Method and device for determining maintenance plan of power equipment

Abstract

The present application discloses a method and device for determining a maintenance plan for electric power equipment, and relates to the technical field of equipment maintenance. The electronic device can input the multimodal data of the electric power equipment into a fault diagnosis model, and obtain a first maintenance plan for the abnormal state of the electric power equipment output by the fault diagnosis model. Then, the electronic device can determine a target maintenance plan for the abnormal state based on the first maintenance plan and a large language model. Since there is no need to manually determine the target maintenance plan based on work experience, on the one hand, the problem of low reliability of the determined target maintenance plan due to manual misjudgment or omission can be avoided; on the other hand, there is no need to spend a long time to identify the problem, and there is no need to make decisions based on intuition, so the reliability of the determined target maintenance plan can be further improved, and the efficiency of determining the target maintenance plan can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of equipment maintenance, and in particular to a method and device for determining a maintenance plan for electric power equipment. Background Art

[0002] The power equipment in the power grid will inevitably fail during long-term operation, and the failure of the power equipment will have a great impact on the stable operation of the power grid. Therefore, it is necessary to inspect and repair the power equipment on a daily basis to ensure the stable operation of the power equipment.

[0003] In the related art, the staff can determine the maintenance plan for the power equipment based on their own work experience and use the maintenance plan to repair the power equipment. However, in the related art, the maintenance plan is determined based on work experience, and the reliability of the determined maintenance plan may be low due to the staff's misjudgment, thereby resulting in low reliability of the maintenance of the power equipment. Summary of the invention

[0004] The present application provides a method and device for determining a maintenance plan for electric power equipment, which can solve the problem of low reliability of maintenance plans determined based on work experience in related technologies. The technical solution is as follows:

[0005] In one aspect, a method for determining a maintenance plan for electric power equipment is provided, the method comprising:

[0006] Acquiring multimodal data of the electric power equipment;

[0007] Inputting the multimodal data into a fault diagnosis model to obtain a first maintenance plan for the abnormal state of the power equipment output by the fault diagnosis model;

[0008] Based on the first maintenance plan and the large language model, a target maintenance plan for the abnormal state is determined.

[0009] Optionally, determining a target maintenance plan for the abnormal state based on the first maintenance plan and the large language model includes:

[0010] Acquire a second maintenance plan for the abnormal state through a large language model;

[0011] If the second maintenance plan is obtained based on the first maintenance plan, determining the second maintenance plan as a target maintenance plan for the abnormal state;

[0012] If the second maintenance plan is not obtained based on the first maintenance plan, a target maintenance plan for the abnormal state is determined based on the first maintenance plan and the second maintenance plan.

[0013] Optionally, obtaining a second maintenance plan for the abnormal state through a large language model includes:

[0014] Based on the target problem for the abnormal state, obtaining reference maintenance knowledge related to the abnormal state;

[0015] Generate prompt words based on the target problem and the reference maintenance knowledge;

[0016] The prompt word is input into a large language model to obtain a second maintenance plan for the abnormal state output by the large language model.

[0017] Optionally, based on the target problem for the abnormal state, obtaining reference maintenance knowledge related to the abnormal state includes:

[0018] Based on the target question, a plurality of candidate text vectors are obtained from a vector database, wherein the similarity between each candidate text vector and the vector result of the target question is greater than a similarity threshold;

[0019] Based on the multiple candidate text vectors, obtaining at least two target text vectors;

[0020] Based on the at least two target text vectors, reference maintenance knowledge related to the abnormal state is acquired from a maintenance knowledge base.

[0021] Optionally, obtaining at least two target text vectors based on the multiple candidate text vectors includes:

[0022] sorting the plurality of candidate text vectors;

[0023] Obtain at least two target text vectors from the sorted plurality of candidate text vectors;

[0024] Among them, the similarity between each of the target text vectors and the vector result is greater than the similarity between other text vectors and the vector result, and the other text vectors are text vectors other than the target text vector among the multiple candidate text vectors.

[0025] Optionally, before obtaining a plurality of candidate text vectors from a vector database based on the target question, the method further includes:

[0026] Acquire multiple maintenance knowledge texts of the electric power equipment;

[0027] Splitting each of the maintenance knowledge texts into a plurality of text blocks;

[0028] Based on the multiple text blocks, the vector database is obtained.

[0029] Optionally, acquiring the vector database based on the multiple text blocks includes:

[0030] Obtaining a text summary of each of the text blocks;

[0031] Each of the text summaries is converted into a text vector to obtain the vector database.

[0032] Optionally, converting each of the text summaries into a text vector to obtain the vector database includes:

[0033] Each of the text summaries is input into a text vectorization model to obtain a text vector of the text summaries output by the text vectorization model.

[0034] Optionally, the text vectorization model is a pre-trained language model.

[0035] Optionally, determining a target maintenance plan for the abnormal state based on the first maintenance plan and the second maintenance plan includes:

[0036] The first maintenance plan and the second maintenance plan are integrated to obtain a target maintenance plan for the abnormal state.

[0037] On the other hand, a device for determining a maintenance plan for electric power equipment is provided, the device comprising:

[0038] An acquisition module, used for acquiring multimodal data of the electric power equipment;

[0039] A diagnosis module, used for inputting the multimodal data into a fault diagnosis model to obtain a first maintenance plan for the abnormal state of the power equipment output by the fault diagnosis model;

[0040] A determination module is used to determine a target maintenance plan for the abnormal state based on the first maintenance plan and a large language model.

