A power distribution network maintenance intelligent cruise method oriented to full-node business flow transfer

Abstract

The application relates to a power distribution network maintenance intelligent cruise method for a whole-node business flow, and comprises the following specific steps: discrete business treatment nodes in the whole process of maintenance, including batch answering, ticket drafting, power-off operation, work permission, completion verification and power-on, are connected in series into a complete maintenance execution business cruise chain; the system initiates a telephone interaction process instead of a dispatcher, and develops telephone interaction of each link of the whole maintenance process with on-site personnel; through full-process automatic processing of input data processing, semantic vectorization, vector matching, thought chain construction, operation ticket generation and multi-layer checking, intelligent generation of power grid operation tickets is realized. The intelligent level is high, system data of OMS, SCADA, 5200 and the like can be automatically crawled, the consistency of data and on-site conditions is intelligently checked through intelligent voice recognition, data flow interaction and other functions, and closed-loop management of the power distribution network maintenance intelligent cruise task is autonomously completed.

Description

Technical Field

[0001] This invention relates to the field of distribution network maintenance, and more specifically, to a smart patrol method for distribution network maintenance oriented towards the business flow of all nodes. Background Technology

[0002] With the continuous expansion of the distribution network, the increasing penetration rate of distributed energy, and the frequent occurrence of extreme weather events, the structure of the distribution network is becoming increasingly complex, and its operating conditions are dynamically changing, placing higher demands on the accuracy, efficiency, and safety of maintenance and dispatching. Traditional distribution network maintenance and dispatching models are primarily manual, resulting in several technical bottlenecks: First, the maintenance process is highly fragmented, requiring manual completion of multiple discrete nodes such as maintenance ticket approval, operation ticket drafting and review, power outage and restoration operations, and commencement and completion confirmation. Data fragmentation across systems leads to delays in process connections, and the numerous manual interventions significantly extend outage time and maintenance cycles. Second, manual operation relies on experience-based judgment, which can easily lead to operational logic oversights or compliance deviations in complex scenarios such as parallel maintenance of multiple lines, network topology changes, and multi-power source coordination, resulting in insufficient safety and error prevention capabilities. Third, manual processing is inefficient; maintenance ticket analysis and operation ticket drafting are time-consuming, making it difficult to meet the needs of efficiently handling massive maintenance tasks. Furthermore, the high labor costs and significant labor shortages fail to meet the dispatching demands of the large-scale development of the distribution network. The traditional "manually-led, segmented" maintenance model can no longer meet the development needs of the intelligent transformation of the distribution network. There is an urgent need for a technical solution that can achieve standardized and automated progress throughout the entire maintenance process. Intelligent cruise technology, based on data flow interaction, the Guangming Power big data model, process engine, automated control, and intelligent voice recognition, breaks down the entire distribution network maintenance process into standardized nodes. Through multi-system data integration and intelligent algorithm-driven operation, it achieves automatic triggering of maintenance tasks, automatic process advancement, automatic operation verification, and automatic status monitoring. This effectively solves the problems of low efficiency, poor adaptability, and high safety risks associated with the traditional manual model. It provides full-chain automated support for distribution network maintenance scheduling and is a key technology for promoting the transformation of distribution network scheduling from "fragmented management and control" to "full-chain dynamic monitoring." It is of great significance for improving the reliability of power supply in the distribution network, reducing scheduling manpower costs, and ensuring the safe and stable operation of the power grid. Summary of the Invention

[0003] To address the aforementioned shortcomings or improvement needs of existing technologies, a smart patrol method for distribution network maintenance that is oriented towards full-node business flow is provided.

[0004] To achieve the above objectives, this application provides the following technical solution:

[0005] This application provides a method for intelligent patrol of distribution network maintenance oriented towards full-node business flow, including the following specific steps:

[0006] The various discrete business processing nodes in the entire process of maintenance, from approval, drafting, power outage operation, commencement permit, completion verification, and power restoration, are linked together into a complete maintenance execution business patrol chain.

[0007] The system replaces the dispatcher in initiating telephone interaction processes and conducts telephone interactions with on-site personnel at all stages of the maintenance process;

[0008] Through fully automated processing of input data, semantic vectorization, vector matching, thought chain construction, operation ticket generation, and multi-layer verification, intelligent generation of power grid operation tickets has been achieved.

[0009] The system initiates telephone interaction processes on behalf of the dispatcher, conducting telephone interactions with on-site personnel at each stage of the maintenance process.

