Power failure work order processing method and device, electronic equipment and storage medium

Abstract

The invention discloses a power failure work order processing method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a to-be-processed failure work order corresponding to a power service area, and carrying out the preprocessing of the to-be-processed failure work order, and determining a target failure work order; performing aspect-level text classification on the target fault work order, and determining a fault aspect word corresponding to the target fault work order; and determining fault alarm information based on the fault aspect word, and performing fault prompt according to the fault alarm information. Based on the above technical scheme, by fusing aspect-level text classification and an automation method, the efficiency and accuracy of fault first-aid repair work order processing are improved, and the technical effect of improving the user service level of a power enterprise is achieved.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a method, device, electronic equipment and storage medium for processing an electric power fault work order. Background Art

[0002] With the continuous improvement of power service standards, efficient and rapid response to fault repair work orders has become one of the key factors in improving customer satisfaction. In order to provide efficient fault outage repair services, repair personnel must handle fault repair work orders in a timely manner, carry out fault repairs, communicate and appease customers, and avoid escalation of demands and concentrated outbreaks.

[0003] However, the traditional fault repair work order processing model relies on staff to visually identify and manually classify the target fault work order to see whether there is a fault and remind the repair personnel. This method has many problems, including being time-consuming and labor-intensive, incomplete classification due to manual negligence, omissions, and thus insufficient notification. Summary of the invention

[0004] The present invention provides a method, device, electronic device and storage medium for processing power fault work orders. By integrating aspect-level text classification and automation methods, the efficiency and accuracy of fault repair work order processing are improved, achieving the technical effect of improving the level of service provided by power companies to users.

[0005] According to one aspect of the present invention, a method for processing a power failure work order is provided, which includes:

[0006] Obtaining pending fault work orders corresponding to the power service area, and pre-processing the pending fault work orders to determine target fault work orders;

[0007] Performing aspect-level text classification on the target fault work order to determine a fault aspect word corresponding to the target fault work order;

[0008] Fault warning information is determined based on the fault aspect words, and a fault prompt is given according to the fault warning information.

[0009] According to another aspect of the present invention, a device for processing a power failure work order is provided, comprising:

[0010] A data acquisition module, used to acquire pending fault work orders corresponding to the power service area, and pre-process the pending fault work orders to determine target fault work orders;

[0011] A text classification module, used to perform aspect-level text classification on the target fault work order and determine the fault aspect words corresponding to the target fault work order;

[0012] The fault prompt module is used to determine fault warning information based on the fault aspect words, and to provide fault prompts according to the fault warning information.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for processing a power fault work order described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for processing a power fault work order described in any embodiment of the present invention when executed.

[0018] The technical solution of the embodiment of the present invention obtains the pending fault work order corresponding to the power service area, pre-processes the pending fault work order to determine the target fault work order, and then performs aspect-level text classification on the target fault work order to determine the fault aspect words corresponding to the target fault work order, and finally determines the fault alarm information based on the fault aspect words, and performs a fault prompt based on the fault alarm information. Based on the above technical solution, by integrating aspect-level text classification and automation methods, the efficiency and accuracy of fault repair work order processing are improved, achieving the technical effect of improving the level of service provided by power companies to users.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 This is a flow chart of a method for processing a power failure work order provided according to Embodiment 1 of the present invention;

[0022] Figure 2 2 is a schematic diagram of the structure of a stacked bidirectional long short-term memory network layer provided according to Embodiment 1 of the present invention;

[0023] Figure 3 is a structural diagram of a semantic learning layer provided according to Embodiment 1 of the present invention;

[0024] Figure 4 This is a structural diagram of a device for processing a power failure work order provided in accordance with Embodiment 2 of the present invention;

[0025] Figure 5 It is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] Embodiment 1

[0029] Figure 1 A flowchart of a method for processing a power fault work order is provided for the first embodiment of the present invention. This embodiment can be applied to perform aspect-level text classification on power fault work orders, extract fault aspect words corresponding to power fault work orders, and issue fault alarms based on the aspect words. The method can be executed by a power fault work order processing device, which can be implemented in the form of hardware and / or software, and can be configured in an electronic device. Figure 1 As shown, the method includes:

[0030] S110, obtaining pending fault work orders corresponding to the power service area, and pre-processing the pending fault work orders to determine target fault work orders.

