice, the executing entity can first input the aforementioned change semantic association vector set into a trained operation anomaly identification model to obtain an operation anomaly degree value set. The operation anomaly degree values ​​in this set characterize the degree of anomaly between the sample enhancement information and the change element information at the local semantic and operational structure levels. The operation anomaly identification model can be a model that performs nonlinear feature transformation processing on the input change semantic association vector set and outputs the operation anomaly degree value set. For example, the operation anomaly identification model can be an MLP (Multi-Layer Perceptron) model composed of a first fully connected layer, a ReLU activation function, a second fully connected layer, and a Sigmoid activation function sequentially connected in series. Next, the mean, maximum, minimum, and variance of the operation anomaly degree value set are determined as statistical feature vectors. These statistical feature vectors can characterize the overall distribution and dispersion characteristics of the operation anomaly degree value set. Then, max pooling is performed on the aforementioned change semantic association vector set to obtain a semantic convergence vector. This semantic convergence vector can represent the global semantic features of the aforementioned change semantic association vector set. Finally, the semantic convergence vector and the statistical feature vector are concatenated and input into the trained change risk identification model to obtain the change risk confidence score. The aforementioned change risk identification model can be a model that performs linear weighted fusion and probability mapping on the input vectors to output the change risk confidence score. For example, the aforementioned change risk identification model can be an LR (Logistic Regression) model.

[0080] The fourth step is to construct an operation and maintenance change execution graph based on the aforementioned change element information. This execution graph can be a directed graph with each operation node in the operation node set as a node and the dependencies between these nodes as edges. Each operation node in the operation node set can include an operation node identifier and an operation node vector. The operation node vector can be a change segment vector from the change segment vector set corresponding to the change element information. The operation node identifier can be the position number of the change segment corresponding to the operation node vector in the change segment sequence. The dependencies between operation nodes can be a path from an operation node with a smaller position number to an operation node with a larger position number. In practice, the execution entity can utilize the NetworkX library to construct the operation and maintenance change execution graph based on the aforementioned change element information.

[0081] Fifth, based on the aforementioned target terminal information set, perform heterogeneous constraint update processing on the aforementioned operation and maintenance change execution graph to obtain a change constraint heterogeneous execution graph. This change constraint heterogeneous execution graph can be obtained by adding nodes reflecting target terminal information and edges reflecting the association between operation nodes and target terminal information to the aforementioned operation and maintenance change execution graph. In practice, the executing entity can first input the aforementioned target terminal information set into the aforementioned semantic encoding model to obtain a terminal constraint vector set. The terminal constraint vectors in the aforementioned terminal constraint vector set can represent the semantic features of the target terminal information. Then, the terminal device identifiers corresponding to each target terminal information in the aforementioned target terminal information set are determined as a terminal constraint node identifier set, and the aforementioned terminal constraint vector set and the aforementioned terminal constraint node identifier set are determined as a terminal constraint node set. Next, using the cosine similarity algorithm, the similarity between the operation node vectors of each operation node in the aforementioned operation and maintenance change execution graph and the terminal constraint node vectors of each terminal constraint node in the aforementioned terminal constraint node set is determined to obtain a node association similarity matrix. Then, undirected edges are established between the operation nodes with similarity values ​​greater than or equal to a preset association threshold (e.g., 0.5) in the above node association similarity matrix and the terminal constraint nodes, resulting in a terminal affiliation association edge set. These terminal affiliation association edges can represent the terminal devices corresponding to the terminal constraint nodes involved in the change operation of the operation node. Finally, using the NetworkX library, the above terminal constraint node set and the above terminal affiliation association edge set are added to the above operation and maintenance change execution graph to obtain a change constraint heterogeneous execution graph.