[0041] On the other hand, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for determining a maintenance plan for the power equipment as described in the above aspects is implemented.

[0042] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for determining a maintenance plan for power equipment as described in the above aspects is implemented.

[0043] The beneficial effects of the technical solution claimed in this application include at least:

[0044] The embodiment of the present application provides a method and device for determining a maintenance plan for an electric device, wherein the electronic device can input the multimodal data of the electric device into a fault diagnosis model to obtain a first maintenance plan for the abnormal state of the electric device output by the fault diagnosis model. Then, the electronic device can determine the target maintenance plan for the abnormal state based on the first maintenance plan and the large language model. Since there is no need to manually determine the target maintenance plan based on work experience, on the one hand, the problem of low reliability of the determined target maintenance plan due to manual misjudgment or omission can be avoided; on the other hand, there is no need to spend a long time to identify the problem, and there is no need to make decisions based on intuition, so the reliability of the determined target maintenance plan can be further improved, and the efficiency of determining the target maintenance plan can be improved. In addition, since the target maintenance plan can be obtained based on the first maintenance plan and the large language model, that is, the target maintenance plan can be obtained through multiple models, it can be further ensured that the reliability of the determined target maintenance plan is high.

[0045] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of a method for determining a maintenance plan for electric power equipment provided in an embodiment of the present application;

[0047] Figure 2 is a flow chart of another method for determining a maintenance plan for electric power equipment provided in an embodiment of the present application;

[0048] Figure 3 is a flow chart of a method for constructing a vector database provided in an embodiment of the present application;

[0049] Figure 4 It is a flow chart of a method for acquiring reference maintenance knowledge related to abnormal status of electric power equipment provided by an embodiment of the present application;

[0050] Figure 5 It is a structural schematic diagram of a device for determining a maintenance plan for electric power equipment provided in an embodiment of the present application;

[0051] Figure 6 This is a structural diagram of another device for determining a maintenance plan for electric power equipment provided in an embodiment of the present application.

[0052] Figure 7 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0054] Power equipment is a basic component of the power grid. There are many types of power equipment, which are huge in number and widely distributed in all power supply areas. Power equipment will inevitably fail in long-term operation. Since power equipment plays a very critical role in the power grid, no matter which equipment fails, it will have a significant impact on the stable operation of the entire power grid. Therefore, it is necessary to inspect and maintain power equipment on a daily basis to ensure its stable operation.

[0055] At present, the maintenance plan is mainly determined based on the work experience of technicians, and the power equipment is maintained based on the maintenance plan. Specifically, technicians continuously accumulate relevant knowledge on the maintenance of power equipment through learning, training and mutual communication. In the course of work, technicians accumulate relevant practical experience through actual operation and guidance from experienced masters. As a result, technicians can obtain maintenance plans based on their work experience and complete the maintenance of power equipment based on the maintenance plan.

[0056] However, maintenance based on work experience lacks standardized processes and methods, resulting in uneven quality and low reliability of the maintenance plans determined; and misjudgments or omissions may occur based on work experience (especially when dealing with complex or rare problems), further resulting in low reliability of maintenance plans. In addition, maintenance based on work inspection may require more time to identify and solve problems, and the decision-making process may be based on intuition and trial and error rather than scientific methods and data, resulting in low efficiency and low reliability in determining maintenance plans.

[0057] The embodiment of the present application provides a method for determining a maintenance plan for power equipment, and the method is applied to an electronic device. Optionally, the electronic device can be a mobile terminal, a fixed terminal or a server. The mobile terminal can be a mobile phone, a tablet computer or a laptop computer. The fixed terminal can be a desktop computer. The server can be a single server, or a server cluster composed of several servers, or a cloud computing service center. Figure 1 , the method comprising:

[0058] Step 101: Acquire multimodal data of power equipment.

[0059] The electronic device can obtain the operating status data and monitoring data of the power equipment, and fuse the operating status data and the monitoring data to obtain the multimodal data of the power equipment. The operating status data may include at least one of operating current, operating voltage, operating temperature, load, displacement, speed and acceleration. The monitoring data may include one of monitoring images and monitoring videos. For example, the multimodal data may include operating current, operating voltage, operating temperature, load, displacement, speed and acceleration. The monitoring data may include monitoring images and monitoring videos.

[0060] In an embodiment of the present application, an electronic device can establish a communication connection with a sensor group for collecting operating status data and monitoring data of an electric power device. After the sensor group collects the operating status data and monitoring data of the electric power device, the operating status data and monitoring data can be uploaded to the electronic device in real time through the communication connection. Accordingly, the electronic device can obtain the operating status data and monitoring data in real time. In other words, the electronic device can obtain the real-time operating status data and real-time monitoring data of the electric power device.

[0061] Step 102: Input the multimodal data of the electric power equipment into a fault diagnosis model to obtain a first maintenance plan for the abnormal state of the electric power equipment output by the fault diagnosis model.

[0062] The electronic device being in an abnormal state may mean that the power device has been operating abnormally, or the power device has signs of operating abnormally. That is, the fault diagnosis model provided by the embodiment of the present application can not only diagnose the fault of the electronic device, but also predict the fault.

