A method and device for generating a report on simulation results based on graphics and text

Abstract

The application provides a simulation result evaluation report generation method and device based on graphics and text. First, heterogeneous data is obtained, and vector data is obtained by processing the heterogeneous data. The heterogeneous data includes index files and scenario files. Then, a drawing logic summary is generated based on the vector data, a visualization instruction code is generated based on the drawing logic summary, and the visualization instruction code is run in an isolated execution environment to generate a target chart. Finally, the vector data and the corresponding structured chart description of the target chart are input into a large language model to obtain qualitative evaluation text corresponding to the target chart. The semantic and numerical alignment of data, charts and text is realized, and the accuracy, interpretability, maintainability and self-adaptive ability of the evaluation report are significantly improved.

Description

Technical Field

[0001] This application relates to the field of large model technology, and in particular to a method and apparatus for generating simulation result evaluation reports based on graphics and text. Background Technology

[0002] In critical decision-making processes such as equipment evaluation and mission planning, professional assessment reports must balance data accuracy with contextual rationale. Currently, the generation of professional assessment reports mainly relies on manual writing, which is inefficient. Automated generation tools often suffer from several drawbacks: First, there is a disconnect between text and graphics, with charts and text generated separately, lacking semantic and numerical alignment, frequently resulting in mismatched text and correct figures. Second, conclusions can be distorted; large models can easily create illusions detached from data, lacking underlying fact-checking mechanisms. Third, interactive mechanisms fail; manual modifications cannot automatically identify the scope of logical impact, leading to logical breaks and making the entire report difficult to maintain. These problems are particularly prominent in high-consequence scenarios, urgently requiring an intelligent generation method that enables cross-verification of text and graphics and full traceability. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method and apparatus for generating simulation result evaluation reports based on graphics and text. By vectorizing the index file and the scenario file into numerical summary features and background knowledge vectors, and combining the self-healing generation of isolated environment charts, DAG-driven graphics-text alignment text generation, and memory-enhanced bias optimization, the method achieves strict alignment of the semantics and numerical values ​​of data, charts, and text, which significantly improves the accuracy, interpretability, maintainability, and adaptability of the evaluation report.

[0004] In a first aspect, embodiments of this application provide a method for generating simulation result evaluation reports based on graphics and text, the method comprising: Heterogeneous data is acquired and processed to obtain vector data; wherein the heterogeneous data includes indicator files and scenario files; wherein the vector data includes numerical summary features and background knowledge vectors; A drawing logic summary is generated based on the vector data, visualization instruction code is generated based on the drawing logic summary, and the visualization instruction code is run in an isolated execution environment to generate the target chart. The vector data and the structured chart description corresponding to the target chart are input into the large language model to obtain the qualitative evaluation text corresponding to the target chart.

[0005] Furthermore, the numerical summary features are used for chart generation and numerical verification, and the background knowledge vectors are used for semantic retrieval and contextual constraints; the processing of the heterogeneous data to obtain vector data includes: After converting the index file from spreadsheet format to structured data format, it is divided into data segments corresponding to multiple time steps according to the time sequence of the simulation run. Then, a set of key points reflecting the changing trend and statistical features characterizing the distribution characteristics are extracted from the data fragments to obtain the numerical summary features; The hypothetical file is segmented into semantic units, and the hypothetical file is mapped to a vector space through a text embedding function to obtain the background knowledge vector; The numerical summary features and the background knowledge vector together constitute the vector data.

[0006] Furthermore, the step of running the visualization instruction code in an isolated execution environment to generate the target chart includes: The initial script is generated by calling a preset code template based on the drawing logic summary. The anchor agent runs the initial script in the isolated execution environment and monitors the execution status, structure information and chart rendering effect of the initial script in real time. When the process fails or the chart does not meet the visual structure constraints, the reasoning agent combines the error message with the visual structure constraints to iteratively rewrite the script until a target chart that meets the requirements is generated.

[0007] Furthermore, the qualitative assessment text is generated through a report generation process organized using a directed acyclic graph structure; In the directed acyclic graph structure, each node represents an evaluation subtask, and the nodes are connected by directed edges to indicate the order of evaluation subtasks. The generation of each node is performed by a reasoning agent, and the execution of the reasoning agent is based on the content of the node's predecessor node, the background information matched from the knowledge vector base, and the historical similar reasoning patterns retrieved from the continuously evolving memory card network.

