A method and system for intelligent management of power spare parts cabinet

Abstract

The present invention provides an intelligent management method and system for electric power spare parts cabinets, belonging to the technical field of electric power spare parts management. It is achieved through the following technical solutions: identifying and recording the spare parts in the current cargo compartment; reading the identification results and updating the spare parts data through the data update system; predicting the maintenance plan through the matching model according to the updated data; according to the maintenance plan prediction results, enhancing the original representation of the semantic matching model and the relationship matching model through the multi-perspective collaborative learning model, and realizing autonomous decision-making of the maintenance plan. The present invention improves the safety of spare parts storage and the convenience of warehousing and warehousing operations, and automatically forms a fault maintenance plan for reference by maintenance personnel, thereby reducing the pressure of emergency repairs on operators and improving emergency repair efficiency.

Description

Technical Field

[0001] The invention relates to an intelligent management method and system for an electric power spare parts cabinet, belonging to the technical field of electric power spare parts management. Background Art

[0002] The power maintenance department has a heavy workload during the spring and autumn inspections, and the pressure of emergency repair and fault elimination in substations is huge. Whether the spare parts needed for emergency repair and fault elimination can be found quickly and accurately directly affects the time for defect elimination, which in turn affects the safe operation of the power grid. In order to improve the efficiency of emergency repair and fault elimination, it is crucial to realize the management of power spare parts.

[0003] At present, the management of spare parts for emergency repair and fault elimination of substations mainly has the following problems: first, there is a lot of manual participation in the entry and exit of the warehouse, and the inventory efficiency is low; second, it is difficult to quickly and accurately find the required spare parts when facing emergency repair tasks. Third, the analysis of the use of spare parts is not comprehensive, making it difficult to purchase them in a targeted manner.

[0004] There is a need for an intelligent management system for power spare parts cabinets that combines the management needs of power companies with digital intelligence technology and has intelligent warehousing and access functions with fuzzy search, autonomous decision-making functions for maintenance plans, and data analysis functions. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent management method and system for electric power spare parts cabinets, which can realize the intelligent storage and warehousing function, autonomous decision-making function of maintenance plan and data analysis function of electric power spare parts management.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0007] Use the YOLOv5m image detection model and weight sensing module to identify and record the spare parts in the current cargo compartment;

[0008] The data update system reads the identification results and updates the spare parts data, including name, model, applicable fault, applicable site, consumption within a certain period of time and current balance;

[0009] According to the updated data, a maintenance plan is predicted by a matching model, wherein the matching model includes a semantic matching model based on hierarchical Transformer text representation and a relational matching model based on an R-GCN neural network;

[0010] According to the prediction results of the maintenance plan, the original representations of the semantic matching model and the relationship matching model are enhanced through a multi-view collaborative learning model to achieve autonomous decision-making on the maintenance plan.

[0011] Preferably, the specific steps of predicting the maintenance plan through the matching model are as follows:

[0012] The first-layer encoder based on the semantic matching model of hierarchical Transformer text representation encodes the statements of fault type and maintenance plan, and the second-layer encoder embeds the statements output by the first-layer encoder into the fault document and maintenance document;

[0013] A fault type-maintenance plan relationship graph for capturing implicit associations was constructed based on fault documents and maintenance documents. The relation matching model based on the R-GCN neural network collected the incoming information of adjacent nodes in the relationship graph and performed aggregation operations to obtain the final matching fault type and maintenance plan.

[0014] Preferably, the first-layer encoder of the semantic matching model based on hierarchical Transformer text representation embeds a special tag CLS at the beginning of the sentence of fault type and maintenance plan, and the embedded special tag is accumulated to a single input vector and fed back to the encoder; the tag embedding method includes tag embedding and position embedding, the tag embedding represents the meaning of each tag, and the position embedding represents the position of each tag in the text sequence.

[0015] Preferably, the second-layer encoder update formula of the semantic matching model based on the hierarchical Transformer text representation is as follows:

[0016] ,

[0017] ,

[0018] ,

[0019] ,

[0020] in, For fault documentation, To review the documentation, and For the The fault vector and maintenance vector of the layer input, LN is the layer normalization operation, MHAtt is the multi-head attention operation, Indicates the number of layers of the Transformer network;

[0021] The output layer is a sigmoid classifier, the specific formula is as follows:

[0022] ,

[0023] in, and The last layer and The representation form, is the parameter matrix for transforming the concatenated fault and repair document representations, For a deviation, express The degree of match with the maintenance documentation, It is the last layer of Transformer network.