[0010] Workflow Engine Driven: A full-node workflow model for intelligent maintenance patrols is built based on the Activiti workflow engine. Telephone interactions such as maintenance approval, pre-order notification, power outage / restoration monitoring authorization, and start / completion confirmation are preset as automated trigger nodes in the patrol process. Each node is configured with standardized execution rules of "trigger condition - interaction execution - result feedback - process jump". When the patrol process flows to the corresponding node and the preset trigger condition is met, the workflow engine automatically issues an interaction execution command, triggering the system's telephone interaction function without manual operation. After the interaction is completed, the result is fed back to the workflow engine, which automatically completes the node jump or anomaly warning based on the verification result.

[0011] Automatic control: It adopts a distributed automated control architecture, integrating a telephone interface call module, a voice processing module, a content verification module, and a business data backfilling module. Each module is linked with the process engine and the power distribution network scheduling core system through standardized interfaces to achieve fully automated closed-loop control of "command issuance - telephone call - voice interaction - data processing - business flow".

[0012] Voice interaction: The voice content processing adopts a four-step processing algorithm of voice denoising, voice recognition, text vectorization and rule verification to ensure accurate extraction of on-site feedback information and compliance judgment.

[0013] The specific process of the four-step processing algorithm—speech denoising, speech recognition, text vectorization, and rule verification—is as follows:

[0014] Speech denoising: A wavelet transform-based speech denoising algorithm is used to filter out environmental noise and line interference in on-site telephone interactions, retaining effective speech signals and making the speech data conform to the input standards for speech recognition.

[0015] Speech Recognition: Based on the State Grid Guangming large model and combined with the inference capabilities of the DeepSeek Qwen 32B model, a dedicated speech corpus is built for distribution network dispatching business. Lightweight fine-tuning technology is used for training to obtain a small dispatching speech recognition model. The denoised speech signal is mapped to "speech-text". The post-processing module performs standardization correction for professional terminology and equipment numbering to generate text that conforms to the dispatching business specifications. At the same time, it is equipped with optimized fault tolerance mechanisms such as sentence completion and fuzzy re-recognition. It also links with various modules of the patrol system through standardized interfaces and triggers manual intervention in case of abnormalities, so as to achieve accurate and real-time recognition of the interactive voice of distribution network dispatching telephone.

[0016] Text vectorization: First, the identified standardized scheduling text undergoes customized preprocessing for the power industry, completing professional and accurate word segmentation and core content word extraction. Then, the processed word sequence is input into a Word2Vec word embedding model optimized and trained with scheduling business corpus to generate 128-dimensional low-dimensional dense vectors for core words. After concatenation and normalization, a unified overall text feature vector is obtained. At the same time, special optimizations are made for the characteristics of scheduling text, constructing a word vector library dedicated to professional vocabulary, realizing semantic completion of short sentences, and performing dynamic vector weight allocation to strengthen the expression of core business information features. The output standardized feature vector will be transmitted to the content verification module for similarity calculation.

[0017] Rule verification: Using the feature vector output from text vectorization as input, a two-layer architecture of "vector matching + rule verification" is built. First, the cosine similarity of the interactive text feature vector with the core business data vector of the maintenance ticket / operation ticket is calculated to complete the consistency verification of business information. Then, relying on the structured scheduling rule engine built according to the scheduling business links, the text vector is searched and matched according to rules to verify whether it meets the requirements of mandatory confirmation, prohibited expressions, and standardized wording in the scheduling procedure. At the same time, optimization mechanisms such as dynamic threshold adjustment, dynamic update of the rule base, and precise location of anomaly causes are set. If all verifications pass, the subsequent automated business process is triggered. If any one fails, a multi-level anomaly warning is immediately triggered and the site is prompted to reconfirm. If multiple failures occur, the process is suspended and manual intervention is required.

[0018] The specific steps for speech denoising are as follows:

[0019] Noisy speech signals are decomposed into different scales using wavelet basis functions to achieve separation of noise and effective speech in the time-frequency domain; whereby... Noisy speech, For pure voice signals, The superimposed noise includes both ambient noise and line interference:

[0020] For noisy speech signals conduct The scaled discrete wavelet transform decomposes the components into low-frequency approximate components. and Layered high-frequency detail components ,in The wavelet decomposition scale is set to 3 in engineering practice. It is a low-pass decomposition filter. For high-pass decomposition filters, This is the low-frequency approximation component of the previous layer. The initial decomposition is into the original noisy signal:

[0021] ,

[0022] High-frequency detail components are filtered through threshold quantization, setting or attenuating the high-frequency coefficients corresponding to noise, while retaining the high-frequency coefficients corresponding to speech features. For symbolic functions, ; The threshold for unbiased risk estimation. , This represents the standard deviation of the noise.

