Power supply service order intelligent monitoring and early warning method and system based on large model

Abstract

The present application relates to the technical field of electric data processing, in particular to a power supply service order intelligent monitoring and early warning method and system based on a large model. The method comprises the following steps: filtering a plurality of executed orders according to the similarity between the address vocabulary of a to-be-processed order and the address vocabulary of the plurality of executed orders to obtain a plurality of reference orders corresponding to the to-be-processed order; analyzing the content repetition degree between the to-be-processed order and the plurality of reference orders corresponding thereto to obtain a plurality of repetition values corresponding to the to-be-processed order; judging the order merging of the to-be-processed order and the plurality of reference orders corresponding thereto based on the plurality of repetition values corresponding to the to-be-processed order to obtain a judgment result of the to-be-processed order; and in the case that the judgment result indicates risk identification, using a large model to identify the safety risk of the to-be-processed order to obtain a risk identification result. The present application can improve the accuracy of power supply service order intelligent monitoring and early warning.

Description

Technical Field

[0001] This invention relates to the field of electrical data processing technology, specifically to a method and system for intelligent monitoring and early warning of power supply service work orders based on a large model. Background Technology

[0002] By identifying (monitoring) the fault types and urgency levels in power supply service work orders, user requests (such as power outages and voltage anomalies) can be identified in a timely manner, providing a basis for subsequent risk warnings and work order dispatch. However, with the expansion of the power network and the increasing complexity of the operating environment, the number of work orders and fault types has surged, making them increasingly difficult to handle.

[0003] In related technologies, the main methods for identifying fault event types and urgency in power supply service work orders are manual screening by staff or keyword matching. However, manual screening is prone to omissions, and keyword matching cannot recognize the complex semantics and emotions in the work order content, making it difficult to identify fault types and urgency levels. This leads to missed identifications and misjudgments, and it cannot distinguish whether the current work order is duplicated with an existing work order, resulting in duplicate dispatching of multiple work orders triggered by the same fault event, causing resource waste.

[0004] This shows that the accuracy of intelligent monitoring and early warning of power supply service work orders in related technologies is relatively low. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for intelligent monitoring and early warning of power supply service work orders based on a large model, in order to solve the technical problem of low accuracy in intelligent monitoring and early warning of power supply service work orders in related technologies.

[0006] In a first aspect, embodiments of the present invention provide a method for intelligent monitoring and early warning of power supply service work orders based on a large model, the method comprising: Based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple executed work orders, multiple executed work orders are filtered to obtain multiple reference work orders corresponding to the work order to be processed. Among them, the reference work order is the executed work order that points to the same work area as the work order to be processed among the multiple executed work orders. The degree of content duplication between the work order to be processed and its corresponding multiple reference work orders is analyzed to obtain multiple duplication values ​​corresponding to the work order to be processed. The duplication value indicates the degree of similarity between the fault problems between the corresponding reference work order and the work order to be processed. Based on multiple duplicate values ​​corresponding to the work order to be processed, the work order to be processed and its corresponding multiple reference work orders are merged and judged to obtain the judgment result of the work order to be processed. The judgment result indicates that the work order to be processed should not be dispatched or should be identified for safety risks. When the judgment result indicates that risk identification is required, a large model is used to identify the safety risks of the work order to be processed, and the risk identification result is obtained. The risk identification result indicates whether the work order to be processed is an urgent work order or a non-urgent work order, and the dispatch priority of urgent work orders is higher than that of non-urgent work orders.

[0007] In one embodiment, the step of filtering multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders to obtain multiple reference work orders corresponding to the work order to be processed includes: Based on the address words of the work order to be processed and the address words of multiple executed work orders, similarity matching is performed to obtain multiple address similarity values ​​corresponding to the work order to be processed. Based on the differences in the similarity values ​​and similarity thresholds of multiple addresses corresponding to the work order to be processed, multiple execution work orders are filtered to obtain multiple reference work orders corresponding to the work order to be processed.

[0008] In one embodiment, analyzing the degree of content duplication between the work order to be processed and its corresponding multiple reference work orders to obtain multiple duplication values ​​corresponding to the work order to be processed includes: Analyze the semantic similarity between the work order to be processed and the corresponding multiple reference work orders to obtain multiple semantic similarity values ​​corresponding to the work order to be processed, wherein the semantic similarity values ​​correspond one-to-one with the reference work orders; Analyze the risk similarity between the work order to be processed and the corresponding multiple reference work orders to obtain multiple risk similarity values ​​corresponding to the work order to be processed, wherein the risk similarity values ​​correspond one-to-one with the reference work orders; Multiple semantic similarity values ​​and multiple risk similarity values ​​corresponding to the work order to be processed are fused to obtain multiple duplicate values, wherein each duplicate value corresponds one-to-one with a reference work order.

