Power supply scheme generation method, device and computer equipment

Abstract

The application discloses a power supply scheme generation method, device and computer equipment. The method comprises the following steps: constructing a power supply scheme template knowledge base; using a multi-modal large model to perform multi-modal fusion and semantic understanding on collected user electricity demand and field investigation data to generate a fused semantic vector; matching a target power supply scheme template from the power supply scheme template knowledge base based on the fused semantic vector; based on the fused semantic vector, using an expert model to dynamically fill and rule embed key fields of the target power supply scheme template to generate a preliminary power supply scheme; performing multi-dimensional index evaluation and multi-layer audit on the preliminary power supply scheme, and adjusting the preliminary power supply scheme based on the evaluation and audit results to generate and output a target power supply scheme with risk identification and credibility classification. The method of the application can generate a power supply scheme that meets technical specifications and policy standards, significantly improves the efficiency, accuracy and compliance of scheme compilation, and reduces the degree of human participation and the design cycle.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method, apparatus and computer equipment for generating power supply schemes. Background Technology

[0002] With the deepening of digital transformation in power marketing, power supply scheme preparation plays a crucial role in power grid companies' engineering application, capacity expansion and renovation, and distribution network planning. Traditional power supply scheme design mainly relies on manual preparation by engineers based on experience, design specifications, and templates, which is characterized by large workload, long cycle, and strong subjectivity.

[0003] In addition, traditional solution generation methods usually rely on static templates and fixed logic, which makes it difficult to dynamically respond to the diverse electricity needs of customers. They also cannot effectively integrate on-site survey information, geographical environment data and engineering constraints, resulting in high content duplication, poor customizability, and low efficiency in reviewing the rationality and compliance of engineering projects, making it difficult to meet the business needs of intelligent marketing and lean power supply.

[0004] In recent years, although some systems have introduced auxiliary tools based on rule bases or knowledge graphs, there are still problems such as single data sources and insufficient model versatility. In particular, when processing multi-source information (such as customer text, GIS geographic information, and on-site images), it is impossible to achieve unified understanding and integration at the semantic level. Summary of the Invention

[0005] In view of this, this application provides a power supply scheme generation method, apparatus, and computer equipment to solve the problems mentioned in the background art.

[0006] According to a first aspect of this application, a method for generating a power supply scheme is provided, the method comprising: Construct a power supply solution template knowledge base, which includes multiple power supply solution templates with key fields marked with logical dependencies, as well as the relationship between power supply solution templates, scenario conditions, and user requirements; A multimodal large model is used to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data to generate a fused semantic vector; The target power supply scheme template is matched from the power supply scheme template knowledge base based on the fused semantic vector; Based on the fused semantic vector, an expert model is used to dynamically fill and embed rules into the key fields of the target power supply scheme template to generate a preliminary power supply scheme. The preliminary power supply scheme is evaluated using multi-dimensional indicators and reviewed at multiple levels. Based on the evaluation and review results, the preliminary power supply scheme is adjusted to generate and output a target power supply scheme with risk identification and credibility rating.

[0007] According to a second aspect of this application, a power supply scheme generation apparatus is provided, the apparatus comprising: The knowledge base construction module is used to build a power supply scheme template knowledge base, wherein the power supply scheme template knowledge base includes multiple power supply scheme templates marked with key fields with logical dependencies, as well as the relationship between power supply scheme templates, scenario conditions and user requirements; The feature fusion module is used to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data using a multimodal large model, and generate a fused semantic vector; The template matching module is used to match a target power supply scheme template from the power supply scheme template knowledge base based on the fused semantic vector; The scheme generation module is used to dynamically fill and embed rules into the key fields of the target power supply scheme template based on the fused semantic vector using an expert model to generate a preliminary power supply scheme; and to evaluate the preliminary power supply scheme with multi-dimensional indicators and conduct multi-level review, and adjust the preliminary power supply scheme based on the evaluation and review results, and generate and output a target power supply scheme with risk identification and credibility rating.

[0008] According to a third aspect of this application, a readable storage medium is provided on which a program or instructions are stored, which, when executed by a processor, implement the steps of the power supply scheme generation method described above.

[0009] According to a fourth aspect of this application, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the power supply scheme generation method described above.

[0010] By employing the aforementioned technical solution, a multimodal large-scale model is used to perform semantic understanding and multimodal fusion of user needs and on-site survey data. This extracts deep-level fusion semantic vectors from heterogeneous data, accurately expressing the user's actual needs and specific on-site conditions. Based on these fusion semantic vectors, a corresponding target power supply scheme template is matched in the power supply scheme template knowledge base. This ensures that the target power supply scheme template meets both enterprise power supply specifications and user electricity needs. After obtaining the target power supply scheme template, an expert model, combined with power industry expertise and rules, automatically and intelligently populates the template and embeds relevant rules. Multidimensional evaluation and review are then conducted to generate a final target power supply scheme with risk identification and credibility rating.

[0011] This application's embodiments, on the one hand, overcome the limitations of traditional solutions that rely solely on static templates and fixed logic. By combining dynamic management of the template knowledge base with a semantic-driven generation mechanism, reliable power supply solutions can be dynamically generated based on multi-dimensional needs, significantly reducing manual intervention and improving the efficiency and relevance of solution generation. On the other hand, it employs a collaborative mechanism between a multimodal large model and an expert model. The large model is responsible for semantic matching and initial template selection, while the expert model is responsible for professional rule embedding, field filling, and solution review. This leverages the powerful contextual understanding and information integration capabilities of the large model while ensuring the accuracy and reliability of the expert model in professional calculations and rule verification, greatly improving the intelligence, efficiency, and accuracy of power supply solution generation.