[0023] ,

[0024] Perform inverse wavelet transform on the processed low-frequency approximation components and high-frequency detail components, iterating repeatedly until... The denoised clean speech signal is reconstructed. :

[0025] .

[0026] Through fully automated processing of input data, semantic vectorization, vector matching, thought chain construction, operation ticket generation, and multi-layer verification, intelligent generation of power grid operation tickets has been achieved. The specific steps are as follows:

[0027] Data Acquisition and Preprocessing: By collecting maintenance tickets and moderation slips, key elements such as equipment name, outage range, work content, and maintenance type are accurately extracted using large-scale model semantic understanding. The feeder to which the equipment belongs and the operation boundary are automatically identified to obtain work order demand data. Historical tickets are collected from the typical operation ticket case library and processed by large-scale model structure to extract business experience such as scheduling procedure constraints and equipment operation sequence to obtain historical experience data. Feeder diagrams of the same system power grid are collected, and a dynamic and computable power grid network model is established through the topology calculation engine to retain the core features of equipment connection relationships and electrical connectivity to obtain physical network structure data.

[0028] Semantic Vectorization: Structured historical tickets and pending maintenance tickets / method slips are transformed into 1024-dimensional dense semantic vectors according to unified rules, laying the foundation for vector matching. First, features are extracted from equipment entities, operation objects, power grid status, and scheduling rules, and each feature is vectorized. These features are then concatenated and weighted to form a single ticket vector, generating a 1024-dimensional vector. This single ticket vector is then standardized and transformed into structured JSON data. Using encoding, combination, and normalization rules completely consistent with historical tickets, a 1024-dimensional vector to be matched is generated.

[0029] Vector matching: Calculate the vector of the vote to be formed using the cosine similarity algorithm. With historical document vector Based on semantic distance, the top-3 historical vectors with a similarity ≥ 0.95 are retrieved. Then, the top-3 historical vector retrieval results are further filtered according to the maintenance service type to ensure consistency with the pending ticket service type.

[0030] ,

[0031] Mind Chain Construction: Using matched historical tickets as a reference, a structured, closed-loop operational mind chain is constructed. First, the top-3 historical tickets are analyzed and broken down into atomic operation steps, verification instructions, and dependencies. The steps are mapped with vectorized features to form standardized units, and the operation sequence logic is restored to obtain the historical reference chain. Then, based on the core elements of the pending ticket, information such as equipment and maintenance scope in the historical reference chain is automatically replaced with the actual information of the pending ticket to form the basic operational mind chain. Finally, the integrity of the basic chain is checked, and risk points such as power outage without power restoration, no phase calibration, and no restoration of the method are identified. Missing / unclosed risk points are automatically generated to fill compliance and complete the closed-loop repair of the mind chain.

[0032] Structured operation ticket generation, verification, and push: First, the thought chain is broken down into standardized instructions of "equipment name + operation action + target status," strictly solidifying the temporal dependencies between each instruction; second, the feasibility of the operation is verified based on a real-time power grid model, including equipment existence, power grid connection relationship, power outage range, and load transfer scheme verification, ensuring no equipment errors, range deviations, or overload / isolated grid operation issues; then, the mandatory logic of the dispatching procedure is matched line by line, including the step-by-step transition of equipment status, the operation sequence of primary / secondary equipment for power outage and restoration, and the verification of standardized operation timing for lines / buses; finally, the instruction sequence that passes the double verification is standardized and formatted to generate a structured operation ticket containing ticket number, power outage / restoration time, operation steps, risk warnings, and signature field, which is then pushed to the networked command system.

[0033] Compared with the prior art, the present invention has significant beneficial effects.