[0031] Among them, the power service area can be understood as the power supply area currently served by the power fault work order processing system. It should be noted that the power service area can be divided according to actual needs, for example, it can be divided according to a preset area, or it can be divided according to a planned power supply area. The fault work order to be processed can be obtained from the power grid management platform. The fault work order can be filled in and reported by the user according to the fault information of the power equipment. The user can be the maintenance personnel of the power equipment or the resident in the power service area. The target fault work order can be understood as the fault description content corresponding to the fault work order to be processed obtained after data preprocessing, that is, after data and processing, the corresponding fault description content can be extracted from the fault work order to be processed.

[0032] Specifically, according to data extraction tools, such as selenium and HttpClient, etc., connect to various data sources, which can be power grid management platforms, marketing systems, and innovation management systems, etc., to obtain pending fault work orders in different systems and modules. It should be noted that the fault work order covers multiple data fields such as the fault order number, the power supply station to which it belongs, the source of the fault report, the location of the fault, and the target fault work order. After obtaining the pending fault work order, the fault work order can be pre-processed according to the pre-set data pre-processing method to determine the target fault work order. For example, data cleaning, data association, and data standardization can be performed on the pending fault work order. For example, First, the data is cleaned. Special characters such as punctuation marks in the target fault tickets have no effect on type classification and will waste computing resources. They are removed through regular expressions and replaced with spaces to facilitate subsequent type classification to ensure data quality and availability. In addition, the data names of the same data field in the fault ticket may be different in different systems. It is necessary to associate data from different systems and unify the data fields to facilitate subsequent identification and processing. Further, tools such as FastJson, JsonPath and regular expressions are used to standardize the extracted work order data, convert the values ​​of the data fields, and standardize them into a consistent format.

[0033] S120: Perform aspect-level text classification on the target fault work order to determine a fault aspect word corresponding to the target fault work order.

[0034] Among them, aspect-level text classification can be understood as a classification method for identifying and classifying different aspects mentioned in the target fault ticket and their corresponding emotions or evaluations. It should be noted that, unlike traditional text classification, aspect-level text classification is more fine-grained and focuses on specific evaluation dimensions in the text. Fault aspect words can be the fault aspect words corresponding to the target fault ticket and the fault.

[0035] Specifically, the target fault work order of the work order may contain multiple aspect words. Different professionals are required to handle different aspect words. The aspect words (also called aspect words) contained in the target fault work order and other contents are called context. By introducing natural language processing technology and deep learning algorithms, aspect-level text classification is used to determine whether there is a fault in the target fault work order containing multiple aspect words, and the corresponding staff are reminded to handle it in time according to the aspect words where the fault exists.

[0036] On the basis of the above technical solution, the aspect-level text classification of the target fault work order includes: inputting the target fault work order into a global word frequency statistical model to determine the fault work order word vector corresponding to the target fault work order; inputting the fault work order word vector into a semantic learning layer to determine the fault description semantic features corresponding to the target fault work order; and performing aspect-level text classification on the target fault work order based on the fault description semantic features.

[0037] Among them, the global word frequency statistics model can be a model for inputting text for processing and extracting word vectors. The fault work order word vector can be understood as the result of converting the text information in the fault work order into a vector form. The fault description semantic feature can be a semantic feature corresponding to the fault work order extracted by the semantic learning layer. The semantic learning layer is formed by combining a stacked bidirectional long short-term memory network layer and a random dense residual connection layer; the stacked bidirectional long short-term memory network layer is used to extract semantic information from the fault work order word vector.

