Service processing method, device and storage medium

Abstract

Embodiments of the present application provide a service processing method and device and a storage medium. In the embodiments of the present application, by generating a fault feature vector based on original fault data and performing fault diagnosis, when the confidence score is lower than a preset threshold, incremental fault data is introduced to update the fault feature vector and perform iterative diagnosis. Through data supplement and iterative diagnosis, the diagnostic result is gradually converged, and the diagnostic accuracy and reliability are improved. Furthermore, by searching in a fault maintenance knowledge graph containing knowledge nodes related to maintenance cost and price influence factor edges, and generating a multi-dimensional price correlation parameter set using dynamic weights dynamically updated based on price time series data of an external data source, the price generation model introduces the dynamic weights of the knowledge nodes and the price influence factor edges associated with the target fault category, thereby generating a price record information set matching the target fault category and the price change situation, and improving the rationality of the price generation result.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a service processing method, device and storage medium. Background Technology

[0002] In repair service scenarios, repair personnel typically diagnose equipment faults based on on-site inspection or user-provided fault descriptions, leveraging their experience. They then estimate the required materials, labor, and costs based on the diagnosed fault type to determine a quote. Because this method relies on the repair personnel's personal experience, the diagnostic results and quotes are significantly influenced by individual skill differences, leading to issues such as inaccurate fault classification and insufficiently reasonable pricing.

[0003] Currently, automated fault diagnosis systems and repair quote generation systems exist. However, existing fault diagnosis systems typically only support single-modality fault data input, and the fault data relies entirely on user submission. When the user-submitted fault data is incomplete or insufficient, the fault diagnosis results generated by the system are often inaccurate. Existing repair quote generation systems usually generate quotes based on static preset information such as parts prices and labor costs. When factors such as parts prices and labor costs fluctuate dynamically, the static preset information cannot reflect these changes in a timely manner, leading to discrepancies between the generated quote and the actual situation. Summary of the Invention

[0004] This application provides a service processing method, device, and storage medium to improve the accuracy of fault diagnosis and thus improve the rationality of the maintenance price generation results.

[0005] This application provides a service processing method, comprising: receiving a service request uploaded by a terminal device, the service request including original fault data of a target fault entity; performing multi-dimensional feature extraction and fusion on the original fault data to generate a fault representation vector; based on the fault representation vector, calling a diagnostic model to perform fault diagnosis, and outputting an initial fault category and a corresponding confidence score; if the confidence score is lower than a preset threshold, sending an information supplementation request to the terminal device so that the terminal device can return incremental fault data; updating the fault representation vector based on the incremental fault data, and using the diagnostic model to perform fault diagnosis on the updated fault representation vector, until the confidence score meets the iteration termination condition. The process involves: obtaining the target fault category; retrieving target knowledge nodes associated with the target fault category from the fault repair knowledge graph; the fault repair knowledge graph includes multiple knowledge nodes related to repair costs, with at least some knowledge nodes connected by price influence factor edges, the price influence factor edges having dynamic weights, the dynamic weights being dynamically updated based on price time-series data from an external data source; generating a multi-dimensional price association parameter set based on the attribute information of the target knowledge nodes and the dynamic weights of the price influence factor edges connected to the target knowledge nodes; calling the price generation model, using the multi-dimensional price association parameter set as reference data input, and generating a price record information set; and sending the price record information set to the terminal device.

[0006] This application also provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor is coupled to the memory to execute the computer program to implement the steps in the above-described method.

[0007] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps in the methods described above.

[0008] In this embodiment, a fault representation vector is generated based on the original fault data, and fault diagnosis is performed. When the confidence score of the diagnosis result is lower than a preset threshold, incremental fault data is introduced to update the fault representation vector and perform iterative diagnosis. Through multiple data supplements and iterative diagnoses, the diagnosis results gradually converge, improving the accuracy and reliability of the diagnosis. Furthermore, by searching in a fault maintenance knowledge graph containing knowledge nodes related to maintenance costs and price influencing factors, and using dynamic weights dynamically updated based on price time-series data from external data sources to generate a multi-dimensional price association parameter set, the price generation model incorporates knowledge nodes associated with the target fault category and dynamic weights of price influencing factors. This generates a price record information set that matches the target fault category and price changes, improving the rationality of the price generation results. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a service processing method provided in an exemplary embodiment of this application; Figure 2 This is a schematic diagram illustrating the input method recognition process in an exemplary embodiment of this application; Figure 3 A schematic diagram of the structure of a diagnostic model provided in an exemplary embodiment of this application; Figure 4 A schematic diagram of the architecture of a fault repair knowledge graph provided as another exemplary embodiment of this application; Figure 5 A schematic diagram illustrating the process of generating a price record information set as provided in yet another exemplary embodiment of this application; Figure 6 A flowchart illustrating the matching of maintenance personnel information is provided as another exemplary embodiment of this application; Figure 7 A schematic diagram illustrating yet another exemplary embodiment of price composition information provided in this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] It should be noted that, in the cases involving user information in the embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to language models or large models) comply with relevant laws and standards.

[0012] Additionally, it should be noted that when user interaction operations or triggering operations are involved in the embodiments of this application, these operations include, but are not limited to, various interaction methods such as touch operations, gesture operations, voice operations, head movement operations, and eye movement operations. Touch operations include, but are not limited to, click operations, double-click operations, long-press operations, swipe operations, pinch operations, or mouse hover operations. Swipe operations include, but are not limited to, straight-line swipes and curved-line swipes.

[0013] Furthermore, it should be noted that, in the cases where the embodiments of this application involve jumping between the first interface and the second interface, the jumping methods involved in the embodiments of this application include, but are not limited to: jumping directly from the first interface to the second interface, or jumping from the first interface to the task interface and completing the corresponding task operation on the task interface before jumping to the second interface; completing the corresponding task operation on the task interface includes, but is not limited to: completing the game operation on the game interface when the task interface is implemented as a game interface; completing identity authentication on the identity authentication interface when the task interface is implemented as an identity authentication interface; completing the recharge operation on the recharge interface when the task interface is implemented as a recharge interface; and so on.

[0014] To address the technical problem of inaccurate fault category diagnosis leading to unreasonable pricing, this application embodiment generates a fault representation vector based on original fault data and performs fault diagnosis. When the confidence score of the diagnosis result is lower than a preset threshold, incremental fault data is introduced to update the fault representation vector and perform iterative diagnosis. Through multiple data supplements and iterative diagnoses, the diagnosis results gradually converge, improving the accuracy and reliability of the diagnosis. Furthermore, by searching a fault maintenance knowledge graph containing knowledge nodes related to maintenance costs and price influencing factors, and using dynamic weights dynamically updated based on price time-series data from external data sources to generate a multi-dimensional price association parameter set, the price generation model incorporates dynamic weights of knowledge nodes and price influencing factors associated with the target fault category. This generates a price record information set that matches the target fault category and price changes, improving the rationality of the price generation results.

[0015] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart illustrating a service processing method provided for an exemplary embodiment of this application. Figure 1 As shown, the method includes the following steps: S101: Receive a service request uploaded by the terminal device. The service request includes the original fault data of the target fault entity. S102: Perform multi-dimensional feature extraction and fusion on the original fault data to obtain the fault representation vector; S103: Based on the fault representation vector, call the diagnostic model to perform fault diagnosis and output the initial fault category and the corresponding confidence score; S104: If the confidence score is lower than the preset threshold, send an information supplementation request to the terminal device so that the terminal device can return incremental fault data; S105: Update the fault representation vector based on incremental fault data, and use the diagnostic model to diagnose the fault in the updated fault representation vector until the confidence score meets the iteration termination condition to obtain the target fault category. S106: In the fault repair knowledge graph, retrieve the target knowledge node associated with the target fault category; the fault repair knowledge graph includes multiple knowledge nodes related to repair costs, and at least some knowledge nodes are connected by price influence factor edges, which have dynamic weights, and the dynamic weights are dynamically updated based on external data source price time series data. S107: Generate a multi-dimensional price association parameter set based on the attribute information of the target knowledge node and the dynamic weights of the price influencing factor edges connected to the target knowledge node; S108: Call the price generation model, take the multi-dimensional price correlation parameter set as reference data input, and generate a price record information set; S109: Send the price record information set to the terminal device.

[0017] In this embodiment, a service request uploaded by a terminal device is received. The service request includes the original fault data of the target fault entity. This embodiment supports processing various information inputs such as text, images, and audio. Users can submit original fault data through various methods such as taking photos, voice descriptions, or text input. Optionally, the original fault data can be multimodal fault data. Multimodal fault data includes, but is not limited to, audio data, video data, text data, and image data including the target fault entity.

[0018] In this embodiment, the "target fault entity" is the entity object where the fault occurs. It can be a complete device or a component or part within the device. For example, it can be a household appliance such as a refrigerator, washing machine, or air conditioner, or it can be a specific component inside it (such as a compressor, motor, bearing, etc.).