[0024] Preferably, the fault type-maintenance solution relationship diagram creates an implicit link between two nodes including:

[0025] Create links using category labels. If the fault nodes or solution nodes share the same category label, a link is created between them. The formed link is attached with the category label as the relationship label.

[0026] Create links using keywords. When the text of two nodes contains the same keyword, create a link between them and attach the keyword as the relationship label.

[0027] Preferably, the keyword generation algorithm is as follows:

[0028] Use the tf-idf weighting method to select the candidate word list;

[0029] By calculating the frequency of two words co-occurring in fault documents or maintenance documents, a word co-occurrence graph is constructed;

[0030] Run the PageRank algorithm on the word co-occurrence graph to obtain the PageRank score for each word;

[0031] Sort the words according to their PageRank scores and keep only the top-ranked words as keywords.

[0032] Preferably, the relationship matching model based on the R-GCN neural network collects the incoming information of adjacent nodes in the relationship graph and performs aggregation operations, specifically in the following manner:

[0033] Get the node representation of each layer of the relationship graph:

[0034] ,

[0035] in, is a node placeholder, indicating a fault type node or a maintenance solution node. , It is Layer Node The expression, express In the relationship type The neighbor set below, Is relationship specific The parameter matrix of is the parameter matrix represented for the original node;

[0036] After the node representation is obtained, the matching scores between the various maintenance plans of the fault type are calculated :

[0037] ,

[0038] Among them, the predicted and Fault Documentation and maintenance documentation In the last layer representation, and are parameter matrices and vectors;

[0039] The relationship graph is aggregated according to the matching score to obtain the final matching fault type and maintenance plan.

[0040] Preferably, the specific manner of enhancing the original representation of the semantic matching model and the relationship matching model through the multi-view collaborative learning model is as follows:

[0041] Enhance the semantic matching model, concatenate the original text-based representation with the relation-based representation, and initialize the node states of the relation matching model with the learned representation obtained by the semantic matching model;

[0042] By selecting examples through data quality enhancement and retaining them in the training phase, the overall performance of the autonomous decision-making algorithm can be improved; the specific methods are as follows:

[0043] For each batch update, the batch dataset Randomly split into two equal-sized subsets and , where A and B represent the enhanced semantic matching model and relation matching model respectively;

[0044] and Filter the noisy samples through peer learners A and B;

[0045] Based on the derived new dataset and , update the parameters of A and B using stochastic gradient descent.

[0046] An intelligent management system for power spare parts cabinets, including a main cabinet, a sub-cabinet, and an intelligent analysis and decision-making module;

[0047] The main cabinet includes a sensor and a display module; the sensor has a built-in intelligent verification module for verifying user identity information; the display module is used to display the information of the warehousing system and the data statistical analysis module;

[0048] The auxiliary cabinet is used to store spare parts, and the spare parts information is collected through industrial cameras and weight sensors;

[0049] The intelligent analysis and decision-making module generates a maintenance plan through the intelligent management method of the power spare parts cabinet; it includes an in-and-out storage system and a data statistical analysis module. The in-and-out storage system is used to obtain the spare parts information of the auxiliary cabinet and display it through the display module. The data statistical analysis module is used to obtain the in-and-out storage information of the spare parts of the auxiliary cabinet, count the common fault types, substation maintenance frequency, equipment maintenance frequency, and provide maintenance suggestions and procurement suggestions.

[0050] The advantages of the present invention are: the present invention adopts an image recognition model based on YOLOv5m combined with a weight sensing module to realize the intelligent warehousing and out-of-warehouse function, thereby improving the safety of spare parts storage and the convenience of warehousing and out-of-warehouse operations; a method for capturing multi-view data signals is designed to improve the reliability of decision-making, and the representation is enhanced by aggregating similar fault types or maintenance plans, fully exploring the potential connections between different fault types and maintenance plans, effectively integrating different data signals, and obtaining a more effective and reliable matching model, which can automatically form a fault maintenance plan for reference by maintenance personnel, thereby reducing the emergency repair pressure of operators and improving emergency repair efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0052] Figure 1 It is a schematic diagram of the process of the present invention.

[0053] Figure 2 It is a schematic diagram of the overall architecture of the intelligent management system for power spare parts cabinets of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Example 1

[0056] like Figure 1 As shown, a method for intelligent management of power spare parts cabinets is implemented through the following technical solutions:

[0057] S1: Use the YOLOv5m image detection model and weight sensing module to identify and record the spare parts in the current cargo compartment.

[0058] S2: The data update system reads the identification results and updates the spare parts data, including the name, model, applicable fault, applicable site, consumption within a certain period of time, and current balance.