[0063] Step 103: Determine a target maintenance plan for the abnormal state of the power equipment based on the first maintenance plan and the large language model.

[0064] In an embodiment of the present application, the electronic device can obtain a second maintenance plan for the abnormal state through a large language model. The second maintenance plan may be obtained based on the first maintenance plan, or may not be obtained based on the first maintenance plan. If the second maintenance plan is obtained based on the first maintenance plan, the electronic device can determine the second maintenance plan as the target maintenance plan for the abnormal state. If the second maintenance plan is not obtained based on the first maintenance plan, the electronic device can determine the target maintenance plan for the abnormal state based on the first maintenance plan and the second maintenance plan.

[0065] In summary, the embodiment of the present application provides a method for determining a maintenance plan for electric equipment, and the electronic device can input the multimodal data of the electric equipment into the fault diagnosis model to obtain the first maintenance plan for the abnormal state of the electric equipment output by the fault diagnosis model. Then, the electronic device can determine the target maintenance plan for the abnormal state based on the first maintenance plan and the large language model. Since there is no need to manually determine the target maintenance plan based on work experience, on the one hand, the problem of low reliability of the determined target maintenance plan due to manual misjudgment or omission can be avoided; on the other hand, there is no need to spend a long time to identify the problem, and there is no need to make decisions based on intuition, so the reliability of the determined target maintenance plan can be further improved, and the determination efficiency of the target maintenance plan can be improved. In addition, since the target maintenance plan can be obtained based on the first maintenance plan and the large language model, that is, the target maintenance plan can be obtained through multiple models, it can be further ensured that the reliability of the determined target maintenance plan is high.

[0066] Figure 2 1 is a flow chart of another method for determining a maintenance plan for electric power equipment provided in an embodiment of the present application, and the method can be applied to electronic equipment. Figure 2 , the method may include:

[0067] Step 201: Build a vector database.

[0068] The vector database may include: a plurality of text vectors related to maintenance knowledge of electric power equipment. Optionally, the vector database may be a milvus vector library or a faiss vector library.

[0069] In the examples of this application, see Figure 3 , the process of constructing a vector database by an electronic device may include:

[0070] Step 2011: Obtain multiple maintenance knowledge texts of power equipment.

[0071] In an embodiment of the present application, the electronic device can obtain multiple maintenance knowledge texts of the power equipment, and store each maintenance knowledge text in a database to obtain a maintenance knowledge base (i.e., a maintenance knowledge database). Each maintenance knowledge text records knowledge content related to the maintenance of the power equipment. The knowledge content may include: equipment information, maintenance records, fault data, maintenance plans, data under normal operation and abnormal operation (the data includes operating status data and monitoring data), safety specifications, training materials and case data, etc. of the power equipment.

[0072] The equipment information may include the model, manufacturer, date of manufacture, main technical parameters, structure diagram, circuit diagram and user manual of the power equipment. The maintenance record may include: the time, reason, maintenance personnel, accessories used for maintenance, maintenance process, maintenance results and time when the equipment resumes operation for each maintenance of the power equipment. The fault data may include: the fault mode (also called fault type), fault cause, fault impact, fault handling method and preventive measures of the power equipment.

[0073] The maintenance plan may include standardized maintenance procedures, maintenance cycles, preventive maintenance recommendations, and items for regular inspections. Data under normal and abnormal operation may include: operating status data of power equipment, normal range and abnormal threshold of operating status data, monitoring data of power equipment under normal operation, and monitoring data of power equipment under abnormal operation. Both the operating status data and monitoring data can be collected by sensors.

[0074] Safety regulations may include industry standards, safe operating procedures, emergency response plans, and personal protective equipment requirements. Training materials may include equipment operation guides, maintenance training materials, safety training materials, and fault diagnosis tutorials. Case data may include successful maintenance cases, complex fault resolution processes, and lessons learned from the cases.

[0075] It can be seen that the maintenance knowledge text not only includes theoretical texts related to the maintenance of power equipment (such as maintenance plans, full specifications, training materials and case data, etc.), but also includes data during the actual operation of power equipment. That is, the maintenance knowledge text obtained by the electronic equipment is more comprehensive and has richer levels, which can improve the accuracy and comprehensiveness of the generated second maintenance plan. In addition, the electronic equipment can periodically update the maintenance text in the maintenance knowledge base, thereby ensuring that the timeliness and accuracy of the constructed maintenance knowledge base are high, thereby effectively avoiding the maintenance effect affected by errors or inaccuracies in data quality.

[0076] Step 2012: divide each maintenance knowledge text into multiple text blocks.

[0077] It is understandable that, since the length of the maintenance knowledge text is generally long, the electronic device can divide each maintenance knowledge text into multiple text blocks after acquiring each maintenance knowledge text. In this way, the accuracy and efficiency of subsequent retrieval can be improved. Among them, each text block can be stored in the maintenance knowledge base.

[0078] In an embodiment of the present application, for each maintenance knowledge text among multiple maintenance knowledge texts, the electronic device can use recursive character text segmentation technology to segment the maintenance knowledge text into multiple text blocks.

[0079] Step 2013: Based on the multiple text blocks, obtain a vector database.