[0012] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0013] 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 the power supply scheme generation method provided in an embodiment of this application is shown; Figure 2 A structural block diagram of the power supply scheme generation device provided in an embodiment of this application is shown; Figure 3 A schematic diagram of the electronic structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0014] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0015] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0016] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “attached” to another element, it can be directly connected or attached to the other element, or there may be intermediate elements. Furthermore, “connected” or “attached” as used herein can include wireless connections or wireless interconnections. The term “and / or” as used herein includes all or any unit and all combinations of one or more associated listed items.

[0017] Exemplary embodiments according to this application will now be described in more detail with reference to the accompanying drawings. However, these exemplary embodiments may be implemented in many different forms and should not be construed as being limited to the embodiments set forth herein. It should be understood that these embodiments are provided so that the disclosure of this application is thorough and complete, and that the concept of these exemplary embodiments is fully conveyed to those skilled in the art.

[0018] The power supply scheme generation method provided in this embodiment can be applied to a power supply system, which constructs a layered interactive architecture: the top layer is a multimodal large model layer, responsible for semantic parsing, context understanding and document content generation; the middle layer is an expert model cluster, including load calculation model, line selection model, distribution transformer capacity assessment model and economic decision-making model; the bottom layer is a task scheduling and data management module, used to realize data flow and instruction distribution between various sub-models.

[0019] During the interaction, the large model encodes user requirement descriptions, on-site survey texts, image features, and other information into semantic vectors through a semantic understanding module, and dynamically distributes them to the corresponding expert models based on the task type. After receiving the task, the expert model performs professional calculations based on its domain knowledge. For example, the load calculation model calculates the transformer capacity based on load type, climate coefficient, and power factor; the line selection model selects the conductor cross-section based on power supply radius, conductor resistivity, and economic current density; and the transformer evaluation model performs a rationality check on the scheme based on the construction environment and power supply level.

[0020] This embodiment provides a method for generating a power supply scheme, such as Figure 1 As shown, the method includes: Step 101: Construct a knowledge base for power supply scheme templates.

[0021] The power supply solution template knowledge base includes multiple power supply solution templates with key fields marked with logical dependencies, as well as the relationships between power supply solution templates, scenario conditions, and user requirements. This facilitates the subsequent retrieval of templates that match actual user needs and on-site conditions through these relationships, and the population of key field content through logical dependencies.

[0022] In one embodiment, step 101, namely, constructing a power supply scheme template knowledge base, specifically includes the following steps: Step 101-1: Based on different power supply project types, create standard templates and collect historical case documents.

[0023] The power supply project types include customer power supply type, project attribute type, and voltage level type. For example, the customer power supply type can be a residential power supply project, an industrial and commercial power supply project, or a special project power supply project, etc.; the project attribute type can be a newly built power supply project or a renovation project; and the voltage level type can be a low-voltage power supply project or a medium- and high-voltage power supply project, etc.

[0024] Step 101-2: Use a natural language segmentation algorithm to identify key fields within the standard template.

[0025] Key fields include project scale, load type, transformer capacity, and construction period.

[0026] Step 101-3: Perform structured annotation of slots on the standard template and establish logical dependencies between key fields.

[0027] Logical dependencies are used to indicate multiple key fields that influence each other. For example, in a power system, transformer capacity is affected by load category, so logical dependencies can be used to associate the first key field "load category" with the second key field "transformer capacity".

[0028] In this embodiment, by identifying key fields and labeling slots, the standard template is transformed from unstructured text into a machine-readable structured framework. This facilitates automated field filling by the expert model, improving the efficiency of dynamic solution generation. Simultaneously, logical dependencies between key fields are established. These dependencies constrain the automated field compilation logic, allowing the expert model to automatically verify the consistency of business rules between data during field filling. This enhances compilation efficiency and solution accuracy, ensuring that the generated power supply solution is technically self-consistent, compliant, and complete.

[0029] Step 101-4: Based on historical case documents, the annotated standard templates are standardized for multiple regions to form power supply scheme templates that are compatible with the standards of different power supply companies.

[0030] In this embodiment, multi-regional standardization is implemented by incorporating historical cases, enabling the power supply scheme template to adapt to different power supply company standards, policy requirements, and engineering practices through a dynamic tagging system. This ensures the template's universality and scalability across different regions and company standards, thereby enhancing the flexibility, intelligence, and engineering adaptability of the power supply scheme generation system.

[0031] Step 101-5: Analyze historical case documents and extract the scenario conditions and user requirements of historical power supply schemes.

[0032] Step 101-6: Associate the power supply scheme templates that conform to the historical power supply schemes with the scenario conditions and user needs, and store the power supply scheme templates hierarchically according to the business scenario to form a power supply scheme template knowledge base.

[0033] In this embodiment, scenario conditions (such as geographical environment and engineering constraints) and user requirements (such as power capacity and power supply reliability requirements) are extracted from historical cases and associated with power supply scheme templates for storage. This allows the multimodal large model to quickly match the optimal template based on user requirements. Compared with the traditional method of manually compiling templates, the template update efficiency is improved by more than 50%, the field extraction accuracy reaches more than 95%, and the degree of scheme customization is greatly improved.

[0034] Step 102: Use a multimodal large model to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data to generate a fused semantic vector.

[0035] User electricity demand data includes electricity consumption and load type, while on-site survey data includes geographical coordinates, climate conditions, power supply distance, and on-site image information. User electricity demand and on-site survey data can be obtained from multiple sources, such as customer electricity application forms, load type descriptions, power supply radius, geographical coordinates, local climate conditions, and on-site images and videos. The image data is acquired through high-definition cameras installed on the survey terminal or by drones.

[0036] In one embodiment, step 102, which involves using a multimodal large model to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data to generate a fused semantic vector, specifically includes the following steps: Step 102-1: Perform feature encoding on user electricity demand and on-site survey data respectively to obtain text feature vectors and visual feature vectors; Step 102-2: A multimodal fusion algorithm based on attention mechanism is used to perform weighted fusion of text feature vectors and visual feature vectors to generate multimodal features.

