Method and system for generating visual instruction chain based on fine-tuning large language model

Abstract

The invention discloses a method and a system for generating a visual instruction chain based on a fine-tuning large language model. The method comprises the following steps: receiving a task target described in a natural language; analyzing the natural language description on the basis of understanding the power data characteristics and the power business context related to the task target; according to an analysis result, generating an instruction chain for visual presentation through a fine-tuned large language model; according to the method, it can be ensured that conversion from the user intention to the visualization result is accurate and efficient, and a visual and friendly display effect is provided for data analysis in the power field.

Description

Technical Field

[0001] This invention relates to the field of technology, and in particular to a method and system for generating visual instruction chains based on fine-tuning of a large language model. Background Technology

[0002] With the continuous expansion of the power system and the acceleration of digital transformation, power companies have accumulated a large amount of heterogeneous and multi-source business data, including equipment operating parameters, inspection records, geospatial information, and drone inspection images. In daily operation and maintenance, fault analysis, and planning decision-making, there is an urgent need to visualize this complex data in an intuitive and effective way to help frontline personnel and managers promptly grasp equipment status, potential hazard distribution, and business dynamics, thereby improving work efficiency and the scientific nature of decision-making. Traditional data visualization processes are highly dependent on manual labor, often requiring professional data analysts to manually design charts, write scripts, and gradually debug them according to needs. This is inefficient and unable to flexibly respond to changing business scenarios on-site. Especially in scenarios closely related to geospatial conditions, such as power line inspection and substation site selection, it is necessary to intelligently process and analyze massive amounts of spatial data, and to make comprehensive judgments by combining inspection images, real-time monitoring, and other multimodal data. Existing automated visualization solutions are mostly limited to templates, static rules, or limited syntax parsing, making it difficult to meet the needs of complex business objectives, personalized interaction, and deep integration of multi-source heterogeneous data.

[0003] In recent years, with the rapid development of large language models and deep learning technologies, natural language processing and artificial intelligence reasoning capabilities have been significantly improved. By fine-tuning large language models for specific domains, the models can not only understand abstract business intentions but also transform complex user requests expressed in natural language into specific data processing and visualization operation instruction chains, effectively improving automation and intelligence levels. Against this backdrop, some technical solutions have attempted to utilize large language models to achieve an automatic closed loop of natural language-data analysis-visualization, initially realizing interactive analysis with users. However, current mainstream solutions generally suffer from the following shortcomings when implemented in the actual power industry: First, insufficient processing capabilities for industry business context and spatial data, failing to automatically visualize complex geographic information such as inspection risk point location and site selection analysis; second, limited support for human-computer interaction and business iteration, making it difficult to dynamically adjust chart presentation and analysis processes based on user feedback; and third, weak integration capabilities for multimodal business scenarios such as drone image recognition and path planning. How to properly solve these problems has become an urgent issue for the industry. Summary of the Invention

[0004] This invention provides a method and system for generating visualized instruction chains based on a fine-tuned large language model. The technical solution of this invention achieves intelligent parsing of complex power business intentions and automatic generation of visualized instruction chains by fine-tuning a large language model, significantly improving the visualization efficiency and accuracy of multi-source heterogeneous data. The system supports deep integration with multiple scenarios such as natural language interaction, geographic information processing, image recognition, and path planning, and can dynamically optimize the display effect, effectively meeting the needs of intelligent operation and maintenance and auxiliary decision-making in the power industry.

[0005] According to one aspect of the present invention, a method for generating a visual instruction chain based on a fine-tuned large language model is provided, comprising:

[0006] Receive task objectives described in natural language;

[0007] Based on an understanding of the power data characteristics and power business context involved in the task objectives, the natural language description is analyzed.

[0008] Based on the analysis results, an instruction chain for visualization is generated using the fine-tuned large language model.

[0009] Execute the instruction chain for visualization to generate a visualization result of the business data.

[0010] According to another aspect of the present invention, a system for generating visual instruction chains based on fine-tuning of a large language model is provided, comprising:

[0011] The receiving module is used to receive task objectives described in natural language;

[0012] The analysis module is used to analyze the natural language description based on an understanding of the power data characteristics and power business context involved in the task objective;

[0013] The generation module is used to generate an instruction chain for visualization based on the analysis results and the fine-tuned large language model.

[0014] The visualization module is used to execute the instruction chain for visualization presentation in order to generate the visualization result of the business data.

[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0016] At least one processor;

[0017] and memory that is communicatively connected to at least one processor;

[0018] The memory stores a computer program that can be executed by at least one processor, which enables the at least one processor to execute the method for generating a visual instruction chain based on a fine-tuned large language model according to any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute a method for generating a visual instruction chain based on a fine-tuning large language model according to any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method for generating a visual instruction chain based on a fine-tuned large language model according to any embodiment of the present invention.