[0009] In one embodiment, analyzing the semantic similarity between the work order to be processed and its corresponding multiple reference work orders to obtain multiple semantic similarity values ​​corresponding to the work order to be processed includes: A threshold segmentation algorithm is used to filter multiple service request terms in the work order to be processed, thereby obtaining multiple target service request terms for the work order to be processed. Feature extraction is performed on multiple target service request words of the work order to be processed to obtain multiple word vectors of the work order to be processed; The weights of multiple target service request terms and multiple word vectors of the work order to be processed are weighted to obtain the service request vector of the work order to be processed. The similarity between the service request vector of the pending work order and the service request vector of the target reference work order is analyzed to obtain the semantic similarity value between the pending work order and the target reference work order. The target reference work order is any one of multiple reference work orders, and the multiple semantic similarity values ​​corresponding to the pending work order are composed of the semantic similarity values ​​between the pending work order and each reference work order.

[0010] In one embodiment, analyzing the risk similarity between the work order to be processed and the corresponding multiple reference work orders to obtain multiple risk similarity values ​​corresponding to the work order to be processed includes: Analyze the similarity between the first word and the second word in multiple word pairs to obtain multiple word similarity values. The first word is the target service request word of the work order to be processed, and the second word is the service request word of the target reference work order. The word similarity value corresponds one-to-one with the word pair. Any word pair includes a first word and a second word, and at least one word is different between any two word pairs. Based on the relative size of multiple word similarity values, multiple word pairs are filtered to obtain target word pairs, wherein the target word pair is the word pair corresponding to the word pair with the largest word similarity value among multiple word similarity values; Obtain the context word vectors corresponding to the first word and the second word in the target word pair; By analyzing the similarity between the context word vectors corresponding to the first word and the context word vectors corresponding to the second word, the risk similarity value between the work order to be processed and the target reference work order is obtained. The multiple risk similarity values ​​corresponding to the work order to be processed are composed of the risk similarity values ​​between the work order to be processed and each reference work order.

[0011] In one embodiment, the step of merging the work order to be processed with its corresponding multiple reference work orders based on multiple duplicate values ​​corresponding to the work order to be processed, to obtain the judgment result of the work order to be processed, includes: The degree of content duplication between the work order to be processed and its multiple associated work orders is analyzed to obtain multiple baseline duplication values ​​corresponding to the work order to be processed. The associated work orders are used to characterize work order samples in the historical work order database that have been confirmed to belong to the same fault problem as the work order to be processed or point to the same work area as the work order to be processed. The repetition threshold is obtained by averaging the multiple benchmark repetition values. If multiple duplicate values ​​of the pending work order are greater than the duplicate threshold, the judgment result of the pending work order indicates that no work order will be dispatched. If any duplicate value in the multiple duplicate values ​​of the pending work order is less than the duplicate threshold, the work order is filtered according to the identifier words of the pending work order and the identifier words of multiple reference work orders to obtain the work order filtering result. The filtering result indicates that the pending work order has a corresponding candidate similar work order, or indicates that there is no corresponding candidate similar work order. The candidate similar work order is the reference work order that points to the same user as the pending work order among the multiple reference work orders. If the screening result indicates that there is a corresponding candidate similar work order for the work order to be processed, it is determined that the judgment result indicates that no work order should be dispatched. If the screening result indicates that there is no corresponding candidate similar work order for the work order to be processed, the judgment result indicates that a safety risk identification should be performed.

[0012] In one embodiment, the step of screening candidate similar work orders includes: Based on the degree of matching between the identifier words of the work order to be processed and the identifier words of multiple reference work orders corresponding to the work order to be processed, the multiple reference work orders corresponding to the work order to be processed are filtered to obtain the filtering results, wherein the identifier words of the candidate similar work orders are the same as the identifier words of the work order to be processed.

[0013] In one embodiment, the process of using a large model to identify security risks in the work order to obtain risk identification results includes: A large language model classifier is used to identify safety risks based on the target service request vocabulary of the work order to be processed and the standard service request vocabulary in the power professional knowledge base, and the risk identification results are obtained.

[0014] In one embodiment, before filtering multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders to obtain multiple reference work orders corresponding to the work order to be processed, the method further includes: The original work orders to be processed are preprocessed to obtain work orders to be processed. The preprocessing includes at least: text extraction processing, text error correction processing and word segmentation processing. Named entity recognition is performed on the work order to be processed to obtain the set of named entities corresponding to the work order to be processed. The set of named entities includes the address entity, date entity, identifier entity and service request entity of the work order to be processed. The service request entity is used to characterize the service request type and urgency.

[0015] Secondly, another embodiment of the present invention provides a power supply service work order intelligent monitoring and early warning system based on a large model, the system comprising: The work order filtering module is used to filter multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders, and obtain multiple reference work orders corresponding to the work order to be processed. The reference work orders are the execution work orders that point to the same work area as the work order to be processed among the multiple execution work orders. The similarity analysis module is used to analyze the degree of content duplication between the work order to be processed and the corresponding multiple reference work orders, and obtain multiple duplication values ​​corresponding to the work order to be processed. The duplication value indicates the degree of similarity between the fault problems between the corresponding reference work order and the work order to be processed. The merging judgment module is used to perform a merging judgment on the work order to be processed and its corresponding multiple reference work orders based on multiple duplicate values ​​corresponding to the work order to be processed, and to obtain the judgment result of the work order to be processed. The judgment result indicates that the work order to be processed should not be dispatched or should be identified for safety risks. The risk identification module is used to identify the safety risks of the work order to be processed using a large model when the judgment result indicates that risk identification is required. The risk identification result indicates whether the work order to be processed is an urgent work order or a non-urgent work order. The dispatch priority of urgent work orders is higher than that of non-urgent work orders.