[0037] Step 102-3: Construct a fused semantic vector based on multimodal features.

[0038] The fusion semantic vector includes power demand patterns, construction condition constraints, and reliability levels.

[0039] In this embodiment, by separating and encoding user requirement text and on-site survey visual information and then adaptively fusing them based on an attention mechanism, deep semantic alignment and accurate context-aware fusion of cross-modal data are achieved. This transforms multi-source heterogeneous data into a knowledge representation that can drive intelligent solution generation, providing reliable input with both scene awareness and professional interpretability for subsequent template matching and expert decision-making. This significantly improves the adaptability and accuracy of power supply solutions to complex on-site conditions.

[0040] For example, after obtaining multi-source information such as user electricity demand and on-site survey data, feature encoding is performed. Image data is processed by a visual encoder to extract visual features such as line direction, tower distribution, terrain slope, and obstacle information, and a spatial vector representation is generated. Text data undergoes word segmentation, noise reduction, and structured classification through a preprocessing module, and demand constraints (such as power capacity, planned construction period, access voltage level, and construction environment requirements) are extracted through a semantic understanding model. GIS data is converted into coordinate vectors through a spatial mapping module and fused with the spatial features of the image data. A multimodal fusion algorithm is used to achieve a unified vector representation of image, text, and GIS spatial data. Subsequently, an attention-based fusion algorithm is used to weight and combine visual, text, and spatial vectors to generate unified multimodal features. During the fusion process, the system sets dynamic weights based on information confidence to improve fusion stability in cases of missing data or blurred images.

[0041] It's worth noting that the large model is responsible for semantic parsing, content generation, and task distribution. Through a capability decoupling module, semantic understanding, information fusion, content generation, and rule judgment are modularized and layered. This encapsulates the semantic understanding, information fusion, content generation, and rule judgment of the multimodal large model into independent functional units. These units interact via interface calls rather than parameter sharing, achieving structural isolation between the semantic and rule layers, thus preventing over-generation or semantic drift of the large model. Furthermore, a consistency detection engine is employed during the interaction between units to verify the logical consistency, data boundaries, and constraints of the model output, ensuring that voltage levels, line capacity, and economic indicators meet the constraints. After consistency verification, the outputs of each model are fused into a comprehensive scheme intent vector. This vector encompasses semantic objectives, technical parameters, and logical constraint information, serving as the core input for subsequent template filling and scheme generation.

[0042] Step 103: Match the target power supply scheme template from the power supply scheme template knowledge base based on the fused semantic vector.

[0043] In this embodiment, the semantic understanding capabilities of a multimodal large model are first utilized to deeply fuse and unify the semantic understanding of user electricity demand text with multi-source heterogeneous data such as on-site survey images and geographic information, generating accurate fused semantic vectors. This provides contextual constraints and semantic support for template filling and content generation. Then, based on these vectors, the most similar target power supply scheme template is matched from the template knowledge base. This overcomes the limitations of mechanical matching based on keywords or fixed rules, generating power supply schemes that comply with technical specifications and policy standards.

[0044] Understandably, similarity calculation combines two methods: semantic distance measurement and business feature matching. Semantic distance is determined by the cosine similarity of BERT vectors, while business feature matching is based on a weighted comparison according to load type, project scale, and regional standards. The system prioritizes templates with a similarity higher than 0.85 as initial candidates to ensure that the scheme structure is consistent with the project type.

[0045] Step 104: Based on the fused semantic vector, an expert model is used to dynamically fill in and embed rules into the key fields of the target power supply scheme template to generate a preliminary power supply scheme.

[0046] Among them, the expert model is responsible for load calculation, line selection, and distribution transformer scheme evaluation.

[0047] In this embodiment, the expert model, based on its understanding of user intent through a multimodal large model, combined with power industry expertise and rules, automatically and intelligently fills in the template and embeds corresponding rules, using the logical dependencies between key fields in the template as constraints. This ensures that the automatically filled content is highly adapted to the user's actual electricity usage scenario, while avoiding parameter inconsistencies and compliance flaws caused by insufficient experience in manual compilation, significantly improving the efficiency, accuracy, and customization of the solution generation.

[0048] In one embodiment, step 104, namely, based on the fused semantic vector, using an expert model to dynamically fill and embed rules into the key fields of the target power supply scheme template to generate a preliminary power supply scheme, specifically includes the following steps: Step 104-1: Extract the text information of the first key field in the corresponding target power supply scheme template from the user intent vector for initial filling, and mark the suspicious fields with confidence scores below the confidence threshold.

[0049] Understandably, the allowed range of field values ​​is usually marked in the key fields of the target power supply scheme template. If the text information of the key field being filled out exceeds the field value range, the confidence level is calculated based on the difference between the text information and the field value range. For suspicious fields with low confidence levels, the use of text information corresponding to the user intent vector for filling can be cancelled to avoid filling in illegal information.

[0050] Step 104-2: Based on the logical dependencies and the already filled fields, calculate the text information of the remaining second key fields and perform secondary filling.

[0051] The first key field is a key field that can be directly filled in using the user intent vector, while the second key field is a key field that is associated with the first key field through logical dependencies and whose content needs to be calculated and filled in using the content of the first key field.

[0052] Step 104-3: During the filling process, embed at least one engineering constraint from the power distribution access principle, power supply radius limit, and wire diameter selection constraint through the rule engine.

[0053] Step 104-4: The document pipeline layout algorithm is used to optimize the hierarchy and adjust the layout of the filled text information to form a preliminary power supply scheme.