[0019] Furthermore, multi-dimensional feature extraction and fusion are performed on the original fault data to obtain a fault representation vector. For example... Figure 2As shown, when the original fault data contains text data, word embedding encoding is performed on the text data to encode it into a text fault feature vector. For example, the text fault feature vector can be used to represent semantic information related to the fault. When the original fault data contains image data, image feature extraction based on a convolutional neural network is performed on the image data to generate an image fault feature vector. For example, the image fault feature vector can be used to represent visual feature information related to the target fault entity, such as the appearance or degree of damage of the target fault entity. When the original fault data contains audio data, acoustic feature transformation is performed on the audio data to generate an audio fault feature vector of the target fault entity. For example, the audio fault feature vector can be used to represent sound-related feature information, such as the sound feature information of the target fault entity during operation. When the original fault data contains video data, keyframes are extracted from the video data, and feature extraction processing is performed based on the keyframes to generate a video fault feature vector.

[0020] Given feature vectors corresponding to at least one modality, text fault feature vectors, image fault feature vectors, audio fault feature vectors, and / or video fault feature vectors are mapped to a unified semantic space and feature fusion processing is performed to generate fault representation vectors.

[0021] In this embodiment, a diagnostic model is invoked based on the fault characterization vector to perform fault diagnosis, and the initial fault category and corresponding confidence score are output.

[0022] In this embodiment, as Figure 3 As shown, the diagnostic model includes a feature analysis module, a feature weight calculation module, a fault category determination module, and a confidence assessment module. In this embodiment, the feature analysis module is used to extract multi-dimensional features from the fault representation vector and analyze and process these multi-dimensional features to obtain the analysis results corresponding to the multi-dimensional features. These multi-dimensional features may include one or more of noise features, visual features, behavioral features, and environmental features. The analysis process of the feature analysis module includes: performing sound spectrum analysis based on noise features to obtain sound spectrum features; assessing the degree of damage based on visual features to obtain damage assessment features; identifying abnormal actions based on behavioral features to obtain abnormal action features; and assessing the environmental situation based on environmental features to obtain environmental features.

[0023] In this embodiment, the feature analysis module inputs the analysis results of extracted multi-dimensional features into the feature weight calculation module. The feature weight calculation module performs weight calculation processing on the analysis results of the multi-dimensional features respectively, and outputs the feature weight representation used to characterize the importance of each feature dimension. Further, the feature weight representation of the importance of each feature dimension is input into the fault category determination module. The fault category determination module performs fault category determination processing based on the feature representation of each dimension and outputs at least one candidate fault category. The at least one candidate fault category is input into the confidence evaluation module. The confidence evaluation module performs confidence evaluation on the at least one candidate fault category respectively and outputs the confidence score corresponding to each candidate fault category. The candidate fault category with the higher confidence score is selected from the candidate fault categories as the initial fault category, and the initial fault category and its corresponding confidence score are output as the diagnostic result of the diagnostic model.

[0024] In this embodiment, each initial fault category is associated with a confidence score, representing the reliability of the diagnostic model for that initial fault category. The confidence score is the score assigned by the model to the initial fault category, indicating the probability that the initial fault category is a true fault. For example, the confidence score for "compressor fault" might be 85%, while the confidence score for "circuit problem" might be 50%.

[0025] In this embodiment, after obtaining the initial fault category and its corresponding confidence score, the confidence score is judged. If the confidence score is lower than a preset threshold, it indicates that the confidence level corresponding to the initial fault category is insufficient, triggering an information supplementation mechanism. Specifically, an information supplementation request is sent to the terminal device. Upon receiving the information supplementation request, the terminal device guides the user to supplement the fault data of the target fault entity, i.e., incremental fault data. Subsequently, the terminal device returns the incremental fault data.

[0026] In one optional embodiment, the incremental fault data includes text data, image data, audio data, and / or video data; updating the fault representation vector based on the incremental fault data includes: performing word embedding encoding on the text data to obtain a text fault feature vector; and / or performing convolutional neural network feature extraction on the image data to obtain an image fault feature vector; and / or performing acoustic feature transformation on the audio data to obtain an audio fault feature vector; and / or performing temporal keyframe encoding on the video data to obtain a video fault feature vector; and then layering and fusing the text fault feature vector, image fault feature vector, audio fault feature vector, and / or video fault feature vector with the fault representation vector in a unified semantic space to obtain the updated fault representation vector.

[0027] In one optional embodiment, after the terminal device returns incremental fault data, the existing fault representation vector is updated based on the incremental fault data. Specifically, the incremental fault data may include text data, image data, audio data, and / or video data. For incremental fault data of different modalities, feature extraction is performed using a feature extraction method that matches its modality.

[0028] In this embodiment, when the incremental fault data includes text data, word embedding encoding is performed on the text data to map it into a text fault feature vector. For example, the text fault feature vector can be used to represent semantic information related to the fault. When the incremental fault data includes image data, features are extracted from the image data using a convolutional neural network to obtain an image fault feature vector. For example, the image fault feature vector can be used to represent visual feature information related to the target fault entity, such as the appearance or damage level of the target fault entity. When the incremental fault data includes audio data, acoustic feature transformation is performed on the audio data to generate an audio fault feature vector. For example, the audio fault feature vector can be used to represent sound-related feature information, such as the sound features of the target fault entity. When the incremental fault data includes video data, keyframes are extracted from the video data, and feature extraction processing is performed based on the keyframes to generate a corresponding video fault feature vector.

[0029] Furthermore, in this embodiment, the text fault feature vector, image fault feature vector, audio fault feature vector and / or video fault feature vector are mapped to a unified semantic space and then fused with the current fault representation vector to obtain an updated fault representation vector.

[0030] In this embodiment, by introducing incremental fault data during the fault diagnosis process and performing corresponding feature extraction methods for different modalities, the newly added fault data in the form of text, images, audio and / or video is converted into a fault feature vector that can be uniformly represented. This vector is then fused and updated with the existing fault representation vector in a unified semantic space. This allows the fault representation vector to be dynamically supplemented and corrected, improving the completeness and accuracy of the fault representation. This is beneficial for improving the reliability of fault category identification in the subsequent fault diagnosis process.

[0031] In this embodiment, after the terminal device returns incremental fault data, the fault representation vector is dynamically updated. Specifically, the incremental fault data undergoes the same feature extraction process as the original fault data to generate a corresponding incremental fault feature vector. In some implementations, the incremental fault feature vector is validated for consistency based on the feature similarity between the incremental fault feature vector and the current fault representation vector; when the feature similarity meets a preset similarity condition, the incremental fault feature vector is determined to have passed the consistency validation. If the consistency validation passes, the fusion weight between the incremental fault feature vector and the current fault representation vector is determined based on the time-series decay factor and data quality score, and the fault representation vector is updated using a weighted fusion method. Through weighted fusion, the updated fault representation vector retains historical diagnostic information while enhancing its ability to express newly added feature information. Furthermore, the updated fault representation vector is input into the diagnostic model, fault diagnosis is re-executed, and the updated fault category and corresponding confidence score are output.

[0032] In this embodiment, to improve the accuracy of fault diagnosis, the incremental data type and guidance content to be requested from the user are determined based on a preset guidance strategy, and a corresponding information supplementation request is sent to the user. Specifically, the preset guidance strategy can make comprehensive decisions based on the current fault representation vector, the internal state of the diagnostic model, and a pre-built fault knowledge graph. The guidance strategy includes at least one of the following methods: modal guidance based on attention weights, candidate differentiation guidance based on the fault maintenance knowledge graph, and information gain guidance based on feature fuzziness. These will be described in detail below.

[0033] Method A1: Modal guidance method based on attention weights: In method A1, when the diagnostic model employs a multimodal fusion model, the attention weights of each modality corresponding to the current fault representation vector are obtained from the diagnostic model. The contribution of each modality to the current diagnostic result is determined based on these attention weights. When the attention weight of a certain modality is lower than the corresponding preset weight threshold, it indicates that the information of that modality is insufficient to contribute to the current decision, which may be due to poor quality or missing information in the original fault data. Accordingly, an information supplementation request for that modality is generated to guide the user to provide clearer and more comprehensive data for that modality.

[0034] In one example, during air conditioner fault diagnosis, the original fault data only provided text data, describing "the air conditioner is not cooling." Attention weighting showed that the text modality contributed 80%, while the image modality contributed only 5%. Therefore, it was determined that the image modality's information contribution was insufficient, and a request for supplementary information regarding the image modality was sent to the user. In one example, the supplementary information request was: "Please take a clear photo of the current state of the outdoor unit of the air conditioner, especially the compressor area, to help determine if there is frost or foreign object blockage."

[0035] Method A2: Candidate differentiation guidance method based on fault repair knowledge graph: In method A2, a fault repair knowledge graph for the fault diagnosis domain is pre-constructed to store knowledge nodes such as fault categories, fault phenomena, and spare parts, as well as their relationships. In some embodiments of this application, the knowledge nodes of the fault repair knowledge graph can also be simply referred to as nodes. The node types of the fault repair knowledge graph include: fault category nodes (such as "motor fault", "loose belt"), fault phenomenon nodes (such as "humming sound during operation", "drum not turning", "outdoor unit frosting"), and spare parts nodes (such as "motor", "belt", "compressor"), etc. Nodes are connected through various relationship types. Further details about the fault repair knowledge graph can be found in subsequent embodiments.