[0016] The present invention has the following beneficial effects: This invention filters multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders, obtaining multiple reference work orders corresponding to the work order to be processed. Then, it analyzes the degree of content duplication between the work order to be processed and the corresponding multiple reference work orders, obtaining multiple duplicate values ​​corresponding to the work order to be processed. Based on this, it performs work order merging judgment on the work order to be processed and its corresponding multiple reference work orders, obtaining the judgment result of the work order to be processed. When the judgment result indicates risk identification, a large model is used to identify the safety risks of the work order to be processed, obtaining the risk identification result. On the one hand, it realizes the identification of fault problems in the work order to be processed through address filtering and duplicate value matching, and can identify the fault problems of execution work orders that duplicate the work order to be processed as the fault problems of the work order to be processed, improving the accuracy of identifying fault problems in the work order to be processed. On the other hand, merging judgment through duplicate values ​​avoids duplicate dispatching, improving the utilization efficiency of dispatching resources. Furthermore, through safety risk identification, it further determines the urgency of the fault in the work order to be processed, improving the response efficiency of urgent work orders. Therefore, this invention can improve the accuracy of intelligent monitoring and early warning of power supply service work orders. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an intelligent monitoring and early warning method for power supply service work orders based on a large model, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a power supply service work order intelligent monitoring and early warning system based on a large model, provided in an embodiment of the present invention. Detailed Implementation

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The specific scheme of the intelligent monitoring and early warning method and system for power supply service work orders based on a large model provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0020] This invention proposes an intelligent monitoring and early warning method for power supply service work orders based on a large model. The method provided by this invention can be executed by an intelligent monitoring and early warning system for power supply service work orders based on a large model, specifically by the controller within that system. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of an intelligent monitoring and early warning method for power supply service work orders based on a large model, according to an embodiment of the present invention. The method includes the following steps: Step S1: Based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple executed work orders, filter the multiple executed work orders to obtain multiple reference work orders corresponding to the work order to be processed.

[0021] Among them, the reference work order is the work order that points to the same work area as the work order to be processed among multiple work orders. The work order to be processed is the work order that is being processed but has not yet been settled. In other words, the work order to be processed is the work order that has been dispatched but has not yet been completed.

[0022] Specifically, the step of filtering multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders to obtain multiple reference work orders corresponding to the work order to be processed includes: Based on the address words of the work order to be processed and the address words of multiple executed work orders, similarity matching is performed to obtain multiple address similarity values ​​corresponding to the work order to be processed. Based on the differences in the similarity values ​​and similarity thresholds of multiple addresses corresponding to the work order to be processed, multiple execution work orders are filtered to obtain multiple reference work orders corresponding to the work order to be processed.

[0023] The controller can standardize the address vocabulary of pending work orders and multiple executed work orders into a standardized address format. This avoids discrepancies in similarity calculations caused by different ways of expressing the same address (such as "Building No. 3" and "Building 3"). It also prevents incorrect matching of different addresses due to similar expressions (e.g., both being Zhongshan Road but belonging to different districts). This improves the accuracy of address similarity calculations. For example, the controller can standardize address vocabulary into the format of province + city + district + street + community + building + unit. Furthermore, it can unify "Building No. XX", "Building XX", and "Building XX" into "Building XX".

[0024] Subsequently, the controller obtains the word vectors of the address words of the work order to be processed and the word vectors of the address words of each executed work order, and calculates the cosine similarity value between the word vectors of the address words of the work order to be processed and the word vectors of the address words of each executed work order. This cosine similarity value is the address similarity value.

[0025] The controller pre-stores a similarity threshold, which ranges from [0.8, 0.9], for example, a similarity threshold of 0.85. The controller compares each address similarity value with the similarity threshold. If the address similarity value is greater than or equal to the similarity threshold, the controller identifies the execution work order corresponding to that address similarity value as a reference work order for the work order to be processed. If the address similarity value is less than the similarity threshold, the controller determines that the execution work order corresponding to that address similarity value is not a reference work order for the work order to be processed.

[0026] In this embodiment, before filtering multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders to obtain multiple reference work orders corresponding to the work order to be processed, the controller may also: The original work orders to be processed are preprocessed to obtain work orders to be processed. The preprocessing includes at least: text extraction processing, text error correction processing and word segmentation processing. Named entity recognition is performed on the work order to be processed to obtain the set of named entities corresponding to the work order to be processed. The set of named entities includes the address entity, date entity, identifier entity and service request entity of the work order to be processed. The service request entity is used to characterize the service request type and urgency.

[0027] The controller pre-stores a named entity recognition (NER) algorithm and a rule-based error correction algorithm. The controller can use the NER algorithm to extract text from the original work order to be processed, and use the rule-based error correction algorithm to correct the text extracted by the NER algorithm. Subsequently, the controller can call the Jieba word segmentation tool to segment the text after text correction to obtain multiple words of the work order to be processed.