[0054] In this embodiment, the expert model first extracts the text information corresponding to the user intent vector based on field confidence to initially fill the template slots of the first key field and marks suspicious fields. Then, relying on the logical dependencies of key fields, it calculates and fills the text information of the remaining unfilled fields using auxiliary information such as already filled fields, the built-in power knowledge base, and GIS geographic environment data. This ensures the internal logical consistency of the entire solution, avoids data conflicts and unreasonable designs, and improves the efficiency and accuracy of solution development. Furthermore, during the field filling process, the rule engine is invoked to introduce and verify various national standards, industry specifications, safety distances, technical parameter limitations, and other engineering constraints for the currently filled fields in real time. Once the filled value of a field violates the rules, the rule engine can immediately trigger prompts, warnings, or suggestions for backtracking and modification, greatly reducing the risk that the solution will be found to be non-compliant or infeasible during subsequent reviews. Finally, a document pipeline layout algorithm is used to optimize and adjust the layout of the filled text information, so that the generated preliminary power supply plan can automatically organize, format and present complex professional information in a standardized way, making the plan easier to understand and review. This not only reduces the tedious work and time consumption caused by manual post-layout, but also ensures that all generated power supply plans have a unified presentation format and hierarchical structure, laying a high-quality foundation for subsequent evaluation and review.

[0055] For example, after the target power supply scheme template is selected, the model's filling module sorts the template slots according to field confidence levels and fills them in layers. For high-confidence fields (such as project name, design capacity, number of distribution transformers, etc.), the system directly calls the corresponding information in the semantic intent vector. For low-confidence fields (such as construction period, material selection recommendations), the system uses expert model reasoning to complete the information or leaves them blank with a "requires verification" label. A field conflict detection mechanism is implemented during the filling process. If field values ​​provided by multiple data sources are inconsistent, the system will record the conflict and output a prompt to ensure the traceability of the filled content. Subsequently, the rule logic control module embeds engineering constraints according to power supply design specifications, including distribution access principles, power supply radius limits, wire diameter selection, and transformer capacity verification. This module combines logical judgment with text generation through a rule engine. When a violation of the power supply radius limit is detected, the system adjusts the power supply access point description or adds explanatory statements to maintain the scheme's compliance. In the content generation stage, the document pipeline layout algorithm optimizes the text structure hierarchically. This algorithm adaptively adjusts the paragraph order and heading level based on chapter importance and field relevance to ensure the generated document is logically clear and semantically coherent. The layout results are optimized in style by the natural language processing module, ensuring the solution maintains professionalism while maintaining high readability. Finally, the model outputs a preliminary draft of the solution and marks potentially abnormal or uncertain fields as suspicious items. The marked information and corresponding confidence values ​​are presented together in the expert review interface for manual confirmation or to trigger model retraining and optimization, thus achieving a closed loop between power supply solution generation and human-machine collaborative correction.

[0056] Step 105: Conduct multi-dimensional indicator evaluation and multi-level review of the preliminary power supply plan, and adjust the preliminary power supply plan based on the evaluation and review results to generate and output a target power supply plan with risk identification and credibility rating.

[0057] Among them, the risk label reflects potential sources of uncertainty (such as terrain complexity and load forecast fluctuations), and the credibility rating is calculated based on the model confidence level and the consistency of expert verification, which is used as a quantitative reference for subsequent manual review and project approval stages.

[0058] The power supply scheme generation method provided in this application utilizes a multimodal large model to perform semantic understanding and multimodal fusion of user needs and on-site survey data. This extracts deep-level fusion semantic vectors from heterogeneous data, accurately expressing the user's actual needs and specific on-site conditions. Based on the fusion semantic vectors, a corresponding target power supply scheme template is matched in the power supply scheme template knowledge base, ensuring that the target power supply scheme template meets both enterprise power supply specifications and user electricity needs. After obtaining the target power supply scheme template, an expert model, combined with power industry expertise and rules, automatically and intelligently fills the template and embeds corresponding rules. Multi-dimensional evaluation and review are then conducted to generate a final target power supply scheme with risk identification and credibility rating.

[0059] This application's embodiments, on the one hand, overcome the limitations of traditional solutions that rely solely on static templates and fixed logic. By combining dynamic management of the template knowledge base with a semantic-driven generation mechanism, reliable power supply solutions can be dynamically generated based on multi-dimensional needs, significantly reducing manual intervention and improving the efficiency and relevance of solution generation. On the other hand, it employs a collaborative mechanism between a multimodal large model and an expert model. The large model is responsible for semantic matching and initial template selection, while the expert model is responsible for professional rule embedding, field filling, and solution review. This leverages the powerful contextual understanding and information integration capabilities of the large model while ensuring the accuracy and reliability of the expert model in professional calculations and rule verification, greatly improving the intelligence, efficiency, and accuracy of power supply solution generation.

[0060] In one embodiment, step 105, which involves evaluating the preliminary power supply plan using multi-dimensional indicators and conducting multi-level reviews, and adjusting the preliminary power supply plan based on the evaluation and review results, specifically includes the following steps: Step 105-1: Conduct a quality assessment of the preliminary power supply scheme based on a multi-dimensional indicator system and a multi-layered review mechanism.

[0061] The multidimensional indicator system includes three main dimensions: (1) Technical rationality indicators: By verifying key technical parameters such as line length, voltage level, conductor cross-section, and transformer capacity, it is determined whether the scheme design meets the requirements of electrical safety and operational stability. (2) Economic indicators: taking into account factors such as material costs, construction period and investment recovery period, evaluate the degree of balance between economic input and benefits of the plan; (3) Compliance indicators: The legality and standardization of the design scheme are verified by comparing the latest national standards, industry norms and local power supply policies.

[0062] Multi-layered review mechanism: The first layer is the system self-checking layer, which performs logical consistency checks by the generated model itself, including format verification, geographic consistency checks, field integrity checks, field association checks, data constraint checks, and template semantic matching degree evaluation. The second layer is the expert model rule review layer, where the power expert model conducts a rule-based review of the key engineering parameters involved in the scheme (such as power supply radius, voltage drop, line economic current density, etc.). The third layer is the regulatory standard comparison layer. The system compares the plan with the policy knowledge base and historical review samples to ensure that the plan fully complies with the current power supply design guidelines and local approval specifications.