[0034] 1. Significantly improved operational efficiency: By connecting various discrete business processing nodes into a complete maintenance execution business patrol chain, the top layer provides a visual display of the entire process of core nodes and a processing interaction platform. The bottom layer integrates existing functional services through network commands at some processing nodes, reducing unnecessary operation steps and improving maintenance efficiency. 2. Significantly enhanced security: The system can handle multiple maintenance and control tasks simultaneously. Combined with intelligent voice recognition, it broadcasts power grid event reports and cross-references information, achieving multi-system information connectivity and avoiding potential risks such as operational errors and inadequate implementation of safety measures. This significantly improves security compared to traditional manual control. 3. High level of intelligence: The system can automatically crawl data from OMS, SCADA, 5200, and other systems. Through intelligent voice recognition and data stream interaction, it intelligently verifies the consistency between data and on-site conditions, autonomously completing the closed-loop management of distribution network maintenance intelligent patrol tasks. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 The overall patrol process for power distribution network maintenance;

[0038] Figure 2 Flowchart for intelligent ticket generation and push. Detailed Implementation

[0040] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0041] The terms “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0042] The terms “first,” “second,” etc., are used only to distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance, nor as requiring or implying any such actual relationship or order between these entities or operations.

[0043] like Figure 1 As shown in the embodiments of this application, a method for intelligent patrol of distribution network maintenance oriented towards full-node business flow is provided. The autonomous maintenance patrol, centered on "planned maintenance," connects the various discrete business handling nodes in the entire process of maintenance—from approval, ticket drafting, power outage operation, commencement permitting, completion verification, and power restoration—into a complete maintenance execution business patrol chain. The top layer provides a visual display and interactive platform for the entire process of core nodes. The bottom layer integrates existing functional services through network commands at some handling nodes, and simultaneously broadcasts power grid events and interacts with information through intelligent voice reminders. Based on topology analysis and operation rule reasoning technologies, intelligent ticketing is achieved; safety verification is provided based on intelligent error prevention technology; and networked automatic command is used to drive the flow of various businesses. Telephone recording and transcription technology enables the commencement and completion of maintenance business via telephone, achieving closed-loop management of autonomous driving tasks and effectively improving the efficiency of maintenance plan execution. Figure 1 This refers to the overall patrol process for power distribution network maintenance.

[0044] Interactive Input and Processing Approach

[0045] In this invention, the system replaces the dispatcher in initiating telephone interaction processes, conducting telephone interactions with on-site personnel at all stages of the maintenance process. The core relies on the node-driven capabilities of the process engine and automated control technology to achieve full automation of interaction triggering, voice processing, content verification, and process flow. It can complete on-site information exchange for key stages such as maintenance approval, pre-order notification, power outage / restoration monitoring authorization, and start / completion confirmation without human intervention, while ensuring the compliance, accuracy, and consistency of the interaction content with dispatch procedures. The following details the triggering logic of the interaction input, the technical solution for the entire processing process, and the specific implementation ideas for each interaction stage.

[0046] Driven by a process engine: A full-node workflow model for intelligent maintenance patrols is built based on the Activiti process engine. Telephone interactions such as maintenance approvals, pre-order notifications, power outage / restoration monitoring authorizations, and start / completion confirmations are preset as automated trigger nodes in the patrol process. Each node is configured with standardized execution rules of "trigger condition - interaction execution - result feedback - process jump". When the patrol process flows to the corresponding node and the preset trigger conditions are met, the process engine automatically issues an interaction execution command, triggering the system's telephone interaction function without manual operation. After the interaction is completed, the result is fed back to the process engine, which automatically completes node jumps or anomaly warnings based on the verification results. The specific trigger conditions for each node are as follows.

[0047]

[0048] Automatic Control: Employing a distributed automated control architecture, the system integrates a telephone interface call module, a voice processing module, a content verification module, and a business data backfilling module. Each module collaborates with the process engine and the distribution network dispatching core system (OMS, 5200, SCADA, and networked command system) through standardized interfaces to achieve fully automated closed-loop control from "command issuance - telephone call - voice interaction - data processing - business flow." For switches requiring remote control in the operation ticket, automatic remote control can be implemented in the distribution network dispatching system after on-site verification confirms remote control capability. Upon completion of each node, operations such as changing the status of the maintenance / operation ticket, backfilling fields, and streamlining process nodes are automatically performed.

[0049] Voice interaction: The voice content processing adopts a four-step processing algorithm of "voice denoising - speech recognition - text vectorization - rule verification" to ensure accurate extraction of on-site feedback information and compliance judgment.