[0021] The technical solution of this invention, through intelligent visualization that integrates large language model reasoning, natural language interaction, multi-source data processing, and complex business intent parsing, meets the needs of digital operation and maintenance and decision support in the power industry. It achieves intelligent visualization of power business data by using task objectives described in natural language. The system receives a natural language task description from the user; based on a thorough understanding of the power data characteristics and business context involved in the task, it performs semantic parsing on the user description to extract key requirement elements; then, it generates an instruction chain for visualization presentation through a pre-tuned industry large language model; finally, it executes the instruction chain to process the relevant business data and output interactive visualization results. This ensures accurate and efficient conversion from user intent to visualization results, providing an intuitive and user-friendly display effect for data analysis in the power field. It solves technical problems such as insufficient processing capabilities for industry business context and spatial data, inability to automatically visualize complex geographic information such as inspection risk point location and site selection analysis; limited support for human-computer interaction and business iteration, making it difficult to dynamically adjust chart presentation and analysis processes based on user feedback; and weak integration capabilities for multimodal business scenarios such as UAV image recognition and path planning, significantly improving operation and maintenance efficiency.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 1 of the present invention;

[0025] Figure 2 This is a flowchart of a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 2 of the present invention.

[0026] Figure 3 A flowchart of a method for generating a visual instruction chain based on a large language model with fine-tuning, provided in Embodiment 3 of the present invention;

[0027] Figure 4 A flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 4 of the present invention;

[0028] Figure 5 A flowchart of a preferred method for generating a visual instruction chain based on a fine-tuned large language model, provided in Embodiment 5 of the present invention;

[0029] Figure 6 A flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment Six of the present invention;

[0030] Figure 7 This is a schematic diagram of the structure of a system for generating a visual instruction chain based on a large language model with fine-tuning, provided in Embodiment 7 of the present invention.

[0031] Figure 8 A schematic diagram of the structure of an electronic device for implementing the method of generating a visual instruction chain based on a large language model with fine-tuning according to an embodiment of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0033] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0034] Figure 1 This is a flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 1 of the present invention. Figure 1 As shown, the method includes steps S11-S14:

[0035] In step S11, the task objective described in natural language is received;

[0036] In step S12, the natural language description is analyzed based on an understanding of the power data characteristics and power business context involved in the task objective.

[0037] In step S13, based on the analysis results, an instruction chain for visualization is generated using the fine-tuned large language model;

[0038] In step S14, the instruction chain for visualization is executed to generate a visualization result of the business data.

[0039] In one embodiment, intelligent visualization can integrate large language model reasoning, natural language interaction, multi-source data processing, and complex business intent parsing to meet the needs of digital operation and maintenance and decision support in the power industry. It uses natural language to describe task objectives, enabling intelligent visualization of power business data. The system receives a natural language task description from the user; based on a thorough understanding of the power data characteristics and business context involved in the task, it performs semantic parsing on the user description to extract key requirement elements; then, it generates an instruction chain for visualization presentation through a pre-tuned industry-specific large language model; finally, it executes the instruction chain to process the relevant business data and output interactive visualization results. This ensures accurate and efficient conversion from user intent to visualization results, providing an intuitive and user-friendly display for data analysis in the power sector.

[0040] Upon receiving the natural language task description, the process enters the task parsing phase. Based on power industry expertise and a terminology database, key entities in the description (such as equipment names, measurement indicators, time ranges, and geographical regions) and the user's intent are identified. For example, for the description "show the monthly electricity load trend of a certain region over the past year," the parsing module extracts elements such as "past year," "certain region," "monthly," and "electricity load," and combines this with the power business context to understand the user's true needs. A context-preserving mechanism ensures consistency in parsing throughout the process, retaining relevant context from the dialogue and maintaining the continuity of the task objective across multiple rounds of questioning. Simultaneously, considering the common seasonal peak-valley variations in power data, the parsing process incorporates this domain knowledge into the understanding, providing a more refined grasp of the user's intent. By combining semantic parsing with context preservation, this method accurately captures the inherent meaning of user needs, laying the foundation for subsequent instruction chain generation.

[0041] Based on the task parsing results, a corresponding visual instruction chain is generated through a fine-tuned large language model. This fine-tuned large language model, jointly trained with power industry data and visualization knowledge, can accurately convert complex requests into executable instruction sequences. The generated instruction chain typically covers the entire process from data acquisition and preprocessing analysis to chart creation. For example, for the aforementioned electricity load trend analysis requirement, the model might generate the following steps: the data extraction step refers to acquiring raw load data for a specified region and time period; the spatiotemporal aggregation step refers to summarizing and calculating the data by month and region to obtain the regional load summary value for each month; the chart selection step refers to selecting a line chart as the primary display format to reflect the load trend over time; and the page presentation step refers to setting the chart mapping rules (mapping time to the horizontal axis and load values ​​to the vertical axis, and adjusting the units and scales to be uniform).

[0042] After generating the instruction chain, an instruction chain execution optimization mechanism is used to improve execution efficiency and reliability. On one hand, this mechanism performs static analysis on the instruction chain, optimizes the execution order, merges redundant steps, and pre-loads necessary data cache to reduce latency. On the other hand, it monitors the output of each step during execution; if data loss or anomalies are detected, subsequent instructions are dynamically adjusted to ensure the entire instruction chain is executed robustly. Through intelligent instruction generation and execution optimization using a large language model, this method can flexibly adapt to different task requirements and quickly produce high-quality, visualized instruction sequences.

