Power document generation method, system and equipment based on sentence drawing retrieval and medium

Abstract

The invention discloses an electric power document generation method, system and device based on sentence division retrieval and a medium, which are applied to the field of document generation and comprise the steps of encoding an electric power equipment image through a visual encoder based on a VLLM model structure to obtain a high-dimensional dense visual feature, performing dimension reduction processing on the high-dimensional dense visual feature by using an abstract module, and obtaining a high-dimensional dense visual feature; extracting low-dimensional visual embedding, performing theme recognition on the low-dimensional visual embedding through a theme division encoder, generating a common theme query vector and a rare theme query vector, retrieving sentences in a power report training set, extracting a statement set matched with the query vector, and generating requirements according to the statement set and a power document. And selecting related statements through a preset statement screening rule to generate the power document. According to the method, efficient monitoring and fault diagnosis of the state of the power equipment are realized, the power document with high pertinence and clear logic is automatically generated, and the intelligent level and the working efficiency of operation and maintenance of the power equipment are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of document generation, and in particular to a power document generation method, system, device and medium based on sentence segmentation retrieval. BACKGROUND

[0002] With the digital transformation of the power industry, the operation and maintenance data of power equipment are growing explosively, generating a large amount of structured and unstructured document resources. These documents contain rich device status, fault information and operation and maintenance records, which have important value for the intelligent management of the power system.

[0003] The existing power document generation method has many inconveniences when processing these massive data. The traditional document generation method usually relies on manual writing or simple template filling. This method not only consumes time and effort, but also cannot guarantee the accuracy and consistency of the document. In addition, the existing technology lacks an effective integration mechanism when processing power equipment images and text information, resulting in the generated document often failing to fully reflect the real state of the equipment. Especially in the face of complex and variable power equipment failure and operation and maintenance scenarios, the limitations of the existing technology are more obvious, and it is unable to quickly generate high-quality document reports.

[0004] Therefore, how to efficiently and accurately generate power documents and realize intelligent processing of power equipment images and text information has become a technical problem to be solved by those skilled in the art. SUMMARY

[0005] The present application provides a power document generation method, system, device and medium based on sentence segmentation retrieval to realize accurate generation of power documents.

[0006] To solve the above technical problems, the present application provides a power document generation method based on sentence segmentation retrieval, comprising: The power equipment image is encoded by a visual encoder based on the VLLM model structure to obtain high-dimensional dense visual features of the power equipment image.

[0007] The high-dimensional dense visual features are processed by dimension reduction to extract low-dimensional visual embedding information.

[0008] The low-dimensional visual embedding information is subjected to field recognition to generate a first field query vector and a second field query vector.

[0009] In the power report training set, the first field query vector and the second field query vector are respectively used as the reference for retrieval to extract a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector.

[0010] generate a power document corresponding to the power equipment image according to the first sentence set and the second sentence set based on a selected power document generation requirement.

[0011] Further, the power equipment image is encoded by the visual encoder based on the VLLM model structure to obtain high-dimensional dense visual features of the power equipment image, including: The power equipment image is input into the visual encoder based on the VLLM model structure.

[0012] The feature extraction layer of the visual encoder extracts features of the power equipment image to obtain initial feature representations.

[0013] The encoding layer of the visual encoder encodes the initial feature representations to obtain high-dimensional dense visual features.

[0014] Further, the high-dimensional dense visual features are processed by dimension reduction to extract low-dimensional visual embedding information, including: The high-dimensional dense visual features are input into the abstract module.

[0015] The multi-layer Transformer decoder of the abstract module processes the high-dimensional dense visual features to generate intermediate feature representations.

[0016] The projection layer of the abstract module projects the intermediate feature representations to a low-dimensional space to extract low-dimensional visual embedding information.

[0017] Further, the low-dimensional visual embedding information is subjected to field recognition to generate a first field query vector and a second field query vector, including: The multi-head self-attention mechanism of the topic division encoder extracts features of the low-dimensional visual embedding information to generate topic feature representations.