[0063] Step 105-2: When non-standard content is detected in the preliminary power supply scheme, locate the abnormal field and regenerate or complete the abnormal field.

[0064] Understandably, during the partial regeneration of abnormal fields, the system re-invokes the large model and expert model to correct the target paragraph content, and performs a second consistency verification after generation to ensure that the corrected content meets all constraints.

[0065] Step 105-3: Adjust the tolerance range of technical parameters in the preliminary power supply scheme based on regional policies and construction conditions through a dynamic threshold adjustment mechanism.

[0066] Specifically, tolerances in technical parameters include appropriately relaxing the upper limit of the power supply radius in high-altitude areas and strictly controlling the line load rate in urban centers.

[0067] In this embodiment, a multi-dimensional indicator system and multi-layered review mechanism are constructed using an expert model. Based on multi-dimensional indicators such as technology, economy, and compliance, the preliminary power supply scheme is dynamically and meticulously evaluated and systemic risks are identified through different review layers. When non-compliant content is detected, the abnormal fields are accurately located and regenerated or supplemented, which improves the efficiency and accuracy of scheme correction and ensures the rigor of the scheme from its underlying logic to its implementation. Simultaneously, combined with a dynamic threshold adjustment mechanism, the system flexibly adapts the tolerance range of technical parameters based on regional policies and construction conditions. This allows the system to maintain the optimal scheme standard in different application scenarios, ultimately forming a power supply scheme that is technologically advanced, cost-controllable, compliant, and adaptable to local needs. This significantly improves the efficiency and stability of scheme implementation, reduces the risk of later rectification and operation and maintenance costs, and ultimately achieves high-quality, high-efficiency, and low-risk power supply scheme decisions.

[0068] In one embodiment, the power supply scheme generation method further includes: constructing a collaborative training dataset; inputting multimodal data of the power industry, power system operating parameters, and semantic annotation information from the collaborative training dataset into a multimodal large model to obtain semantic results; inputting power industry rule labels and enhanced input features from the collaborative training dataset into an expert model to obtain a structured knowledge representation containing industry rule logic and power index calculation results; calculating the difference between the semantic results and the structured knowledge representation as a first loss based on a first distillation loss function, and updating the parameters of the multimodal large model based on the first loss so that the multimodal large model learns the industry rules of the power expert model; and calculating the difference between the power index calculation results and the true values ​​of the power index as a second loss based on a second distillation loss function, and updating the parameters of the expert model based on the second loss to correct the calculation bias of the expert model.

[0069] The collaborative training dataset includes multimodal data from the power industry, power system operating parameters and corresponding industry rule labels, semantic annotation information, and real values ​​of power indicators. Enhanced input features are obtained by concatenating semantic results with power system operating parameters.

[0070] In this embodiment, a bidirectional distillation mechanism enables the co-evolution and complementary enhancement of the two models. This allows the multimodal large model to learn the decision boundaries and rule weights of the expert model through distillation, enhancing its understanding of power industry expertise and reasoning ability, and ensuring adherence to industry standards during the generation process. Conversely, the expert model receives semantic distribution information from the large model to correct its computational biases under non-standard inputs, enhancing its adaptability and computational accuracy. This not only simultaneously improves the specialization of the large language model and the intelligence of the expert model but also forms a hybrid augmented intelligence paradigm combining data-driven and knowledge-guided approaches. Ultimately, while reducing reliance on massive amounts of high-quality labeled data, it outputs reliable results with both semantic understanding depth and industry-standard computational accuracy, providing a more solid and interpretable model foundation for intelligent applications in power systems.

[0071] In one embodiment, after step 105, the power supply scheme generation method further includes: displaying the power supply parameters, engineering layout, power supply path and load curve of the target power supply scheme in chapters through a visualization engine; receiving user feedback results, and updating the learning samples of the multimodal large model and expert model based on the user feedback results to optimize the multimodal large model and expert model.

[0072] In this embodiment, a visual interface for the solution is provided on the user console, supporting chapter expansion, floating parameter explanations, and synchronized chart display. Furthermore, the visualization engine can generate engineering layouts, power supply paths, and load curves, which are then displayed in chapters. Users can upload feedback on the displayed content to the data loop in real time to update the model's learning samples, thereby forming a closed-loop learning mechanism of "generation-evaluation-correction-regeneration" for continuous optimization.

[0073] For example, the system provides a visual interface for the power supply design through a front-end user console. This interface adopts a modular layout design, mainly including a section display area, a parameter floating explanation area, and a dynamic chart display area. Users can expand the structure of the design step-by-step in the section display area, browsing the chapter content of the power supply design. When the mouse hovers over key parameters (such as conductor type, power supply radius, transformer capacity), the system pops up a parameter explanation window, displaying the calculation source, rule constraints, and corresponding confidence values. Simultaneously, the system provides a multi-version design comparison function. The version management module records the historical versions, generation time, and corresponding model version number of each round of design generation or modification. Users can compare the differences between two or more versions of the design on the interface. The system uses a difference highlighting algorithm to mark changed fields, facilitating review and tracking. The system embeds a visualization engine to generate spatial and topological visualizations of the power supply project. This engine draws engineering layout diagrams, power line route diagrams, and dynamic load curve diagrams based on GIS coordinate information and topological relationships. The route diagram uses dynamic rendering and supports zooming and rotation; the load curve can be refreshed in real time according to changes in the time axis, providing users with intuitive load distribution and capacity analysis.