[0038] Specifically, the target fault work order is input into the global word frequency statistics model to determine the fault work word vector corresponding to the target fault work order, and then the fault work word vector is input into the semantic learning layer to determine the fault description semantic features corresponding to the target fault work order, and finally the target fault work order is subjected to aspect-level text classification based on the fault description semantic features. Exemplarily, since the target fault work order is a text, the computer cannot process it directly, and the target fault work order needs to be vectorized first, so the target fault work order is input into the text representation layer. The text representation layer converts words into vector representations through a specific model, and can convert text into a form that can be understood by the computer. It should be noted that GloVe is introduced in the text representation layer. GloVe is a model based on global word frequency statistics. The sliding window is used to count the co-occurrence times of word pairs that appear in the same sliding window in all target fault work orders in the work order corpus, forming a co-occurrence matrix of size n*n, where n represents the number of independent and non-repeating words in the vocabulary. Each element in the co-occurrence matrix represents the number of times the words in the corresponding row and column appear together in the sliding window. In addition to using a sliding window to obtain local features, the overall features are extracted by statistics on all word pairs, that is, the local and the overall features are learned together. Through text representation layer processing, the target fault ticket is converted into a vector representation that can be recognized by a computer, and the semantic information of the target fault ticket can be preliminarily extracted.

[0039] Furthermore, the semantic learning layer is formed by stacking a bidirectional long short-term memory network layer and a random dense residual connection layer. It can not only obtain the deep semantic information of the target fault ticket through the stacked BiLSTM layer, but also effectively solve the overfitting problem caused by the stacked BiLSTM layer through the random dense residual connection layer. For example, the sentence is processed by the text representation layer to obtain the initial word vector representation. Since the target fault tickets of many tickets are not only lengthy, but also unclear in expression, there may be polysemy problems. Therefore, it is solved by stacking BiLSTM, such as Figure 2 As shown in the figure, the target fault ticket is first processed by GloVe to obtain the initial word vector representation. The initial word vector representation is then processed by the first layer of BiLSTM to obtain the shallow semantic feature representation of the sentence context. Then the shallow semantic feature representation is processed by the second layer of BiLSTM to obtain the higher-level semantic feature representation of the context, and so on. Specifically, the sentence processed by the word embedding layer is first used as the input of the first layer of BiLSTM. The first layer of BiLSTM generates a hidden state vector, which is used as the input of the second layer of BiLSTM. The hidden state vector is represented as: h c t =BiLSTM(x t ,h c t-1 )if c=1,h ct =BiLSTM(h c-1 t ,h c t-1 )if c>1; where c represents the number of BiLSTM layers, and the best effect can be achieved when it is set to 7. After encoding the word vector using stacked BiLSTM, the sentence context corresponding to the target fault ticket is represented as:

[0040] It should be noted that as the number of stacked BiLSTM layers increases, although the deep semantic features of the context can be extracted, it also brings about a serious overfitting problem, which reduces the generalization of the model. This is reflected in the high accuracy of fault type recognition during training, but low accuracy during testing. In order to solve the above problem, a random dense residual connection is designed for the stacked BiLSRM layer, such as Figure 3 As shown in Figure 1. By recursively passing the hidden state of each BiLSTM layer to all subsequent BiLSTM layers and performing linear stacking as the input of the subsequent BiLSTM layer, and randomly discarding the dense residual connection during the training process, the context-related semantic features are finally obtained. The original dense residual connection of the stacked BiLSTM is: c t =BiLSTM(h c-1 t +h c-2 t +···+x t ,h c t-1 ), in order to improve the generalization and solve the overfitting problem of the model. Random parameters are added to each layer of dense residual connections (except the first layer), and each layer decides to delete or retain dense residual connections. After adding random parameters, it is as follows: h c t =BiLSTM(r c-1 *h c-1 t +r c-2 *h c-2 t +···+x t ,h c t-1 ), where the random parameter r c Only 0 or 1. c = 0, the corresponding dense residual connection is deleted. c = 1, the corresponding dense residual connection is retained. c It is generated by Bernoulli distribution. It should be noted that the higher the BiLSTM layer, the more important the extracted semantic features are, and the greater the weight they should occupy in the final output.