[0043] After the instruction chain is executed, a visualization of power business data is generated. The effectiveness and visual appeal of the charts are improved in several ways during the visualization phase. Regarding page presentation rules, a unified page presentation standard is followed, ensuring that similar information uses consistent graphic elements and color mapping (e.g., different voltage levels are distinguished by fixed colors, and time series data is uniformly represented by line charts), avoiding user misunderstandings due to inconsistent representation. In terms of aesthetic adaptation, an aesthetic adaptation strategy is introduced. The system automatically adjusts the visual style based on the data volume and chart complexity, including elements such as theme color, grid line density, and label font size, ensuring that the charts maintain good aesthetics and readability under different data densities. Regarding layer overlay, multi-layer information overlay is supported, integrating multiple data dimensions in a single view. For example, the power grid topology and load distribution thermal layers can be presented simultaneously on a geographic map, or load and voltage curves can be overlaid on the same timeline chart, achieving a multi-dimensional data presentation in a single chart. Regarding interactive logic, rich interactive logic is incorporated, giving the generated charts dynamic responsiveness. Users can interact with the charts, such as zooming in and out of the timeline to view details, and clicking on equipment nodes to obtain inspection details. When user interaction or new data inflow occurs, the system uses a dynamic chart reconstruction strategy to partially update the chart content to reflect the latest data and context, ensuring that the visualization content is always timely and context-consistent. Through these detailed enhancements, the final visualization not only accurately conveys business information but also provides a good user experience and a unified aesthetic style.

[0044] To address the characteristics of power industry data, the system enhances the parsing and adaptation capabilities of the large language model and instruction chain for industry scenarios. Considering the long-term, spatially distributed nature of power data, the system strengthens its spatiotemporal aggregation and analysis capabilities. It can aggregate data at any desired time granularity (hour, day, month, quarter, etc.) and geographical scope. For example, it can automatically aggregate the time-of-use load of each substation and generate a spatiotemporal distribution map of the entire network load, providing comprehensive data support for visualization. In power equipment inspection and maintenance scenarios, it can transform a large number of unstructured inspection records into structured information. Through natural language processing of inspection reports and intelligent recognition of inspection images, key fields such as equipment number, inspection date, defect type, and severity are extracted and stored in a unified inspection database. Therefore, when it is necessary to display equipment health status or defect distribution, the model-generated instruction chain can directly utilize this structured data to draw statistical charts of different defect categories or equipment failure timelines. To meet the needs of power grid planning and site selection, a multi-factor evaluation model for substation site selection was pre-built and learned, covering factors such as load density, grid connection conditions, geographical environmental risks, and land costs. When a user submits a site selection evaluation request, the system automatically identifies relevant factors and calls the model to comprehensively score candidate locations. Subsequently, the large language model incorporates the evaluation results into the instruction chain, generating visual charts (such as bar charts or map heatmaps) to intuitively present the advantages and disadvantages of each candidate scheme. Even in complex power scenarios, it can generate accurate and effective visualization results, further enhancing the system's practicality and decision-making value in power operations.

[0045] The technical solution in this embodiment achieves intelligent parsing of complex power business intentions and automatic generation of visual instruction chains by fine-tuning a large language model, significantly improving the visualization efficiency and accuracy of multi-source heterogeneous data. The system supports deep integration of multiple scenarios such as natural language interaction, geographic information processing, image recognition, and path planning, and can dynamically optimize the display effect, effectively meeting the needs of intelligent operation and maintenance and auxiliary decision-making in the power industry.

[0046] Figure 2 This is a flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 2 of the present invention. Figure 2 As shown, the method includes steps S21-S23:

[0047] In step S21, semantic parsing is performed on the natural language description to extract initial visualization requirement information;

[0048] In step S22, at least one clarifying question is generated through the fine-tuned large language model and interacted with the user in natural language to eliminate ambiguity or supplement missing information.

[0049] In step S23, the user's answer to the clarifying question is received, and the description of the task objective and the corresponding visual instruction chain are updated based on the answer.

[0050] In one embodiment, in the method of generating visualization instruction chains by fine-tuning a large language model, multi-level and progressive information acquisition and intelligent interaction are implemented for the task objectives input by the user in natural language. This can greatly improve the system's understanding of business needs and the accuracy of visualization results. Specifically, semantic parsing is performed on the natural language description to automatically extract initial visualization requirement information, including but not limited to the analysis object, data range, display dimension, and expected output type. Through semantic parsing, the system can automatically identify and structure the core elements and business intentions contained in the original task description, and accurately map the professional terms and scenario expressions in the description by combining the domain knowledge unique to power data. For example, in the power inspection scenario, the description "display the defect trend of key areas in the past three months" can be automatically parsed to extract key information such as "key areas," "past three months," and "defect trend," and then matched to relevant data tables, time fields, and business indicators.

[0051] After initially extracting the requirements, an intelligent interaction mechanism based on a fine-tuned large language model is introduced to provide conversational feedback to users by automatically generating clarifying questions. Based on its understanding of the business context, the system proactively judges whether the initially parsed requirements are ambiguous, incomplete, or vaguely expressed, and automatically generates one or more clarifying questions accordingly. It interacts with users in natural language, prompting them to supplement, confirm, or correct their true intentions. For example, when the system parses "abnormal equipment distribution" but cannot determine the required equipment type, it can proactively ask, "Do you only need to show the abnormal distribution of transformers, or should all equipment be included?"; or when the time range is unclear, it automatically asks, "Do you need to collect data from the past year, or focus on the most recent three months?" This mechanism effectively eliminates the static and limited understanding of requirements in traditional automated systems, significantly improving the responsiveness to complex, ambiguous, and dynamic business requirements.