[0036] When the diagnostic model outputs multiple candidate fault categories with similar confidence levels, the system queries the fault maintenance knowledge graph for nodes related to each candidate fault category. Furthermore, based on the retrieved nodes, attribute information related to each candidate fault category is extracted, and the attribute information corresponding to different candidate fault categories is compared and analyzed to determine the discriminative features that can distinguish each candidate fault category.

[0037] Once discriminative features are identified, a large language model can be used to analyze these features, determine the target information and corresponding modalities that need further acquisition, and generate user-guided content based on the target information, thereby generating an information supplementation request. For example, a large language model can be used to analyze and determine the target information and corresponding modalities that need to be supplemented, and generate user-guided content. For instance, based on the results of a fault repair knowledge graph retrieval, including but not limited to: the name of the current candidate fault category, fault symptoms, suggested supplementary modalities, information on relevant spare parts, and optional example guidance content, the large language model can generate an information supplementation request for obtaining incremental fault data.

[0038] In one example, for a washing machine fault diagnosis scenario, the diagnostic model outputs two candidate fault categories with similar confidence levels: "motor fault" (confidence level 0.45) and "loose belt" (confidence level 0.40). Based on these candidate fault categories, nodes related to each category are retrieved from the fault repair knowledge graph, and their corresponding attribute information is extracted for comparative analysis. The analysis results show that "motor fault" is usually accompanied by symptoms such as "humming sound during operation but the drum does not rotate," while "loose belt" manifests as "slipping sound during startup and slow drum rotation." Based on these discriminative features, a large language model is used to generate supplementary information requests, such as: "Please record the sound of the washing machine starting up and describe whether the drum rotation is normal," to guide users to provide audio data and / or text data as incremental fault data, thereby further improving the accuracy of fault diagnosis.

[0039] Method A3: Information gain guidance based on feature fuzziness: In method A3, the feature distribution of the current fault representation vector across each mode is analyzed, and its information entropy or variance is calculated. If the entropy value of the feature vector of a certain mode is too high or the variance is too small, it indicates that the features of that mode are fuzzy and have low discriminative power. In this case, the user is guided to supplement more specific or more critical data in that mode. In addition, the gradient information of the diagnostic model can be combined to analyze the input features to determine the degree of influence of different input features on the diagnostic results. For example, the contribution of each input feature to the output results can be calculated using the gradient attribution method, and modes with higher influence can be prioritized for supplementation.

[0040] In one example, a user uploaded an image of the refrigerator's interior. However, due to the image being too dark and blurry, the corresponding image feature vector had a high information entropy, resulting in low feature discrimination for this image mode. Furthermore, combined with the feature analysis results of the diagnostic model, it was identified that this image data had a high impact on determining "ice buildup in the refrigerator compartment." Based on this analysis, an information supplement request was generated, such as: "The current image clarity is insufficient. Please retake the image of the back wall area of ​​the refrigerator compartment, ensuring sufficient lighting and clear focus on the ice-covered area," to guide the user to provide higher-quality image data.

[0041] Optionally, a combination of the above strategies can be dynamically selected based on currently collected data, user historical behavior, and trends in diagnostic confidence. For example, method A1 can be prioritized in the first round of follow-up; if the effect is not significant, method A2 can be used in the second round for more refined follow-up questions. Simultaneously, the information gain after each follow-up question can be recorded to optimize subsequent guidance strategies.

[0042] In this embodiment, after the terminal device returns incremental fault data, the fault representation vector is updated based on the incremental fault data. After the fault representation vector update is completed, the updated fault representation vector is input into the diagnostic model, the fault diagnosis process described in the above embodiment is re-executed, and the corresponding fault category and its corresponding confidence score are output. If the re-output confidence score does not meet the preset iteration termination condition, new incremental fault data is acquired and the previously updated fault representation vector is updated, repeating the above update and diagnosis process; when the confidence score meets the iteration termination condition, the iteration stops, and the corresponding fault category is determined as the target fault category.

[0043] In this embodiment, the specific implementation of the iteration termination condition is not limited. For example, the iteration termination condition may be that the confidence score output by the diagnostic model reaches or exceeds a preset threshold. Another example is that the iteration termination condition may be that the output fault category of the diagnostic model is the same for a preset number of consecutive iterations.

[0044] Furthermore, given the identified target fault category, the target knowledge node associated with the target fault category is retrieved from the fault maintenance knowledge graph. The knowledge graph includes multiple knowledge nodes related to maintenance costs, with at least some knowledge nodes connected by price influence factor edges. These price influence factor edges have dynamic weights, which are weight attributes of the price influence factor edges and are used to represent the degree of price influence between the knowledge nodes connected by the price influence factor edge. The dynamic weights are dynamically updated based on price time-series data from an external data source.

[0045] In this embodiment, a structured fault repair knowledge graph for fault diagnosis and price generation is constructed. This fault repair knowledge graph includes multiple knowledge nodes and their interrelationships, thus forming a structured fault repair knowledge graph that provides an interpretable and reasonable knowledge foundation. In one example, such as... Figure 4 As shown, the fault repair knowledge graph uses fault categories, repair methods, tool requirements, and spare parts information as top-level knowledge nodes, and further refines the corresponding sub-knowledge nodes under each knowledge node.

[0046] Furthermore, based on the attribute information of the target knowledge node and the dynamic weights of the price influencing factor edges connected to the target knowledge node, a multi-dimensional price correlation parameter set is generated.

[0047] In this embodiment, the attribute information of each target knowledge node and the dynamic weights of the price influencing factor edges connected to it are obtained. The attribute information is used to characterize attributes related to maintenance costs, such as spare parts prices and labor costs. Then, on a per-target knowledge node basis, the attribute information of the target knowledge node is associated with the dynamic weights of the price influencing factor edges connected to it, and this association is structured and encapsulated to form price association parameter records. These price association parameter records of the target knowledge nodes are collected to obtain a multi-dimensional price association parameter set, which serves as reference data input for the subsequent price generation model.

[0048] Furthermore, such as Figure 5 As shown, the price generation model is invoked, using a multi-dimensional price association parameter set as reference data input to generate a price record information set. Specifically, upon receiving the multi-dimensional price association parameter set, the price generation model determines the base price corresponding to the target fault category based on the attribute information of the target knowledge nodes contained in the price association parameter set. Subsequently, the price generation model dynamically adjusts the base price by combining the dynamic weights of the price influencing factor edges connected to the target knowledge nodes contained in the price association parameter set, in order to reflect the impact of external price changes on repair prices, thereby generating the corresponding repair price result.

[0049] In this embodiment, the price generation model outputs at least one repair price result, each repair price result serves as a price record information, and at least one price record information constitutes a price record information set.

[0050] Furthermore, the price record information set is sent to the terminal device.

[0051] In this embodiment, a fault representation vector is generated based on the original fault data, and fault diagnosis is performed. When the confidence score of the diagnosis result is lower than a preset threshold, incremental fault data is introduced to update the fault representation vector and perform iterative diagnosis. Through multiple data supplements and iterative diagnoses, the diagnosis results gradually converge, improving the accuracy and reliability of the diagnosis. Furthermore, by searching in a fault maintenance knowledge graph containing knowledge nodes related to maintenance costs and price influencing factors, and using dynamic weights dynamically updated based on price time-series data from external data sources to generate a multi-dimensional price association parameter set, the price generation model incorporates knowledge nodes associated with the target fault category and dynamic weights of price influencing factors. This generates a price record information set that matches the target fault category and price changes, improving the rationality of the price generation results.

[0052] Further optional, such as Figure 5As shown, before obtaining the output price record information set, a reasonableness check can be performed on the price record information set based on reasonableness constraints. Optionally, the reasonableness check is to ensure that the price record information in the price record information set is within a reasonable range. During the reasonableness check process, it is checked whether the price record information in the price record information set exceeds the reasonable range. If it exceeds the preset minimum or maximum value, it can be dynamically adjusted to bring the price record information back to the reasonable range.

[0053] In this embodiment, the reasonable range is a constraint boundary used to determine whether price record information is within an acceptable range. Its construction integrates the influence of internal cost data, external market information, and dynamic adjustment factors. In some implementations, the benchmark value of the reasonable range is determined through at least one of the following methods: based on external price information, based on historical transaction data, and based on a cost-plus model. These are described below. In the implementation based on external price information, repair price information for the same fault category from similar service platforms is obtained through a preset data interface or public channels. The obtained repair price information is cleaned and deduplicated to form a market repair price dataset. Statistical characteristics of price distribution are calculated based on the market repair dataset. Based on these statistical characteristics, the upper and lower boundaries of the reasonable range are determined, where the lower limit is a first preset proportion of the statistical characteristics, and the upper limit is a second preset proportion of the statistical characteristics.

[0054] In the implementation based on historical transaction data, repair price information of transactions with the same or similar fault categories and fault entity models is extracted from the historical order database. The weighted average price and standard deviation are calculated using a time decay weighted algorithm. The standard deviation of the weighted average price plus or minus a preset multiple is used as the upper and lower boundaries of the reasonable range.