[0008] Furthermore, after obtaining the qualitative evaluation text, the generation method further includes: In response to a user's modification of any segment in the qualitative assessment text, the modified content is obtained; Perform causal logic verification on the modified content to determine whether it conflicts with the underlying physical data subset; If no conflict is determined, then based on the modified content and the topological connection relationship of the directed acyclic graph structure, all successor nodes that depend on the modified node are automatically identified, triggering the regeneration of the successor nodes.

[0009] Furthermore, after obtaining the qualitative evaluation text, the generation method further includes: In response to the automatic optimization instruction for the qualitative evaluation text, the anchor agent performs reverse numerical consistency verification on the underlying data corresponding to the qualitative evaluation text. When a discrepancy is found between the text description and the data facts, a historical logical pattern matching the type of the discrepancy is retrieved from the memory card network, and this logical pattern is input as a hard constraint signal into the reasoning agent to drive the reasoning agent to regenerate the discrepancy paragraph. The correction records generated during the regeneration process are extracted and stored in the memory card network to form new memory cards.

[0010] Secondly, embodiments of this application also provide a simulation result evaluation report generation device based on graphics and text, the generation device comprising: A data processing module is used to acquire heterogeneous data and process the heterogeneous data to obtain vector data; wherein the heterogeneous data includes indicator files and scenario files; wherein the vector data includes numerical summary features and background knowledge vectors; The chart generation module is used to generate a drawing logic summary based on the vector data, generate visualization instruction code based on the drawing logic summary, and run the visualization instruction code in an isolated execution environment to generate the target chart. The evaluation text generation module is used to input the vector data and the structured chart description corresponding to the target chart into the large language model to obtain the qualitative evaluation text corresponding to the target chart.

[0011] Furthermore, the numerical summary features are used for chart generation and numerical verification, and the background knowledge vectors are used for semantic retrieval and contextual constraints; when the data processing module processes the heterogeneous data to obtain vector data, the data processing module is also used for: After converting the index file from spreadsheet format to structured data format, it is divided into data segments corresponding to multiple time steps according to the time sequence of the simulation run. Then, a set of key points reflecting the changing trend and statistical features characterizing the distribution characteristics are extracted from the data fragments to obtain the numerical summary features; The hypothetical file is segmented into semantic units, and the hypothetical file is mapped to a vector space through a text embedding function to obtain the background knowledge vector; The numerical summary features and the background knowledge vector together constitute the vector data.

[0012] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the simulation result evaluation report generation method described above are performed.

[0013] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the simulation result evaluation report generation method described above.

[0014] This application provides a method, apparatus, device, and medium for generating a simulation result evaluation report. First, heterogeneous data is acquired and processed to obtain vector data; wherein the heterogeneous data includes indicator files and scenario files. Then, a drawing logic summary is generated based on the vector data, and visualization instruction code is generated based on the drawing logic summary. The visualization instruction code is then run in an isolated execution environment to generate a target chart. Finally, the vector data and the structured chart description corresponding to the target chart are input into a large language model to obtain a qualitative evaluation text corresponding to the target chart.

[0015] This application achieves strict alignment of semantics and numerical values ​​among data, charts, and text by vectorizing the indicator file and scenario file into numerical summary features and background knowledge vectors, combined with self-healing generation of isolated environment charts, DAG-driven image-text alignment text generation, and memory-enhanced bias optimization. This significantly improves the accuracy, interpretability, maintainability, and adaptability of the evaluation report.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a method for generating simulation result evaluation reports based on graphics and text, provided in an embodiment of this application; Figure 2A flowchart illustrating a method for generating simulation result evaluation reports based on graphics and text, provided in an embodiment of this application; Figure 3 This is one of the structural schematic diagrams of a simulation result evaluation report generation device based on graphics and text provided in an embodiment of this application; Figure 4 This is a second schematic diagram of a device for generating simulation result evaluation reports based on graphics and text, provided in an embodiment of this application. Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0020] First, the applicable scenarios for this application will be introduced. This application can be applied to the field of large-scale model technology.