[0050] 1. Speech Denoising: A wavelet transform-based speech denoising algorithm is employed to filter out environmental noise and line interference in live telephone interactions, preserving the effective speech signal and ensuring the speech data meets the input standards for speech recognition. The noisy speech signal is decomposed into different scales (frequency bands) using wavelet basis functions, achieving separation of noise and effective speech in the time-frequency domain; whereby... Noisy speech, For pure voice signals, For superimposed noise (including ambient noise and line interference):

[0051]

[0052] For noisy speech signals conduct The scaled discrete wavelet transform decomposes the components into low-frequency approximate components. and Layered high-frequency detail components ,in The wavelet decomposition scale is set to 3 in engineering practice. It is a low-pass decomposition filter. For high-pass decomposition filters, This is the low-frequency approximation component of the previous layer. The initial decomposition is into the original noisy signal:

[0053]

[0054]

[0055] High-frequency detail components are filtered through threshold quantization, setting or attenuating the high-frequency coefficients corresponding to noise, while retaining the high-frequency coefficients corresponding to speech features. For symbolic functions, ; The threshold for unbiased risk estimation. , This represents the standard deviation of the noise.

[0056]

[0057] Perform inverse wavelet transform on the processed low-frequency approximation components and high-frequency detail components, iterating repeatedly until... The denoised clean speech signal is reconstructed. :

[0058]

[0059] 2. Speech Recognition: Based on the State Grid Guangming large model and combined with the inference capabilities of the DeepSeek Qwen 32B model, a dedicated speech corpus is built for distribution network dispatching business. Lightweight fine-tuning technology is used for training to obtain a small dispatching speech recognition model. The denoised speech signal is mapped to "speech-text", and then the post-processing module performs standardized corrections such as professional terminology and equipment numbering to generate text that conforms to the dispatching business specifications. At the same time, it is equipped with optimized fault tolerance mechanisms such as sentence completion and fuzzy re-recognition. It also links with various modules of the patrol system through standardized interfaces and triggers manual intervention in case of anomalies, so as to achieve accurate and real-time recognition of the interactive voice of distribution network dispatching telephone.

[0060] 3. Text Vectorization: First, the identified standardized dispatch text undergoes customized preprocessing for the power industry, completing professional and accurate word segmentation and core content word extraction. Then, the processed word sequence is input into a Word2Vec word embedding model optimized and trained with dispatch business corpus to generate 128-dimensional low-dimensional dense vectors for core words. After concatenation and normalization, a unified overall text feature vector is obtained. At the same time, special optimizations are made for the characteristics of dispatch text, constructing a word vector library dedicated to professional vocabulary, realizing semantic completion of short sentences, and performing dynamic vector weight allocation to strengthen the expression of core business information features. The output standardized feature vector will be transmitted to the content verification module for similarity calculation.

[0061] 4. Rule Verification: Using the feature vectors output by text vectorization as input, a two-layer architecture of "vector matching + rule verification" is built. First, the cosine similarity of the interactive text feature vectors with the core business data vectors of the maintenance ticket / operation ticket is calculated to complete the consistency verification of business information. Then, relying on the structured scheduling rule engine built according to the scheduling business links, the text vectors are searched and matched according to rules to verify whether they meet the requirements of mandatory confirmation, prohibited expressions, and standardized wording in the scheduling procedures. At the same time, optimization mechanisms such as dynamic threshold adjustment, dynamic update of the rule base, and precise location of anomaly causes are set up. If all verifications pass, the subsequent automated business process is triggered. If any one fails, a multi-level anomaly warning is immediately triggered and the site is prompted to reconfirm. If multiple failures occur, the process is suspended and manual intervention is required.

[0062] Intelligent ticket generation and push

[0063] This module is used for the ticket drafting part of intelligent navigation, replacing manual ticket drafting by dispatchers. Through fully automated processing of input data, semantic vectorization, vector matching, thought chain construction, operation ticket generation, and multi-level verification, it realizes the intelligent generation of power grid operation tickets. The specific steps are as follows:

[0064] Data Acquisition and Preprocessing: This section collects maintenance tickets and moderation slips, and uses a large-scale model semantic understanding to accurately extract key elements such as equipment name, outage range, work content, and maintenance type. It automatically identifies the feeder and operation boundary of the equipment to obtain work order requirement data. Historical tickets are collected from the typical operation ticket case library, and after large-scale model structured processing, business experience such as scheduling procedure constraints and equipment operation sequence is extracted to obtain historical experience data. Feeder diagrams of the same system power grid are collected, and a dynamic and computable power grid network model is established through a topology calculation engine, retaining core features such as equipment connection relationships and electrical connectivity to obtain physical network structure data.