[0018] The classification layer of the topic division encoder classifies the topic feature representations to generate the first field query vector and the second field query vector.

[0019] Further, in the power report training set, the first field query vector and the second field query vector are respectively used as benchmarks for retrieval to extract a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector, including: The first field query vector and the second field query vector are respectively subjected to similarity calculation with sentence vectors of sentences in the power report training set.

[0020] According to the similarity calculation result, sentences with a matching degree higher than a preset threshold value are screened out from the first field query vector and the second field query vector, and are combined to form a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector.

[0021] Further, the document generation module is configured to generate, according to the first sentence set and the second sentence set, a power document corresponding to the power equipment image based on a selected power document generation requirement. According to the power document generation requirement, a set of key information keywords is determined.

[0022] From the first sentence set and the second sentence set, sentences containing any keyword in the set of key information keywords are screened out.

[0023] According to a preset sentence combination template, the screened-out sentences are combined to generate the power document.

[0024] Another embodiment of the present application provides a power document generation system based on sentence retrieval, comprising: An image encoding module is configured to encode a power equipment image by a visual encoder based on a VLLM model structure to obtain high-dimensional dense visual features of the power equipment image.

[0025] A feature dimension reduction module is configured to perform dimension reduction processing on the high-dimensional dense visual features to extract low-dimensional visual embedding information.

[0026] A vector generation module is configured to perform field recognition on the low-dimensional visual embedding information to generate a first field query vector and a second field query vector.

[0027] A sentence retrieval module is configured to perform retrieval in a power report training set by taking the first field query vector and the second field query vector as a reference respectively to extract a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector.

[0028] A document generation module is configured to generate, according to the first sentence set and the second sentence set, a power document corresponding to the power equipment image based on a selected power document generation requirement.

[0029] Further, the vector generation module is configured to: The low-dimensional visual embedding information is subjected to feature extraction by a multi-head self-attention mechanism of a topic division encoder to generate topic feature representation.

[0030] The topic feature representation is classified by a classification layer of the topic division encoder, to generate a first field query vector and a second field query vector.

[0031] A further embodiment of the present application provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the power document generation method based on sentence retrieval as described above when executing the computer program.

[0032] A further embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the device where the computer readable storage medium is located implements the power document generation method based on sentence retrieval as described above when executing the computer program.

[0033] Compared with the prior art, the beneficial effects of the embodiments of the present application are at least one of the following: By efficiently processing power equipment images through the visual encoder and the abstract module, low-dimensional visual embeddings are extracted, redundant information is significantly reduced, feature expression capability is enhanced, and common and rare topic query vectors are accurately generated in combination with the topic division encoder, thereby realizing differentiated retrieval, improving retrieval accuracy and efficiency, and effectively making up for the shortcomings of traditional methods on long-tail data. The rare topics in the power field are crucial but scarce, and the topic division encoder is used to accurately identify relevant sentences, which fully reflects the operation status of the power system and improves the professionalism and integrity of the report. The relevant sentences are selected from the retrieved sentence set through the preset sentence screening rules, and the sentences are combined using the preset sentence combination template to generate high-quality power documents, ensuring that the document structure is clear, the semantics is coherent, and the document meets the industry standards. The workload of manual writing is greatly reduced, the work efficiency is improved, and the labor cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 A step flowchart of the power document generation method based on sentence retrieval in one of the embodiments of the present application; Figure 2 A structural block diagram of the power document generation system based on sentence retrieval in one of the embodiments of the present application; Figure 3 A structural diagram of the computer device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0035] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described, obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application, and the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0036] In the description of the present application, the terms "first", "second", "third" and the like are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" and the like can be explicitly or implicitly included one or more of the features. In the description of the present application, unless otherwise stated, the meaning of "a plurality of" is two or more.

[0037] In the description of the present application, it should be noted that, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or indirectly connected through an intermediate medium, or it can be connected inside two elements. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are for the purpose of description only, and cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood in specific cases.