[0074] During the feedback phase, the system incorporates a user interaction loop. Users can provide feedback on the technical rationale, layout standards, or risk warnings of the proposed solution. All feedback results are uploaded to the backend data loop in real time, processed through feature extraction and labeling, and then written into the model training sample set. The model training module performs weight fine-tuning and parameter correction based on the feedback data, achieving a closed loop from solution generation to feedback learning. This forms an intelligent learning mechanism of "generation-evaluation-correction-regeneration." Through this closed-loop mechanism, the solution generation model can be continuously optimized over long-term use, not only improving the accuracy and readability of solution generation but also gradually adapting to regional differences, business updates, and user preferences, achieving the self-evolutionary goal of intelligent power supply solution generation.

[0075] It should be noted that for multimodal large models and expert models, the latest power supply project documents can be used periodically as training samples for updates. Specifically, the system establishes a training sample collection module, which regularly collects the latest power supply project documents from the enterprise power supply file management system, project acceptance database, and user application data. The collected raw documents undergo semantic cleaning, format standardization, and metadata annotation before being input into the sample management module. The sample management module uses a semantic difference analysis algorithm to compare the semantic vector distance between new samples and existing training corpora, extracting change features, including policy wording changes, parameter constraint range updates, and changes in economic evaluation indicators, to determine the latest semantic increments that the model needs to learn. The training module uses reinforcement learning algorithms to dynamically adjust the parameters of the power supply scheme generation model. By constructing a policy network based on a reward function, the system calculates reward values ​​in real time based on the compliance score, semantic consistency, and user feedback score of the generated results, adjusting the model's generation strategy positively or negatively, thereby making the generated text more aligned with industry semantics, professional terminology, and policy guidance. Meanwhile, during training, for samples deemed unqualified by manual or expert models, the system traces their generation path, labels the corresponding parameter weight contribution values, and optimizes the reward function of the policy network through backpropagation, enabling the model to avoid similar errors in similar tasks. Furthermore, the system continuously integrates external industry information, including policy documents, design specifications, and dynamic databases of material prices. During training, the model vectorizes and embeds this external data into the input feature space to correct the economic and compliance weights during solution generation, ensuring that the model's output solutions balance cost control and standard adaptation. After training, the system uses an evaluation module to screen the output of the updated model based on confidence levels, retaining only high-quality samples whose technical rationality, semantic consistency, and compliance scores all reach the thresholds for inclusion in the incremental training set, ensuring the stability and controllability of the model's evolution process.

[0076] In one embodiment, the power supply scheme generation method further includes: encrypting and transmitting data streams and implementing hierarchical access control during the generation of the target power supply scheme; generating a unique security token for the multimodal large model or expert model in response to the calling instructions of the multimodal large model or expert model; detecting abnormal calls and unauthorized access behaviors in the generation process of the target power supply scheme in real time through a behavior monitoring module; and generating security audit logs based on the abnormal calls and unauthorized access behaviors.

[0077] In this embodiment, encrypted transmission and hierarchical access control of the data stream effectively prevent data leakage, tampering, and unauthorized use during transmission and access, ensuring the confidentiality and integrity of core power supply scheme data (such as technical parameters, cost budgets, and compliance indicators). A unique security token is generated for the multimodal large model or expert model, enabling precise traceability of the entity calling each model and avoiding risks to scheme generation caused by unauthorized model access or unauthorized calls. Real-time detection of abnormal calls and unauthorized access behaviors through the behavior monitoring module allows for timely identification and warning of security threats such as unauthorized operations and abnormal instruction injection, blocking security risks at the source. Security audit logs are generated based on abnormal behavior, providing complete evidence for post-event security incident tracing and responsibility determination, and providing data support for subsequent security strategy optimization. Thus, a multi-layered information security protection mechanism ensures data security, access controllability, and operational traceability during the generation and transmission of the power supply scheme, preventing issues such as sensitive information leakage, unauthorized access, and model abuse.

[0078] The power supply scheme generation method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0079] It should be noted that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0080] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant 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.

[0081] Furthermore, such as Figure 2As shown, as a specific implementation of the above power supply scheme generation method, this application embodiment provides a power supply scheme generation device 200, which includes: a knowledge base construction module 201, a feature fusion module 202, a template matching module 203, and a scheme generation module 204.

[0082] Among them, the knowledge base construction module 201 is used to build a power supply scheme template knowledge base. The power supply scheme template knowledge base includes multiple power supply scheme templates with key fields marked with logical dependencies, as well as the relationship between power supply scheme templates, scenario conditions and user requirements. The feature fusion module 202 is used to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data using a multimodal large model, and generate a fused semantic vector; Template matching module 203 is used to match the target power supply scheme template from the power supply scheme template knowledge base based on the fused semantic vector; The scheme generation module 204 is used to dynamically fill and embed rules into the key fields of the target power supply scheme template based on the fused semantic vector and the expert model to generate a preliminary power supply scheme; and to evaluate the preliminary power supply scheme with multi-dimensional indicators and multi-level review, and adjust the preliminary power supply scheme based on the evaluation and review results, and generate and output a target power supply scheme with risk identification and credibility rating.

[0083] Furthermore, the knowledge base construction module 201 is specifically used to create standard templates and collect historical case documents based on different power supply project types; to identify key fields in the standard templates using a natural language segmentation algorithm, including project scale, load type, transformer capacity, and construction period; to perform slot structure annotation on the standard templates and establish logical dependencies between key fields; to perform multi-regional standardization processing on the annotated standard templates based on historical case documents to form power supply scheme templates adapted to the standards of different power supply companies; to parse historical case documents and extract the scenario conditions and user requirements of historical power supply schemes; to associate power supply scheme templates that conform to historical power supply schemes with scenario conditions and user requirements, and to store power supply scheme templates hierarchically according to business scenarios to form a power supply scheme template knowledge base.