[0041] Therefore, a weight parameter w is set for each layer c The higher the BiLSTM layer, the larger the corresponding weight parameter. c The expression is: Where c represents the cth layer, C is the total number of layers, which is 7, and q is a global parameter, which is set to 0.4. The sentence context is added with the weight coefficient w c After that, it is expressed as: h c (x t )=w c *BiLSTM(r c-1 *h c-1 t +r c-2 *h c-2 t +···+x t ,h c t-1 ).

[0042] On the basis of the above technical solution, the target fault work order is subjected to aspect-level text classification based on the semantic features of the fault description, including: processing the semantic features of the fault description based on a multi-channel graph convolutional layer to determine the aspect word vector to be screened corresponding to the target fault work order; determining the attention weight corresponding to each aspect word vector to be screened based on a gated attention mechanism; and determining the fault aspect word vector from each aspect word vector to be screened according to the attention weight.

[0043] Among them, the multi-channel graph convolution layer can be a graph convolutional neural network layer that extracts the relationship vector in the fault ticket based on the relationship between the text. The gated attention mechanism can be regarded as a neural network mechanism that combines the advantages of the gating mechanism and the attention mechanism, which can improve the performance and accuracy of the model when processing complex tasks. The gating mechanism can be a mechanism that uses learnable parameters to dynamically adjust the flow of information in the network. The attention mechanism can be to assign different weights to each part of the input sequence, and based on the weight information, let the model selectively focus on different parts of the input sequence, so as to highlight the information that is more critical to the task.

[0044] Specifically, the attention mechanism extracts important words related to aspect words from the context of the target fault work order, and sets attention weights for each word. The importance is different, and the attention weights assigned are also different. However, many attention weights with small values ​​in the attention mechanism are not only ineffective for judging whether the aspect word has a fault, but also bring noise. For example, a customer on Nanjiao Road reported that a truck scratched the electric meter and the meter fell, and hoped that it would be handled as soon as possible. In this target fault work order, words such as "Nanjiao Road", "Customer" and "As soon as possible" have nothing to do with judging whether the aspect word meter is faulty. Their attention weight values ​​are very small. By designing a gating unit to filter them out, it can not only avoid the influence of irrelevant words on judging whether the aspect word is faulty, but also reduce redundancy and improve processing speed. To address this problem, a gating unit is designed for the attention mechanism, which only passes the attention weights that meet the conditions and resets the attention weights that do not meet the conditions to zero. The attention weight calculation of the gating unit is as follows: Wherein, c represents the threshold of the gating unit in the attention mechanism, and the purpose of updating the attention weight is achieved by judging the threshold. The technical solution provided by the embodiment of the present invention designs a gating unit in the attention mechanism, which effectively reduces the redundancy and noise of the attention mechanism, can focus on the words related to the aspect words, give the words with high relevance a greater weight, ignore the irrelevant words, and thus strengthen the connection between the aspect words and the context.

[0045] On the basis of the above technical scheme, the multi-channel graph convolution layer includes a syntactic dependency graph convolution channel, a relative aspect distance relationship convolution channel, a random edge elimination convolution channel and a distance relationship convolution channel. The multi-channel graph convolution layer is used to process the semantic features of the fault description, including: obtaining the syntactic dependency tree of the target fault work order based on the dependency parser, and converting the syntactic dependency tree into a directed graph; processing the directed graph according to the dependency graph convolution channel to determine the syntactic dependency vector; processing the directed graph according to the relative aspect distance relationship convolution channel to determine the relative type distance relationship vector; processing the directed graph according to the random edge elimination convolution channel to determine the random edge elimination vector; processing the syntactic dependency tree according to the distance relationship convolution channel to determine the distance relationship vector.