[0065] Semantic Vectorization: Structured historical tickets and pending maintenance tickets / method slips are transformed into 1024-dimensional dense semantic vectors according to unified rules, laying the foundation for vector matching. First, features are extracted from equipment entities, operation objects, power grid status, and scheduling rules, and each feature is vectorized. These features are then concatenated and weighted to form a single ticket vector, generating a 1024-dimensional vector (128-dimensional equipment entity + 256-dimensional operation characteristics + 56-dimensional power grid status + 384-dimensional safety constraint timing). Next, the single ticket vector is standardized and transformed into structured JSON data. Using encoding, combination, and normalization rules completely consistent with historical tickets, a 1024-dimensional matching vector is generated.

[0066] Vector matching: Calculate the vector of the vote to be formed using the cosine similarity algorithm. With historical document vector Based on semantic distance, the top-3 historical vectors with a similarity ≥ 0.95 are retrieved. Then, the top-3 historical vector retrieval results are further filtered according to the maintenance service type (line maintenance, relocation, business expansion, etc.) to ensure consistency with the pending ticket service type.

[0067]

[0068] Mindset Chain Construction: Using matched historical tickets as a reference, a structured, closed-loop operational mindset chain is constructed. First, the top-3 historical tickets are analyzed and broken down into atomic operation steps, verification instructions, and dependencies. The steps are mapped with vectorized features to form standardized units, and the operation sequence logic is restored to obtain the historical reference chain. Then, based on the core elements of the pending ticket, information such as equipment and maintenance scope in the historical reference chain is automatically replaced with the actual information of the pending ticket to form the basic operational mindset chain. Finally, the integrity of the basic chain is checked, and missing / unclosed risk points such as power outage without power restoration, no phase calibration, and no mode restoration are identified. Compliance filling steps are automatically generated to complete the closed-loop repair of the mindset chain.

[0069] Structured operation ticket generation, verification, and push: First, the thought chain is broken down into standardized instructions of "equipment name + operation action + target state," strictly solidifying the temporal dependencies between each instruction; second, the feasibility of the operation is verified based on a real-time power grid model, including equipment existence, power grid connection relationship, power outage range, and load transfer scheme verification, ensuring no equipment errors, range deviations, overload / isolated operation issues; then, the mandatory logic of the scheduling procedure is matched line by line, including step-by-step equipment state transitions (prohibiting cross-state operations), the operation sequence of primary / secondary equipment power outages and restorations, and standardized operation timing verification of lines / buses; finally, the instruction sequence that passes the double verification is standardized and formatted to generate a structured operation ticket containing ticket number, power outage / restoration time, operation steps, risk warnings, and signature field, which is then pushed to the networked command system.

[0070] The intelligent patrol technology for distribution network maintenance, oriented towards full-node business flow, described in this invention has the following significant advantages compared to existing technologies: 1. Significantly improved operational efficiency: By connecting various discrete business handling nodes into a complete maintenance execution business patrol chain, the top layer provides a visual display and interactive platform for the entire process of core nodes, while the bottom layer integrates existing functional services through network commands at some handling nodes, reducing unnecessary operation steps and improving maintenance efficiency. 2. Significantly improved security: The system can handle multiple maintenance and control tasks simultaneously. Combined with intelligent voice recognition, it broadcasts and cross-references power grid event reports, achieving information connectivity across multiple systems and avoiding potential risks such as operational errors and inadequate implementation of safety measures, significantly improving security compared to traditional manual control. 3. High level of intelligence: The system can automatically crawl data from systems such as OMS, SCADA, and 5200. Through intelligent voice recognition and data flow interaction, it intelligently verifies the consistency between data and on-site conditions, autonomously completing the closed-loop management of intelligent patrol tasks for distribution network maintenance.

[0071] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent patrol maintenance of distribution networks oriented towards full-node business flow, characterized in that, The specific steps include the following: The various discrete business processing nodes in the entire process of maintenance, from approval, drafting, power outage operation, commencement permit, completion verification, and power restoration, are linked together into a complete maintenance execution business patrol chain. The system replaces the dispatcher in initiating telephone interaction processes and conducts telephone interactions with on-site personnel at all stages of the maintenance process; Through fully automated processing of input data, semantic vectorization, vector matching, thought chain construction, operation ticket generation, and multi-layer verification, intelligent generation of power grid operation tickets has been achieved.