[0080] In an embodiment of the present application, for each text block among multiple text blocks, the electronic device can directly vectorize the text to convert the text block into a text vector, and record the text vector in a database to obtain a vector database.

[0081] Alternatively, for each text block, the electronic device may first obtain a text summary of the text block. Then, the electronic device may perform vectorization processing on the text summary to convert the text summary into a text vector, and record the text vector in a database, thereby obtaining a vector database. The text summary may be recorded in a maintenance knowledge base.

[0082] Compared with text blocks, text summaries have less redundant information, that is, text summaries are more concise and more targeted. Therefore, obtaining a vector database based on text summaries can improve the efficiency and accuracy of obtaining reference maintenance knowledge.

[0083] Optionally, the electronic device can use graph neural networks (GNN) to perform structured storage on the text vectors, so that the electronic device can support more intelligent knowledge reasoning and retrieval, and improve the retrieval efficiency of reference maintenance knowledge.

[0084] In an embodiment of the present application, for each text block (or text summary), the electronic device may perform one-hot encoding on the text block (or text summary) to vectorize the text block (or text summary). Alternatively, the electronic device pre-stores a text vectorization model. For each text block (or text summary), the electronic device may input the text block (or text summary) into the text vectorization model to obtain a text vector of the text block output by the text vectorization model.

[0085] Optionally, the text vectorization model can be a deep learning model. For text blocks (or text summaries) that record data under normal operation and abnormal operation, the text vectorization model can be a multimodal Transformer model. The multimodal Transformer model can jointly vectorize multimodal data, so that data such as images, text (such as text recording operation data) and videos can be integrated into a unified vector space, thereby improving the accuracy of vectorizing text blocks (or text summaries).

[0086] For text blocks (or text summaries) that record knowledge content other than data under normal operation and abnormal operation, the text vectorization model can be a pre-trained language model. The pre-trained language model can be a BERT model or a RoBERTa model. Since the pre-trained language model can capture the deep semantic relationship and contextual information of the vocabulary in the text block (or text summary), it is ensured that the obtained text vector can contain deeper semantic information. Thus, it can be ensured that the reference maintenance knowledge obtained by the electronic device based on the target problem of the abnormal state of the power equipment is highly reliable, and then it can be ensured that the large language model is based on more reliable reference maintenance knowledge, and the second maintenance plan output by the model is more accurate.

[0087] Step 202: Acquire multimodal data of the power equipment.

[0088] The electronic device can obtain the operating status data and monitoring data of the power equipment, and fuse the operating status data and the monitoring data to obtain the multimodal data of the power equipment. The operating status data may include at least one of operating current, operating voltage, operating temperature, load, displacement, speed and acceleration. The monitoring data may include one of monitoring images and monitoring videos. For example, the multimodal data may include operating current, operating voltage, operating temperature, load, displacement, speed and acceleration. The monitoring data may include monitoring images and monitoring videos.

[0089] In an embodiment of the present application, after the electronic device obtains the operating status data and monitoring data of the power equipment, it can first pre-process the operating status data and monitoring data, and then fuse the pre-processed operating status data and monitoring data to obtain multimodal data.

[0090] Specifically, the electronic device can clean and preprocess the operation status data through a data processing library, etc., and then extract the time series features of the operation status data through a time series model for subsequent fault prediction and diagnosis. In addition, the electronic device can scale, normalize and enhance the monitoring image using an image processing library to preprocess the monitoring image, and then extract the image features of the monitoring image through a convolutional neural network (CNN) to identify the appearance and internal structure of the power equipment. In addition, the electronic device can extract and preprocess the video frames of the monitoring video using a multimedia tool set, and extract the dynamic features in the monitoring video using a temporal convolutional network (TCN) to identify the appearance and internal structure of the power equipment. Afterwards, the electronic device can fuse the time series features extracted from the operation status data, the image features extracted from the monitoring image, and the dynamic features extracted from the monitoring video to obtain multimodal data.

[0091] Among them, the data processing library can be Pandas data analysis tool library or Numpy scientific computing library. The time series model can be a long short-term memory network (long short-term memory, LSTM) or a gated recurrent unit (gated recurrent unit, GRU). The image processing library can be an open source computer vision library (open source computer vision library, OpenCV) or Python imaging library (Python imaging library, PIL). The multimedia tool set can be FFmpeg.

[0092] Optionally, the electronic device may use a deep learning model that integrates a multilayer perceptron (MLP) and an attention mechanism to perform feature fusion, thereby obtaining multimodal data. The deep learning model may be a multimodal Transformer model or a joint neural network.

[0093] In an embodiment of the present application, the electronic device can establish a communication connection with a sensor group for collecting operating status data and monitoring data of the power equipment. After the sensor group collects the operating status data and monitoring data of the power equipment, the operating status data and monitoring data can be uploaded to the electronic device in real time through the communication connection. Accordingly, the electronic device can obtain the operating status data and monitoring data in real time.

[0094] The sensor group may include: a current sensor for collecting operating current, a voltage sensor for collecting operating voltage, a temperature sensor for collecting operating temperature, a load sensor for collecting load of the power equipment, a displacement sensor for collecting displacement and / or velocity and / or acceleration of the power equipment, and an image acquisition component for collecting monitoring images and monitoring videos of the power equipment. Optionally, the image acquisition component may be a camera or a webcam.