[0055] In the implementation based on the cost-plus model, the lower and upper limits of the theoretical cost-plus price are calculated based on the cost of spare parts required for repair, labor cost, labor cost corresponding to the skill level of repair personnel, and a preset profit margin range, and the theoretical price range is taken as the reasonable range.

[0056] In some implementations, multiple reasonable intervals obtained through the above method are weighted and merged to obtain a comprehensive reasonable interval.

[0057] In one optional embodiment, when the price record information is within a reasonable range, it is determined that the price record information passes the verification; when the price record information exceeds the reasonable range, the deviation of the price record information from the reasonable range is calculated, and it is classified according to the deviation. A flexible adjustment strategy is preferentially adopted in the case of slight deviation, while a price reconstruction or manual review process is triggered in the case of severe deviation. Simultaneously, possible causes of deviation are analyzed, including checking whether the dynamic adjustment factor is abnormal, verifying whether the base price is deviated due to missing fault maintenance knowledge graph data, and analyzing whether any special service items were not fully considered in the range construction. For the price record information in the generated price record information set, a reasonableness verification is performed on each price record information separately, and the range affiliation status of each price record information is marked, providing a basis for subsequent optimal selection.

[0058] In this embodiment, for price record information that fails the reasonableness check, a variety of flexible adjustment strategies are adopted to bring it back to a reasonable range, while preserving as much of the original price record information's personalized characteristics and dynamic factor adjustment logic as possible.

[0059] The first strategy is dynamic factor readjustment, which re-examines the calculation logic of the dynamic adjustment factor and adjusts it to bring price records back to a reasonable range. Specific implementation methods for dynamic factor readjustment include proportional scaling, component optimization, and constraint solving. Proportional scaling keeps the proportions of each component of the adjustment factor unchanged, scaling the overall adjustment factor value. Component optimization identifies the specific adjustment component causing the exceedance and reduces its weight or value individually, while preserving the dynamic characteristics of other adjustment factors. Constraint solving models the problem as a constrained optimization problem, with the objective function being the minimum adjustment range of the adjustment factor and the constraint being that the price records fall within a reasonable range. Numerical optimization algorithms are then used to find the optimal adjustment factor.

[0060] The second strategy is price record information reconstruction. In this embodiment, when the price record information fails the reasonableness check, the price generation process is reconstructed based on the original price generation model to regenerate price record information that meets the reasonable range constraint. In some implementations, price record information reconstruction includes hard constraint generation, knowledge-enhanced generation, and multi-round iterative optimization.

[0061] In the hard-constraint generation method, a reasonable range is embedded as a hard constraint, prompting the model to indicate that the generated price records fall within this reasonable range. In other words, the reasonable range is introduced as a constraint into the price generation process, controlling the generated price records to ensure they fall within the reasonable range.

[0062] In the knowledge-enhanced generation method, supplementary information such as market repair price information and historical transaction cases are added to the prompt words. By introducing the above-mentioned supplementary information, the price generation model can refer to the market repair price distribution and historical transaction situation when generating price record information, so that the generated price record information falls within a reasonable range.

[0063] In the multi-round iterative optimization approach, when the initially generated price record information fails the rationality check, the corresponding deviation information is obtained based on the check result and used as feedback input to the price generation process to adjust the generation process. Based on the feedback information, the price generation process is re-executed, and the newly generated price record information is checked for rationality again; if it still fails the check, the above process continues until the generated result passes the rationality check or the preset number of iterations is reached.

[0064] The third strategy involves selecting and combining multiple options. When price records that pass the reasonableness check exist, the price record with the highest confidence score that passes the check is prioritized as the final output. When multiple price records pass the reasonableness check, user profile information can be further combined to perform matching analysis on the price records, and the price record that matches the user's needs is selected as the output. When all price records fail the reasonableness check, the price record with the smallest deviation is selected and optimized. In some implementations, optimization includes: mapping and adjusting the price based on the boundary of the reasonable range to bring it back to the reasonable range; and readjusting the dynamic adjustment factor based on the dynamic weight of the price influencing factor edge to correct the price deviation.

[0065] Furthermore, in some implementations, a strategy for combining price record information is supported, that is, based on the price components in multiple price record information, different price record information is combined to generate new price record information, thereby improving flexibility and adaptability.

[0066] In one optional embodiment, retrieving target knowledge nodes associated with the target fault category in the fault repair knowledge graph includes: determining fault entity nodes in the fault repair knowledge graph based on the target fault entity, wherein the fault entity nodes correspond to subgraphs in the fault repair knowledge graph; the subgraphs include at least fault category nodes, fault phenomenon nodes, repair method nodes, spare parts nodes, and tool demand nodes; and recursively retrieving the target knowledge nodes in the subgraphs by using a path retrieval algorithm with dynamic weights corresponding to the edges of the connected price influencing factors as edge weights, in conjunction with the target fault category, to obtain the target knowledge nodes.

[0067] In this embodiment, the fault repair knowledge graph includes multiple knowledge nodes centered around a fault entity node, and forms a graph structure usable for reasoning through the relationships between these knowledge nodes. After determining the target fault category, the fault entity node corresponding to the target fault entity is located in the fault repair knowledge graph, and its corresponding subgraph is determined starting from this fault entity node. This subgraph is used to describe the scope of repair knowledge related to the target fault entity. This subgraph includes at least fault category nodes, fault phenomenon nodes, repair method nodes, spare parts nodes, and tool requirement nodes related to fault repair.

[0068] Furthermore, after determining the subgraph, knowledge nodes are retrieved within the subgraph based on the target fault category. Specifically, a path retrieval algorithm is used to recursively search the subgraph. During the retrieval process, the dynamic weights of the price influence factor edges connected to each knowledge node are used as edge weights for path retrieval, thus differentiating the weights of different retrieval paths. By introducing dynamic weights as edge weights, the path retrieval process can comprehensively consider the influence of different knowledge nodes on price correlations. Through the above recursive path retrieval process, target knowledge nodes associated with the target fault category are selected from the subgraph.

[0069] In this embodiment, a path retrieval algorithm using dynamic weights corresponding to price influence factors as edge weights is employed for recursive retrieval within the subgraph. This ensures that the retrieval process for target knowledge nodes considers not only the correlation between knowledge nodes and target fault categories but also the degree of influence of different knowledge nodes on price correlation. Through this method, target knowledge nodes with high correlation to target fault categories and significant price reference value can be selected from maintenance-related knowledge nodes, providing a more reasonable reasoning basis for the subsequent price generation process.

[0070] In one optional embodiment, the dynamic weight is dynamically updated based on price time-series data from an external data source, including: for any faulty entity, obtaining price time-series data from a third-party service data source and a regional service capacity index within a set time window; calculating the price change rate corresponding to the knowledge node associated with the faulty entity based on the price time-series data; normalizing the price change rate, the scenario feature parameters to which the faulty entity belongs, and the regional service capacity index to obtain the dynamic weight corresponding to the knowledge node; and establishing the correlation between the dynamic weight and the price influence factor edge connected to the knowledge node.

[0071] In this embodiment, for any faulty entity, price time-series data (including multi-dimensional price time-series data such as spare parts market prices, labor costs in different regions, transportation costs in different regions, and the market value of the faulty entity) from a third-party service data source is acquired within a preset time window. A regional service capacity index related to the faulty entity is also acquired to reflect the service supply situation within that time window (such as repair technician density and average order volume). Furthermore, based on the acquired price time-series data, the price changes corresponding to the knowledge nodes associated with the faulty entity are analyzed, and the corresponding price change rate is calculated. The price change rate is used to characterize the trend of price changes over time within the time window. Simultaneously, scenario characteristic parameters of the faulty entity, such as equipment type, usage environment, and repair level, are extracted.

[0072] In this embodiment, the price change rate, the scene feature parameters corresponding to the target fault entity, and the regional service capacity index are all normalized together, and the normalization result is used as a dynamic weight. Then, a correlation is established between this dynamic weight and its corresponding price influencing factor edge to achieve dynamic updating of the dynamic weight.

[0073] Optionally, the price generation model includes: a feature encoder, a price record generator, and a price record discriminator. The price generation model is invoked, and a multi-dimensional price association parameter set is input as reference data to generate a price record information set. This includes: S1. Inputting the multi-dimensional price association parameter set into the feature encoder, and performing the following operations in the feature encoder: For each target knowledge node, the dynamic weight of the price influence factor edge corresponding to the target knowledge node is used as the attention weight adjustment coefficient of the internal attributes of the target knowledge node; the association features between the internal attributes of the target knowledge node are calculated using a self-attention mechanism to obtain the same-node feature vector; For different target knowledge nodes, the dynamic weight of the price influence factor edge between target knowledge nodes is used as the cross-node attention weight; the association features between the attributes of different target knowledge nodes are calculated using a cross-attention mechanism to obtain the cross-node feature vector; the same-node feature vector and cross-node feature vector corresponding to each target knowledge node are concatenated to obtain the sub-fusion feature of each target knowledge node; the sub-fusion features of all target knowledge nodes are pooled to obtain the fusion feature vector. S2. Input the fused feature vector into the price record generator, and use the fused feature vector as the generation condition to generate diverse price records through multi-dimensional noise injection; S3. Input the price records into the price record discriminator and, based on the historical price records corresponding to the target fault category, determine whether the price records meet the requirements of distribution consistency and rationality. S4. If not satisfied, repeat steps S2-S3 to continue generating new price records. If satisfied, confirm that the price record meets the requirements. S5. When the number of qualified price records reaches the preset number, stop generating and use the qualified price records as a price record information set.