[0052] Upon receiving user responses to clarifying questions, the system dynamically updates the task objective description and corresponding visualization instruction chain based on these responses, thereby achieving precise fitting and intelligent iterative optimization of business requirements. This is reflected not only in the dynamic adjustment of the semantic level of the task description but also in the adaptive correction of the instruction chain's structure and content. On one hand, a "context preservation" mechanism is employed during instruction chain updates to remember key information from multiple rounds of interaction, ensuring the task objective remains consistent and effectively utilizing historical input. On the other hand, combined with an "instruction chain execution optimization mechanism," the instruction chain can be dynamically reconstructed based on new user input, supporting the skipping, merging, or refinement of certain steps to maximize the satisfaction of personalized user needs. Furthermore, the system can adaptively select page presentation methods, aesthetic parameters, or layer overlay strategies based on the response content. For example, it can automatically adjust the legend color scheme or label display based on the user-specified device type, thus optimizing the final visualization result in terms of expressiveness and interactivity. Through this progressive information acquisition and intelligent adaptive optimization, the practicality, flexibility, and advanced intelligence level of intelligent visualization in the power industry are enhanced.

[0053] Figure 3 This is a flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 3 of the present invention. Figure 3 As shown, the method includes steps S31-S33:

[0054] In step S31, geospatial business data corresponding to the task objective is obtained. Specifically, when the task is power line inspection, the geographical location information of the power line and related equipment is obtained; when the task is substation site selection, the geographical information data of the candidate site area is obtained.

[0055] In step S32, the geospatial business data is analyzed. When the task is power line inspection, the location information of the line and equipment is analyzed to identify key areas or risk points for inspection. When the task is substation site selection, the candidate site data is analyzed to filter target areas that meet predetermined conditions.

[0056] In step S33, the target geographic information elements obtained from the analysis are integrated into the visualization instruction chain, and the corresponding layers are presented on the map visualization interface.

[0057] In one embodiment, in a power industry data visualization method driven by a large language model, for scenarios such as power line inspection and substation site selection, it is necessary to acquire geospatial business data corresponding to the task objectives. The intelligent agent can automatically identify the required spatial data type and scope based on the actual business content of the task objectives and accurately match relevant databases. For power line inspection tasks, it automatically retrieves the spatial location information of power lines and auxiliary equipment, including but not limited to line paths, tower locations, equipment identification, and operating status. For substation site selection tasks, it focuses on acquiring geographic information data of each candidate site area, including terrain, land use attributes, existing power grid access conditions, and surrounding load distribution. This not only requires the system to have efficient data retrieval and integration capabilities but also needs to embed professional knowledge in the field of power GIS to achieve automatic fusion and standardized preprocessing of multi-source heterogeneous spatial data.

[0058] The acquired geospatial business data is analyzed in depth, and different spatial analysis strategies are adopted according to specific business scenarios. In the case of power line inspection, it is necessary to analyze the spatial distribution and status information of lines and equipment, and combine multi-dimensional data such as operation and maintenance history and environmental risk parameters to intelligently identify key inspection areas and risk points. New technologies such as spatiotemporal clustering algorithms, automatic labeling of environmentally sensitive areas, and historical defect thermal modeling can be combined to achieve high-precision positioning of areas with abnormal spatial distribution and potential hazard clusters. For substation site selection, a multi-factor evaluation model needs to be established based on the geographic information data of candidate sites. Indicators such as terrain flatness, land availability, nearby load density, grid access convenience, and environmental safety are used as screening conditions. Spatial weight analysis and candidate area optimization ranking methods are used to intelligently select target areas that meet predetermined conditions. This not only reflects the deep embedding of business logic within the power industry but also enables the system to adapt to complex and diverse spatial data processing needs.

[0059] After completing geospatial data analysis, the analyzed target geographic information elements (such as high-risk areas, key inspection sections, and preferred site locations) are automatically integrated into a visualization instruction chain. This visualization instruction chain not only clarifies the specific processes for data retrieval and layer generation but also integrates technical features such as spatial layer overlay, element highlighting, interactive map zooming, and layered rendering. For example, instructions can be automatically generated to present high-incidence areas of line defects on the base map in the form of a heatmap, highlighting key equipment with icons or coloring; or in substation site selection applications, the results of candidate site selection can be intuitively displayed using hierarchically colored map layers, while also allowing users to click to view the comprehensive evaluation details of each site. To further enhance the scalability and usability of the intelligent agent, functions such as dynamic control of layer visibility, real-time map data updates, and adaptive legend generation are introduced, supporting efficient fusion and flexible interaction of multi-source spatial information. In summary, with task-driven data acquisition, intelligent spatial analysis, and adaptive layer visualization as the main focus, the intelligence level and industry adaptability of data visualization in power business scenarios are significantly enhanced.

[0060] Figure 4 This is a flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 4 of the present invention. Figure 4 As shown, the method includes steps S41-S43:

[0061] In step S41, on-site or regional image data related to the task is acquired, and image recognition is performed to extract target features. Specifically, when the task is power line inspection, defects in power equipment or obstacles in the line passage are detected in the image; when the task is substation site selection, terrain and environmental features in the image are extracted.

[0062] In step S42, path planning is performed based on the image recognition results and geospatial data. Specifically, when the task is power line inspection, an inspection route is planned; when the task is substation site selection, a transportation route for the power grid connection line or site is planned.

[0063] In step S43, the identified features and path planning are incorporated into the visualization instruction chain and presented on the map visualization interface as an overlay layer.