[0095] Step 203: input the multimodal data of the power equipment into the fault diagnosis model to obtain a first maintenance plan for the abnormal state of the power equipment output by the fault diagnosis model.

[0096] In an embodiment of the present application, after the electronic device inputs the multimodal data of the power device into the fault diagnosis model, the fault diagnosis result output by the fault diagnosis model can be obtained. The fault diagnosis result may include: indication information for indicating the state of the power device. The state may be an abnormal state or a normal state. The abnormal state may refer to: a state in which the power device has been operating abnormally, or a state in which the power device has signs of abnormal operation.

[0097] It is understandable that if the power equipment is in a normal state, the electronic device can perform step 202 again, that is, the electronic device can obtain new multimodal data and perform fault diagnosis prediction based on the new multimodal data. If the power equipment is in an abnormal state, the fault diagnosis result can also include: a first maintenance plan for the abnormal state of the power equipment. In addition, the fault diagnosis result can also include: a fault that causes the power equipment to be in an abnormal state. In this way, it can be ensured that the reference maintenance knowledge obtained is highly targeted and accurate, thereby ensuring that the second maintenance method obtained through the large language model is highly accurate.

[0098] Since multimodal data is more comprehensive and can more accurately reflect the operating status of power equipment, electronic equipment uses multimodal data of power equipment for fault diagnosis. On the one hand, it can ensure that the obtained fault diagnosis results are highly accurate; on the other hand, it can quickly locate the fault, thereby improving the maintenance efficiency of power equipment; on the other hand, it can discover the operating hazards of power equipment in advance, and prevent the occurrence of equipment failures through maintenance, thereby achieving preventive maintenance, so that power equipment can operate reliably and stably, thereby reducing the downtime caused by equipment failures, and then reducing the losses caused by failures.

[0099] Optionally, the fault diagnosis model may be a rule base or a decision tree model.

[0100] Step 204: When the power device is in an abnormal state, obtain a target problem for the abnormal state of the power device.

[0101] After the electronic device determines that the power device is in an abnormal state, it can issue a prompt message. The prompt message is used to prompt that the power device is in an abnormal state. After the staff learns of the prompt message, they can enter the target problem for the abnormal state in the electronic device. Correspondingly, the electronic device can obtain the target problem for the abnormal state. Among them, the target problem can be: a description text of a solution to the abnormal state.

[0102] Step 205: Based on the target problem and the vector database, obtain reference maintenance knowledge related to the abnormal state of the power equipment.

[0103] In the examples of this application, see Figure 4 , the process of the electronic device acquiring reference maintenance knowledge related to the abnormal state of the power equipment may include:

[0104] Step 2051: Based on the target question, obtain multiple candidate text vectors from the vector database.

[0105] The electronic device may perform vectorization processing on the target question to obtain a vector result of the target question. Then, the electronic device may obtain multiple candidate text vectors from a vector database based on the vector result. The similarity between each candidate text vector and the vector result of the target question is greater than a similarity threshold.

[0106] In an embodiment of the present application, the electronic device can obtain the vector result of the target question and the similarity between the vector result and each text vector in the vector database, and can determine the text vector in the vector database whose similarity is greater than the similarity threshold as the candidate text vector. The similarity threshold can be pre-stored by the electronic device.

[0107] For example, the electronic device may calculate the distance between the vector result and each text vector in the vector database, and determine the similarity between the vector result and the text vector based on the distance. The similarity is negatively correlated with the distance, such as the similarity is the reciprocal of the distance. Optionally, the distance may be a cosine distance, a Euclidean distance, or a Manhattan distance.

[0108] Step 2052: Based on the multiple candidate text vectors, obtain at least two target text vectors.

[0109] In an optional implementation, the electronic device can directly determine each candidate text vector in the multiple candidate text vectors as a target text vector, thereby obtaining at least two target text vectors. That is, at this time, the total number of the at least two target text vectors is equal to the total number of the multiple candidate text vectors.

[0110] In another optional implementation, after the electronic device obtains multiple candidate text vectors, it can sort the multiple candidate text vectors and obtain at least two target text vectors from the sorted multiple candidate text vectors. The similarity between each target text vector and the vector result of the target question is greater than the similarity between other text vectors and the vector result. The other text vector is a text vector other than the target text vector among the multiple candidate text vectors. That is to say, at this time, the total number of at least two target text vectors is less than the total number of multiple candidate text vectors.

[0111] In an embodiment of the present application, the electronic device may sort the multiple candidate text vectors in descending order of similarity. At this time, the electronic device may determine the first at least two candidate text vectors among the sorted multiple candidate text vectors as at least two target text vectors.

[0112] Alternatively, the electronic device may sort the multiple candidate text vectors in order of similarity from low to high. At this time, the electronic device may determine the last at least two candidate text vectors among the sorted multiple candidate text vectors as at least two target text vectors.

[0113] Step 2053: Based on at least two target text vectors, obtain reference maintenance knowledge related to the abnormal state of the power equipment from the maintenance knowledge base.

[0114] The power equipment can obtain a text block (or text summary) associated with the multiple at least two target text vectors from the maintenance knowledge base, and determine the text block (or text summary) as reference maintenance knowledge related to the abnormal state of the power equipment.