[0074] In this embodiment, the price generation model includes a feature encoder, a price record generator, and a price record discriminator. When generating the price record information set, the price generation model is invoked, with a multi-dimensional price correlation parameter set as reference data input, and feature encoding, price record generation, and price record discriminator are executed sequentially.

[0075] In S1, a multi-dimensional price correlation parameter set is input into the feature encoder. Within the feature encoder, for each target knowledge node, the dynamic weights corresponding to the price influence factor edges connected to that node are used to weight and model the attribute information within that node. By using the dynamic weights of the price influence factor edges corresponding to that target knowledge node as attention weight adjustment coefficients for the attributes within that node, and employing a self-attention mechanism, the correlation features between the attributes within that node are calculated, resulting in the same-node feature vector for that target knowledge node. This approach ensures that the same-node feature vector reflects the differences in price correlations between the attributes within the target knowledge node.

[0076] In S1, for the correlation between different target knowledge nodes, the dynamic weights corresponding to the price influence factor edges between target knowledge nodes are used as cross-node attention weights. A cross-attention mechanism is used to calculate the correlation features between the attributes of different target knowledge nodes, resulting in a cross-node feature vector. This cross-node feature vector reflects the mutual influence between different target knowledge nodes in price correlation. Furthermore, the same-node feature vector corresponding to each target knowledge node is concatenated with the cross-node feature vector to form a sub-fusion feature for each target knowledge node. Pooling is then applied to all sub-fusion features of the target knowledge nodes to obtain a fusion feature vector. This fusion feature vector characterizes the overall price correlation between multiple target knowledge nodes.

[0077] Furthermore, in S2, the aforementioned fused feature vector is input into the price record generator. Using the fused feature vector as a generation condition, multidimensional noise is introduced to generate diverse price records. This multidimensional noise injection aims to generate different price records while maintaining the reasonableness of price correlation.

[0078] In step S3, the generated price record is input into the price record discriminator, which judges whether the generated price record satisfies the distribution consistency and rationality based on the historical price record corresponding to the target fault category.

[0079] In step S4, if the judgment result is not satisfied, then return to steps S2 and S3 to continue generating a new price record; if the judgment result is satisfied, then determine that the corresponding price record meets the requirements.

[0080] In step S5, when the number of qualified price records reaches a preset number, the price record generation process is stopped, and the qualified price records are summarized to form a price record information set.

[0081] Further optionally, the original fault data is multimodal data; before sending an information supplementation request to the terminal device, the above method includes: evaluating the feature completeness of each feature dimension contained in the fault representation vector based on the multimodal data, and obtaining the feature completeness evaluation result; based on the feature completeness evaluation result, identifying the missing feature dimensions in the fault representation vector, and generating an information supplementation request for requesting the supplementation of the missing feature dimensions.

[0082] In this embodiment, the original fault data is multimodal data, meaning it may include text data, image data, audio data, and / or video data. Before sending an information supplementation request to the terminal device, the feature completeness of the currently constructed fault representation vector is evaluated based on the acquired multimodal data.

[0083] In this embodiment, the feature dimensions included in the fault characterization vector include noise features, visual features, behavioral features, and environmental features.

[0084] In this embodiment, one modality may correspond to multiple feature dimensions, or multiple modalities may be used together to characterize the same feature dimension.

[0085] Among them, noise features are mainly used to characterize the sound anomalies of the target faulty entity during operation, and the extraction source includes at least audio data; in addition, when video data contains visual cues that can reflect the source of the sound (such as vibration, collision), or when the user describes the sound features in text, noise features can also be constructed.

[0086] Visual features are primarily used to characterize the appearance, structural integrity, degree of damage, or instrument readings of a target faulty entity. Their extraction sources include keyframes in image and video data. When text data contains a detailed description of the appearance, it can also serve as supplementary information to visual features.

[0087] Behavioral features are mainly used to characterize the dynamic operation process or state changes of the target faulty entity. The sources of their extraction include the temporal information of video data and continuous image sequences (such as keyframes). In addition, the sound change patterns in audio data (such as the sound frequency change corresponding to the speed change) or sensor data can also be used to assist in the analysis of behavioral features.

[0088] Environmental features are mainly used to characterize the external conditions of the target faulty entity, including temperature, humidity, light, spatial layout, etc. The extraction sources include image data (environmental photos), video data (environmental dynamics), text data (user's verbal or written description of the environment), and audio data (environmental background sound).

[0089] Furthermore, based on the acquired multimodal data, the completeness of each feature dimension contained in the fault representation vector is evaluated to obtain a feature completeness evaluation result. In this embodiment, the completeness evaluation result is an assessment of the information completeness of each feature dimension in the fault representation vector. Specifically, the completeness evaluation result can be a score value, level label, or status identifier corresponding to each feature dimension, used to reflect whether there is any missing information in that feature dimension, so as to determine whether further supplementation is needed. In this embodiment, after obtaining the feature completeness evaluation result, the fault representation vector is analyzed according to the feature completeness evaluation result to identify missing feature dimensions. For the identified missing feature dimensions, an information supplementation request is generated, which is used to instruct the terminal device to supplement the fault data corresponding to the missing feature dimension.

[0090] In some implementations, after sending the price record information set to the terminal device, the price record information can be personalized based on user-related information. Specifically, multiple price record information sets can be matched and analyzed by combining the user's historical service records, consumption preferences, device usage habits, and / or consumption preference analysis results. Furthermore, multiple price record information sets can be personalized for recommendation display; for example, price record information can be tagged as an economy recommendation plan, a standard recommendation plan, or a premium recommendation plan. In some implementations, the terminal device compares and displays different personalized recommendation plans to assist users in choosing among multiple personalized recommendations. For example, for consumption-sensitive customers, the economy recommendation plan can be prioritized; for customers who prefer service quality or long-term device maintenance, the premium recommendation plan can be prioritized; and for customers who consider both price and service effectiveness, the standard recommendation plan can be prioritized.

[0091] In some implementations, the terminal device can differentiate the display of different price records based on the difference between the price record information and the upper and lower limits of the reasonable range. For example, price records above the upper limit of the reasonable range can be marked as "too high"; price records below the lower limit of the reasonable range can be marked as "too low"; and price records within the reasonable range can be marked as "reasonable". Further optionally, "too high", "too low", and "reasonable" prices can be distinguished and displayed using different labels, prompts, or sorting methods.

[0092] In an optional embodiment, after sending the price record information set to the terminal device, the above method further includes: receiving an order request sent by the terminal device, the order request including target price record information in the price record information set; matching maintenance personnel information based on the target price record information and the target fault category, combined with the maintenance personnel knowledge graph; when a maintenance personnel information is matched, generating a service processing order corresponding to the target fault entity, and sending the service processing order to the terminal device.

[0093] In this embodiment, after the price record information set is sent to the terminal device, a corresponding service processing order can be generated based on the terminal device's operation. Specifically, when the terminal device initiates an order request based on the displayed price record information set, the order request carries the target price record information selected by the user. Upon receiving the order request, the corresponding repair personnel information is matched according to the target price record information and the target fault category, combined with the repair personnel knowledge graph. The repair personnel knowledge graph stores relevant information about repair personnel. When repair personnel information that meets the conditions is matched, a service processing order corresponding to the target fault entity is generated. The service processing order includes at least the matched repair personnel information, and the generated service processing order is sent to the terminal device.

[0094] In one optional embodiment, the maintenance personnel knowledge graph dynamically maintains the skill information, current location information, and current load information of each maintenance personnel, with these information residing in different information layers. In this optional embodiment, when matching maintenance personnel information based on target price record information and target fault category, combined with the maintenance personnel knowledge graph, the process includes: parsing the current maintenance service request based on the target price record information and target fault category to obtain the fault category, service location, and urgency level involved in the current maintenance service; determining the query priority between information layers based on the urgency level of the current maintenance service; and retrieving information layer by layer in the maintenance personnel knowledge graph according to the fault category, service location, and urgency level involved in the current maintenance service, in descending order of query priority, until maintenance personnel information matching the fault category, service location, and urgency level involved in the current maintenance service is obtained.

[0095] In this embodiment, the skill information, current location information, and current load information of maintenance personnel are dynamically maintained based on a maintenance personnel knowledge graph. The skill information, current location information, and current load information are each set in different information layers. Each information layer describes different information dimensions of the maintenance personnel, thereby supporting hierarchical retrieval according to different information dimensions during the matching process.