2. The intelligent patrol method for distribution network maintenance oriented towards full-node business flow as described in claim 1, characterized in that, The system initiates telephone interaction processes on behalf of the dispatcher, conducting telephone interactions with on-site personnel at each stage of the maintenance process. Process Engine Driven: Based on the Activiti process engine, a full-node flow model for intelligent maintenance patrol is built. Telephone interaction links such as maintenance approval, pre-order notification, power outage / power restoration monitoring authorization, and start / completion confirmation are preset as automated trigger nodes in the patrol process. Each node is configured with standardized execution rules of "trigger condition - interaction execution - result feedback - process jump". When the patrol process flows to the corresponding node and the preset trigger condition is met, the process engine automatically issues an interaction execution command to trigger the system's telephone interaction function without manual operation. After the interaction is completed, the result is fed back to the process engine, which automatically completes the node jump or anomaly warning based on the verification result; Automatic control: It adopts a distributed automated control architecture, integrating a telephone interface call module, a voice processing module, a content verification module, and a business data backfilling module. Each module is linked with the process engine and the power distribution network scheduling core system through standardized interfaces to achieve fully automated closed-loop control of "command issuance - telephone call - voice interaction - data processing - business flow". Voice interaction: The voice content processing adopts a four-step processing algorithm of voice denoising, voice recognition, text vectorization and rule verification to ensure accurate extraction of on-site feedback information and compliance judgment.

3. The intelligent patrol method for distribution network maintenance oriented towards full-node business flow as described in claim 2, characterized in that, The specific process of the four-step processing algorithm—speech denoising, speech recognition, text vectorization, and rule verification—is as follows: Speech denoising: A wavelet transform-based speech denoising algorithm is used to filter out environmental noise and line interference in on-site telephone interactions, retaining effective speech signals and making the speech data conform to the input standards for speech recognition. Speech Recognition: Based on the State Grid Guangming large model and combined with the inference capabilities of the DeepSeek Qwen 32B model, a dedicated speech corpus is built for distribution network dispatching business. Lightweight fine-tuning technology is used for training to obtain a small dispatching speech recognition model. The denoised speech signal is mapped to "speech-text". The post-processing module performs standardization correction for professional terminology and equipment numbers to generate text that conforms to the dispatching business specifications. At the same time, it is equipped with optimized fault tolerance mechanisms such as sentence completion and fuzzy re-recognition. It also links with various modules of the patrol system through standardized interfaces and triggers manual intervention in case of abnormalities, so as to achieve accurate and real-time recognition of the interactive voice of distribution network dispatching telephone. Text vectorization: First, the identified standardized scheduling text undergoes customized preprocessing for the power industry, completing professional and accurate word segmentation and core content word extraction. Then, the processed word sequence is input into a Word2Vec word embedding model optimized and trained with scheduling business corpus to generate 128-dimensional low-dimensional dense vectors for core words. After concatenation and normalization, a unified overall text feature vector is obtained. At the same time, special optimizations are made for the characteristics of scheduling text, constructing a word vector library dedicated to professional vocabulary, realizing semantic completion of short sentences, and performing dynamic vector weight allocation to strengthen the expression of core business information features. The output standardized feature vector will be transmitted to the content verification module for similarity calculation. Rule verification: Using the feature vector output from text vectorization as input, a two-layer architecture of "vector matching + rule verification" is built. First, the cosine similarity of the interactive text feature vector with the core business data vector of the maintenance ticket / operation ticket is calculated to complete the consistency verification of business information. Then, relying on the structured scheduling rule engine built according to the scheduling business links, the text vector is searched and matched according to rules to verify whether it meets the requirements of mandatory confirmation, prohibited expressions, and standardized wording in the scheduling procedure. At the same time, optimization mechanisms such as dynamic threshold adjustment, dynamic update of the rule base, and precise location of anomaly causes are set. If all verifications pass, the subsequent automated business process is triggered. If any one fails, a multi-level anomaly warning is immediately triggered and the site is prompted to reconfirm. If multiple failures occur, the process is suspended and manual intervention is required.