[0084] Furthermore, the power supply scheme generation device 200 also includes: The data acquisition module (not shown in the figure) is used to construct a collaborative training dataset, which includes multimodal data of the power industry, power system operating parameters and corresponding industry rule labels, semantic annotation information and real values ​​of power indicators. The collaborative optimization module (not shown in the figure) is used to input multimodal data of the power industry, power system operating parameters, and semantic annotation information from the collaborative training dataset into the multimodal large model to obtain semantic results; and to input power industry rule labels and enhanced input features from the collaborative training dataset into the expert model to obtain structured knowledge representation containing industry rule logic and power index calculation results, wherein the enhanced input features are obtained by concatenating the semantic results and power system operating parameters; and, based on the first distillation loss function, to calculate the difference between the semantic results and the structured knowledge representation as the first loss, and to update the parameters of the multimodal large model based on the first loss so that the multimodal large model learns the industry rules of the power expert model; and, based on the second distillation loss function, to calculate the difference between the power index calculation results and the actual power index values ​​as the second loss, and to update the parameters of the expert model based on the second loss so as to correct the calculation bias of the expert model.

[0085] Furthermore, the template matching module 203 is specifically used to encode the user's electricity demand and the on-site survey data to obtain text feature vectors and visual feature vectors respectively; a multimodal fusion algorithm based on attention mechanism is used to weight and fuse the text feature vectors and visual feature vectors to generate multimodal features; a fused semantic vector is constructed based on the multimodal features, wherein the fused semantic vector includes electricity demand patterns, construction condition constraints and reliability levels.

[0086] Furthermore, the scheme generation module 204 is specifically used to extract the text information of the first key field in the corresponding target power supply scheme template from the user intent vector for preliminary filling, and to mark suspicious fields with confidence scores below the confidence threshold; based on logical dependencies and the filled fields, to calculate the text information of the remaining second key field for secondary filling; through the rule engine, to embed at least one engineering constraint condition from the power distribution access principle, power supply radius limit, and wire diameter selection constraint during the filling process; and to use a document pipeline layout algorithm to perform hierarchical optimization and layout adjustment on the filled text information to form a preliminary power supply scheme.

[0087] Furthermore, the scheme generation module 204 is specifically used to conduct quality assessments of the preliminary power supply scheme based on a multi-dimensional indicator system and a multi-layer review mechanism. The multi-layer review mechanism includes a self-inspection layer, a rule review layer, and a regulatory standard comparison layer. The multi-dimensional indicator system includes technical rationality, economy, and compliance. When non-compliant content is detected in the preliminary power supply scheme, abnormal fields are located, and abnormal fields are regenerated or supplemented. Through a dynamic threshold adjustment mechanism, the tolerance range of technical parameters in the preliminary power supply scheme is adjusted based on regional policies and construction conditions.

[0088] Furthermore, the power supply scheme generation device 200 also includes: The visualization module (not shown in the figure) is used to display the power supply parameters, engineering layout, power supply path and load curve of the target power supply scheme in chapters through the visualization engine; The model optimization module (not shown in the figure) is used to receive user feedback results and update the learning samples of the multimodal large model and expert model based on the user feedback results to optimize the multimodal large model and expert model.

[0089] Furthermore, the power supply scheme generation device 200 also includes: The security protection module (not shown in the figure) is used to encrypt and control the data stream during the generation of the target power supply scheme; generate a unique security token for the multimodal large model or expert model in response to the calling instructions of the multimodal large model or expert model; detect abnormal calling and unauthorized access behavior in the generation process of the target power supply scheme in real time through the behavior monitoring module; and generate security audit logs based on abnormal calling and unauthorized access behavior.

[0090] Specific limitations regarding the power supply scheme generation device can be found in the limitations of the power supply scheme generation method described above, and will not be repeated here. Each module in the aforementioned power supply scheme generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0091] Based on the above, Figure 1 Accordingly, embodiments of this application also provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 The method for generating the power supply scheme is shown.

[0092] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), and includes several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the methods described in the various implementation scenarios of this application.

[0093] Based on the above, Figure 1 The method shown, and Figure 2 The virtual device embodiment shown is designed to achieve the above objectives, such as... Figure 3 As shown in the figure, this application embodiment also provides a computer device 300, which includes a processor 301 and a memory 302. The memory 302 stores a program or instructions that can run on the processor 301. When the program or instructions are executed by the processor 301, they implement the above-mentioned... Figure 1 The method for generating the power supply scheme is shown.

[0094] The memory 302 can be used to store software programs and various data. The memory 302 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 302 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 302 in the embodiments of this application includes, but is not limited to, these and any other suitable types of memory.

[0095] Processor 301 may include one or more processing units; optionally, processor 301 integrates an application processor and a modem processor, wherein the application processor mainly handles operations involving the operating system, user interface, and applications, and the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into processor 301.

[0096] Computer equipment can specifically include personal computers, servers, network devices, etc.

[0097] Optionally, the computer device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB ports, card reader ports, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Bluetooth interfaces, Wi-Fi interfaces), etc.

[0098] Those skilled in the art will understand that the computer device structure provided in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0100] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0101] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for generating a power supply scheme, characterized in that, Applied to a power supply system, the method includes: Construct a power supply solution template knowledge base, which includes multiple power supply solution templates with key fields marked with logical dependencies, as well as the relationship between power supply solution templates, scenario conditions, and user requirements; A multimodal large model is used to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data to generate a fused semantic vector; The target power supply scheme template is matched from the power supply scheme template knowledge base based on the fused semantic vector; Based on the fused semantic vector, an expert model is used to dynamically fill and embed rules into the key fields of the target power supply scheme template to generate a preliminary power supply scheme. The preliminary power supply scheme is evaluated using multi-dimensional indicators and reviewed at multiple levels. Based on the evaluation and review results, the preliminary power supply scheme is adjusted to generate and output a target power supply scheme with risk identification and credibility rating.