[0096] In this embodiment, as Figure 6As shown, when matching repair personnel information based on target price records and target fault categories, and combining this with a repair personnel knowledge graph, the repair service request is analyzed. Specifically, based on the target price records and target fault categories, the fault categories, service locations, and urgency levels involved in this repair service are analyzed.

[0097] In this embodiment, after completing the analysis of the maintenance service requirements, the query priority among the information layers in the maintenance personnel knowledge graph is determined based on the urgency. The query priority indicates the retrieval order among information layers during subsequent information retrieval. After determining the query priority, information is retrieved layer by layer in the maintenance personnel knowledge graph according to the fault type, service location, and urgency of the maintenance service, in descending order of query priority. After each information layer is retrieved, the selected maintenance personnel information is used as the retrieval object for the next information layer. In this way, the candidate range of maintenance personnel information is gradually narrowed as the information layers are retrieved until maintenance personnel information that matches the fault type, service location, and urgency of the maintenance service is obtained.

[0098] In this embodiment, as Figure 6 As shown, a list of repair personnel is generated based on the matched repair personnel information, and service orders are automatically assigned to them, thus realizing order dispatch. After a repair personnel receives an order, it is determined whether they confirm the order: if the repair personnel confirm the order, the order status is updated to "assigned," and the subsequent on-site service process is initiated; if the repair personnel do not respond within the specified time or refuse to accept the order, other repair personnel are automatically selected from the repair personnel list to reassign the order, until the order is accepted or the manual dispatch mechanism is triggered.

[0099] In an optional embodiment, after sending the price record information set to the terminal device, the method further includes: receiving a price generation request sent by the terminal device, the price generation request including supplementary parameter information submitted to affect price generation; combining the parameter information, invoking the price generation model, using the multi-dimensional price association parameter set and parameter information as reference data input, and regenerating a new price record information set.

[0100] In this embodiment, after sending the price record information set to the terminal device, the terminal device is also allowed to initiate a price generation request based on the currently displayed price result. The price generation request includes supplementary parameter information that influences price generation, and this parameter information is used to form constraints during the price generation process.

[0101] Upon receiving a price generation request, the parameter information and the multi-dimensional price association parameter set are input as reference data into the price generation model. This allows the price generation model to generate a new set of price record information based on the supplementary parameter information and the price association parameter set. By using both the parameter information and the multi-dimensional price association parameter set as reference data input to the price generation model, the price generation result can be dynamically adjusted without changing the original generation process of the price generation model. This ensures that the regenerated price record information set reflects the impact of the parameter information on price generation. The process of generating the price record information set can be referred to the above embodiment and will not be repeated here.

[0102] In an optional embodiment, the above method further includes: generating maintenance assistance information based on target knowledge nodes and / or multimodal data input by the user; and sending the maintenance assistance information to the maintenance personnel terminal according to the maintenance personnel information to support the maintenance personnel in completing on-site maintenance operations of the target faulty entity.

[0103] In this embodiment, maintenance assistance information can be generated based on target knowledge nodes and / or multimodal data input by the user. This maintenance assistance information provides support for the maintenance process of the target faulty entity and may include, but is not limited to, maintenance method information, spare parts information, or tool requirements information related to the target fault category.

[0104] In this embodiment, after generating maintenance assistance information, the information is sent to the corresponding maintenance personnel's terminal based on the matched maintenance personnel information. The maintenance assistance information can be displayed on the maintenance personnel's terminal, facilitating on-site maintenance operations on the target faulty entity and assisting them in completing on-site maintenance of the target faulty entity more accurately and efficiently.

[0105] In this embodiment, when maintenance personnel arrive at the user's site, they can scan the device's QR code using a mobile terminal to obtain the unique identification information of the corresponding device. Based on the device identification information, they can query the basic attributes, historical maintenance records, and other auxiliary information corresponding to the device identification information from the fault maintenance knowledge graph, and then display maintenance assistance information to the maintenance personnel accordingly.

[0106] In this embodiment, the maintenance guidance information may include, but is not limited to: (1) Step-by-step maintenance instructions, used to guide maintenance personnel to carry out operations step by step according to the recommended standard maintenance procedures; (2) A list of required tools, used to indicate in advance the tools and protective equipment needed for this maintenance; (3) Safety precautions, used to highlight potential risks at the maintenance site and necessary protective requirements; (4) Frequently Asked Questions (FAQ) are used to provide common solutions for corresponding fault scenarios.

[0107] Maintenance personnel can perform on-site repairs based on the above information and use terminal devices to take photos or record the repair process in real time. The collected repair images or videos are uploaded, and a quality verification model automatically analyzes the repair results, such as determining whether the repair steps are complete, whether the parts are installed correctly, and whether the equipment's operating status meets standards after repair. If the verification is successful, a repair report is automatically generated. Maintenance personnel can guide users to view the repair results on the terminal and complete electronic signature confirmation. The order status is then updated based on the signed information, ultimately completing the evaluation of the repair order.

[0108] In this embodiment, the output format of the price generation model is not limited. Two output formats are given below, but are not limited to these.

[0109] Mode B1: Structured Detailed Output Mode. In Mode B1, any price record in the price record information set is presented in structured data form, which can directly include various price details. Price details are used to characterize the composition of the price record. For example, this may include, but is not limited to, spare parts costs, labor costs, service fees, and dynamically adjusted premiums (such as peak-hour surcharges and emergency service surcharges). In this embodiment, each price detail can be associated with a corresponding knowledge node in the fault repair knowledge graph (such as a spare parts node or repair labor hour node) to facilitate subsequent reasonableness verification and user traceability. When output to the terminal device, the price details are displayed visually to allow users to view the price composition item by item.

[0110] Mode B2: Total Price Output + Dynamic Decomposition Mode. In Mode B2, any price record in the price record information set is output as a single total price. No price record in the set includes detailed price information. In this mode, after generating the total price, the price decomposition engine is invoked to perform price composition decomposition processing on the price record information. This process splits the price record information according to preset price composition rules, resulting in price composition information containing various detailed price information. The price composition information describes the composition of the price record information. After obtaining the price composition information, it can be sent to the terminal device so that the terminal device can access the composition of the price record information. Figure 7 The image shown is an example of price detail information included in price composition information provided in this embodiment.

[0111] In this embodiment, the price decomposition engine decomposes the total price based on preset price composition rules and cost-related data in the fault repair knowledge graph. Cost-related data may include baseline information on spare parts costs, labor costs, and other expense items associated with the target fault category. The baseline information defines the price components of various repair services and their default percentage ranges (e.g., spare parts costs typically account for 40%-60% of the total price, and labor costs account for 20%-30%). Furthermore, the percentage of each price component can be dynamically adjusted by combining the dynamic weights of the current price influencing factors, generating a composition ratio suitable for the current scenario. Then, using the total price as a constraint, the total price is decomposed into each price component by solving a linear programming or proportional allocation algorithm, resulting in price composition information including detailed price information such as spare parts costs and labor costs. During the decomposition process, to ensure that the value of each price component conforms to the reasonable range of the corresponding node in the fault repair knowledge graph (e.g., the market price range of a certain model compressor is 300-500 yuan), if it exceeds the range, iterative correction is performed. Finally, the breakdown of price composition information is sent to the terminal device and displayed in a visual format, allowing users to clearly understand the logic behind the price quote.

[0112] Furthermore, the output mode can be intelligently selected based on user profile or terminal device type. For example, for enterprise users or high-value service processing orders, mode B1 is prioritized to provide fully transparent price details; for quick quotation scenarios or low-end service processing orders, mode B2 can be used to simplify the display, while dynamically breaking down and providing price details when the user clicks "View Details".

[0113] Through the above implementation, not only are flexible quotation output formats provided, but the price breakdown engine also enables reversible calculation from total price to details, ensuring that users can obtain transparent and understandable quotation information regardless of the output mode, thereby improving user experience and platform credibility.

[0114] In an optional embodiment, the price generation model may further include a price prediction and optimization mechanism for dynamically predicting repair price trends based on historical pricing data, market characteristic data, and user feedback data, and generating an optimized price generation model to improve the accuracy, stability, and adaptability of price generation.

[0115] First, historical pricing data can be obtained from the fault repair knowledge graph, including price records for different fault categories at different times. Simultaneously, market change data and customer feedback data are integrated to reflect external price fluctuations and user sensitivity to different pricing strategies. Time series analysis is performed on the historical pricing data to identify regular trends in price changes over time. Then, external factors such as market environment factors, regional differences, and material cost fluctuations are input into the external factor modeling module, generating structured variables that can be used for trend prediction through feature modeling.

[0116] After acquiring the aforementioned multi-source data, price trend prediction is performed based on time series characteristics and external factor variables, generating a price trend curve for the future time window. To ensure the reliability of the prediction results, the accuracy of the price trend prediction results can be verified. The verification results are used to drive the model parameter adjustment module, improving prediction accuracy and generalization ability by adjusting key parameters of the price generation model.

[0117] After accuracy verification and parameter adjustment, the pricing strategy is optimized based on the prediction results. This pricing strategy guides the model to price within a more reasonable price range and according to user sensitivity during subsequent price generation. To verify the actual effectiveness of the optimized pricing strategy, A / B testing can be conducted on a small scale. After obtaining the A / B test results, the strategy's effectiveness is comprehensively evaluated, and this evaluation serves as the basis for continuous optimization and iteration of the price generation model and pricing strategy.