[0064] In one embodiment, a power data visualization method driven by fine-tuning a large language model acquires task-related on-site or regional image data and performs image recognition. On-site image data is acquired automatically or semi-automatically, and can originate from various sources such as UAV inspections, ground cameras, satellite remote sensing, or historical photo archives. Deep learning image recognition technology is used to extract features from the acquired images. In power line inspection scenarios, potential defects (such as corrosion, broken strands, loosening), obstacles (such as tree branches, foreign object intrusion), and environmental hazards are automatically detected in the images of power equipment (such as conductors, towers, insulators, etc.). In substation site selection scenarios, the focus is on extracting topographic features (such as slope, water distribution), land use status, vegetation cover, and traffic accessibility from the images. Advanced image processing algorithms such as multi-scale target detection, semantic segmentation, and target change monitoring can be introduced to improve the accuracy and robustness of feature extraction, enabling automatic discrimination of multiple types of features in complex scenarios.

[0065] After image feature extraction, the system further combines the recognition results with existing geospatial data to perform business-goal-oriented path planning. For power line inspection tasks, the system can automatically plan the optimal inspection route based on information such as the detected defect distribution, risk point location, line direction, and site terrain. This optimal inspection route not only optimizes inspection coverage and reduces redundant inspections, but also dynamically adjusts the inspection sequence and route nodes based on real-time weather, road conditions, and operational safety requirements. For substation site selection, based on the extraction of site and surrounding environmental features, the system comprehensively considers multiple factors such as the existing power grid structure, terrain obstacles, transportation network, and construction convenience to automatically plan the power line connection route or traffic access route between the new substation site and the existing power grid. The system can incorporate "multi-constraint path optimization algorithms," "dynamic weight adaptive adjustment mechanisms," and "real-time risk avoidance logic" to improve the practicality and intelligence of path planning. Furthermore, to support multi-scenario adaptation for business decisions, various planning strategy templates can be preset for users to select as needed or for the system to automatically match the optimal strategy.

[0066] The identified image features and path planning results are incorporated into the visualization instruction chain and presented as overlay layers on the map interface. This not only achieves deep fusion of multimodal data in spatial visualization but also provides users with an intuitive and interactive decision support tool. Specifically, it can automatically overlay information such as defect points, obstacle locations, and optimal paths on the base map, and use visualization techniques such as highlighting, animation, and layered display to enhance the recognizability of key data. Furthermore, it can combine new features such as "dynamic control of layer visibility," "adaptive adjustment of visualization styles," and "multi-user collaborative annotation and review" to further improve the flexibility and operational feasibility of map visualization. This enhances the intelligent analysis and dynamic decision-making capabilities for complex scenarios in the power industry, providing innovative data visualization and decision support methods for industry operation and maintenance planning.

[0067] Figure 5 This is a flowchart illustrating a preferred method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment 5 of the present invention. Figure 5 As shown, the method includes steps S51-S53:

[0068] In step S51, the visualization results are evaluated based on predetermined evaluation indicators or user feedback to determine the degree of matching between the visualization results and the expected task objectives.

[0069] In step S52, when the evaluation result of the matching degree indicates that the visualization result does not meet expectations, an instruction chain update for improving the visualization effect is generated through the fine-tuned large language model.

[0070] In step S53, the updated instruction chain is executed to generate improved visualization results, and the evaluation and update process is repeated until the visualization results meet the preset visualization quality requirements.

[0071] In one embodiment, the data visualization generation method based on fine-tuning a large language model introduces an automatic evaluation and dynamic optimization mechanism for the visualization results. This ensures that the visualization output continuously meets user needs and achieves high-quality standards in actual business scenarios. Specifically, the generated visualization results are systematically evaluated based on predetermined evaluation indicators or real-time user feedback. These evaluation indicators can include visualization accuracy, information completeness, visual aesthetics, and user-friendliness. They can also be set according to the specific needs of the power industry, such as data coverage, the prominence of key business indicators, and the accuracy of geospatial information representation. Multiple evaluation rule templates can also be preset for flexible selection in different business scenarios. The user feedback section obtains subjective opinions through natural language interaction, interactive operation, or multi-user collaborative evaluation. This integrates quantitative and qualitative information to determine the degree of matching between the current visualization results and the task objectives, improving the objectivity and intelligence of the evaluation.

[0072] When evaluation results indicate that the current visualization results fail to meet expectations, the system automatically invokes a fine-tuned large language model to intelligently analyze the task objectives, user feedback, and the execution process of previous instruction chains, generating a new round of instruction chain updates to improve the visualization effects. During instruction chain optimization, strategies can be adjusted by integrating multi-dimensional information sources: on the one hand, the system can intelligently fine-tune detailed parameters such as visualization mapping logic, chart types, color schemes, and layer layouts; on the other hand, it can automatically introduce new data sources, aggregate new analytical indicators, and highlight key business elements based on user feedback or business scenarios. This ensures that each improvement to the visualization results continuously converges with business objectives and user expectations, achieving an intelligent and iterative process of visualization quality optimization.

[0073] After executing the updated instruction chain and outputting new visualization results, the system automatically enters the next round of evaluation and optimization. This process repeats until the visualization results meet the system's preset or user-specified quality standards. This closed-loop mechanism enables intelligent management of the entire data visualization process, from automatic generation, evaluation, and optimization to achieving satisfactory output, greatly improving the scientific rigor, flexibility, and user satisfaction of business operations and maintenance decisions. It not only breaks through the passive adaptation mode of traditional static visualization but also achieves continuous quality optimization and personalized visualization services driven by advanced intelligence.