[0028] The controller can use the NER algorithm to obtain the named entity set of the work orders to be processed. Then, based on this named entity set, it extracts the corresponding entity terms from the work orders to determine the address terms, date terms, identifier terms, and service request terms for the work orders. It is understandable that the controller obtains the address terms, date terms, identifier terms, and service request terms for multiple execution work orders in a similar way to it obtains the address terms, date terms, identifier terms, and service request terms for the work orders to be processed; therefore, this will not be elaborated upon here.

[0029] In this embodiment, if there is only one execution work order, the controller can also use the method in this step to determine whether the execution work order is a reference work order for the work order to be processed.

[0030] Step S2: Analyze the degree of content duplication between the work order to be processed and the corresponding multiple reference work orders to obtain multiple duplication values ​​corresponding to the work order to be processed.

[0031] The repetition value indicates the degree of similarity between the fault issues in the corresponding reference work order and the pending work order. The degree of content repetition can be represented by the degree of repetition of service request terms, such as: power outage, meter, burnout, need, repair, inspection and processing, etc.

[0032] Specifically, the analysis of the degree of content duplication between the work order to be processed and its corresponding multiple reference work orders, to obtain multiple duplication values ​​corresponding to the work order to be processed, includes: Analyze the semantic similarity between the work order to be processed and the corresponding multiple reference work orders to obtain multiple semantic similarity values ​​corresponding to the work order to be processed, wherein the semantic similarity values ​​correspond one-to-one with the reference work orders; Analyze the risk similarity between the work order to be processed and the corresponding multiple reference work orders to obtain multiple risk similarity values ​​corresponding to the work order to be processed, wherein the risk similarity values ​​correspond one-to-one with the reference work orders; Multiple semantic similarity values ​​and multiple risk similarity values ​​corresponding to the work order to be processed are fused to obtain multiple duplicate values, wherein each duplicate value corresponds one-to-one with a reference work order.

[0033] This invention identifies reference work orders that overlap with the work order to be processed by analyzing the semantic and risk similarity between the work order to be processed and multiple corresponding reference work orders. This allows for a more comprehensive assessment of the actual overlap relationship between the work order to be processed and the reference work orders. Furthermore, it can determine the fault problems of the work order to be processed based on the fault problems (including fault type and fault urgency) of the reference work orders that overlap with the work order to be processed, thereby improving the accuracy of identifying the fault problems of the work order to be processed.

[0034] The step of analyzing the semantic similarity between the work order to be processed and its corresponding multiple reference work orders, and obtaining multiple semantic similarity values ​​corresponding to the work order to be processed, includes: A threshold segmentation algorithm is used to filter multiple service request terms in the work order to be processed, thereby obtaining multiple target service request terms for the work order to be processed. Feature extraction is performed on multiple target service request words of the work order to be processed to obtain multiple word vectors of the work order to be processed; The weights of multiple target service request terms and multiple word vectors of the work order to be processed are weighted to obtain the service request vector of the work order to be processed. The similarity between the service request vector of the pending work order and the service request vector of the target reference work order is analyzed to obtain the semantic similarity value between the pending work order and the target reference work order. The target reference work order is any one of multiple reference work orders, and the multiple semantic similarity values ​​corresponding to the pending work order are composed of the semantic similarity values ​​between the pending work order and each reference work order.

[0035] The controller pre-stores threshold segmentation algorithms, such as Otsu's method, and term frequency-inverse document frequency (TF-IDF) algorithms. The controller can use the TF-IDF algorithm to calculate the weight value of each service request term in the work order to be processed, mapping each discrete text term (i.e., service request term) to a one-dimensional numerical value representing its importance. Thus, multiple service request terms in the work order to be processed can be converted into a numerical sequence composed of multiple one-dimensional values. Subsequently, the controller can use Otsu's method to perform binarization segmentation on this numerical sequence to obtain a weight threshold. The controller can then filter the multiple service request terms in the work order to be processed based on this weight threshold, removing action modifiers and interference words to obtain multiple target service request terms for the work order to be processed. Specifically, the controller can remove service request terms with weight values ​​below the weight threshold, and determine service request terms with weight values ​​greater than or equal to the weight threshold as the target service request terms for the work order to be processed. Action modifiers include, for example, "repair," "inspect," "need," and "perform." Distractor words include: downstairs, neighbor, should, and possibly.

[0036] Subsequently, the controller can invoke the BERT model and input multiple target service request terms into it, obtaining multiple word vectors output by the BERT model. For any target service request term, the controller can count the number of times it appears in the pending work orders and count the total number of all target service request terms. The controller can determine the weight of the target service request term by the ratio of the number of occurrences to the total number. Then, the controller can obtain the word vectors of multiple target service request terms and their weights, and perform weighted processing to obtain the service request vector of the pending work order. For example, if the service request terms of the pending work order include "power outage" and "smoke," where the word vector for "power outage" is [0.8, 0.2] and the word vector for "smoke" is [0.1, 0.9], and the weights for both "power outage" and "smoke" are 0.5, then the service request vector of the pending work order is [0.8×0.5+0.1×0.5, 0.2×0.5+0.9×0.5], which is [0.45, 0.55]. Similarly, the controller can obtain the service request vector for each reference work order in the manner described above.