[0082] The sixth step involves performing multi-objective scheduling mapping on the aforementioned heterogeneous execution graph of change constraints to obtain the automation adaptation values ​​for operation and maintenance. In practice, the execution entity can first input the aforementioned heterogeneous execution graph of change constraints into a trained graph feature extraction model to obtain an updated set of operation node vectors. The updated operation node vectors in the updated set of operation node vectors can represent the semantic features of the operation nodes after incorporating information from associated neighboring operation nodes and terminal constraint nodes associated with terminal-owned edge associations. The aforementioned graph feature extraction model can be a model that performs node neighborhood message propagation and feature aggregation processing on the input heterogeneous execution graph of change constraints, outputting an updated set of operation node vectors. For example, the aforementioned graph feature extraction model can be an R-GCN (Relational Graph Convolutional Network) model. Next, the aforementioned updated set of operation node vectors is input into a trained adaptation recognition model to obtain a set of operation adaptation degree values. The operation adaptation degree values ​​in the aforementioned set of operation adaptation degree values ​​can represent the degree of automation adaptation of the change operations corresponding to each operation node in the aforementioned operation and maintenance change execution graph in the operation and maintenance change dimension. The aforementioned adaptation identification model can be a model that performs margin maximization classification processing on the updated set of input operation node vectors and outputs a set of operation adaptation degree values. For example, the aforementioned adaptation identification model can be an SVM (Support Vector Machine) model. Then, the mean of the aforementioned set of operation adaptation degree values ​​is determined as the first target value. Subsequently, for each terminal constraint node in the terminal constraint node set of the aforementioned heterogeneous execution graph of change constraints, the number of operation nodes associated with the aforementioned terminal constraint node is determined as the terminal association count, resulting in a terminal association count set. Then, the ratio of the number of terminal constraint nodes corresponding to the terminal association counts in the aforementioned terminal association count set that are greater than a preset terminal constraint threshold (e.g., 1) to the total number of terminal constraint nodes in the aforementioned terminal constraint node set is determined as the initial second target value. The aforementioned initial second target value can characterize the degree of conflict in the aforementioned operation and maintenance change information where multiple operation steps compete for the same terminal device resource. Finally, the difference between the preset value and the aforementioned initial second target value is determined as the second target value, and the aforementioned first target value and the aforementioned second target value are weighted and summed to obtain the operation and maintenance automation adaptation value. The above preset value can be 1.

[0083] Step 7: Based on the aforementioned change risk confidence level and the aforementioned operation and maintenance automation adaptation value, perform information fusion processing on the aforementioned change element information and the aforementioned target operation and maintenance change sample set to obtain change analysis text and change control information. In practice, the executing entity can first fill the aforementioned change risk confidence level, the aforementioned operation and maintenance automation adaptation value, the aforementioned change element information, and the aforementioned target operation and maintenance change sample set into a preset information fusion prompt template to obtain fusion prompt text. The aforementioned preset information fusion prompt template can be a pre-defined structured instruction text containing change risk confidence level slots, operation and maintenance automation adaptation value slots, change element information slots, and target operation and maintenance change sample slots. Then, input the aforementioned fusion prompt text into the information fusion generation model to obtain change analysis text and change control information. The aforementioned information fusion generation model can be a model that generates natural language from the input fusion prompt text and outputs change analysis text and change control information. For example, the aforementioned information fusion generation model can be Qwen-14B or ChatGLM3-6B.

[0084] Step 8: The above-mentioned change risk confidence level, the above-mentioned operation and maintenance automation adaptation value, the above-mentioned change analysis text, and the above-mentioned change control information are identified as operation and maintenance change detection information.

[0085] Step 9: Based on the aforementioned maintenance change detection information, perform maintenance control on the terminal device set corresponding to the target terminal information set. As an example, the implementation of this step can refer to the implementation of step 107, and will not be repeated here.