2. The power supply scheme generation method according to claim 1, characterized in that, The knowledge base for constructing power supply scheme templates includes: Based on different power supply project types, standard templates were created and historical case documents were collected. The key fields within the standard template are identified using a natural language segmentation algorithm, wherein the key fields include project scale, load type, transformer capacity, and construction period; The standard template is annotated with slot structure, and the logical dependencies between the key fields are established. Based on the historical case documents, the standard template after annotation is standardized for multiple regions to form a power supply scheme template that is compatible with the standards of different power supply companies. Analyze the historical case documents to extract the scenario conditions and user requirements of the historical power supply schemes; The power supply scheme templates that conform to the historical power supply schemes are associated with the scenario conditions and user requirements, and the power supply scheme templates are stored hierarchically according to business scenarios to form the power supply scheme template knowledge base.

3. The power supply scheme generation method according to claim 1, characterized in that, The method further includes: Construct a collaborative training dataset, wherein the collaborative training dataset includes multimodal data of the power industry, power system operating parameters and corresponding industry rule labels, semantic annotation information and real values ​​of power indicators; The multimodal data of the power industry, power system operating parameters, and semantic annotation information in the collaborative training dataset are input into the multimodal large model to obtain semantic results; The power industry rule labels and enhanced input features in the collaborative training dataset are respectively input into the expert model to obtain a structured knowledge representation containing industry rule logic and power index calculation results. The enhanced input features are obtained by concatenating the semantic results with the power system operating parameters. Based on the first distillation loss function, the difference between the semantic result and the structured knowledge representation is calculated as the first loss, and the parameters of the multimodal large model are updated based on the first loss so that the multimodal large model learns the industry rules of the power expert model. Based on the second distillation loss function, the difference between the calculated result of the power index and the actual value of the power index is calculated as the second loss, and the parameters of the expert model are updated based on the second loss to correct the calculation deviation of the expert model.

4. The power supply scheme generation method according to claim 1, characterized in that, The method employs a multimodal large model to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data, generating a fused semantic vector, including: The user's electricity demand and the on-site survey data are respectively encoded to obtain text feature vectors and visual feature vectors; A multimodal fusion algorithm based on an attention mechanism is used to weight and fuse the text feature vector and the visual feature vector to generate multimodal features; The fused semantic vector is constructed based on the multimodal features, wherein the fused semantic vector includes power demand patterns, construction condition constraints, and reliability levels.

5. The power supply scheme generation method according to claim 1, characterized in that, Based on the fused semantic vector, an expert model is used to dynamically fill in and embed rules into the key fields of the target power supply scheme template to generate a preliminary power supply scheme, including: The text information of the first key field in the corresponding target power supply scheme template is extracted from the user intent vector for initial filling, and suspicious fields with confidence scores below the confidence threshold are marked. Based on the logical dependencies and the already filled fields, the text information of the remaining second key fields is calculated and filled a second time. Through the rule engine, at least one engineering constraint from the power distribution access principle, power supply radius limit, and wire diameter selection constraint is embedded during the filling process; The document pipeline layout algorithm is used to optimize the hierarchy and adjust the layout of the filled text information to form the preliminary power supply scheme.

6. The power supply scheme generation method according to claim 1, characterized in that, The process of evaluating the preliminary power supply plan using multi-dimensional indicators and conducting multi-level reviews, and adjusting the preliminary power supply plan based on the evaluation and review results, includes: The preliminary power supply scheme is evaluated based on a multi-dimensional indicator system and a multi-layered review mechanism. The multi-layered review mechanism includes a self-inspection layer, a rule review layer, and a regulatory standard comparison layer. The multi-dimensional indicator system includes technical rationality, economy, and compliance. When non-compliant content is detected in the preliminary power supply scheme, the abnormal field is located, and the abnormal field is regenerated or completed. The tolerance range of the technical parameters in the preliminary power supply scheme is adjusted based on regional policies and construction conditions through a dynamic threshold adjustment mechanism.

7. The power supply scheme generation method according to claim 1, characterized in that, The method further includes: The power supply parameters, engineering layout, power supply path and load curve of the target power supply scheme are displayed in chapters using a visualization engine. Receive user feedback results, and update the learning samples of the multimodal large model and the expert model based on the user feedback results to optimize the multimodal large model and the expert model.

8. The power supply scheme generation method according to claim 1, characterized in that, The method further includes: During the generation of the target power supply scheme, the data stream is encrypted and subject to hierarchical access control. In response to the invocation instruction of the multimodal large model or the expert model, a unique security token is generated for the multimodal large model or the expert model; The behavior monitoring module detects abnormal calls and unauthorized access behaviors in the generation process of the target power supply scheme in real time. Based on the abnormal calls and the unauthorized access behaviors, a security audit log is generated.

9. A power supply scheme generation device, characterized in that, The device includes: The knowledge base construction module is used to build a power supply scheme template knowledge base, wherein the power supply scheme template knowledge base includes multiple power supply scheme templates marked with key fields with logical dependencies, as well as the relationship between power supply scheme templates, scenario conditions and user requirements; The feature fusion module is used to perform multimodal fusion and semantic understanding on the collected user electricity demand and on-site survey data using a multimodal large model, and generate a fused semantic vector; The template matching module is used to match a target power supply scheme template from the power supply scheme template knowledge base based on the fused semantic vector; The scheme generation module is used to dynamically fill in and embed rules into the key fields of the target power supply scheme template based on the fused semantic vector using an expert model, thereby generating a preliminary power supply scheme; and... The preliminary power supply scheme is evaluated using multi-dimensional indicators and reviewed at multiple levels. Based on the evaluation and review results, the preliminary power supply scheme is adjusted to generate and output a target power supply scheme with risk identification and credibility rating.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the power supply scheme generation method as described in any one of claims 1 to 8.

Images