[0037] In one example, the controller can calculate the cosine similarity value between the service request vector of the work order to be processed and the service request vector of the target reference work order. This cosine similarity value is the semantic similarity value between the service request vector of the work order to be processed and the service request vector of the target reference work order.

[0038] The step of analyzing the risk similarity between the work order to be processed and its corresponding multiple reference work orders to obtain multiple risk similarity values ​​corresponding to the work order to be processed includes: Analyze the similarity between the first word and the second word in multiple word pairs to obtain multiple word similarity values. The first word is the target service request word of the work order to be processed, and the second word is the service request word of the target reference work order. The word similarity value corresponds one-to-one with the word pair. Any word pair includes a first word and a second word, and at least one word is different between any two word pairs. Based on the relative size of multiple word similarity values, multiple word pairs are filtered to obtain target word pairs, wherein the target word pair is the word pair corresponding to the word pair with the largest word similarity value among multiple word similarity values; Obtain the context word vectors corresponding to the first word and the second word in the target word pair; By analyzing the similarity between the context word vectors corresponding to the first word and the context word vectors corresponding to the second word, the risk similarity value between the work order to be processed and the target reference work order is obtained. The multiple risk similarity values ​​corresponding to the work order to be processed are composed of the risk similarity values ​​between the work order to be processed and each reference work order.

[0039] The controller can permutate and combine multiple target service request terms from a work order to be processed and multiple service request terms from a target reference work order to obtain multiple term pairs. For example, the multiple target service request terms of the work order to be processed include: term 1 and term 2, and the multiple service request terms of the target reference work order include: term A and term B, where term 1 and term 2 are both first terms, and term A and term B are both second terms. The controller can obtain four term pairs corresponding to the work order to be processed and the target reference work order, including: (term 1, term A), (term 1, term B), (term 2, term A), and (term 2, term B).

[0040] Similar to the previous example, for each word pair, the controller can calculate the cosine similarity value between the first and second words in the pair. This cosine similarity value is the word similarity value of the word pair. Subsequently, the controller can compare multiple word similarity values ​​and determine the word pair corresponding to the largest word similarity value as the target word pair.

[0041] The controller can acquire a preset number of characters from both sides of the first word in the target word pair, and combine the acquired characters to obtain the context window text of the first word. Subsequently, the controller can obtain the word vector of this context window text using the BERT model; this word vector is the context word vector corresponding to the first word. The preset number is pre-stored by the controller, preferably 8. Similarly, the controller can acquire the context word vector corresponding to the second word using the same method as acquiring the context word vector corresponding to the first word. Then, the controller can calculate the cosine similarity value between the context word vectors corresponding to the first and second words; this cosine similarity value is the risk similarity value between the work order to be processed and the target reference work order.

[0042] In one example, the controller pre-stores a first weight for semantic similarity and a second weight for risk similarity; for example, the first weight is 0.7 and the second weight is 0.3. The controller can calculate the duplicate value between the work order to be processed and the target reference work order by performing a weighted summation based on the semantic similarity, risk similarity, first weight, and second weight.

[0043] Step S3: Based on the multiple duplicate values ​​corresponding to the work order to be processed, perform work order merging judgment on the work order to be processed and its multiple corresponding reference work orders to obtain the judgment result of the work order to be processed.

[0044] The judgment result indicates that the work order to be processed should not be dispatched or that a safety risk identification should be performed.

[0045] Specifically, the step of merging the work order to be processed with its corresponding multiple reference work orders based on multiple duplicate values ​​corresponding to the work order to be processed, to obtain the judgment result of the work order to be processed, includes: The degree of content duplication between the work order to be processed and its multiple associated work orders is analyzed to obtain multiple baseline duplication values ​​corresponding to the work order to be processed. The associated work orders are used to characterize work order samples in the historical work order database that have been confirmed to belong to the same fault problem as the work order to be processed or point to the same work area as the work order to be processed. The repetition threshold is obtained by averaging the multiple benchmark repetition values. If multiple duplicate values ​​of the pending work order are greater than the duplicate threshold, the judgment result of the pending work order indicates that no work order will be dispatched. If any duplicate value in the multiple duplicate values ​​of the pending work order is less than the duplicate threshold, the work order is filtered according to the identifier words of the pending work order and the identifier words of multiple reference work orders to obtain the work order filtering result. The filtering result indicates that the pending work order has a corresponding candidate similar work order, or indicates that there is no corresponding candidate similar work order. The candidate similar work order is the reference work order that points to the same user as the pending work order among the multiple reference work orders. If the screening result indicates that there is a corresponding candidate similar work order for the work order to be processed, it is determined that the judgment result indicates that no work order should be dispatched. If the screening result indicates that there is no corresponding candidate similar work order for the work order to be processed, the judgment result indicates that a safety risk identification should be performed.

[0046] In this embodiment of the invention, if there is no corresponding associated work order for the work order to be processed, the controller can determine that the repetition threshold is a preset fixed value, such as 0.75.