[0021] In critical decision-making processes such as equipment evaluation and mission planning, professional assessment reports must balance data accuracy with contextual rationale. Research has found that current professional assessment report generation primarily relies on manual writing, which is inefficient. Automated generation tools often suffer from several drawbacks: First, there is a disconnect between text and graphics, with charts and text generated separately, lacking semantic and numerical alignment, frequently resulting in mismatched text and correct figures. Second, conclusions can be distorted; large models can easily create illusions detached from data, lacking underlying fact-checking mechanisms. Third, interactive mechanisms fail; manual modifications cannot automatically identify the scope of logical impact, leading to logical breaks and difficulty in maintenance. These problems are particularly pronounced in high-consequence scenarios, necessitating an intelligent generation method that enables cross-verification of text and graphics and full traceability.

[0022] Based on this, the embodiments of this application provide a method for generating simulation result evaluation reports, which achieves strict alignment of semantics and numerical values ​​among data, charts, and text, and significantly improves the accuracy, interpretability, maintainability, and adaptability of the evaluation report.

[0023] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for generating a simulation result evaluation report, as provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the generation method includes: S101, Obtain heterogeneous data and process the heterogeneous data to obtain vector data.

[0024] In specific implementation of step S101 above, heterogeneous data is acquired and processed to obtain vector data.

[0025] Here, according to an embodiment of this application, the heterogeneous data includes indicator files and scenario files. Specifically, the input includes two types of heterogeneous data: one is the indicator file output by the simulation engine, which is usually converted from Excel format to JSON format; the other is the scenario file (XML) describing the initial situation, combat intentions, and mission plans.

[0026] Furthermore, the vector data includes numerical summary features and background knowledge vectors. The numerical summary features are used for chart generation and numerical verification, and the background knowledge vectors are used for semantic retrieval and contextual constraints. Regarding step S101 above, processing the heterogeneous data to obtain vector data includes: Step 1011: After converting the indicator file from spreadsheet format to structured data format, divide it into multiple data segments corresponding to multiple time steps according to the time sequence of the simulation run.

[0027] Step 1012: Extract the set of key points reflecting the changing trend and the statistical features characterizing the distribution from the data fragment to obtain the numerical summary features.

[0028] For steps 1011-1012 above, the input index file is first read and converted into a structured data format using a standard parsing tool, ensuring consistent field names, clear marking of missing values, and preservation of unit information. Then, based on the inherent time step of the simulation, the entire time series data is divided into several continuous and equal-length data segments in chronological order. Each segment corresponds to a specific time step (e.g., second 1, second 2, etc.), thus constructing a discretized data sequence strictly aligned with the simulation time axis. Specifically, the simulation indices extracted from the index file are abstracted into discrete time series according to time division, expressed by the following formula:

[0029] in, Indicates the simulation step size. This represents the corresponding state observation value, and statistical features are extracted from it to form a structured numerical summary.

[0030] For each data segment corresponding to each time step obtained in step 1011, each indicator is analyzed separately to extract two types of core numerical features: First, a set of key points reflecting the changing trend. For example, the set of key points may include the global maximum and minimum values ​​within the segment and their occurrence times, the precise time point when the preset tactical threshold is first broken, or the start and end intervals that satisfy the steady-state condition, etc. This application does not make specific limitations on this. Second, statistical features characterizing the distribution characteristics. For example, statistical features may include the mean, standard deviation, skewness, kurtosis, etc. of the indicator within the segment, etc. This application does not make specific limitations on this. All the above key point coordinates and statistics are arranged and concatenated in order according to the indicator dimension to obtain the numerical summary features.

[0031] Step 1013: Segment the hypothetical file into semantic units and map the hypothetical file to a vector space using a text embedding function to obtain the background knowledge vector.

[0032] Regarding step 1013 above, in specific implementation, during the target file processing stage, the XML file is segmented according to semantic units and processed using a text embedding function. Mapping to a vector space to construct a vector knowledge base It also supports subsequent background constraint retrieval based on similarity.

[0033] Step 1014: Combine the numerical summary features with the background knowledge vector to form the vector data.

[0034] Regarding step 1014 above, in specific implementation, the numerical summary features and background knowledge vectors are combined to form vector data.