[0038] In the description of the present application, it should be noted that, unless otherwise defined, all technical and scientific terms used by the present application are the same as those commonly understood by those skilled in the art of the present application. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments, and are not intended to limit the present application. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood in specific cases.

[0039] An embodiment of the present application provides a power document generation method based on sentence segmentation retrieval, specifically, please refer to Figure 1 , Figure 1 The step flow chart of the power document generation method based on sentence segmentation retrieval in one embodiment of the present application is shown, which includes steps S11-S15: S11, encode the power equipment image through a visual encoder based on a VLLM model structure to obtain high-dimensional dense visual features of the power equipment image.

[0040] With the continuous development of the power system and the acceleration of the intelligentization process, the operation and maintenance management of power equipment faces higher requirements. The health status of power equipment is directly related to the stable operation of the power system, and traditional equipment detection methods often rely on manual inspection, which is not only inefficient but also easily affected by human factors. Therefore, using advanced technical means to automatically detect and assess the status of power equipment has become an important research direction. In recent years, computer vision technology has made significant progress in many fields, and its application in power equipment detection has gradually attracted attention. By encoding the power equipment image through a visual encoder based on a VLLM (Vision-Language Large Model) model structure, high-dimensional dense visual features can be effectively extracted, providing strong support for power equipment status monitoring and fault diagnosis. The specific implementation process is as follows: In the actual application of power equipment detection, the image data of the power equipment first needs to be input into the visual encoder based on the VLLM model structure. The VLLM model is a large model that combines visual and language processing capabilities, and its visual encoder part is specifically designed to process image data. This encoder can preliminarily process and analyze the input power equipment image, laying the foundation for subsequent feature extraction and encoding work. Specifically, the power equipment image of this embodiment contains rich information, including the appearance of the equipment, the status of the components, the surrounding environment, etc. Inputting these images into the visual encoder will convert the complex information in the images into processable feature data.

[0041] The feature extraction layer inside the visual encoder is the key part of realizing image feature extraction. When the power equipment image enters the visual encoder, the feature extraction layer will perform a series of processing on it to extract important features in the image.

[0042] This process is similar to the preliminary perception of the human visual system for objects, which can identify basic visual elements such as edges, textures, shapes, etc. in the image. Through the combination and analysis of these basic elements, the initial feature representation of the image can be obtained. The feature extraction in this stage is mainly to filter out the key information related to the status of the power equipment from the original image, remove irrelevant noise and redundant data, and thus provide a clearer and more valuable data basis for the subsequent encoding process. For example, for an image of a transformer, the feature extraction layer can identify the shape of the transformer's shell, the texture of the cooling fins, the position of the connecting lines, etc.

[0043] After obtaining the initial feature representation, the encoding layer of the visual encoder will further encode these features. The role of the encoding layer is to transform the initial feature representation into a more abstract and dense high-dimensional feature vector. This process can be seen as a kind of deep processing and compression of features, through complex neural network structure and algorithm, the information in the initial features is recombined and optimized, so that it can better reflect the state characteristics of power equipment. High-dimensional dense visual features have stronger expression and discrimination ability, which can more accurately describe the image feature differences of power equipment in different states. For example, for normally running power equipment and power equipment with faults, the high-dimensional dense visual features will have obvious differences in numerical value, and these differences can be used by subsequent analysis models to accurately judge the state of power equipment and diagnose faults.

[0044] S12, dimensionality reduction processing is performed on the high-dimensional dense visual features to extract low-dimensional visual embedding information.

[0045] In the image processing flow of power equipment, the high-dimensional dense visual features processed by the visual encoder need to be further processed to extract key information, so this embodiment inputs these high-dimensional features into a specially designed abstraction module. The role of the abstraction module is to remove redundant information in high-dimensional features through a series of complex processing steps, while retaining the most critical information for power equipment state evaluation. The abstraction module can be regarded as an information filter, which can identify which features are useful for subsequent analysis and which can be ignored.