[0047] In this embodiment of the invention, the controller analyzes the degree of content duplication between the work order to be processed and its corresponding multiple associated work orders to obtain multiple baseline duplication values ​​corresponding to the work order to be processed. This method is similar to the method of analyzing the degree of content duplication between the work order to be processed and its corresponding multiple reference work orders to obtain multiple duplication values ​​corresponding to the work order to be processed, and will not be described in detail here.

[0048] In one example, the steps for screening candidate similar work orders include: Based on the degree of matching between the identifier words of the work order to be processed and the identifier words of multiple reference work orders corresponding to the work order to be processed, the multiple reference work orders corresponding to the work order to be processed are filtered to obtain the filtering results, wherein the identifier words of the candidate similar work orders are the same as the identifier words of the work order to be processed.

[0049] Identifier terms are used to uniquely identify a work order among multiple work orders. Optionally, the identifier term can be a phone number. The controller can use a string matching algorithm to obtain the matching degree between the identifier terms of the work order to be processed and the identifier terms of its multiple corresponding reference work orders. For each reference work order corresponding to the work order to be processed, if the identifier terms of the reference work order completely match the identifier terms of the work order to be processed, the controller can determine that the reference work order is a candidate similar work order corresponding to the work order to be processed. If the identifier terms of the reference work order do not completely match the identifier terms of the work order to be processed, the controller can determine that the reference work order is not a candidate similar work order corresponding to the work order to be processed.

[0050] If the screening results indicate that there are corresponding candidate similar work orders for the work order to be processed, it means that the work order to be processed was issued by the same user. Since the candidate similar work orders are execution work orders, the work order to be processed will not be dispatched to avoid the waste of resources caused by duplicate dispatch.

[0051] If the screening results indicate that there is no corresponding candidate similar work order for the work order to be processed, it means that the work order to be processed was issued by different users. At this time, the judgment result indicates that security risk identification is performed to identify the risk of the work order to be processed and thus determine whether the work order to be processed is an urgent work order or a non-urgent work order.

[0052] Step S4: If the judgment result indicates that risk identification is required, a large model is used to identify the safety risks of the work order to be processed, and the risk identification results are obtained.

[0053] Among them, the risk identification result indicates whether the work order to be processed is an urgent work order or a non-urgent work order, and the dispatch priority of urgent work orders is higher than that of non-urgent work orders.

[0054] Specifically, the controller can employ a large language model classifier to identify security risks based on the target service request vocabulary of the work order to be processed and the standard service request vocabulary in the power professional knowledge base, thereby obtaining the risk identification result. Optionally, the large language model classifier in this embodiment can be an autoencoder-only model or an autoregressive decoder-only model.

[0055] The controller pre-stores a power industry knowledge base, which includes standard service request terms. These terms include, for example, power outages, voltage anomalies, and meter malfunctions. The controller can invoke a large language model classifier, inputting the target service request terms of the work order to be processed and the standard service request terms from the power industry knowledge base into the classifier, and obtain the risk identification results output by the classifier.

[0056] In this embodiment of the invention, before inputting the target service request vocabulary of the work order to be processed and the standard service request vocabulary from the power professional knowledge base into the large language model classifier, the controller can acquire multiple training data sets. Each training data set includes: a sample of service request vocabulary and a corresponding risk label (such as an urgent work order or a non-urgent work order). Then, the controller can train the model using these multiple training data sets to obtain the large language model classifier.

[0057] In summary, this invention filters multiple execution work orders based on the similarity between the address terms of the work order to be processed and the address terms of multiple execution work orders, obtaining multiple reference work orders corresponding to the work order to be processed. Then, it analyzes the degree of content duplication between the work order to be processed and its corresponding multiple reference work orders, obtaining multiple duplicate values ​​corresponding to the work order to be processed. Based on this, it performs work order merging judgment on the work order to be processed and its corresponding multiple reference work orders, obtaining the judgment result of the work order to be processed. When the judgment result indicates risk identification, a large model is used to perform security risk identification on the work order to be processed, obtaining the risk identification result. On the one hand, it realizes the identification of fault problems in the work order to be processed through address filtering and duplicate value matching, and can identify the fault problems of execution work orders that duplicate the work order to be processed as the fault problems of the work order to be processed, improving the accuracy of identifying fault problems in the work order to be processed. On the other hand, it avoids duplicate dispatching by merging judgment through duplicate values, improving the utilization efficiency of dispatching resources. Furthermore, it further determines the urgency of the fault in the work order to be processed through security risk identification, improving the response efficiency of urgent work orders. Therefore, this invention can improve the accuracy of intelligent monitoring and early warning of power supply service work orders.