[0035] In this way, for the processing of heterogeneous input data, this application constructs a dual-track data abstraction mechanism. By extracting statistical features from the time series of simulation indicators and using text embedding functions to vectorize the scenario files, it achieves deep fusion of multi-source data at the feature level, providing a data foundation for generating targeted operational background constraints.

[0036] S102, generate a drawing logic summary based on the vector data, generate visualization instruction code based on the drawing logic summary, and run the visualization instruction code in an isolated execution environment to generate the target chart.

[0037] Regarding step S102 above, in specific implementation, a drawing logic summary is generated based on the vector number constructed in step S101 above, and Python visualization instruction code is automatically generated based on the drawing logic summary. At the same time, the visualization instruction code is run in an isolated execution environment to generate the target chart.

[0038] Please refer to Table 1 below, which is a table showing the correspondence between intelligent agents and responsibilities provided in the embodiments of this application.

[0039] Table 1. Correspondence between an intelligent agent and its responsibilities.

[0040] As shown in Table 1 above, the intelligent agents used in this application embodiment include: Anchor Agent, responsible for executing code, extracting chart structure, and performing numerical consistency verification; Reasoner Agent, responsible for generating drawing logic, rewriting code, and generating / rewriting text; Controller Agent, responsible for determining whether user modifications conflict with data and deciding whether to trigger subsequent node updates; and Archivist Agent, responsible for extracting correction records and writing them into the memory card network.

[0041] As an optional embodiment, regarding step S102 above, running the visualization instruction code in an isolated execution environment to generate the target chart includes: Step 1021: Generate an initial script by calling a preset code template based on the drawing logic summary.

[0042] Regarding step 1021 above, in specific implementation, the plotting logic summary generated in step S102 above is input into the template matching engine. The engine retrieves and instantiates the most matching visualization template from the preset code template library based on the keywords in the summary, automatically fills in parameters such as indicator field names, time range, threshold values, and annotation text, and generates a syntactically correct and executable initial version of the Python visualization script.

[0043] Step 1022: The anchor agent runs the initial script in the isolated execution environment and monitors the execution status, structure information, and chart rendering effect of the initial script in real time.

[0044] Step 1023: When the operation fails or the chart does not meet the visual structure constraints, the reasoning agent combines the error message with the visual structure constraints to iteratively rewrite the script until a target chart that meets the requirements is generated.

[0045] Here, according to the embodiments provided in this application, to alleviate the "illusion" problem in large language model generation, a reasoning agent is introduced as the responsible agent for the restricted execution and structural constraint extraction module. Its embedded mathematical feature extraction function... This is used to extract deterministic structural constraints from data. Simultaneously triggered. Extracting visual structural constraints This constraint contains key structural information, such as the first-order difference trend. With the set of extreme points Ultimately, the report text generation process is modeled as a controlled generation problem: .in, Together, they serve as global objective factual constraints, ensuring that the generated content is strictly aligned with the underlying data.

[0046] Regarding steps 1022-1023 above, in specific implementation, the anchor agent runs the initial script, while simultaneously collecting three types of information in real time: first, the execution status, including process exit code, exception type, and stack information; second, structural information, i.e., the internal data of the chart object actually rendered by the script, such as the coordinates of each point in the line chart, axis scales, legend text, etc.; and third, the chart rendering effect, referring to the result after verification by preset visual structural constraint rules. When the anchor agent detects that the script has failed to run, or although it has run successfully but the rendered chart does not meet any visual structural constraint, it immediately submits the error message and the specific visual structural constraint conditions to the inference agent. The inference agent, combining all the above context, rewrites the initial script in a targeted manner to generate a new version of the script. The above steps are executed cyclically, with each iteration being re-verified by the anchor agent until the generated chart simultaneously meets all visual structural constraints and has no runtime errors. At this point, the output target chart is the final usable result.

[0047] Thus, this application introduces an anchor agent with an embedded mathematical feature extraction function. By executing plotting code and simultaneously extracting key structural parameters such as trend derivatives and extreme points, these objective features are transformed into prior hard constraints for the generation of a large language model. Since the text generation process is modeled as a conditional probability problem based on multidimensional chart features, this approach forces the generated qualitative text fragments to refer to the mathematical laws of quantitative data. This, to some extent, solves the illusion that the model is detached from reality and ensures the alignment between chart and textual conclusions.