[0115] According to the description of steps 2051 to 2053, after the electronic device obtains the target problem, it can recall the maintenance knowledge related to the target problem from the maintenance knowledge base according to the target problem. In addition, during the implementation process, the electronic device can first perform a rough recall to recall multiple candidate text vectors from the vector database, and then select at least two target text vectors from the multiple candidate text vectors to obtain reference maintenance knowledge based on the at least two target text vectors.

[0116] That is, the electronic device can recall reference maintenance knowledge related to the abnormal state from the vector database in a two-step recall method, so as to ensure that the recalled reference maintenance knowledge is more relevant to the fault and has higher accuracy.

[0117] Step 206: Generate prompt words based on the target problem and reference maintenance knowledge.

[0118] The electronic device can fuse the target question about the abnormal state of the power equipment with the reference maintenance knowledge to obtain the prompt word Prompt. The prompt word can not only accurately express the target question, but also guide the large language model to generate the second maintenance plan of the power equipment according to the recalled reference maintenance knowledge.

[0119] In an embodiment of the present application, the electronic device may generate prompt words based only on the target problem and reference maintenance knowledge. Alternatively, the electronic device may generate prompt words based on the target problem, reference maintenance knowledge and the first maintenance plan. That is, the electronic device may use the output of the fault diagnosis model as the input of the large language model to make higher-level decisions. For example, the electronic device may directly generate prompt words based on the complete first maintenance plan. Alternatively, the electronic device may generate prompt words based on a portion of the first maintenance plan.

[0120] It is understandable that, when the technician is able to eliminate some simple problems of the power equipment based on the first maintenance plan, the electronic device can generate prompt words based on some solutions in the first maintenance plan, that is, some problems that were not solved.

[0121] For example, assuming that the prompt word is obtained only based on the target problem and reference maintenance knowledge, the style of the prompt word can be as follows:

[0122] “<Target question>, please answer based on the following:

[0123] <text block (or text summary) 1>;

[0124] <text block (or text summary) 2>;

[0125] …

[0126] <text block (or text summary) k2>."

[0127] Wherein, k2 is the total number of at least two target text vectors mentioned above.

[0128] Step 207: input the prompt word into the large language model to obtain a second maintenance plan for the power equipment output by the large language model.

[0129] The electronic device pre-stores a large language model. After the electronic device obtains the prompt word, the prompt word can be input into the large language model. The large language model can generate and output a second maintenance plan for the power equipment based on the target problem in the prompt word and the reference maintenance knowledge.

[0130] Since the large language model has excellent text generation and conversation capabilities, it can help technicians quickly and accurately query relevant maintenance knowledge of power equipment.

[0131] Optionally, the large language model is an open source model. The large language model may be a Llama model, a Qwen model, or a ChatGLM model.

[0132] According to the description of the above steps 204 to 207, the electronic device provided in the embodiment of the present application can generate the second maintenance plan by combining the retrieval-augmented generation (RAG) technology with the large language model. In this way, the accuracy of the second maintenance plan output by the large language model can be higher.

[0133] Moreover, the question-answering method of the intelligent agent based on the large language model is simple and fast, which simplifies the knowledge query and decision-making process in the maintenance process, so that technicians can quickly obtain relevant and accurate maintenance solutions through the large language model, thereby improving the work efficiency of technicians. In addition, the skill requirements for technicians are relatively low, reducing the dependence on highly skilled technicians (especially when dealing with complex or rare problems), ensuring the consistency and repeatability of maintenance quality.

[0134] Step 208: Based on the second maintenance plan, determine a target maintenance plan for the abnormal state of the power equipment.

[0135] In an embodiment of the present application, if the second maintenance plan is determined based on the first maintenance plan, the electronic device can determine the second maintenance plan as the target maintenance plan for the abnormal state. If the second maintenance plan is not obtained based on the first maintenance plan, the electronic device can determine the target maintenance plan for the abnormal state based on the first maintenance plan and the second maintenance plan.

[0136] Specifically, the electronic device can display the first maintenance plan and the second maintenance plan for the technician to view and compare. Then, in response to the technician's selection operation for the better maintenance plan between the first maintenance plan and the second maintenance plan, the electronic device can determine the maintenance plan selected by the technician as the target maintenance plan. At this time, the target maintenance plan is one of the first maintenance plan and the second maintenance plan.

[0137] Alternatively, the electronic device may fuse the first maintenance solution with the second maintenance solution to obtain a target maintenance solution. For example, the electronic device may process the first maintenance solution and the second maintenance solution using a weighted fusion technique to fuse the first maintenance solution with the second maintenance solution.

[0138] It is understandable that the electronic equipment can also collect feedback data from staff after they have repaired the power equipment based on the target maintenance plan, and update the maintenance knowledge base and fault diagnosis model based on the feedback data to optimize the maintenance knowledge base and fault diagnosis model, thereby improving the reliability of the maintenance knowledge base and fault diagnosis model.

[0139] That is to say, the electronic device provided in the embodiment of the present application can continuously optimize and adjust the maintenance knowledge base and fault diagnosis model through an online learning mechanism, ensuring that the fault diagnosis model and the large language model can also provide effective and accurate maintenance solutions when facing new equipment or unknown faults, thereby improving the adaptability and reliability of the fault diagnosis model and the large language model.