[0046] Inside the abstraction module, the multi-layer Transformer decoder is the core component for feature processing. The Transformer architecture is widely adopted due to its outstanding performance in natural language processing and computer vision. In this stage, high-dimensional dense visual features are input into the multi-layer Transformer decoder, and each layer of decoder processes features through self-attention mechanism. The self-attention mechanism allows the model to dynamically focus on other related features when processing each feature, thereby better capturing the relationships and dependencies between features.

[0047] In this way, the multi-layer Transformer decoder can gradually extract more abstract and advanced feature representations, which not only contain the information of the original high-dimensional features, but also enhance the semantic association between features through the processing of the decoder. For example, in power equipment images, some features may be related to the normal running state of the equipment, while other features may be related to potential fault patterns.

[0048] After generating the intermediate feature representation, the next step is to project these features into a low-dimensional space to extract low-dimensional visual embedding information, which is achieved through the projection layer in the abstraction module. The role of the projection layer is to map high-dimensional intermediate feature representations into a low-dimensional space while trying to preserve the semantic information and structural characteristics of the original features. In the low-dimensional space, although the dimensionality of the features has been reduced, the low-dimensional embedding can still effectively represent the state of the power equipment through the designed projection method. Projection layers usually use linear or nonlinear transformation methods to achieve this goal. For example, principal component analysis (PCA) or other advanced dimensionality reduction techniques can be used to reduce the dimensionality of the features while maximizing the variance information in the features. In this way, low-dimensional visual embedding information can represent the key features of power equipment images in a more compact form, enabling more efficient computation and analysis in subsequent applications such as equipment state evaluation and fault diagnosis.

[0049] S13, field recognition is performed on the low-dimensional visual embedding information to generate a first field query vector and a second field query vector.

[0050] Although the low-dimensional visual embedding has effectively compressed the feature dimension, in the analysis of power equipment images, in order to further improve the understanding and analysis ability of the state of the power equipment, it is still necessary to conduct more in-depth field recognition on these low-dimensional features, so as to realize topic recognition.

[0051] Multi-head self-attention mechanism is a powerful feature extraction tool that can capture complex relationships and dependencies between features. In the topic division encoder of the embodiment, the multi-head self-attention mechanism generates more rich and semantic topic feature representations through processing of the low-dimensional visual embedding. Specifically, the multi-head self-attention mechanism allows the model to independently focus on different aspects of the features in different subspaces, enabling a more comprehensive understanding of the semantic information of the features. For example, in power equipment images, some features may be related to the normal operating state of the equipment, while other features may be related to potential fault patterns.

[0052] After generating the topic feature representation, the next step is to classify these features through the classification layer of the topic division encoder. The role of the classification layer is to further subdivide the topic feature representation into common topics and rare topics.

[0053] Common topics usually refer to frequently occurring feature patterns in power equipment images, which are related to the normal operating state of the equipment. Rare topics are related to abnormal states or potential faults of the equipment. Through the processing of the classification layer, two types of query vectors can be generated: a first field query vector and a second field query vector, wherein the common topic query vector is the first field query vector and the rare topic query vector is the second field query vector.

[0054] The two query vectors are used for subsequent retrieval and analysis, helping the system to more accurately identify and handle the status of power equipment. For example, the common topic query vector can be used to quickly identify the normal operating state of the equipment, while the rare topic query vector can be used to detect and warn potential failures.

[0055] S14, in the power report training set, respectively, with the first field query vector and the second field query vector as the reference for retrieval, and the first sentence set corresponding to the first field query vector is extracted, and the second sentence set corresponding to the second field query vector is extracted.

[0056] In order to make full use of the information in the text data, the generated common topic query vector and rare topic query vector need to be matched with the sentences in the power report training set respectively.

[0057] The power report training set contains a large amount of text information related to the operating state of power equipment, which exists in the form of sentences. Each sentence can be converted into a sentence vector through natural language processing technology. The sentence vector is a numerical representation of the sentence, which can reflect the semantic content of the sentence. By calculating the similarity between the common topic query vector and the rare topic query vector and the sentence vector, the matching degree of the sentence and the query vector can be quantified. In this embodiment, the similarity calculation can use cosine similarity, Euclidean distance or other appropriate distance measurement methods. For example, the cosine similarity can measure the angle between two vectors, and the smaller the angle, the higher the similarity.