[0046] Among them, syntactic parsing refers to analyzing the grammatical structure in the target fault ticket through the syntactic parsing tool Spacy, that is, the syntactic dependency relationship between words, and then converting the syntactic dependency relationship of the target fault ticket into a tree form, and then forming an adjacency matrix based on the relationship between words, so as to facilitate the subsequent graph convolutional neural network to extract grammatical information. Each word in the sentence is used as a node of the tree. A sentence containing n words has n nodes, and the edges of the tree represent the syntactic dependency relationship between words. In order to utilize the syntactic features contained in the syntactic dependency tree, the syntactic dependency tree is usually converted into an adjacency matrix. The elements of the adjacency matrix are composed of 1 or 0. 1 indicates that there is a syntactic dependency relationship between the corresponding words, and 0 indicates that there is no syntactic dependency relationship. The syntactic dependency relationship of the target fault ticket is parsed to prepare for the next step of extracting grammatical information. In addition, considering that there are many different relationship features between the words in the target fault ticket, including syntactic dependency, relative type distance relationship, random edge elimination, and tree-based distance relationship, making good use of these relationship features will help determine whether the aspect word has a fault. In order to extract the four relationship features, a graph convolution channel was built for each relationship feature to perform graph convolution, forming a multi-channel graph convolutional neural network.

[0047] Specifically, although the semantic learning layer can effectively extract the semantic features of the context, it does not take into account the syntactic dependency relationship between the aspect words and the context. With the help of Spacy's dependency parser, the syntactic dependency tree of the target fault work order is obtained and converted into a directed graph. The directed graph contains the syntactic dependency relationship of all words in the fault description. The syntactic dependency tree and graph convolutional neural network can provide syntactic constraints for aspect words, reduce the spatial distance between the context and the related words for judging whether the aspect words are faulty, and enhance the influence of related words on whether the aspect words are faulty. For example, the target fault work order is the feedback from the customer of Chengxi 1st Road that the line near his home was installed in 2002. It has been more than 20 years since then, and it is already very dilapidated and may be unsafe. Among them, the aspect word is line. The judgment of whether the line is faulty is determined by dilapidated and unsafe, which are far away from the aspect words and have little influence on the judgment of aspect words, which is not conducive to the judgment of whether the aspect words are faulty. The syntactic dependency tree can be used to parse that shabby and unsafe are adjectives of the aspect word line, greatly shortening the distance between the line and related words shabby and unsafe. Then, convolution is performed through the graph convolutional neural network to extract the syntactic dependency vector.

[0048] Relative aspect distance relationship: words with different relative aspect word distances in the target fault ticket have different importance to the aspect word. The closer the word is to the aspect word, the more important it is to the judgment of the aspect word. For example, a customer on Nanjiao Road reported that a truck had scratched an electric meter and the meter fell off, and hoped that it would be handled as soon as possible. The word "falling" is closest to the aspect word "electricity meter", and should have the greatest impact on judging whether the meter is faulty. Based on this principle, a relative type distance weight coefficient w is added to each word.i , which is used to enhance the importance of contextual aspect words near type face words. By setting the weight coefficient w closer to the aspect words i The larger the value, the weight coefficient of the aspect word is set to 0, as shown below: Among them, w i Represents the position weight of the word vector representation of the i-th node, and the corresponding word vector Multiply them to get a vector representation containing the relative type distance relationship

[0049] Random edge elimination: By randomly eliminating the neurons in the network layer or the connections between neurons, overfitting can be effectively alleviated and the generalization of the model can be improved. Then, by adding random parameters to the edges of the directed graph of the target fault work order and randomly eliminating the edges in the graph, overfitting and oversmoothing of the graph convolutional neural network can be alleviated, and excessive pursuit of high accuracy in judging whether the aspect words are faulty during training can be avoided. The random parameters are set to 1 or 0. 1 means that the edge of the directed graph is retained, and 0 means that the edge is eliminated. The random parameter value follows the Bernoulli distribution. After random edge elimination, the node vector is represented as

[0050] Tree-based distance relationship: Through the study and analysis of the syntactic dependency tree of the target fault ticket, it is found that the closer the distance between most words and the aspect word tree is, the greater the influence on judging whether the aspect word is faulty. Similar to the relative type distance relationship, a tree-based distance weight coefficient q is added to each word on the same syntactic dependency tree as the aspect word. i , used to enhance the influence of important words on aspect words. The closer the distance to the aspect word tree, the greater the weight coefficient q i The larger the size, as shown below: Where d represents the tree-based distance between the context word and the aspect word, q i Represents the tree-based distance weight coefficient of the word vector representation of the i-th node, and the corresponding word vector Multiply them together to get a vector representation containing the relative aspect distance relationship