[0074] Figure 6 This is a flowchart illustrating a method for generating a visual instruction chain based on a fine-tuned large language model, as provided in Embodiment Six of the present invention. Figure 6 As shown, the method includes steps S61-S63:

[0075] In step S61, based on predetermined visualization specifications, a page presentation scheme for the business data is determined. The page presentation scheme includes color matching, graphic symbol shapes, size ratios, and layout design to ensure the readability of the visualization results.

[0076] In step S62, the visualization instruction chain is adjusted using the fine-tuned large language model, and the page presentation scheme is applied to the visualization result;

[0077] In step S63, the adjusted instruction chain is executed, and the optimized visualization results are output.

[0078] In one embodiment, in the method for generating a data visualization instruction chain based on fine-tuning a large language model, the page presentation scheme is determined by combining business needs and visualization specifications. According to predetermined visualization specifications, the characteristics and display objectives of the business data are analyzed to determine the optimal page presentation scheme. This typically includes color matching, the shape of graphic symbols, the size ratio of various information types, and the overall layout design. For example, for geographic information visualization commonly used in the power industry, hot and cold color gamut can be used to reflect the quality of equipment status; key potential hazards can be highlighted with high-contrast colors and special symbols; important equipment nodes can have larger annotation sizes for quick location; and complex line information can be displayed in layers or groups to avoid information overlap and visual confusion. Simultaneously, based on the data volume, user habits, and business scenario, the optimal chart layout method is automatically selected, such as multi-chart linkage and clear division of primary and secondary areas, to improve overall information delivery efficiency.

[0079] Through a finely tuned large language model, the initially generated visualization instruction chain is intelligently adjusted, efficiently mapping the page presentation scheme to the specific visualization output process. The large language model not only understands and applies visualization specifications but also integrates historical experience and contextual information to intelligently optimize the details of the instruction chain. For example, it can automatically identify graphic symbols or color schemes that need adjustment due to changes in data characteristics, dynamically adapting to the visual needs of different terminals (such as large-screen displays, mobile devices, and printed materials). Furthermore, a visual conflict detection mechanism can be introduced to automatically identify and correct potential visualization defects such as color confusion, symbol ambiguity, and crowded layouts. For the multi-layered spatial overlay and dynamic data updates common in power business, an adaptive layer rendering algorithm can be supported, adjusting the display order and transparency of each layer in real time according to business focus, ensuring that key information is always in the visual focus area. During the execution phase of the finely tuned and optimized instruction chain, the final optimized visualization result is output. This not only closely aligns with business objectives in terms of content accuracy but also fully considers data readability and user experience at the visual expression level. It can automatically generate accompanying legends, interactive prompts, and dynamic highlighting, facilitating efficient understanding and rapid decision-making for users in multi-source data and complex business contexts. In the stages of page presentation design and instruction chain application, it integrates multiple innovative advantages such as intelligence, automation, and personalization, greatly enhancing the business adaptability and user experience of data visualization.

[0080] To overcome the aforementioned shortcomings, in some preferred embodiments, a visualization presentation agent and method are proposed. This method utilizes a fine-tuned large model to recognize high-level visual intents and automatically generates and orchestrates visualization instruction chains following standard syntax (such as Vega-Lite or ECharts configuration). The rendering engine then outputs interactive visualization results across multiple terminals, including Web UI and mobile devices. This achieves automatic mapping from abstract business objectives to effective visual encoding, improving the efficiency, quality, and scalability of visualization generation. Furthermore, in power enterprise operation and maintenance management, it provides an intelligent dispatch visualization implementation based on geographical location, equipment status, and real-time personnel distribution, significantly improving operation and maintenance efficiency.

[0081] 1. System Overall Structure

[0082] The visual representation of the intelligent agent in this invention includes, but is not limited to, the following modules (module numbers are for clarity of description only):

[0083] 101 Preprocessing and Context Injection Module: Receives the user's natural language target and data source description, extracts data patterns (field names, data types, geographic fields, time fields, units of measurement) and business context (industry terms, task objectives, user preferences) and forms a context information package.

[0084] 102 Intent Recognition and Task Decomposition Module (Fine-tuned LLM): Performs visual intent recognition on user goals (such as comparison, distribution, relevance, ranking, geographic heat, anomaly interpretation), and outputs a structured "Visual Intent Representation" (VIL).

[0085] 103 Scheme Planning and Intermediate Representation Generation Module: Based on VIL and data features, it generates a visual planning intermediate language (VPL) to describe candidate view sets, necessary data transformations (filtering, aggregation, binning, window functions), encoding (x / y / color / size / shape), layout (facetization, grid, linkage) and interaction (tooltip, brush, zoom, linking).

[0086] 104 Instruction Generation and Orchestration Module: Compiles VPL into an instruction chain. The instructions follow standard visual syntax (Vega-Lite specification, ECharts configuration, etc.) and establishes dependencies (DAG) to form an executable order and fallback path.

[0087] 105 Syntax / Semantic Validation and Quality Assessment Module: Performs syntax validation, data constraint validation (field existence, type matching, encoding conflicts), and visual accessibility and readability assessment (color contrast, occlusion, legend ambiguity) on the instruction chain, and triggers rollback strategies and secondary planning when necessary.