[0048] S103, input the vector data and the structured chart description corresponding to the target chart into the large language model to obtain the qualitative evaluation text corresponding to the target chart.

[0049] Regarding step S103 above, in specific implementation, the vector data generated in step S101 and the structured icon description corresponding to the target chart generated in step S102 are used as input and fed into the large language model. Here, when generating qualitative evaluation text, the large language model follows the Directed Acyclic Graph (DAG) report generation process, generating logically coherent, factually accurate, and context-appropriate natural language text node by node, ensuring that each description has traceable data support and assumptions, and finally outputting qualitative evaluation text that is completely consistent with the target chart in semantic meaning.

[0050] Here, each node in the directed acyclic graph (DAG) structure represents an evaluation subtask, and the nodes are connected by directed edges to indicate the order of evaluation subtasks. This DAG can be dynamically expanded based on the execution results of preceding nodes; that is, based on the initial DAG template, supplementary nodes are dynamically added according to intermediate inference results, but the newly added nodes must still satisfy the acyclic constraint. The subtasks are hierarchically divided into chapter / section writing, sentence organization, and factual question answering. The generation of each node is executed by the inference agent, whose execution is based on the content generated by the preceding nodes, background information matched from the knowledge vector base, and historical similar inference patterns retrieved from the continuously evolving memory card network.

[0051] In the embodiments provided in this application, a directed acyclic graph (DAG) is used to model the generation process of qualitative evaluation text. Specifically, a set of task nodes is defined. Each node The generation of this is performed by the reasoning agent, which mainly relies on three types of information: 1) a subset of background knowledge. 2) Parent node state set 3) From memory map Search results The node state is recursively defined as follows:

[0052] in, Concat This indicates the context concatenation operator. This represents the node-level generation result. This mechanism supports node-by-node progressive generation and dynamically expands the DAG structure based on the inference depth.

[0053] In this way, the progressive generation architecture based on directed acyclic graphs, by defining the set of task nodes and their recursive reasoning logic, allows each stage of the report to integrate background knowledge, conclusions from preceding nodes, and prior constraints from the memory card network. This progressive generation mode simulates the rigorous reasoning process of human experts analyzing before drawing conclusions, ensuring that each evaluation conclusion is based on a reliable subset of facts and logical foundations. This effectively alleviates the logical discontinuity problem commonly found in long document generation and enhances the closed-loop rigor of the entire report's macro-architecture.

[0054] As an optional embodiment, after obtaining the qualitative evaluation text, the generation method provided in this application further includes: A: In response to a user's modification operation on any segment of the qualitative evaluation text, obtain the modified content, or automatically perform a reverse self-modification check on the child nodes of the directed acyclic graph to obtain the modified content.

[0055] Regarding step A above, in practical implementation, paragraph-level manual interactive modification is supported. In response to a user's modification operation on any segment of the qualitative evaluation text, the system obtains the user-inputted modification content. Alternatively, it automatically performs a reverse self-modification check on the child nodes of the directed acyclic graph to obtain the content to be modified.

[0056] B: Perform causal logic verification on the modified content to determine whether there is a conflict between it and the underlying data subset.

[0057] Here, the fact consistency determination function is defined. When the user enters correction text At that time, the controlling agent makes the judgment: if (If the semantics conflict with the original data, then the update is rejected; if) This allows the original text to be replaced.

[0058] C: If no conflict is determined, based on the modified content and the topological connection relationship of the directed acyclic graph structure, all successor nodes that depend on the modified node are automatically identified, and the regeneration of the successor nodes is triggered.

[0059] Regarding step C above, in practical implementation, when it is determined that there is no conflict in step B, based on the user-inputted modifications and the topological connections of the directed acyclic graph, all successor nodes that depend on the modified node are automatically identified. Specifically, for the modified node... The system automatically identifies its set of successor nodes. And trigger a linked update reminder, with the execution status being recursively pushed: The Reasoner is the reasoning agent that eliminates global inconsistencies caused by local modifications.

[0060] In this way, when manual content interaction and modification operations occur, the DAG topology is used to automatically identify and allow the affected downstream derivative chain nodes to be updated in a coordinated manner. This cascading response ensures the consistency of global logic after local modification, so that the report has both the flexibility of human intervention and the data rigor of automated evaluation.