[0051] On the basis of the above technical solution, the aspect word vector to be screened corresponding to the target fault work order is determined, including: performing average pooling processing on the syntactic dependency relationship vector, the relative type distance relationship vector, the random edge elimination vector and the distance relationship vector to determine the aspect word vector to be screened.

[0052] Specifically, these four node vectors processed by graph convolution are represented as and To perform fusion, average pooling operation is performed on the four node vector representations: in represents the final output of the multi-channel graph convolution, represents the node representation processed by the syntactic dependency tree, represents the node representation after relative aspect distance processing, represents the node representation after random edge elimination. Represents the node representation after the distance processing of the tree.

[0053] On the basis of the above technical solution, the fault aspect word corresponding to the target fault work order is determined, including: inputting the fault aspect word vector into the fully connected layer and the fault probability calculation function to determine the fault probability value corresponding to the fault aspect word vector; determining the fault aspect word from the fault aspect word vector based on the fault probability value and a preset fault probability threshold.

[0054] The fully connected layer may be a neural network structure for integrating local features extracted by previous layers (such as convolutional layers or pooling layers) into global features. The fault probability calculation function may be understood as a pre-set calculation function, such as a softmax function.

[0055] Specifically, in order to check whether the aspect words involved in the target fault work order are faulty, each aspect word vector is input into the fully connected layer and the softmax function to obtain the probability of whether the aspect word a is faulty: p(a) = softmax(W p r+b p ), where W p and b p All of them are learnable parameters. The larger the calculated probability of an aspect word, the more likely a fault is to occur. The application will screen out aspect words with a probability greater than 70% and judge them as aspect words with faults. This effectively avoids omissions caused by manual identification of fault types in lengthy target fault work orders, improves work efficiency, and reduces errors caused by manual labor.

[0056] S130: Determine fault warning information based on the fault aspect words, and provide a fault prompt according to the fault warning information.

[0057] The fault warning information may be understood as information used to warn relevant staff members of the fault.

[0058] Specifically, a pre-set fault information mapping table can be obtained, and the fault aspect words in the fault information mapping table can be matched to determine the fault warning information corresponding to the current fault aspect words, and then a fault prompt can be given according to the fault warning information. It should be noted that the fault information mapping table can be a pre-established table for storing the mapping relationship between fault aspect words and fault warning information. Exemplarily, the aspect words with faults are matched one by one to the corresponding fault handling professionals, and the corresponding aspect word fault warning information is generated, and notified by email, text message, mobile application, etc., to urge them to quickly handle the fault.

[0059] It should be noted that after obtaining the pending fault work order corresponding to the power service area, it also includes: determining the arrival time of the work order corresponding to the pending fault work order, and determining the timeout status of the pending fault work order according to the arrival time of the work order and a preset time threshold; when the timeout status of the pending fault work order is timeout, the timed fault work orders are summarized based on the automatic summary mechanism to generate a timed fault work order table; when the timeout status of the pending fault work order is not timeout, the pending fault work order is pre-processed to determine the target fault work order.

[0060] Specifically, for work orders that have timed out, the traditional statistical method requires manual login to the system to calculate, identify and collect overtime work orders, which is easy to miss overtime work orders, causing work orders to be timed out for a long time, and customer power restoration needs cannot be met, seriously affecting customer experience. In order to improve work efficiency and accuracy and enhance user experience, an automatic summary mechanism is designed to automatically summarize work orders into tables on a daily, weekly and monthly basis, so that detailed information about work order timeouts can be easily understood. In addition, a timed notification mechanism is set up according to the needs of the business department. For example, a table with detailed information about overtime work orders is sent to relevant managers at 10 a.m. every day to urge professionals to handle overtime work orders in a timely manner, forming a closed loop for the entire process.