[0058] According to the results of similarity calculation, screening is carried out. In order to ensure that the extracted sentences have high relevance with the query vector, a preset threshold needs to be set.

[0059] The preset threshold is a standard for judging whether the similarity is high enough. Only when the similarity of the sentence vector of the sentence and the query vector is higher than the threshold, the sentence will be selected. In this way, the system can screen out sentences with high matching degree with the common topic and rare topic query vectors from the power report training set.

[0060] The selected sentences contain detailed information related to the state of the power equipment, such as the operating parameters of the equipment, the description of the failure, the maintenance record, etc. These selected sentences are combined into a sentence set, wherein the sentence set corresponding to the common topic query vector is the first sentence set, and the sentence set corresponding to the rare topic query vector is the second sentence set, which can provide more comprehensive text information support for the maintenance and fault diagnosis of power equipment. For example, if a rare topic query vector has a high matching degree with a sentence describing the failure of the equipment, this sentence can be extracted to provide important clues for fault diagnosis for technicians.

[0061] S15, based on the selected power document generation requirements, generating a power document corresponding to the power equipment image according to the first sentence set and the second sentence set.

[0062] Before generating the power document, the specific requirements of the document need to be clarified first. Different power documents focus on different information, such as the running state of the equipment, fault diagnosis, maintenance records, etc. Therefore, according to the generation requirements of the power document, a set of key information keyword set needs to be determined. These keywords are important basis for filtering sentences in the document generation process. If the purpose of the document is to record the fault diagnosis process of the equipment, the keyword set may include "fault", "abnormal", "repair", "detection", etc. By determining these keywords, it can be ensured that the generated document can accurately reflect the required information.

[0063] After determining the key information keyword set, the next step is to filter the sentences containing these keywords from the extracted first sentence set and second sentence set. Each sentence in the sentence set may contain information related to the state of the power equipment, but not all sentences are directly related to the generation requirements of the document. By keyword filtering, only those sentences highly related to the document requirements can be selected. For example, if the keyword set contains "fault", all sentences containing the word "fault" will be filtered out.

[0064] After filtering the sentences related to the key information keywords, the next step is to combine these sentences into a complete power document. In order to ensure the logicality and readability of the document, pre-set sentence combination templates need to be used. These templates define the structure and format of the document, such as how the beginning, middle and end of the document should be organized. By combining the filtered sentences according to the structure of the template, a complete and logically clear power document can be generated. For example, the template may stipulate that the beginning of the document should contain the basic information of the equipment, the middle part should describe the running state and fault condition of the equipment in detail, and the end part should provide maintenance suggestions and subsequent measures. In this way, the system can automatically generate a power document that meets specific requirements, improving the efficiency and quality of document generation. The final generated power document not only contains the key information related to the state of the power equipment, but also presents it in an easy-to-understand and easy-to-operate format to the user, providing strong support for the maintenance and management of power equipment.

[0065] The power document generation method based on sentence division retrieval provided by the application efficiently processes power equipment images through a visual encoder and an abstraction module, extracts low-dimensional visual embedding, significantly reduces redundant information, enhances feature expression capability, combines a topic division encoder, accurately generates common and rare topic query vectors, realizes differentiated retrieval, improves retrieval accuracy and efficiency, and for the rare topics that are crucial but scarce in the power field, the topic division encoder accurately identifies relevant sentences, effectively makes up for the shortcomings of traditional methods on long-tail data, comprehensively reflects the operation status of the power system, improves the professionalism and integrity of the report, selects relevant sentences from the retrieved sentence set through preset sentence screening rules, combines them using a preset sentence combination template, and generates high-quality power documents, ensuring that the document structure is clear, the semantics is coherent, and it meets industry standards, greatly reducing the workload of manual writing, improving work efficiency, and reducing labor costs.