[0059] S3: According to the updated data, a maintenance plan is predicted through a matching model, wherein the matching model includes a semantic matching model based on hierarchical Transformer text representation and a relational matching model based on an R-GCN neural network.

[0060] S4: Based on the maintenance plan prediction results, the original representations of the semantic matching model and the relationship matching model are enhanced through a multi-view collaborative learning model to achieve autonomous decision-making on the maintenance plan.

[0061] As a refinement of the above embodiment, the specific steps of predicting the maintenance plan by matching the model in step S3 are as follows:

[0062] S3-1: The first-layer encoder of the semantic matching model based on the hierarchical Transformer text representation encodes the statements of the fault type and maintenance plan, and the second-layer encoder embeds the statements output by the first-layer encoder into the fault document and maintenance document.

[0063] Specifically, the first-layer encoder of the semantic matching model based on hierarchical Transformer text representation embeds a special token CLS at the beginning of the sentence of fault type and maintenance plan. The embedded special token is accumulated to a single input vector and fed back to the encoder. The token embedding methods include token embedding and position embedding. The token embedding represents the meaning of each token, and the position embedding represents the position of each token in the text sequence.

[0064] Specifically, the second-layer encoder update formula of the semantic matching model based on hierarchical Transformer text representation is as follows:

[0065] ,

[0066] ,

[0067] ,

[0068] ,

[0069] in, For fault documentation, To review the documentation, and For the The fault vector and maintenance vector of the layer input, LN is the layer normalization operation, MHAtt is the multi-head attention operation, Indicates the number of layers in the Transformer network.

[0070] The output layer is a sigmoid classifier, the specific formula is as follows:

[0071] ,

[0072] in, and The last layer and The representation form of is the parameter matrix for transforming the concatenated fault and repair document representations, For a deviation, express The degree of match with the maintenance documentation, It is the last layer of Transformer network.

[0073] S3-2: A fault type-maintenance solution relationship graph is constructed based on fault documents and maintenance documents to capture implicit associations. The relationship matching model based on the R-GCN neural network collects the incoming information of adjacent nodes in the relationship graph and performs aggregation operations to obtain the final matching fault type and maintenance solution.

[0074] S3-2-1: Construct a fault type-maintenance plan relationship diagram. Before constructing a specific model, first define the fault type-maintenance plan relationship. Where V is the node set and E is the link set. For the matching task, the node set is the union of the fault set and the solution set, that is, , the link set contains all the interaction links between two nodes on the graph, that is, , where T is a relation set, yes , Since there are two types of nodes, there are three possible types of links to be considered in the task, namely, fault-solution, fault-fault and solution-solution. The present invention uses two data signals to create an implicit link between two nodes, one is a category label and the other is a keyword. The fault type-maintenance solution relationship diagram creates an implicit link between two nodes in the following ways:

[0075] (1) Create links using category labels. If fault nodes or solution nodes share the same category label, a link is created between them. The formed link is attached with the category label as the relationship label.

[0076] (2) Create links using keywords. When the text of two nodes contains the same keyword, a link is created between them and the keyword is attached as the relationship label.

[0077] Specifically, the keyword generation algorithm is as follows, as shown in Table 1:

[0078] The tf-idf weighting method is used to select the candidate word list.

[0079] A word co-occurrence graph is constructed by calculating the frequency with which two words co-occur in fault documents or troubleshooting documents.

[0080] Run the PageRank algorithm on the word co-occurrence graph to get the PageRank score for each word.

[0081] Sort the words according to their PageRank scores and keep only the top-ranked words as keywords.

[0082] Table 1 Keyword generation algorithm

[0083]

[0084] S3-2-2: The relationship matching model based on the R-GCN neural network collects the incoming information of adjacent nodes in the relationship graph and performs aggregation operations. The specific method is as follows:

[0085] Get the node representation of each layer of the relationship graph:

[0086] ,

[0087] in, is a node placeholder, indicating a fault type node or a maintenance solution node. , It is Layer Node The expression, express In the relationship type The neighbor set below, Is relationship specific The parameter matrix of is the parameter matrix for the original node representation.

[0088] After the node representation is obtained, the matching scores between the various maintenance plans of the fault type are calculated :

[0089] ,

[0090] Among them, the predicted and Fault Documentation and maintenance documentation In the last layer representation, and are parameter matrices and vectors.

[0091] The relationship graph is aggregated according to the matching score to obtain the final matching fault type and maintenance plan.

[0092] As a refinement of the above embodiment, the specific manner of enhancing the original representation of the semantic matching model and the relationship matching model through the multi-view collaborative learning model in step S4 is as follows:

[0093] S4-1: Enhance the semantic matching model, concatenate the original text-based representation with the relation-based representation, and initialize the node state of the relation matching model with the learned representation obtained by the semantic matching model.