[0061] The technical solution of the embodiment of the present invention obtains the pending fault work order corresponding to the power service area, pre-processes the pending fault work order to determine the target fault work order, and then performs aspect-level text classification on the target fault work order to determine the fault aspect words corresponding to the target fault work order, and finally determines the fault alarm information based on the fault aspect words, and performs a fault prompt based on the fault alarm information. Based on the above technical solution, by integrating aspect-level text classification and automation methods, the efficiency and accuracy of fault repair work order processing are improved, achieving the technical effect of improving the level of service provided by power companies to users.

[0062] Embodiment 2

[0063] Figure 4This is a schematic diagram of the structure of a device for processing a power failure work order provided in the second embodiment of the present invention. Figure 4 As shown, the device includes: a data acquisition module 410, a text classification module 420 and a fault prompt module 430;

[0064] The data acquisition module 410 is used to acquire the pending fault work orders corresponding to the power service area, and pre-process the pending fault work orders to determine the target fault work orders;

[0065] A text classification module 420 is used to perform aspect-level text classification on the target fault work order and determine the fault aspect words corresponding to the target fault work order;

[0066] The fault prompt module 430 is used to determine fault warning information based on the fault aspect words, and to provide fault prompts according to the fault warning information.

[0067] Based on the above technical solution, the text classification module is used to input the target fault work order into the global word frequency statistical model to determine the fault work order text vector corresponding to the target fault work order; input the fault work order text vector into the semantic learning layer to determine the fault description semantic features corresponding to the target fault work order; and perform aspect-level text classification on the target fault work order based on the fault description semantic features.

[0068] On the basis of the above technical solution, the text classification module is used to process the semantic features of the fault description based on a multi-channel graph convolution layer to determine the aspect word vector to be screened corresponding to the target fault work order; determine the attention weight corresponding to each aspect word vector to be screened based on the gated attention mechanism; and determine the fault aspect word vector from each aspect word vector to be screened according to the attention weight.

[0069] Based on the above technical solution, the text classification module is used to input the fault aspect word vector into at least the fully connected layer and the fault probability calculation function to determine the fault probability value corresponding to the fault aspect word vector; and determine the fault aspect word from the fault aspect word vector based on the fault probability value and a preset fault probability threshold.

[0070] On the basis of the above technical scheme, the multi-channel graph convolution layer includes a syntactic dependency graph convolution channel, a relative aspect distance relationship convolution channel, a random edge elimination convolution channel and a distance relationship convolution channel. The text classification module is used to obtain the syntactic dependency tree of the target fault work order based on the dependency parser, and convert the syntactic dependency tree into a directed graph; process the directed graph according to the dependency graph convolution channel to determine the syntactic dependency vector; process the directed graph according to the relative aspect distance relationship convolution channel to determine the relative type distance relationship vector; process the directed graph according to the random edge elimination convolution channel to determine the random edge elimination vector; process the syntactic dependency tree according to the distance relationship convolution channel to determine the distance relationship vector.

[0071] On the basis of the above technical solution, the text classification module is used to perform average pooling processing on the syntactic dependency relationship vector, the relative type distance relationship vector, the random edge elimination vector and the distance relationship vector to determine the aspect word vector to be screened.

[0072] Based on the above technical solution, the semantic learning layer is formed by combining a stacked bidirectional long short-term memory network layer and a random dense residual connection layer; the stacked bidirectional long short-term memory network layer is used to extract semantic information from the fault work order text vector.

[0073] The technical solution of the embodiment of the present invention obtains the pending fault work order corresponding to the power service area, pre-processes the pending fault work order to determine the target fault work order, and then performs aspect-level text classification on the target fault work order to determine the fault aspect words corresponding to the target fault work order, and finally determines the fault alarm information based on the fault aspect words, and performs a fault prompt based on the fault alarm information. Based on the above technical solution, by integrating aspect-level text classification and automation methods, the efficiency and accuracy of fault repair work order processing are improved, achieving the technical effect of improving the level of service provided by power companies to users.