4. The intelligent patrol method for distribution network maintenance oriented towards full-node business flow as described in claim 3, characterized in that, The specific steps for speech denoising are as follows: Noisy speech signals are decomposed into different scales using wavelet basis functions to achieve separation of noise and effective speech in the time-frequency domain; whereby... Noisy speech, For pure voice signals, The superimposed noise includes both ambient noise and line interference: ， For noisy speech signals conduct The scaled discrete wavelet transform decomposes the components into low-frequency approximate components. and Layered high-frequency detail components ,in The wavelet decomposition scale is set to 3 in engineering practice. It is a low-pass decomposition filter. For high-pass decomposition filters, This is the low-frequency approximation component of the previous layer. The initial decomposition is into the original noisy signal: ， ， High-frequency detail components are filtered through threshold quantization, setting or attenuating the high-frequency coefficients corresponding to noise, while retaining the high-frequency coefficients corresponding to speech features. For symbolic functions, ; The threshold for unbiased risk estimation. , The standard deviation of the noise. ， Perform inverse wavelet transform on the processed low-frequency approximation components and high-frequency detail components, iterating repeatedly until... The denoised clean speech signal is reconstructed. : 。 5. The intelligent patrol method for distribution network maintenance oriented towards full-node business flow as described in claim 1, characterized in that, Through fully automated processing of input data, semantic vectorization, vector matching, thought chain construction, operation ticket generation, and multi-layer verification, intelligent generation of power grid operation tickets has been achieved. The specific steps are as follows: Data Acquisition and Preprocessing: By collecting maintenance tickets and moderation slips, key elements such as equipment name, outage range, work content, and maintenance type are accurately extracted using large-scale model semantic understanding. The feeder to which the equipment belongs and the operation boundary are automatically identified to obtain work order demand data. Historical tickets are collected from the typical operation ticket case library and processed by large-scale model structure to extract business experience such as scheduling procedure constraints and equipment operation sequence to obtain historical experience data. Feeder diagrams of the same system power grid are collected, and a dynamic and computable power grid network model is established through the topology calculation engine to retain the core features of equipment connection relationships and electrical connectivity to obtain physical network structure data. Semantic Vectorization: Structured historical tickets and pending maintenance tickets / method slips are transformed into 1024-dimensional dense semantic vectors according to unified rules, laying the foundation for vector matching. First, features are extracted from equipment entities, operation objects, power grid status, and scheduling rules, and each feature is vectorized. These features are then concatenated and weighted to form a single ticket vector, generating a 1024-dimensional vector. This single ticket vector is then standardized and transformed into structured JSON data. Using encoding, combination, and normalization rules completely consistent with historical tickets, a 1024-dimensional vector to be matched is generated. Vector matching: Calculate the vector of the vote to be formed using the cosine similarity algorithm. With historical document vector Based on semantic distance, the top-3 historical vectors with a similarity ≥ 0.95 are retrieved. Then, the top-3 historical vector retrieval results are further filtered according to the maintenance service type to ensure consistency with the pending ticket service type. ， Mind Chain Construction: Using matched historical tickets as a reference, a structured, closed-loop operational mind chain is constructed. First, the top-3 historical tickets are analyzed and broken down into atomic operation steps, verification instructions, and dependencies. The steps are mapped with vectorized features to form standardized units, and the operation sequence logic is restored to obtain the historical reference chain. Then, based on the core elements of the pending ticket, information such as equipment and maintenance scope in the historical reference chain is automatically replaced with the actual information of the pending ticket to form the basic operational mind chain. Finally, the integrity of the basic chain is checked, and risk points such as power outage without power restoration, no phase calibration, and no restoration of the method are identified. Missing / unclosed risk points are automatically generated to fill compliance and complete the closed-loop repair of the mind chain. Structured operation ticket generation, verification, and push: First, the thought chain is broken down into standardized instructions of "equipment name + operation action + target status," strictly solidifying the temporal dependencies between each instruction; second, the feasibility of the operation is verified based on a real-time power grid model, including equipment existence, power grid connection relationship, power outage range, and load transfer scheme verification, ensuring no equipment errors, range deviations, or overload / isolated grid operation issues; then, the mandatory logic of the dispatching procedure is matched line by line, including the step-by-step transition of equipment status, the operation sequence of primary / secondary equipment for power outage and restoration, and the verification of standardized operation timing for lines / buses; finally, the instruction sequence that passes the double verification is standardized and formatted to generate a structured operation ticket containing ticket number, power outage / restoration time, operation steps, risk warnings, and signature field, which is then pushed to the networked command system.

Images