[0066] The application also provides a power document generation system based on sentence division retrieval for executing the power document generation method based on sentence division retrieval as described above, Figure 2 The power document generation system based on sentence division retrieval is a structural block diagram of the application, and the system comprises: An image encoding module 21 is configured to encode the power equipment image through a visual encoder based on a VLLM model structure to obtain high-dimensional dense visual features of the power equipment image.

[0067] An feature dimension reduction module 22 is configured to perform dimension reduction processing on the high-dimensional dense visual features to extract low-dimensional visual embedding information.

[0068] A vector generation module 23 is configured to perform field recognition on the low-dimensional visual embedding information to generate a first field query vector and a second field query vector.

[0069] A sentence retrieval module 24 is configured to perform retrieval in a power report training set by taking the first field query vector and the second field query vector as references respectively to extract a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector.

[0070] A document generation module 25 is configured to generate a power document corresponding to the power equipment image according to the first sentence set and the second sentence set based on selected power document generation requirements.

[0071] The vector generation module is configured to: extract features from the low-dimensional visual embedding information through a multi-head self-attention mechanism of the topic division encoder to generate topic feature representations.

[0072] The topic feature representation is classified by a classification layer of the topic division encoder, to generate a first field query vector and a second field query vector.

[0073] The technical features and effects of the system provided by the embodiments of the present application are the same as those of the method provided by the embodiments of the present application, which are not repeated here. Each module in the above system can be realized by software, hardware and combinations thereof, in whole or in part. Each of the above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each of the above modules.

[0074] Referring to Figure 3 which is a structural block diagram of the computer device provided by the embodiments of the present application. The computer device provided by the embodiments of the present application includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the above power document generation method based on sentence division retrieval embodiments are implemented, such as steps S11-S15 described in the above embodiment. Figure 1 Or, when the processor executes the computer program, the functions of each module in the above device embodiments are implemented, such as modules 21-25 of the power document generation system based on sentence division retrieval.

[0075] For example, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments that can complete a specific function, which are used to describe the execution process of the computer program in the computer device.

[0076] The computer device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the schematic diagram is only an example of the computer device and does not limit the computer device, which can include more or fewer components than the diagram, or combine certain components, or different components, for example, the computer device can also include an input / output device, a network access device, a bus, etc.

[0077] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the computer device, and connects various parts of the computer device through various interfaces and lines.

[0078] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the computer device by running or executing the computer program and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store an operating system, at least one application program required for a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device or other volatile solid-state memory device.

[0079] The modules integrated in the computer device can be stored in a computer readable storage medium in the form of software function units when they are sold or used as independent products. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium. When the processor executes the computer program, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code.

[0080] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned various method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0081] Correspondingly, the embodiment of the present application provides a computer readable storage medium including a stored computer program. When the computer program is running, it controls the device where the computer readable storage medium is located to perform the steps in the power document generation method based on sentence retrieval as described in the above-mentioned embodiment, for example, the steps S11-S15 described in the above-mentioned embodiment. Figure 1

[0082] In summary, the power document generation method, system, computer device, and computer readable storage medium based on sentence retrieval provided by the embodiment of the present application have at least one of the following advantages compared with the prior art: ​The power equipment image is efficiently processed by the visual encoder and the abstraction module, low-dimensional visual embedding is extracted, redundant information is significantly reduced, feature expression capability is enhanced, a theme division encoder is combined, common and rare theme query vectors are accurately generated, differential retrieval is realized, retrieval precision and efficiency are improved, for rare themes which are scarce but crucial in the power field, the theme division encoder is used to accurately identify relevant sentences, effectively making up for the shortcomings of traditional methods on long-tail data, comprehensively reflecting the operation status of the power system, improving the professionalism and integrity of the report, relevant sentences are selected from the retrieved sentence set through preset sentence screening rules, and are combined using a preset sentence combination template, to generate high-quality power documents, ensuring that the document structure is clear, the semantics is coherent, and it meets the industry standards, greatly reducing the workload of manual writing, improving work efficiency and reducing labor costs.