[0058] This invention also proposes an intelligent monitoring and early warning system for power supply service work orders based on a large model. Please refer to [link / reference]. Figure 2 , Figure 2 This diagram illustrates the structure of a power supply service work order intelligent monitoring and early warning system 200 based on a large model, according to an embodiment of the present invention. The system includes: The work order filtering module 201 is used to filter multiple execution work orders based on the similarity between the address words of the work order to be processed and the address words of multiple execution work orders, and obtain multiple reference work orders corresponding to the work order to be processed. The reference work orders are the execution work orders that point to the same work area as the work order to be processed among the multiple execution work orders. Similarity analysis module 202 is used to analyze the degree of content duplication between the work order to be processed and the corresponding multiple reference work orders, and obtain multiple duplication values ​​corresponding to the work order to be processed, wherein the duplication value indicates the degree of similarity of the fault problems between the corresponding reference work order and the work order to be processed; The merging judgment module 203 is used to perform a work order merging judgment on the work order to be processed and its corresponding multiple reference work orders based on multiple duplicate values ​​corresponding to the work order to be processed, and to obtain the judgment result of the work order to be processed. The judgment result indicates that the work order to be processed should not be dispatched or should be identified for safety risks. The risk identification module 204 is used to identify the safety risks of the work order to be processed using a large model when the judgment result indicates that risk identification is required, and to obtain the risk identification result. The risk identification result indicates whether the work order to be processed is an urgent work order or a non-urgent work order, and the dispatch priority of the urgent work order is higher than that of the non-urgent work order.

[0059] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the intelligent monitoring and early warning system for power supply service work orders based on a large model and the intelligent monitoring and early warning method for power supply service work orders based on a large model provided in the above embodiments belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0060] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0061] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for intelligent monitoring and early warning of power supply service work orders based on a large model, characterized in that, The method includes: Based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple executed work orders, multiple executed work orders are filtered to obtain multiple reference work orders corresponding to the work order to be processed. Among them, the reference work order is the executed work order that points to the same work area as the work order to be processed among the multiple executed work orders. The degree of content duplication between the work order to be processed and its corresponding multiple reference work orders is analyzed to obtain multiple duplication values ​​corresponding to the work order to be processed. The duplication value indicates the degree of similarity between the fault problems between the corresponding reference work order and the work order to be processed. Based on multiple duplicate values ​​corresponding to the work order to be processed, the work order to be processed and its corresponding multiple reference work orders are merged and judged to obtain the judgment result of the work order to be processed. The judgment result indicates that the work order to be processed should not be dispatched or should be identified for safety risks. When the judgment result indicates that risk identification is required, a large model is used to identify the safety risks of the work order to be processed, and the risk identification result is obtained. The risk identification result indicates whether the work order to be processed is an urgent work order or a non-urgent work order. The dispatch priority of urgent work orders is higher than that of non-urgent work orders. The process involves merging and judging the work order to be processed with its corresponding reference work orders based on multiple duplicate values, to obtain the judgment result of the work order to be processed. This includes: analyzing the degree of content duplication between the work order to be processed and its corresponding multiple associated work orders to obtain multiple baseline duplication values ​​corresponding to the work order to be processed, wherein associated work orders are used to characterize work order samples in the historical work order database that have been confirmed to belong to the same fault problem as the work order to be processed or point to the same work area as the work order to be processed; averaging the multiple baseline duplication values ​​to obtain a duplication threshold; if multiple duplication values ​​of the work order to be processed are all greater than the duplication threshold, the judgment result of the work order to be processed indicates that no work order should be dispatched; and in the case of the work order to be processed... If any of the multiple duplicate values ​​is less than the duplicate threshold, the work order is filtered based on the identifier words of the work order to be processed and the identifier words of multiple reference work orders to obtain a work order filtering result. The filtering result indicates whether the work order to be processed has a corresponding candidate similar work order, or whether there is no corresponding candidate similar work order. The candidate similar work order is one of the multiple reference work orders that points to the same user as the work order to be processed. If the filtering result indicates that the work order to be processed has a corresponding candidate similar work order, the judgment result indicates that no work order should be dispatched. If the filtering result indicates that the work order to be processed does not have a corresponding candidate similar work order, the judgment result indicates that security risk identification should be performed. The execution work order refers to a work order that has been dispatched but has not yet been completed.

2. The intelligent monitoring and early warning method for power supply service work orders based on a large model as described in claim 1, characterized in that, The process involves filtering multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders, resulting in multiple reference work orders corresponding to the work order to be processed, including: Based on the address words of the work order to be processed and the address words of multiple executed work orders, similarity matching is performed to obtain multiple address similarity values ​​corresponding to the work order to be processed. Based on the differences in the similarity values ​​and similarity thresholds of multiple addresses corresponding to the work order to be processed, multiple execution work orders are filtered to obtain multiple reference work orders corresponding to the work order to be processed.

3. The intelligent monitoring and early warning method for power supply service work orders based on a large model according to claim 1, characterized in that, The analysis determines the degree of content duplication between the work order to be processed and its corresponding multiple reference work orders, resulting in multiple duplication values ​​for the work order to be processed, including: Analyze the semantic similarity between the work order to be processed and the corresponding multiple reference work orders to obtain multiple semantic similarity values ​​corresponding to the work order to be processed, wherein the semantic similarity values ​​correspond one-to-one with the reference work orders; Analyze the risk similarity between the work order to be processed and the corresponding multiple reference work orders to obtain multiple risk similarity values ​​corresponding to the work order to be processed, wherein the risk similarity values ​​correspond one-to-one with the reference work orders; Multiple semantic similarity values ​​and multiple risk similarity values ​​corresponding to the work order to be processed are fused to obtain multiple duplicate values, wherein each duplicate value corresponds one-to-one with a reference work order.