[0086] The above-mentioned technical solution and related content, as an inventive point of this disclosure, solves technical problem four: "reducing the implementation time and power consumption of terminal devices during maintenance changes, and improving the stability of terminal devices." Factors leading to prolonged implementation time, increased power consumption, and reduced stability of terminal devices during maintenance changes are often as follows: Because text-based maintenance change detection methods assess maintenance change information as a whole, they cannot perceive the execution dependencies between various operation steps in the implementation plan or the attribution relationships between each operation step and the target terminal device. This results in the generated maintenance change detection information failing to accurately reflect the risks caused by local operation steps in long-chain operations and the constraint conflicts between operation steps and terminal devices. Consequently, inaccurate maintenance control operations are performed on the terminal devices, leading to prolonged implementation time, increased power consumption, and reduced stability. Solving these factors can reduce the implementation time and power consumption of terminal devices during maintenance changes and improve their stability. To achieve this effect, this disclosure first performs semantic enhancement processing on the target maintenance change sample set to obtain an enhanced sample information set. Next, the aforementioned sample enhancement information set and the aforementioned change element information are subjected to multi-dimensional dynamic association encoding processing to obtain a change semantic association vector set. Then, the aforementioned change semantic association vector set undergoes multi-level feature decoding processing to obtain the change risk confidence score. Here, the obtained change risk confidence score can comprehensively reflect the probability of abnormal impact on terminal equipment after the implementation of operation and maintenance change information from both semantic and operational structure dimensions, providing a quantitative basis for subsequent judgment on whether the operation and maintenance change detection information meets preset control conditions. Next, based on the aforementioned change element information, an operation and maintenance change execution graph is constructed. Then, based on the aforementioned target terminal information set, the aforementioned operation and maintenance change execution graph undergoes heterogeneous constraint update processing to obtain a change constraint heterogeneous execution graph. Subsequently, the aforementioned change constraint heterogeneous execution graph undergoes multi-target scheduling mapping processing to obtain an operation and maintenance automation adaptation value. Here, the obtained operation and maintenance automation adaptation value integrates the execution dependencies between operation steps and the attribution constraints between operation steps and terminal equipment. Compared to the method of evaluating the operation and maintenance change information as a whole, it can more accurately reflect the degree of automation execution adaptation of each local operation step in a long chain operation. Next, based on the aforementioned change risk confidence level and the aforementioned operational automation adaptation value, the aforementioned change element information and the aforementioned target operational change sample set are subjected to information fusion processing to obtain change analysis text and change control information. Then, the aforementioned change risk confidence level, the aforementioned operational automation adaptation value, the aforementioned change analysis text, and the aforementioned change control information are determined as operational change detection information.Here, the obtained operation and maintenance change detection information integrates change risk confidence, operation and maintenance automation adaptation values, change analysis text, and change control information, providing multi-layered decision-making basis for subsequent operation and maintenance control of the terminal device set. Finally, based on the above operation and maintenance change detection information, operation and maintenance control is performed on the terminal device set corresponding to the target terminal information set. This multi-layered operation and maintenance change detection information-based operation and maintenance control of the terminal device set can reduce the occurrence of inaccurate operation and maintenance control operations on the terminal devices, reduce the change implementation time and power consumption of the terminal devices during operation and maintenance changes, and improve the stability of the terminal devices.

[0087] Step 107: In response to the operation and maintenance change detection information meeting the preset control conditions, generate terminal operation and maintenance control instructions based on the operation and maintenance change detection information, and perform operation and maintenance control on the terminal device set corresponding to the target terminal information set based on the terminal operation and maintenance control instructions.

[0088] In some embodiments, the execution entity may, in response to the operation and maintenance change detection information satisfying preset control conditions, generate terminal operation and maintenance control instructions based on the operation and maintenance change detection information, and perform operation and maintenance control on the terminal device set corresponding to the target terminal information set based on the terminal operation and maintenance control instructions. The preset control conditions may be conditions where the operation and maintenance automation adaptation value included in the operation and maintenance change detection information is greater than a preset control threshold (e.g., 0.8) and the change risk confidence level is less than a preset risk threshold (e.g., 0.3). The terminal operation and maintenance control instructions may be a set of parameters for controlling the state changes of the terminal device set. The operation and maintenance control may include, but is not limited to: operational status control (e.g., shutting down or restarting terminal devices), resource and performance control (e.g., rate limiting or load reduction of terminal devices), and configuration and version control (e.g., parameter distribution, certificate switching). In practice, the execution entity may first use the change control information included in the operation and maintenance change detection information as search conditions to perform a matching query on a preset operation and maintenance instruction library to obtain the terminal operation and maintenance control instructions. The aforementioned pre-defined maintenance instruction library can be a database that uses the combination of operation type and terminal device identifier in the change control information as the index key and the executable maintenance control instructions as the index value. Finally, the aforementioned terminal maintenance control instructions are executed to complete the maintenance control operation on the set of terminal devices.

[0089] Optionally, in response to the above-mentioned operation and maintenance change detection information not meeting the preset control conditions, the above-mentioned operation and maintenance change detection information is sent to the manual review client.