[0094] S4-2: Select instances through data quality enhancement and retain them in the training phase to improve the overall performance of the autonomous decision-making algorithm; the specific methods are as follows:

[0095] S4-2-1: For each batch update, the batch data set Randomly split into two equal-sized subsets and , where A and B represent the enhanced semantic matching model and relation matching model respectively.

[0096] S4-2-2: and Noise samples are filtered by peer learners A and B.

[0097] S4-2-3: Based on the derived new dataset and , update the parameters of A and B using stochastic gradient descent.

[0098] Data quality enhancement specifically uses a collaborative learning algorithm, as shown in Table 2.

[0099] Table 2 Collaborative learning algorithm process

[0100]

[0101] Example 2

[0102] like Figure 2 As shown, an intelligent management system for electric power spare parts cabinets includes a main cabinet, a sub-cabinet, and an intelligent analysis and decision-making module.

[0103] The main cabinet includes a sensor and a display module; the sensor has a built-in intelligent verification module for verifying user identity information; the display module is used to display the information of the warehousing system and the data statistical analysis module.

[0104] The auxiliary cabinet is used to store spare parts, and the spare parts information is collected through industrial cameras and weight sensors.

[0105] The intelligent analysis and decision-making module generates a maintenance plan through the intelligent management method of the power spare parts cabinet; it includes an in-and-out storage system and a data statistical analysis module. The in-and-out storage system is used to obtain the spare parts information of the auxiliary cabinet and display it through the display module. The data statistical analysis module is used to obtain the in-and-out storage information of the spare parts of the auxiliary cabinet, count common fault types, substation maintenance frequency, equipment maintenance frequency, and provide maintenance suggestions and procurement suggestions.

[0106] As a refinement of the above system, when the inbound and outbound system enters the warehouse, it first allocates a sub-cabinet cargo compartment for the spare parts, then places a certain number of spare parts, automatically calculates the weight of each piece based on the number of parts placed through image feature recognition, and finally adds the spare parts data to the system. When leaving the warehouse, you can directly open the sub-cabinet door to collect the parts, and the system will automatically display the type and quantity of the parts collected. You can also enter any spare parts data on the main cabinet screen for fuzzy search.

[0107] After the intelligent analysis and decision module inputs the fault site and fault manifestation, the system can aggregate the fuzzy search function to give the components that may need to be repaired and replaced, and recommend spare parts used when similar faults occur at similar sites, and can issue the recommended spare parts with one click. Overall, the function can assist maintenance personnel in handling emergency faults and shorten the preparation time for emergency repairs.

[0108] The hardware part of the intelligent analysis and decision-making module consists of a weight sensor installed in each compartment of the auxiliary cabinet, a built-in industrial camera and an industrial host. The pre-trained YOLOv5m image detection model is used in conjunction with the weight sensing module to identify and record the spare parts in the current compartment. The data statistical analysis function uses the consumption of spare parts within a certain period as a training data set, and uses the designed matching model and multi-view collaborative learning model to realize the autonomous decision-making function of the maintenance plan.

[0109] The embodiment of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to execute the above-mentioned intelligent management method for power spare parts cabinets.

[0110] The computer-readable storage medium mentioned above may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0111] The technical solution of the embodiment of the present disclosure can be embodied in the form of a software product, which is stored in a storage medium and includes one or more instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiment of the present disclosure. The aforementioned storage medium may be a non-transient storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, and other media that can store program codes, or a transient storage medium.

[0112] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for intelligent management of power spare parts cabinets, characterized in that: include: Use the YOLOv5m image detection model and weight sensing module to identify and record the spare parts in the current cargo compartment; Read the identification results and update the spare parts data through the data update system; According to the updated data, a maintenance plan is predicted by a matching model, wherein the matching model includes a semantic matching model based on hierarchical Transformer text representation and a relational matching model based on an R-GCN neural network; According to the maintenance plan prediction results, the maintenance plan can be autonomously decided through the original representation of multi-view collaborative learning model, enhanced semantic matching model and relation matching model; The specific steps of predicting maintenance plans through matching models are as follows: The first-layer encoder based on the semantic matching model of hierarchical Transformer text representation encodes the statements of fault type and maintenance plan, and the second-layer encoder embeds the statements output by the first-layer encoder into the fault document and maintenance document; A fault type-maintenance plan relationship graph for capturing implicit associations was constructed based on fault documents and maintenance documents. The relation matching model based on the R-GCN neural network collected the incoming information of adjacent nodes in the relationship graph and performed aggregation operations to obtain the final matching fault type and maintenance plan.