[0088] 106 Rendering Execution and Multi-Terminal Output Module: Sends the instruction chain to the visualization engine (WebGL / D3 / ECharts / Vega-Lite renderer) to generate interactive views, supporting Web UI, mobile App, mini-program and large screen.

[0089] 107 Interactive Event Acquisition and Incremental Optimization Module: Collects user interactions on the finished view (filtering, zooming, linked selection, view replacement), triggering incremental command optimization and preference learning.

[0090] 108 Security and Resource Management Module: Implements data access control, anonymization, key management, caching and concurrency control, logging and traceability.

[0091] 109 Business Plugins and GIS Capability Modules (Optional): Provides GIS base map management, coordinate projection, spatial aggregation (grid / cellular / administrative region), path calculation and spatiotemporal linkage; integrates industry business plugins (such as power equipment ledger, work order engine, personnel trajectory access).

[0092] 2. Method and Flow

[0093] A visualization method based on the above system includes, but is not limited to, the following steps:

[0094] S201, User Input: Obtain natural language target and data source connection information / data sample.

[0095] S202, Context Construction: Perform data pattern parsing, industry terminology mapping, and assembly of user preferences and display constraints.

[0096] S203, Intent Recognition: Fine-tune the LLM output VIL (e.g., "Comparison + Geographic Aggregation + Anomaly Labeling").

[0097] S204, Solution Planning: Generate VPL (Candidate View Combination and Data Transformation / Encoding / Layout / Interaction Requirements).

[0098] S205, Instruction Generation: Compiles VPL into a standard syntax instruction chain, including dependencies and execution order.

[0099] S206. Verification and Evaluation: Perform grammatical / semantic verification and visual quality scoring. If it fails, return to S204 for further planning.

[0100] S207. Rendering Output: Generate interactive views and dashboards on the Web or mobile devices.

[0101] S208, Interaction Closed Loop: Collect user interactions and feedback to trigger incremental updates or recommend alternative views.

[0102] 3. Fine-tuning and context enhancement

[0103] Training data consists of: paired samples of "natural language objectives - visualization specifications / configurations"; common task intent and data type mapping samples; GIS task corpus (place name disambiguation, coordinate system, spatial aggregation); power business corpus (equipment type, fault level, SLA); and user preference and finished map legend configuration samples.

[0104] Training objectives: To jointly optimize three tasks: intent classification, scheme planning, and instruction generation; and to improve controllability and interpretability by using structured intermediate representations (VIL / VPL) as supervision signals.

[0105] Context injection: At runtime, data schema, field semantics, units of measurement, visualization disabling rules (such as color blindness safety), and business constraints (such as operation and maintenance SLA and skill matching) are injected into the prompting and verification process.

[0106] 4. Key Validation and Rollback

[0107] Semantic conflict detection: Field type and encoding channel mismatch, repeated legends causing ambiguity, insufficient color contrast.

[0108] Rollback strategies include: automatically changing the marker type (scatter to bar), switching the aggregation strategy (count to sum to mean), switching the color scheme or legend position, and downsampling / aggregation to avoid occlusion.

[0109] Quality rating: The optimal instruction chain is selected based on factors such as readability, task completion (whether comparison / sorting / location is supported), and interaction redundancy.

[0110] 5. Multi-engine and multi-terminal output

[0111] Syntax adaptation: Provides compilers for VPL→Vega-Lite and VPL→ECharts;

[0112] Terminal formats: Web dashboard, large screen, mobile (native / mini-program), offline report (export PNG / PDF);

[0113] Interactive features include: multi-view selection and linkage, time slider, geographic layer drill-down, and path playback.

[0114] Compared with the prior art, the present invention has the following advantages:

[0115] 1. By fine-tuning the LLM, accurate visual intent recognition of abstract business objectives can be achieved, significantly reducing manual configuration;

[0116] 2. The instruction chain carries the visualized decision-making process, enabling multi-step orchestration and rollback optimization to ensure the quality and stability of the finished product;

[0117] 3. For complex scenarios such as GIS, it automatically couples multi-dimensional spatial / temporal / indicator data to output interactive, cross-terminal expressions;

[0118] 4. In industries such as power operation and maintenance, plug-and-play visual decision support can be achieved through business plugins and context injection, thereby improving scheduling and operation and maintenance efficiency.

[0119] Figure 7 This is a schematic diagram of the structure of a system for generating a visual instruction chain based on a large language model with fine-tuning, as provided in Embodiment 7 of the present invention. Figure 7 As shown, the system includes a receiving module 71, an analysis module 72, a generation module 73, and a visualization module 74.

[0120] The receiving module 71 is used to receive task objectives described in natural language;

[0121] The analysis module 72 is used to analyze the natural language description based on an understanding of the power data characteristics and power business context involved in the task objective;

[0122] The generation module 73 is used to generate an instruction chain for visualization based on the analysis results and the fine-tuned large language model.

[0123] The visualization module 74 is used to execute the instruction chain for visualization presentation in order to generate the visualization result of the business data.