[0090] The above embodiments of this disclosure have the following beneficial effects: The terminal device operation and maintenance control method based on retrieval enhancement in some embodiments of this disclosure can reduce the implementation time and power consumption of terminal devices during operation and maintenance changes, and improve the stability of terminal devices. Specifically, the reasons for the extended implementation time, increased power consumption, and reduced stability of terminal devices during operation and maintenance changes are as follows: Due to the diverse sources and inconsistent expressions of operation and maintenance change information, the method based on the preset rule engine is difficult to cover the combination changes of different terminal models, versions, and complex scenarios, resulting in low accuracy of rule matching results and easy misjudgment or omission. This leads to inaccurate operation and maintenance control operations on terminal devices, causing unexpected abnormal operating states of terminal devices, requiring additional rollback operations or secondary change operations, thus extending the implementation time, increasing power consumption, and reducing stability of terminal devices during operation and maintenance changes. Based on this, the terminal device operation and maintenance control method based on retrieval enhancement in some embodiments of this disclosure can first, in response to receiving operation and maintenance change information for a set of terminal devices, obtain the terminal attribute information set at the current moment of receiving the aforementioned operation and maintenance change information. Here, the terminal attribute information set at the current moment is obtained only when maintenance change information is received, ensuring that the obtained terminal attribute information is the real-time status information at the current moment and avoiding deviations in subsequent constraint alignment processing due to outdated terminal attribute information. Secondly, the aforementioned maintenance change information undergoes element segmentation extraction processing to obtain the terminal identifier set to be changed and the change element information. Here, the inconsistently expressed maintenance change information is transformed into a structured terminal identifier set to be changed and change element information, reducing noise caused by differences in the format and ambiguity of the original maintenance change information, and providing standardized input for subsequent constraint alignment processing and retrieval-based construction processing. Thirdly, based on the aforementioned maintenance change information, constraint alignment processing is performed on the aforementioned terminal identifier set to be changed and the aforementioned terminal attribute information set to obtain the target terminal information set. Here, the target terminal information set can reflect the status information of the target terminal device at the current moment and within a preset time window, ensuring that the subsequently generated retrieval query information matches the maintenance change detection information and the actual status of the target terminal device, reducing the possibility of inaccurate maintenance control commands due to inconsistent terminal information. Next, the aforementioned changed element information and target terminal information set are processed using a retrieval-based construction method to obtain retrieval query information for the aforementioned operation and maintenance change information. Here, the obtained retrieval query information can simultaneously cover the semantic and keyword-level retrieval needs of operation and maintenance change information, providing accurate retrieval input for subsequent dual-channel fusion recall processing and improving the recall accuracy of the candidate operation and maintenance change sample set. Subsequently, based on a pre-set operation and maintenance retrieval database, the aforementioned retrieval query information is processed using a dual-channel fusion recall method to obtain the candidate operation and maintenance change sample set.Here, dual-channel fusion recall processing, compared to single-channel recall, reduces missed recalls and provides a more comprehensive candidate maintenance change sample set for subsequent multi-layer screening and enhancement processing. Then, multi-layer screening and enhancement processing is applied to the candidate maintenance change sample set to obtain the target maintenance change sample set. Maintenance change detection information is generated based on the aforementioned change element information, the target terminal information set, and the target maintenance change sample set. This multi-layer screening and enhancement processing filters out candidate maintenance change samples with semantic redundancy and contradictions. Finally, in response to the maintenance change detection information meeting preset control conditions, terminal maintenance control instructions are generated based on the maintenance change detection information, and maintenance control is performed on the terminal device set corresponding to the target terminal information set based on these instructions. By generating and executing terminal maintenance control instructions only when the maintenance change detection information meets preset control conditions, inaccurate maintenance control operations on terminal devices are avoided, reducing unexpected abnormal operating states of terminal devices and improving terminal device stability. Therefore, this retrieval-enhanced terminal device operation and maintenance control method can reduce the implementation time and power consumption of terminal devices during operation and maintenance changes, and improve the stability of terminal devices.

[0091] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a terminal device operation and maintenance control device based on retrieval enhancement. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this retrieval-enhanced terminal device operation and maintenance control device can be specifically applied to various electronic devices.

[0092] like Figure 2As shown, a terminal device operation and maintenance control device 200 based on retrieval enhancement includes: an acquisition unit 201, an extraction unit 202, an alignment unit 203, a construction unit 204, a recall unit 205, an enhancement unit 206, and a control unit 207. The acquisition unit 201 is configured to: in response to receiving operation and maintenance change information for a set of terminal devices, acquire a set of terminal attribute information at the current time of receiving the operation and maintenance change information; the extraction unit 202 is configured to: perform element segmentation extraction processing on the operation and maintenance change information to obtain a set of terminal identifiers to be changed and change element information; the alignment unit 203 is configured to: perform constraint alignment processing on the set of terminal identifiers to be changed and the set of terminal attribute information based on the operation and maintenance change information to obtain a target terminal information set; the construction unit 204 is configured to: perform retrieval-based construction processing on the change element information and the target terminal information set to obtain retrieval query information for the operation and maintenance change information; the recall unit 205... The system is configured to: perform dual-channel fusion recall processing on the above-mentioned retrieval query information according to a preset operation and maintenance retrieval database to obtain a candidate operation and maintenance change sample set; the enhancement unit 206 is configured to: perform multi-level screening enhancement processing on the above-mentioned candidate operation and maintenance change sample set to obtain a target operation and maintenance change sample set, and generate operation and maintenance change detection information according to the above-mentioned change element information, the above-mentioned target terminal information set and the above-mentioned target operation and maintenance change sample set; the control unit 207 is configured to: in response to the above-mentioned operation and maintenance change detection information satisfying preset control conditions, generate a terminal operation and maintenance control instruction according to the above-mentioned operation and maintenance change detection information, and perform operation and maintenance control on the terminal device set corresponding to the above-mentioned target terminal information set according to the above-mentioned terminal operation and maintenance control instruction.