2. The intelligent management method of power spare parts cabinet according to claim 1 is characterized in that: The first-layer encoder of the semantic matching model based on hierarchical Transformer text representation embeds a special token CLS at the beginning of the sentence of fault type and maintenance plan. The embedded special token is accumulated to a single input vector and fed back to the encoder. The token embedding methods include token embedding and position embedding. The token embedding represents the meaning of each token, and the position embedding represents the position of each token in the text sequence.

3. The intelligent management method of power spare parts cabinet according to claim 1 is characterized in that: The second-layer encoder update formula of the semantic matching model based on hierarchical Transformer text representation is as follows: ， ， ， ， in, For fault documentation, To review the documentation, and For the The fault vector and maintenance vector of the layer input, LN is the layer normalization operation, MHAtt is the multi-head attention operation, Indicates the number of layers of the Transformer network; The output layer is a sigmoid classifier, the specific formula is as follows: ， in, and The last layer and The representation form, is the parameter matrix for transforming the concatenated fault and repair document representations, For a deviation, express The degree of match with the maintenance documentation, It is the last layer of Transformer network.

4. The intelligent management method of power spare parts cabinet according to claim 1 is characterized in that: The implicit link between two nodes in the fault type-repair plan relationship diagram includes: Create links using category labels. If the fault nodes or solution nodes share the same category label, a link is created between them. The formed link is attached with the category label as the relationship label. Create links using keywords. When the text of two nodes contains the same keyword, create a link between them and attach the keyword as the relationship label.

5. The intelligent management method of power spare parts cabinet according to claim 4 is characterized in that: The keyword generation algorithm is as follows: Use the tf-idf weighting method to select the candidate word list; By calculating the frequency of two words co-occurring in fault documents or maintenance documents, a word co-occurrence graph is constructed; Run the PageRank algorithm on the word co-occurrence graph to obtain the PageRank score for each word; Sort the words according to their PageRank scores and keep only the top-ranked words as keywords.

6. The intelligent management method of power spare parts cabinet according to claim 1 is characterized in that: The relationship matching model based on the R-GCN neural network collects the incoming information of adjacent nodes in the relationship graph and performs aggregation operations as follows: Get the node representation of each layer of the relationship graph: ， in, is a node placeholder, indicating a fault type node or a maintenance solution node. , It is Layer Node The expression, express In the relationship type The neighbor set below, Is relationship specific The parameter matrix of is the parameter matrix represented for the original node; After the node representation is obtained, the matching scores between the various maintenance plans of the fault type are calculated : ， Among them, the predicted and Fault Documentation and maintenance documentation In the last layer representation, and are parameter matrices and vectors; The relationship graph is aggregated according to the matching score to obtain the final matching fault type and maintenance plan.

7. The intelligent management method of power spare parts cabinet according to claim 1 is characterized in that: The specific method of the original representation of the multi-view collaborative learning model, enhanced semantic matching model, and relation matching model is as follows: Enhance the semantic matching model, concatenate the original text-based representation with the relation-based representation, and initialize the node states of the relation matching model with the learned representation obtained by the semantic matching model; By selecting examples through data quality enhancement and retaining them in the training phase, the overall performance of the autonomous decision-making algorithm can be improved; the specific methods are as follows: For each batch update, the batch dataset Randomly split into two equal-sized subsets and , where A and B represent the enhanced semantic matching model and relation matching model respectively; and Filter the noisy samples through peer learners A and B; Based on the derived new dataset and , update the parameters of A and B using stochastic gradient descent.

8. An intelligent management system for power spare parts cabinets that implements the method according to any one of claims 1 to 7, characterized in that: Including main cabinet, sub-cabinet, and intelligent analysis and decision-making module; The main cabinet includes a sensor and a display module; the sensor has a built-in intelligent verification module for verifying user identity information; the display module is used to display the information of the warehousing system and the data statistical analysis module; The auxiliary cabinet is used to store spare parts, and the spare parts information is collected through industrial cameras and weight sensors; The intelligent analysis and decision-making module generates a maintenance plan through the intelligent management method of the power spare parts cabinet as described in any one of claims 1-7; it includes an in-and-out storage system and a data statistical analysis module, the in-and-out storage system is used to obtain the spare parts information of the sub-cabinet and display it through the display module, and the data statistical analysis module is used to obtain the in-and-out storage information of the spare parts of the sub-cabinet, count the common fault types, substation maintenance frequency, equipment maintenance frequency, and provide maintenance suggestions and procurement suggestions.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the program is executed by a processor according to the intelligent management method for electric power spare parts cabinets as described in any one of claims 1-7.

Images