[0074] The power failure work order processing device provided in the embodiment of the present invention can execute the power failure work order processing method provided in any embodiment of the present invention, and has the corresponding functional modules and

[0075] Beneficial Effects

[0076] Embodiment 3

[0077] Figure 5A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0078] like Figure 5 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0079] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0080] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a method for processing a power fault work order.

[0081] In some embodiments, the method for processing a power fault work order may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for processing a power fault work order described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the method for processing a power fault work order in any other appropriate manner (e.g., by means of firmware).

[0082] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0083] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0084] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0085] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0086] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0087] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0088] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0089] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for processing a power failure work order, characterized in that: include: Obtaining pending fault work orders corresponding to the power service area, and pre-processing the pending fault work orders to determine target fault work orders; Performing aspect-level text classification on the target fault work order to determine a fault aspect word corresponding to the target fault work order; Fault warning information is determined based on the fault aspect words, and a fault prompt is given according to the fault warning information.

2. The method according to claim 1, characterized in that The aspect-level text classification of the target fault work order includes: Input the target fault work order into a global word frequency statistical model to determine a fault work word vector corresponding to the target fault work order; Inputting the fault work order word vector into the semantic learning layer to determine the fault description semantic features corresponding to the target fault work order; The target fault work order is subjected to aspect-level text classification based on the fault description semantic features.

3. The method according to claim 2, characterized in that The performing aspect-level text classification on the target fault work order based on the fault description semantic features includes: Processing the semantic features of the fault description based on a multi-channel graph convolutional layer to determine the aspect word vector to be screened corresponding to the target fault work order; Determine the attention weight corresponding to each of the aspect word vectors to be screened based on the gated attention mechanism; Determine a fault aspect word vector from each of the aspect word vectors to be screened according to the attention weight.

4. The method according to claim 3, characterized in that: Determining the fault aspect words corresponding to the target fault work order, including: Inputting the fault aspect word vector into a fully connected layer and a fault probability calculation function to determine a fault probability value corresponding to the fault aspect word vector; The fault aspect word is determined from the fault aspect word vector based on the fault probability value and a preset fault probability threshold.

5. The method according to claim 3, characterized in that: The multi-channel graph convolution layer includes a syntactic dependency graph convolution channel, a relative aspect distance relationship convolution channel, a random edge elimination convolution channel and a distance relationship convolution channel. The processing of the fault description semantic features based on the multi-channel graph convolution layer includes: Obtaining a syntactic dependency tree of the target fault work order based on a dependency parser, and converting the syntactic dependency tree into a directed graph; Processing the directed graph according to the dependency graph convolution channel to determine a syntactic dependency vector; Processing the directed graph according to the relative aspect distance relationship convolution channel to determine a relative type distance relationship vector; Processing the directed graph according to the random edge elimination convolution channel to determine a random edge elimination vector; The syntactic dependency tree is processed according to the distance relationship convolution channel to determine a distance relationship vector.

6. The method according to claim 5, characterized in that Determining the aspect word vector to be screened corresponding to the target fault work order includes: Average pooling is performed on the syntactic dependency relationship vector, the relative type distance relationship vector, the random edge elimination vector and the distance relationship vector to determine the aspect word vector to be screened.

7. The method according to claim 2, characterized in that The semantic learning layer is formed by combining a stacked bidirectional long short-term memory network layer and a random dense residual connection layer; the stacked bidirectional long short-term memory network layer is used to extract semantic information from the fault word vector.

8. A device for processing power failure work orders, characterized in that: include: A data acquisition module, used to acquire pending fault work orders corresponding to the power service area, and pre-process the pending fault work orders to determine target fault work orders; A text classification module, used to perform aspect-level text classification on the target fault work order and determine the fault aspect words corresponding to the target fault work order; The fault prompt module is used to determine fault warning information based on the fault aspect words, and to provide fault prompts according to the fault warning information.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for processing a power fault work order according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for processing a power fault work order according to any one of claims 1 to 7 when executed.

Images