[0118] In one optional embodiment, a fault repair knowledge graph update and model optimization mechanism is provided. When a new repair case is received, the case undergoes data cleaning and standardization. Subsequently, feature extraction and analysis steps are performed to extract structured features from the repair case and compare these features with historical cases in the fault repair knowledge graph to determine if the case contains new knowledge. If the determination result indicates that the case contains new knowledge, an expansion operation is performed, incorporating the newly extracted knowledge into the fault repair knowledge graph. If the determination result indicates that the knowledge is not new, existing knowledge enhancement processing is performed, strengthening the expressive power of existing knowledge by updating weights or supplementing related information. After the update or enhancement is completed, an incremental model training phase is initiated, using the updated knowledge base and new case features to train the model and improve its adaptability to new scenarios.

[0119] After completing incremental training, a model performance verification step is performed to evaluate the trained model in multiple dimensions, including accuracy and stability. If the verification results show that the model performance has improved, the new model is deployed to replace the original model; if the verification results do not meet expectations, the original model is rolled back to avoid performance degradation affecting business processes.

[0120] After the model is deployed, the performance monitoring phase begins. The model's performance in real-world scenarios is monitored in real time, and the monitoring results are used as feedback input into the continuous optimization loop to achieve long-term dynamic optimization of the model and knowledge base.

[0121] The detailed implementation methods and beneficial effects of each step in this embodiment have been described in detail in the foregoing embodiments, and will not be elaborated here.

[0122] It should be noted that the execution subject of each step of the method provided in the above embodiments can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 101 to 103 can be device A; or the execution subject of steps 101 and 102 can be device A, and the execution subject of step 103 can be device B; and so on.

[0123] Furthermore, some processes described in the above embodiments and accompanying drawings include multiple operations appearing in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or they may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0124] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 8 As shown, in practice, this electronic device includes a memory 84 and a processor 85.

[0125] Memory 84 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0126] Processor 85, coupled to memory 84, is used to execute computer programs in memory 84 for: receiving service requests uploaded by terminal devices, the service requests including raw fault data of a target fault entity; performing multi-dimensional feature extraction and fusion on the raw fault data to generate a fault representation vector; based on the fault representation vector, calling a diagnostic model to perform fault diagnosis, and outputting an initial fault category and corresponding confidence score; if the confidence score is lower than a preset threshold, sending an information supplementation request to the terminal device so that the terminal device can return incremental fault data; updating the fault representation vector based on the incremental fault data, and using the diagnostic model to perform fault diagnosis on the updated fault representation vector until the confidence score is lower than the preset threshold. The iteration termination condition is met to obtain the target fault category; in the fault maintenance knowledge graph, the target knowledge node associated with the target fault category is retrieved; the fault maintenance knowledge graph includes multiple knowledge nodes related to maintenance costs, at least some of which are connected by price influence factor edges, which have dynamic weights that are dynamically updated based on price time-series data from an external data source; a multi-dimensional price association parameter set is generated based on the attribute information of the target knowledge node and the dynamic weights of the price influence factor edges connected to the target knowledge node; a price generation model is invoked, and the multi-dimensional price association parameter set is used as reference data input to generate a price record information set; the price record information set is sent to the terminal device.

[0127] In an optional embodiment, the incremental fault data includes text data, image data, audio data, and / or video data; when updating the fault representation vector based on the incremental fault data, the processor 85 specifically performs the following operations: performs word embedding encoding on the text data to obtain a text fault feature vector; and / or performs convolutional neural network feature extraction on the image data to obtain an image fault feature vector; and / or performs acoustic feature transformation on the audio data to obtain an audio fault feature vector; and / or performs temporal keyframe encoding on the video data to obtain a video fault feature vector; and then stacks and fuses the text fault feature vector, the image fault feature vector, the audio fault feature vector, and / or the video fault feature vector with the fault representation vector in a unified semantic space to obtain an updated fault representation vector.

[0128] In an optional embodiment, when the processor 85 retrieves a target knowledge node associated with the target fault category in the fault repair knowledge graph, it specifically performs the following steps: based on the target fault entity, it determines a fault entity node in the fault repair knowledge graph, wherein the fault entity node corresponds to a subgraph in the fault repair knowledge graph; the subgraph includes at least fault category nodes, fault phenomenon nodes, repair method nodes, spare parts nodes, and tool demand nodes; and, in conjunction with the target fault category, recursively searches the subgraph using a path retrieval algorithm that uses the dynamic weights corresponding to the edges of the connected price influencing factors as edge weights to obtain the target knowledge node.

[0129] In an optional embodiment, the dynamic weight is dynamically updated based on price time-series data from an external data source. The processor 85 is specifically configured to: for any faulty entity, acquire price time-series data from a third-party service data source within a set time window and a regional service capacity index; calculate the price change rate corresponding to the knowledge node associated with the faulty entity based on the price time-series data; normalize the price change rate, the scene feature parameters to which the faulty entity belongs, and the regional service capacity index to obtain the dynamic weight corresponding to the knowledge node; and establish the association between the dynamic weight and the price influence factor edge connected to the knowledge node.

[0130] In an optional embodiment, the price generation model includes: a feature encoder, a price record generator, and a price record discriminator; when the processor 85 calls the price generation model and inputs the multi-dimensional price association parameter set as reference data to generate a price record information set, it specifically performs the following operations: S1, inputting the multi-dimensional price association parameter set into the feature encoder, and performing the following operations in the feature encoder: for each target knowledge node, using the dynamic weight of the price influence factor edge corresponding to the target knowledge node as the attention weight adjustment coefficient of the internal attributes of the target knowledge node, and using a self-attention mechanism to calculate the association features between the internal attributes of the target knowledge node to obtain the same node feature vector; for different target knowledge nodes, using the dynamic weight of the price influence factor edge between target knowledge nodes as the cross-node attention weight, and using a cross-attention mechanism to calculate the relationship between the attributes of different target knowledge nodes. The following steps are performed: S1) Concatenate the features of each target knowledge node as a cross-node feature vector; concatenate the same-node feature vector and cross-node feature vector corresponding to each target knowledge node to obtain the sub-fusion feature of each target knowledge node; pool all the sub-fusion features of the target knowledge nodes to obtain a fusion feature vector; S2) Input the fusion feature vector into the price record generator, and use the fusion feature vector as the generation condition to generate diverse price records through multi-dimensional noise injection; S3) Input the price record into the price record discriminator, and determine whether the price record meets the requirements of distribution consistency and rationality based on the historical price records corresponding to the target fault category; S4) If not, repeat steps S2-S3 to continue generating new price records; if satisfied, determine that the price record meets the requirements; S5) When the number of qualified price records reaches a preset number, stop generating and use the qualified price records as a price record information set.

[0131] In an optional embodiment, the original fault data is multimodal data; before sending an information supplementation request to the terminal device, the processor 85 is further configured to: perform feature completeness evaluation on each feature dimension contained in the fault representation vector according to the multimodal data, and obtain a feature completeness evaluation result; based on the feature completeness evaluation result, identify the missing feature dimension in the fault representation vector, and generate an information supplementation request for requesting the supplementation of the missing feature dimension.

[0132] In an optional embodiment, after sending the price record information set to the terminal device, the processor 85 is further configured to: receive an order request sent by the terminal device, the order request including target price record information in the price record information set; based on the target price record information and the target fault category, and in conjunction with the maintenance personnel knowledge graph to match maintenance personnel information, when a maintenance personnel information is matched, generate a service processing order corresponding to the target fault entity, and send the service processing order to the terminal device; or, receive a price generation request sent by the terminal device, the price generation request including supplementary parameter information submitted to affect price generation; invoke the price generation model, input the multi-dimensional price association parameter set and the parameter information as reference data, and regenerate a new price record information set.

[0133] In an optional embodiment, the processor 85 is further configured to: generate maintenance assistance information based on the target knowledge node and / or multimodal data input by the user; and send the maintenance assistance information to the maintenance personnel terminal according to the maintenance personnel information to support the maintenance personnel in completing on-site maintenance operations of the target faulty entity.

[0134] In an optional embodiment, the maintenance personnel knowledge graph dynamically maintains the skill information, current location information, and current load information of each maintenance personnel, and the skill information, current location information, and current load information are located in different information layers; when the processor 85 matches maintenance personnel information based on the target price record information and the target fault category, combined with the maintenance personnel knowledge graph, it specifically performs the following: analyzes the current maintenance service requirement based on the target price record information and the target fault category to obtain the fault category, service location, and urgency involved in the current maintenance service; determines the query priority between information layers based on the urgency involved in the current maintenance service; and performs information retrieval layer by layer in the maintenance personnel knowledge graph according to the fault category, service location, and urgency involved in the current maintenance service, in descending order of the query priority, until maintenance personnel information that matches the fault category, service location, and urgency involved in the current maintenance service is obtained.