[0140] It is understandable that the order of the steps of the method for determining the maintenance plan of the power equipment provided in the embodiment of the present application can be appropriately adjusted, and the steps can be increased or decreased accordingly according to the situation. For example, step 201 can be deleted according to the situation; or, step 201 can also be executed after step 204. Any technician familiar with the technical field can easily think of a change within the technical scope disclosed in this application, and it should be covered within the scope of protection of this application, so it will not be repeated.

[0141] In summary, the embodiment of the present application provides a method for determining a maintenance plan for electric equipment, and the electronic device can input the multimodal data of the electric equipment into the fault diagnosis model to obtain the first maintenance plan for the abnormal state of the electric equipment output by the fault diagnosis model. Then, the electronic device can determine the target maintenance plan for the abnormal state based on the first maintenance plan and the large language model, so on the one hand, it can avoid the problem of low reliability of the determined target maintenance plan due to human misjudgment or omission; on the other hand, it does not need to spend a long time to identify the problem, and it is not necessary to make decisions based on intuition, so the reliability of the determined target maintenance plan can be further improved, and the determination efficiency of the target maintenance plan can be improved. In addition, since the target maintenance plan can be obtained based on the first maintenance plan and the large language model, that is, the target maintenance plan can be obtained through multiple models, it can further ensure that the reliability of the determined target maintenance plan is high.

[0142] The present application provides a device for determining a maintenance plan for electric power equipment, which can execute the method for determining a maintenance plan for electric power equipment provided in the above method embodiment. Figure 5 , the device 300 comprises:

[0143] The acquisition module 301 is used to acquire multi-modal data of the power equipment.

[0144] The diagnosis module 302 is used to input the multimodal data into the fault diagnosis model to obtain a first maintenance plan for the abnormal state of the power equipment output by the fault diagnosis model.

[0145] The determination module 303 is used to determine a target maintenance plan for the abnormal state based on the first maintenance plan and the large language model.

[0146] Optionally, the determination module 303 may be used to:

[0147] Obtain the second maintenance plan for abnormal conditions through a large language model;

[0148] If the second maintenance plan is obtained based on the first maintenance plan, the second maintenance plan is determined as a target maintenance plan for the abnormal state;

[0149] If the second maintenance plan is not obtained based on the first maintenance plan, a target maintenance plan for the abnormal state is determined based on the first maintenance plan and the second maintenance plan.

[0150] Optionally, the determination module 303 may be used to:

[0151] Based on the target questions for abnormal states, obtain reference maintenance knowledge related to abnormal states;

[0152] Generate prompt words based on target problems and reference maintenance knowledge;

[0153] The prompt words are input into the large language model to obtain a second maintenance plan for the abnormal state output by the large language model.

[0154] Optionally, the determination module 303 may be used to:

[0155] Based on the target question, multiple candidate text vectors are obtained from the vector database, and the similarity between each candidate text vector and the vector result of the target question is greater than the similarity threshold;

[0156] Based on the multiple candidate text vectors, obtaining at least two target text vectors;

[0157] Based on at least two target text vectors, reference maintenance knowledge related to the abnormal state is obtained from a maintenance knowledge base.

[0158] Optionally, the determination module 303 may be used to:

[0159] sorting multiple candidate text vectors;

[0160] Obtain at least two target text vectors from the sorted multiple candidate text vectors;

[0161] Among them, the similarity between each target text vector and the vector result is greater than the similarity between other text vectors and the vector result, and the other text vectors are text vectors other than the target text vector among multiple candidate text vectors.

[0162] Optional, see Figure 6 , the apparatus 300 may further include:

[0163] The second acquisition module 304 is used to acquire multiple maintenance knowledge texts of the power equipment before acquiring multiple candidate text vectors from the vector database based on the target problem. For the sake of distinction, the acquisition module for acquiring the multiple maintenance knowledge texts is referred to as the second acquisition module, and the acquisition module for acquiring multimodal data of the power equipment described above is referred to as the first acquisition module.

[0164] The segmentation module 305 is used to segment each maintenance knowledge text into multiple text blocks.

[0165] The third acquisition module 306 acquires a vector database based on the multiple text blocks.

[0166] Optionally, the third acquisition module 306 may be used to:

[0167] Get the text summary of each text block;

[0168] Each text summary is converted into a text vector to obtain a vector database.

[0169] Optionally, the third acquisition module 306 may be used to:

[0170] Each text summary is input into the text vectorization model to obtain the text vector of the text summary output by the text vectorization model.

[0171] Optionally, the text vectorization model is a pre-trained language model.

[0172] Optionally, the determination module 303 may be used to:

[0173] The first maintenance plan is integrated with the second maintenance plan to obtain a target maintenance plan for the abnormal state.

[0174] In summary, the embodiment of the present application provides a device for determining a maintenance plan for an electric device, which can input the multimodal data of the electric device into a fault diagnosis model to obtain a first maintenance plan for the abnormal state of the electric device output by the fault diagnosis model. Then, the device can determine the target maintenance plan for the abnormal state based on the first maintenance plan and the large language model. Since there is no need to manually determine the target maintenance plan based on work experience, on the one hand, the problem of low reliability of the determined target maintenance plan due to manual misjudgment or omission can be avoided; on the other hand, there is no need to spend a long time to identify the problem, and there is no need to make decisions based on intuition, so the reliability of the determined target maintenance plan can be further improved, and the determination efficiency of the target maintenance plan can be improved. In addition, since the target maintenance plan can be obtained based on the first maintenance plan and the second maintenance plan, that is, the target maintenance plan can be obtained through multiple models, it can be further ensured that the reliability of the determined target maintenance plan is high.