[0093] It is understandable that the units described in the retrieval-enhanced terminal equipment operation and maintenance control device 200 are related to the reference. Figure 1 The steps described in the method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the retrieval-enhanced terminal equipment operation and maintenance control device 200 and the units contained therein, and will not be repeated here.

[0094] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device (e.g., an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0095] like Figure 3As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0096] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0097] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0098] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-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 of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, 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 device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0099] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0100] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: in response to receiving maintenance change information for a set of terminal devices, acquire a set of terminal attribute information at the current time of receiving the maintenance change information; perform element segmentation extraction processing on the maintenance change information to obtain a set of terminal identifiers to be changed and change element information; perform constraint alignment processing on the set of terminal identifiers to be changed and the set of terminal attribute information based on the maintenance change information to obtain a target terminal information set; and perform retrieval-based construction processing on the change element information and the target terminal information set to obtain the aforementioned maintenance change... The system retrieves and queries information; based on a preset maintenance retrieval database, it performs dual-channel fusion recall processing on the retrieved and queried information to obtain a candidate maintenance change sample set; it then performs multi-layer screening and enhancement processing on the candidate maintenance change sample set to obtain a target maintenance change sample set; and based on the change element information, the target terminal information set, and the target maintenance change sample set, it generates maintenance change detection information; in response to the maintenance change detection information satisfying preset control conditions, it generates terminal maintenance control instructions based on the maintenance change detection information, and performs maintenance control on the terminal device set corresponding to the target terminal information set based on the terminal maintenance control instructions.

[0101] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and 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 (e.g., via the Internet using an Internet service provider).

[0102] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0103] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an acquisition unit, an extraction unit, an alignment unit, a construction unit, a recall unit, an enhancement unit, and a control unit. The names of these units do not necessarily limit the specific unit itself; for example, an acquisition unit may also be described as "a unit that, in response to receiving maintenance change information for a set of terminal devices, acquires a set of terminal attribute information at the current moment of receiving the maintenance change information."

[0104] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0105] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A terminal device operation and maintenance control method based on retrieval enhancement, comprising: In response to receiving maintenance change information for a set of terminal devices, obtain the set of terminal attribute information at the current moment when the maintenance change information is received; The operation and maintenance change information is processed by element segmentation extraction to obtain the terminal identifier set to be changed and the change element information; Based on the maintenance change information, the set of terminal identifiers to be changed and the set of terminal attribute information are constrained and aligned to obtain the target terminal information set. The change element information and the target terminal information set are subjected to retrieval-based construction processing to obtain retrieval query information for the operation and maintenance change information; Based on the preset operation and maintenance retrieval database, the retrieval query information is processed by dual-channel fusion recall to obtain a candidate operation and maintenance change sample set; The candidate operation and maintenance change sample set is subjected to multi-level screening and enhancement processing to obtain the target operation and maintenance change sample set, and operation and maintenance change detection information is generated based on the change element information, the target terminal information set and the target operation and maintenance change sample set. In response to the maintenance change detection information meeting preset control conditions, a terminal maintenance control instruction is generated based on the maintenance change detection information, and maintenance control is performed on the terminal device set corresponding to the target terminal information set based on the terminal maintenance control instruction.

2. The method according to claim 1, wherein, The maintenance retrieval database is obtained through the following steps: The acquired set of operation and maintenance change cases is processed by text segmentation to obtain a set of change case segment information. An inverted index is constructed by segmenting the information set of the aforementioned change cases to obtain a keyword inverted index; The segmented information set of the change cases is input into the trained semantic coding model to obtain the segmented vector set of the change cases; An approximate nearest neighbor index is constructed from the segmented vector set of the change cases to obtain a semantic vector index; An operation and maintenance retrieval database is generated based on the segmented information set of the change cases, the keyword inverted index, and the semantic vector index.