4. The intelligent monitoring and early warning method for power supply service work orders based on a large model according to claim 3, characterized in that, The analysis of the semantic similarity between the work order to be processed and its corresponding multiple reference work orders yields multiple semantic similarity values ​​corresponding to the work order to be processed, including: A threshold segmentation algorithm is used to filter multiple service request terms in the work order to be processed, thereby obtaining multiple target service request terms for the work order to be processed. Feature extraction is performed on multiple target service request words of the work order to be processed to obtain multiple word vectors of the work order to be processed; The weights of multiple target service request terms and multiple word vectors of the work order to be processed are weighted to obtain the service request vector of the work order to be processed. The similarity between the service request vector of the pending work order and the service request vector of the target reference work order is analyzed to obtain the semantic similarity value between the pending work order and the target reference work order. The target reference work order is any one of multiple reference work orders, and the multiple semantic similarity values ​​corresponding to the pending work order are composed of the semantic similarity values ​​between the pending work order and each reference work order.

5. The intelligent monitoring and early warning method for power supply service work orders based on a large model according to claim 4, characterized in that, The analysis of the risk similarity between the work order to be processed and the corresponding multiple reference work orders yields multiple risk similarity values ​​corresponding to the work order to be processed, including: Analyze the similarity between the first word and the second word in multiple word pairs to obtain multiple word similarity values. The first word is the target service request word of the work order to be processed, and the second word is the service request word of the target reference work order. The word similarity value corresponds one-to-one with the word pair. Any word pair includes a first word and a second word, and at least one word is different between any two word pairs. Based on the relative size of multiple word similarity values, multiple word pairs are filtered to obtain target word pairs, wherein the target word pair is the word pair corresponding to the word pair with the largest word similarity value among multiple word similarity values; Obtain the context word vectors corresponding to the first word and the second word in the target word pair; By analyzing the similarity between the context word vectors corresponding to the first word and the context word vectors corresponding to the second word, the risk similarity value between the work order to be processed and the target reference work order is obtained. The multiple risk similarity values ​​corresponding to the work order to be processed are composed of the risk similarity values ​​between the work order to be processed and each reference work order.

6. The intelligent monitoring and early warning method for power supply service work orders based on a large model according to claim 1, characterized in that, The steps for screening candidate similar work orders include: Based on the degree of matching between the identifier words of the work order to be processed and the identifier words of multiple reference work orders corresponding to the work order to be processed, the multiple reference work orders corresponding to the work order to be processed are filtered to obtain the filtering results, wherein the identifier words of the candidate similar work orders are the same as the identifier words of the work order to be processed.

7. The intelligent monitoring and early warning method for power supply service work orders based on a large model according to claim 1, characterized in that, The process of using a large model to identify security risks in work orders yields risk identification results, including: A large language model classifier is used to identify safety risks based on the target service request vocabulary of the work order to be processed and the standard service request vocabulary in the power professional knowledge base, and the risk identification results are obtained.

8. The intelligent monitoring and early warning method for power supply service work orders based on a large model according to claim 1, characterized in that, Before filtering multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders to obtain multiple reference work orders corresponding to the work order to be processed, the method further includes: The original work orders to be processed are preprocessed to obtain work orders to be processed. The preprocessing includes at least: text extraction processing, text error correction processing and word segmentation processing. Named entity recognition is performed on the work order to be processed to obtain the set of named entities corresponding to the work order to be processed. The set of named entities includes the address entity, date entity, identifier entity and service request entity of the work order to be processed. The service request entity is used to characterize the service request type and urgency.

9. A power supply service work order intelligent monitoring and early warning system based on a large model, characterized in that, Using the early warning method as described in any one of claims 1 to 8, the system comprises: The work order filtering module is used to filter multiple execution work orders based on the similarity between the address keywords of the work order to be processed and the address keywords of multiple execution work orders, and obtain multiple reference work orders corresponding to the work order to be processed. The reference work orders are the execution work orders that point to the same work area as the work order to be processed among the multiple execution work orders. The similarity analysis module is used to analyze the degree of content duplication between the work order to be processed and the corresponding multiple reference work orders, and obtain multiple duplication values ​​corresponding to the work order to be processed. The duplication value indicates the degree of similarity between the fault problems between the corresponding reference work order and the work order to be processed. The merging judgment module is used to perform a merging judgment on the work order to be processed and its corresponding multiple reference work orders based on multiple duplicate values ​​corresponding to the work order to be processed, and to obtain the judgment result of the work order to be processed. The judgment result indicates that the work order to be processed should not be dispatched or should be identified for safety risks. The risk identification module is used to identify the safety risks of the work order to be processed using a large model when the judgment result indicates that risk identification is required. The risk identification result indicates whether the work order to be processed is an urgent work order or a non-urgent work order. The dispatch priority of urgent work orders is higher than that of non-urgent work orders.

Images