[0175] The present application embodiment provides an electronic device, see Figure 7 , the electronic device 400 includes: a processor 401 and a memory 403. The processor 401 and the memory 403 are connected, such as through a bus 402. Optionally, the electronic device 400 may also include a transceiver 404. It should be noted that in actual applications, the transceiver 404 is not limited to one, and the structure of the electronic device 400 does not constitute a limitation on the embodiments of the present application.

[0176] Processor 401 may be a CPU (Central Processing Unit), a general purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 401 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0177] The bus 402 may include a path to transmit information between the above components. The bus 402 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 402 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0178] The memory 403 is used to store a computer program corresponding to the method for determining the maintenance plan of the power equipment provided in the above embodiment of the present application, and the computer program is controlled and executed by the processor 401. The processor 401 is used to execute the computer program stored in the memory 403 to implement the content shown in the above method embodiment.

[0179] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for determining a maintenance plan for power equipment provided in the above method embodiment is implemented. For example, Figure 1 or Figure 2 The method shown.

[0180] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0181] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0182] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0183] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0184] In this application, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0185] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for determining a maintenance plan for electric power equipment, characterized in that: The method comprises: Acquiring multimodal data of the electric power equipment; Inputting the multimodal data into a fault diagnosis model to obtain a first maintenance plan for the abnormal state of the power equipment output by the fault diagnosis model; Based on the first maintenance plan and the large language model, a target maintenance plan for the abnormal state is determined.

2. The method according to claim 1, characterized in that Determining a target maintenance plan for the abnormal state based on the first maintenance plan and the large language model includes: Acquire a second maintenance plan for the abnormal state through a large language model; If the second maintenance plan is obtained based on the first maintenance plan, determining the second maintenance plan as a target maintenance plan for the abnormal state; If the second maintenance plan is not obtained based on the first maintenance plan, a target maintenance plan for the abnormal state is determined based on the first maintenance plan and the second maintenance plan.

3. The method according to claim 2, characterized in that Obtaining a second maintenance plan for the abnormal state through a large language model includes: Based on the target problem for the abnormal state, obtaining reference maintenance knowledge related to the abnormal state; Generate prompt words based on the target problem and the reference maintenance knowledge; The prompt word is input into a large language model to obtain a second maintenance plan for the abnormal state output by the large language model.

4. The method according to claim 2, characterized in that: Based on the target problem for the abnormal state, reference maintenance knowledge related to the abnormal state is obtained, including: Based on the target question, a plurality of candidate text vectors are obtained from a vector database, wherein the similarity between each candidate text vector and the vector result of the target question is greater than a similarity threshold; Based on the multiple candidate text vectors, obtaining at least two target text vectors; Based on the at least two target text vectors, reference maintenance knowledge related to the abnormal state is acquired from a maintenance knowledge base.

5. The method according to claim 4, characterized in that Based on the multiple candidate text vectors, obtaining at least two target text vectors includes: sorting the plurality of candidate text vectors; Obtain at least two target text vectors from the sorted plurality of candidate text vectors; Among them, the similarity between each of the target text vectors and the vector result is greater than the similarity between other text vectors and the vector result, and the other text vectors are text vectors other than the target text vector among the multiple candidate text vectors.

6. The method according to claim 4, characterized in that Before obtaining a plurality of candidate text vectors from a vector database based on the target question, the method further includes: Acquire multiple maintenance knowledge texts of the electric power equipment; Splitting each of the maintenance knowledge texts into a plurality of text blocks; Based on the multiple text blocks, the vector database is obtained.

7. The method according to claim 5, characterized in that Based on the multiple text blocks, obtaining the vector database includes: Obtaining a text summary of each of the text blocks; Each of the text summaries is converted into a text vector to obtain the vector database.

8. The method according to claim 5, characterized in that Converting each of the text summaries into a text vector to obtain the vector database includes: Each of the text summaries is input into a text vectorization model to obtain a text vector of the text summaries output by the text vectorization model.

9. The method according to claim 7, characterized in that: The text vectorization model is a pre-trained language model.

10. The method according to any one of claims 2 to 8, characterized in that: Determining a target maintenance plan for the abnormal state based on the first maintenance plan and the second maintenance plan includes: The first maintenance plan and the second maintenance plan are integrated to obtain a target maintenance plan for the abnormal state.

11. A device for determining a maintenance plan for electric power equipment, characterized in that: The device comprises: An acquisition module, used for acquiring multimodal data of the electric power equipment; A diagnosis module, used for inputting the multimodal data into a fault diagnosis model to obtain a first maintenance plan for the abnormal state of the power equipment output by the fault diagnosis model; A determination module is used to determine a target maintenance plan for the abnormal state based on the first maintenance plan and a large language model.

12. An electronic device, characterized in that: The electronic device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the method according to any one of claims 1 to 10 is implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images