3. The method according to claim 1, wherein, The step of performing element segmentation extraction on the operation and maintenance change information yields a set of terminal identifiers to be changed and change element information, including: The attachment document set included in the operation and maintenance change information is processed into text to obtain the attachment text set; Terminal entity recognition processing is performed on the attached text set and the change description text included in the operation and maintenance change information to obtain the terminal identifier set to be changed; Element field identification is performed on the attachment text set and the change description text to obtain a change element field information set; Semantic segmentation processing is performed on the attachment text set and the change description text included in the operation and maintenance change information to obtain a change segment sequence; The set of changed element field information and the changed segment sequence are determined as changed element information.

4. The method according to claim 1, wherein, The step of performing retrieval-based construction processing on the changed element information and the target terminal information set to obtain retrieval query information for the operation and maintenance change information includes: The changed element information and the target terminal information set are subjected to targeted extraction processing to obtain a retrieval feature item information set; A semantic query information set is generated based on the retrieval feature item information set and the change segment sequence included in the change element information; The search feature information set is weighted and filtered to obtain the target search feature information set, which serves as the keyword query information set. Field filtering is performed on the retrieval feature information set and the target terminal information set to obtain the terminal constraint information set; The keyword query information set, the semantic query information set, and the terminal constraint information set are determined as the retrieval query information.

5. The method according to claim 1, wherein, The step involves performing a dual-channel fusion recall process on the retrieval query information based on a preset operation and maintenance retrieval database to obtain a candidate operation and maintenance change sample set, including: Boolean compilation is performed on the terminal constraint information set included in the search query information to obtain the search filter expression; The semantic query information set included in the retrieval query information is vectorized to obtain a semantic query vector set; Based on the retrieval filtering expression, the semantic query vector set and the operation and maintenance retrieval database are subjected to multi-channel semantic recall processing to obtain a semantic recall sample set. The semantic recall sample set is merged and deduplicated to obtain a semantic candidate sample set; Based on the retrieval filtering expression, the keyword query information set included in the retrieval query information and the operation and maintenance retrieval database are subjected to lexical channel recall processing to obtain a lexical candidate sample set; The semantic candidate sample set and the lexical candidate sample set are sorted and fused to obtain a fused candidate sample set; The fusion candidate sample set and the acquired current time value are subjected to time decay matching processing to obtain a set of time decay degree values; Based on the time decay value set, the fusion candidate sample set is rearranged and filtered to obtain a candidate operation and maintenance change sample set.

6. The method according to claim 1, wherein, The step of performing multi-level screening and enhancement processing on the candidate operation and maintenance change sample set to obtain the target operation and maintenance change sample set includes: The candidate operation and maintenance change sample set is subjected to context completion processing to obtain the first operation and maintenance change sample set; The first maintenance change sample set is subjected to similarity clustering to remove duplicates, resulting in the second maintenance change sample set. The second maintenance change sample set is subjected to sample conflict identification processing to obtain the third maintenance change sample set; Based on the retrieval query information, the third maintenance change sample set is cross-coded and rearranged to obtain the target maintenance change sample set.

7. A terminal device operation and maintenance control device based on retrieval enhancement, comprising: The acquisition unit is configured to acquire the set of terminal attribute information at the current moment when the maintenance change information is received in response to receiving maintenance change information for a set of terminal devices. The extraction unit is configured to perform element segmentation extraction processing on the operation and maintenance change information to obtain the terminal identifier set to be changed and the change element information. The alignment unit is configured to perform constraint alignment processing on the terminal identifier set to be changed and the terminal attribute information set according to the operation and maintenance change information to obtain the target terminal information set; The construction unit is configured to perform retrieval-based construction processing on the changed element information and the target terminal information set to obtain retrieval query information for the operation and maintenance change information. The recall unit is configured to perform dual-channel fusion recall processing on the search query information based on a preset operation and maintenance retrieval database to obtain a candidate operation and maintenance change sample set. The enhancement unit is configured to perform multi-level screening and enhancement processing on the candidate operation and maintenance change sample set to obtain the target operation and maintenance change sample set, and to generate operation and maintenance change detection information based on the change element information, the target terminal information set and the target operation and maintenance change sample set. The control unit is configured to, in response to the maintenance change detection information meeting preset control conditions, generate a terminal maintenance control instruction based on the maintenance change detection information, and perform maintenance control on the terminal device set corresponding to the target terminal information set based on the terminal maintenance control instruction.

8. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images