[0124] The system for generating visual instruction chains based on fine-tuning large language models provided in the embodiments of the present invention can execute the method for generating visual instruction chains based on fine-tuning large language models provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0125] Figure 8 This is a schematic diagram of an electronic device for implementing the method of generating a visual instruction chain based on a fine-tuned large language model according to embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptops, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0126] like Figure 8As shown, the electronic device 80 includes at least one processor 81 and a memory, such as a read-only memory (ROM) 82 and a random access memory (RAM) 83, communicatively connected to the at least one processor 81. The memory stores computer programs executable by the at least one processor. The processor 81 can perform various appropriate actions and processes based on the computer program stored in the ROM 82 or loaded from storage unit 88 into the RAM 83. The RAM 83 can also store various programs and data required for the operation of the electronic device 80. The processor 81, ROM 82, and RAM 83 are interconnected via a bus 84. An input / output (I / O) interface 85 is also connected to the bus 84.

[0127] Multiple components in electronic device 80 are connected to I / O interface 85, including: input unit 86, such as keyboard, mouse, etc.; output unit 87, such as various types of monitors, speakers, etc.; storage unit 88, such as disk, optical disk, etc.; and communication unit 89, such as network card, modem, wireless transceiver, etc. Communication unit 89 allows electronic device 80 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0128] Processor 81 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 81 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 81 performs the various methods and processes described above, such as the method of generating a visual instruction chain based on a fine-tuned large language model.

[0129] In some embodiments, the method for generating a visual instruction chain based on a fine-tuned large language model can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 88. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 80 via ROM 82 and / or communication unit 89. When the computer program is loaded into RAM 83 and executed by processor 81, one or more steps of the method for generating a visual instruction chain based on a fine-tuned large language model described above can be performed. Alternatively, in other embodiments, processor 81 can be configured to perform the method for generating a visual instruction chain based on a fine-tuned large language model by any other suitable means (e.g., by means of firmware).

[0130] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0131] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0132] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0133] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0134] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0135] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0136] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0137] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for generating visual instruction chains based on fine-tuning of a large language model, characterized in that, include: Receive task objectives described in natural language; Based on an understanding of the power data characteristics and power business context involved in the task objectives, the natural language description is analyzed. Based on the analysis results, an instruction chain for visualization is generated using the fine-tuned large language model. Execute the instruction chain for visualization to generate a visualization result of the business data.

2. The method according to claim 1, characterized in that, The method further includes: Semantic parsing is performed on the natural language description to extract initial visualization requirements information; At least one clarifying question is generated using the finely tuned large language model and interacted with the user in natural language to dispel ambiguity or supplement missing information. Receive the user's answer to the clarifying question, and update the description of the task objective and the corresponding visual instruction chain based on the answer.

3. The method according to claim 1, characterized in that, The tasks corresponding to the aforementioned objectives include power line inspection or substation site selection, and the methods specifically include: Obtain geospatial business data corresponding to the task objective, wherein when the task is power line inspection, obtain the geographical location information of the power line and related equipment; when the task is substation site selection, obtain the geographical information data of the candidate site area. The geospatial business data is analyzed. When the task is power line inspection, the location information of the line and equipment is analyzed to identify key areas or risk points for inspection. When the task is substation site selection, the candidate site data is analyzed to filter target areas that meet predetermined conditions. The target geographic information elements obtained from the analysis are integrated into the visualization instruction chain, and the corresponding layers are presented on the map visualization interface.

4. The method according to claim 1, characterized in that, The method further includes: Acquire on-site or regional image data related to the task and perform image recognition to extract target features. Specifically, when the task is power line inspection, detect power equipment defects or line passage obstacles in the image; when the task is substation site selection, extract terrain and environmental features in the image. Path planning is performed based on the image recognition results and geospatial data. Specifically, when the task is power line inspection, an inspection route is planned; when the task is substation site selection, a transportation route for the power grid connection line or site is planned. The identified features and path planning are incorporated into the visualization instruction chain and presented as overlay layers on the map visualization interface.

5. The method according to claim 1, characterized in that, After generating the visualization results of the business data, the method further includes: The visualization results are evaluated based on predetermined evaluation indicators or user feedback to determine the degree of matching between the visualization results and the expected task objectives. When the evaluation results of the matching degree indicate that the visualization results do not meet expectations, the fine-tuned large language model generates an instruction chain update to improve the visualization effect. The updated instruction chain is executed to generate improved visualization results, and the evaluation and update process is repeated until the visualization results meet the preset visualization quality requirements.

6. The method according to claim 1, characterized in that, The visualization results generated from the business data include: Based on predetermined visualization specifications, a page presentation scheme for the business data is determined. The page presentation scheme includes color matching, graphic symbol shapes, size ratios, and layout design to ensure the readability of the visualization results. The visualization instruction chain is adjusted by fine-tuning the large language model, and the page presentation scheme is applied to the visualization result. Execute the adjusted instruction chain and output optimized visualization results.

7. A system for generating visual instruction chains based on fine-tuning of a large language model, characterized in that, include: The receiving module is used to receive task objectives described in natural language; The analysis module is used to analyze the natural language description based on an understanding of the power data characteristics and power business context involved in the task objective; The generation module is used to generate an instruction chain for visualization based on the analysis results and the fine-tuned large language model. The visualization module is used to execute the instruction chain for visualization presentation in order to generate the visualization result of the business data.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the method for generating a visual instruction chain based on a fine-tuning large language model as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the method for generating a visual instruction chain based on a large language model based on fine-tuning, as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method for generating a visual instruction chain based on a large language model with fine-tuning, as described in any one of claims 1-6.

Images