[0135] In one optional embodiment, for any price record information in the price record information set, the any price record information includes multiple price detail information; or, the processor 85 is further configured to: for any price record information in the price record information set, perform price composition decomposition processing on the any price record information to obtain price composition information containing multiple price detail information; and send the price composition information corresponding to the any price record information to the terminal device.

[0136] Furthermore, such as Figure 8As shown, the electronic device also includes other components such as a communication component 86, a display 87, a power supply component 88, and an audio component 89. Figure 8 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 8 The components shown. Additionally... Figure 8 The components within the dashed box are optional, not mandatory, and their specific requirements depend on the product form of the electronic device. The electronic device in this embodiment can be a terminal device such as a desktop computer, laptop computer, smartphone, or IoT device, or a server-side device such as a conventional server, cloud server, or server array. If the electronic device in this embodiment is a terminal device such as a desktop computer, laptop computer, or smartphone, it may include... Figure 8 The components within the dashed box; if the electronic device in this embodiment is implemented as a conventional server, cloud server, or server array, etc., it may be omitted. Figure 8 The component within the dashed box.

[0137] The aforementioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0138] The aforementioned communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel.

[0139] The aforementioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a Touch Panel, the screen can be implemented as a touchscreen to receive input signals from the user. The Touch Panel includes one or more touch sensors to sense touches, swipes, and gestures on the Touch Panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0140] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0141] The aforementioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0142] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above-described method embodiments. The computer-readable storage medium includes volatile or non-volatile components, or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium. Accordingly, this application also provides a computer program product, which includes a computer program or instructions that, when executed by a processor, cause the processor to implement the steps in the above method embodiments. It should be understood that each step or combination of steps in the above method flow can be implemented by the computer program or instructions. Furthermore, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, enabling the processor of the general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to function as an apparatus for implementing the corresponding functions in the above method embodiments.

[0143] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0144] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A service processing method, characterized in that, include: Receive service requests uploaded by terminal devices, the service requests including the original fault data of the target fault entity; Multi-dimensional feature extraction and fusion are performed on the original fault data to generate a fault characterization vector; Based on the fault characterization vector, the diagnostic model is invoked to perform fault diagnosis, and the initial fault category and corresponding confidence score are output. If the confidence score is lower than a preset threshold, an information supplementation request is sent to the terminal device so that the terminal device can return incremental fault data; the fault representation vector is updated based on the incremental fault data, and the diagnostic model is used to diagnose the fault in the updated fault representation vector until the confidence score meets the iteration termination condition, and the target fault category is obtained. In the fault repair knowledge graph, target knowledge nodes associated with the target fault category are retrieved; the fault repair knowledge graph includes multiple knowledge nodes related to repair costs, and at least some knowledge nodes are connected by price influence factor edges, which have dynamic weights, and the dynamic weights are dynamically updated based on price time series data from external data sources. Based on the attribute information of the target knowledge node and the dynamic weights of the price influencing factor edges connected to the target knowledge node, a multi-dimensional price correlation parameter set is generated. The price generation model is invoked, and the multi-dimensional price correlation parameter set is used as reference data input to generate a price record information set; The price record information set is sent to the terminal device.

2. The method according to claim 1, characterized in that, The incremental fault data includes text data, image data, audio data, and / or video data; Updating the fault representation vector based on the incremental fault data includes: The text data is subjected to word embedding encoding to obtain a text fault feature vector; and / or Perform convolutional neural network feature extraction on the image data to obtain image fault feature vectors; and / or Perform acoustic feature transformation on the audio data to obtain an audio fault feature vector; and / or Perform temporal keyframe encoding on the video data to obtain video fault feature vectors; The text fault feature vector, the image fault feature vector, the audio fault feature vector, and / or the video fault feature vector are layered and fused with the fault representation vector in a unified semantic space to obtain the updated fault representation vector.

3. The method according to claim 1, characterized in that, In the fault repair knowledge graph, retrieving target knowledge nodes associated with the target fault category includes: based on the target fault entity, determining fault entity nodes in the fault repair knowledge graph, wherein the fault entity nodes correspond to subgraphs in the fault repair knowledge graph; the subgraphs include at least fault category nodes, fault phenomenon nodes, repair method nodes, spare parts nodes, and tool requirement nodes. Based on the target fault category, a path retrieval algorithm is used, with the dynamic weights corresponding to the edges of the connected price influencing factors as edge weights, to recursively search the subgraph to obtain the target knowledge node.

4. The method according to claim 1, characterized in that, The dynamic weights are dynamically updated based on time-series price data from an external data source, including: For any faulty entity, obtain the price time-series data and regional service capacity index of the third-party service data source within the set time window; Based on the price time-series data, calculate the price change rate corresponding to the knowledge node associated with any faulty entity; The price change rate, the scene feature parameters to which any faulty entity belongs, and the regional service capacity index are normalized to obtain the dynamic weight corresponding to the knowledge node. Establish the correlation between the dynamic weights and the price influence factor edges connected to the knowledge nodes.

5. The method according to claim 1, characterized in that, The price generation model includes: a feature encoder, a price record generator, and a price record discriminator; the price generation model is invoked, and the multi-dimensional price association parameter set is used as reference data input to generate a price record information set, including: S1. Input the multi-dimensional price association parameter set into the feature encoder, and perform the following operations in the feature encoder: For each target knowledge node, the dynamic weight of the price influencing factor edge corresponding to the target knowledge node is used as the attention weight adjustment coefficient of the internal attributes of the target knowledge node. The self-attention mechanism is used to calculate the correlation features between the internal attributes of the target knowledge node to obtain the feature vector of the same node. For different target knowledge nodes, the dynamic weight of the price influence factor edge between target maintenance knowledge nodes is used as the cross-node attention weight. The cross-attention mechanism is used to calculate the correlation features between the attributes of different target knowledge nodes, which are used as the cross-node feature vector. The same-node feature vector and cross-node feature vector corresponding to each target knowledge node are concatenated to obtain the sub-fusion feature of each target knowledge node. The sub-fusion features of all target knowledge nodes are pooled to obtain the fusion feature vector. S2. Input the fused feature vector into the price record generator, and use the fused feature vector as the generation condition to generate diverse price records through multi-dimensional noise injection; S3. Input the price record into the price record discriminator, and determine whether the price record satisfies the distribution consistency and rationality based on the historical price record corresponding to the target fault category; S4. If not satisfied, repeat steps S2-S3 to continue generating new price records. If satisfied, determine that the price record meets the requirements. S5. When the number of qualified price records reaches the preset number, stop generating and use the qualified price records as a price record information set.

6. The method according to any one of claims 1-5, characterized in that, The original fault data is multimodal data; before sending an information supplementation request to the terminal device, the method includes: Based on the multimodal data, the feature completeness of each feature dimension contained in the fault characterization vector is evaluated to obtain the feature completeness evaluation result. Based on the feature completeness assessment results, the missing feature dimensions in the fault characterization vector are identified, and an information supplementation request is generated to request the supplementation of the missing feature dimensions.

7. The method according to any one of claims 1-5, characterized in that, After sending the price record information set to the terminal device, the method further includes: The system receives an order request sent by the terminal device, the order request including target price record information in the price record information set; based on the target price record information and the target fault category, it matches maintenance personnel information using a maintenance personnel knowledge graph, and when maintenance personnel information is matched, it generates a service processing order corresponding to the target fault entity and sends the service processing order to the terminal device. or The system receives a price generation request sent by the terminal device, the price generation request including supplementary parameter information that affects price generation; it then calls the price generation model, inputs the multi-dimensional price association parameter set and the parameter information as reference data, and regenerates a new price record information set.

8. The method according to claim 7, characterized in that, The method further includes: Based on the target knowledge nodes and / or multimodal data input by the user, maintenance assistance information is generated; Based on the maintenance personnel information, the maintenance assistance information is sent to the maintenance personnel's terminal to support the maintenance personnel in completing on-site maintenance operations on the target faulty entity.

9. The method according to claim 7, characterized in that, The maintenance personnel knowledge graph dynamically maintains the skill information, current location information, and current load information of each maintenance personnel, and the skill information, current location information, and current load information are located in different information layers; Based on the target price record information and the target fault category, and combined with the maintenance personnel knowledge graph, maintenance personnel information is matched, including: Based on the target price record information and the target fault category, the repair service requirement is analyzed to obtain the fault category, service location, and urgency of the repair service. Based on the urgency of this repair service, determine the query priority between information layers; Based on the fault type, service location, and urgency of this repair service, information is retrieved layer by layer in the repair personnel knowledge graph according to the query priority from high to low, until repair personnel information that matches the fault type, service location, and urgency of this repair service is obtained.

10. The method according to any one of claims 1-5, characterized in that, For any price record in the price record information set, the any price record includes multiple price details; or The method further includes: For any price record in the price record information set, the price composition is decomposed to obtain price composition information containing multiple price details. The price composition information corresponding to any of the price record information is sent to the terminal device.

11. An electronic device, characterized in that, include: A processor and a memory; the memory stores a computer program, and the processor, after running the computer program, performs the method according to any one of claims 1-10.

12. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, performs the method described in any one of claims 1-10.

Images