[0083] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as limiting the scope of the patent of the present application. It should be noted that, for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A power document generation method based on sentence division search, characterized by, The method comprises the following steps: encoding the power equipment image through a visual encoder based on a VLLM model structure to obtain high-dimensional dense visual features of the power equipment image; dimension reduction processing is performed on the high-dimensional dense visual features to extract low-dimensional visual embedding information; field recognition is performed on the low-dimensional visual embedding information to generate a first field query vector and a second field query vector; in the power report training set, the first field query vector and the second field query vector are respectively taken as the reference for retrieval to extract a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector; based on the selected power document generation requirement, the first sentence set and the second sentence set are used to generate a power document corresponding to the power equipment image.

2. The power document generation method based on sentence division search according to claim 1, wherein, The method comprises the following steps: inputting the power equipment image into the visual encoder based on the VLLM model structure; feature extraction is performed on the power equipment image through a feature extraction layer of the visual encoder to obtain an initial feature representation; encoding is performed on the initial feature representation through an encoding layer of the visual encoder to obtain high-dimensional dense visual features.

3. The power document generation method based on sentence division search according to claim 1, wherein, The method comprises the following steps: inputting the high-dimensional dense visual features into an abstraction module; processing the high-dimensional dense visual features through a multi-layer Transformer decoder of the abstraction module to generate an intermediate feature representation; projecting the intermediate feature representation to a low-dimensional space through a projection layer of the abstraction module to extract low-dimensional visual embedding information.

4. The power document generation method based on sentence division search according to claim 1, wherein, The method comprises the following steps: feature extraction is performed on the low-dimensional visual embedding information through a multi-head self-attention mechanism of a topic division encoder to generate a topic feature representation; classification is performed on the topic feature representation through a classification layer of the topic division encoder to generate a first field query vector and a second field query vector.

5. The power document generation method based on sentence division search according to claim 1, wherein, The method comprises the following steps: similarity calculation is performed between the first field query vector and the second field query vector and the sentence vectors of the sentences in the power report training set; according to the similarity calculation result, sentences with a matching degree higher than a preset threshold value are screened out from the first field query vector and the second field query vector to combine a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector.

6. The power document generation method based on sentence division search according to claim 1, wherein, The power document generation requirement is selected based on the first sentence set and the second sentence set to generate a power document corresponding to the power equipment image, and the power document generation requirement comprises: determining a key information keyword set according to a power document generation requirement; selecting sentences containing any keyword in the key information keyword set from the first sentence set and the second sentence set; combining the selected sentences according to a preset sentence combination template to generate a power document.

7. A power document generation system based on sentence division search, characterized by, Comprise: An image encoding module for encoding the power equipment image through a visual encoder based on a VLLM model structure to obtain high-dimensional dense visual features of the power equipment image; A feature dimension reduction module for dimension reduction processing of the high-dimensional dense visual features to extract low-dimensional visual embedding information; A vector generation module for field recognition of the low-dimensional visual embedding information to generate a first field query vector and a second field query vector; A sentence retrieval module for retrieving in a power report training set respectively with the first field query vector and the second field query vector as a reference to extract a first sentence set corresponding to the first field query vector and a second sentence set corresponding to the second field query vector; A document generation module for generating a power document corresponding to the power equipment image based on a selected power document generation requirement according to the first sentence set and the second sentence set.

8. The power document generation system based on sentence division search according to Claim 7, wherein, The vector generation module is used for: extracting features of the low-dimensional visual embedding information through a multi-head self-attention mechanism of a topic division encoder to generate a topic feature representation; classifying the topic feature representation through a classification layer of the topic division encoder to generate a first field query vector and a second field query vector.

9. A computer device, comprising: The computer readable storage medium stores a computer program, wherein the device where the computer readable storage medium is located executes the computer program to realize the power document generation method based on sentence division retrieval as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, wherein the device where the computer readable storage medium is located executes the computer program to realize the power document generation method based on sentence division retrieval as claimed in any one of claims 1 to 7.

Images