[0061] As another optional embodiment, after obtaining the qualitative evaluation text, the generation method provided in this application further includes: I: In response to the automatic optimization instruction for the qualitative evaluation text, the anchor agent performs reverse numerical consistency verification on the underlying data corresponding to the qualitative evaluation text.

[0062] Regarding step I above, in specific implementation, upon receiving the automatic optimization instruction triggered by the user, the anchor agent automatically locates the subset of underlying simulation index data pointed to by the specific numerical description appearing in the qualitative evaluation text, and compares the text value with the underlying data. II: When a discrepancy is found between the text description and the data facts, a historical logical pattern matching the type of the discrepancy is retrieved from the memory card network, and this logical pattern is input as a hard constraint signal into the reasoning agent to drive the reasoning agent to regenerate the discrepancy paragraph.

[0063] Regarding step II above, in practical implementation, once a deviation is confirmed, based on the type of deviation, memory cards that have historically successfully corrected similar deviations are retrieved from the continuously evolving memory card network. The retrieved historical logical patterns are extracted as structured hard constraint signals, forcing strict adherence to these patterns when rewriting the paragraph, thereby ensuring that the corrected text is both factual and meets domain expression norms. Specifically, for the generated draft Y, when the user selects the automatic optimization mode, the anchor agent is invoked to back-verify numerical consistency and perform game-theoretic correction. When it is detected that the generated text is inconsistent with... When a conflict exists, the corresponding logical pattern is retrieved from the memory graph as a constraint signal to guide the reasoning agent to regenerate the memory card, which then stores this memory card containing the correction process in the memory card network. This process involves three levels: memory card, memory card box, and memory card network. Memory card: A single historical correction record. Memory card box: A cluster of semantically similar memory cards. Memory card network: A graph structure composed of multiple memory card boxes and their relationships.

[0064] III: Extract the correction records generated during the regeneration process and store them in the memory card network to form new memory cards.

[0065] Regarding step III above, after the expert performs interactive modification and passes the factual consistency judgment function, the archiving agent abstracts the error correction behavior into memory cards s and applies the memory feature embedding function. Store it in the memory card network.

[0066] As an optional embodiment, the memory card network is defined as an undirected attribute graph. ,in Represents a collection of memory card boxes (each) This refers to a cluster of semantically similar memory cards, i.e., a card box. For newly acquired memory cards... The following formula is used to calculate its relationship with the center of each card box. Logical similarity:

[0067] when At that time, They are assigned to their respective card boxes, and topological connections are established or strengthened between the relevant card boxes. This enables the structure to evolve automatically. The connecting edge represents the correlation strength between different memory card boxes in terms of error type, indicator object, correction strategy or task scenario. The edge weight increases with the number of historical co-recalls or continuous corrections.

[0068] Next, we will handle new tasks. At that time, the system's semantic and topological joint retrieval mechanism scores the card boxes, and the scoring formula is shown in the following formula:

[0069] in Represents the normalized cosine similarity. This represents the probability of a random walk with restart. Final search results. As a priori constraint input, it is used to suppress hallucinations during the generation process.

[0070] After manual modification passes the consistency check, it triggers updates to the DAG successor nodes and is also stored in the memory card network by the Archivist Agent; automatic optimization, on the other hand, involves the system actively detecting deviations and executing the same memory write process.

[0071] The above embodiments, through a self-driven evolutionary mechanism based on memory card network evolution and joint retrieval scoring, enable the system to accurately recall historical correction paradigms as prior control when processing new tasks, using a semantic and topological joint pathfinding mechanism. This feedback-based closed-loop evolution achieves a leapfrog improvement in the system's generation capabilities, moving from simple data interpretation to continuous convergence of expert knowledge.

[0072] Please see Figure 2 , Figure 2 This is a flowchart illustrating a method for generating simulation result evaluation reports based on graphics and text, provided in an embodiment of this application. Figure 2 As shown, the simulation result evaluation report generation method provided in this application first performs dual-track data ingestion, extracting time series data from heterogeneous time series data and vector knowledge base data from heterogeneous vector knowledge data. Then, anchor agent alignment is performed, using the anchor agent for visual constraints. Subsequently, DAG progressive generation and optimization are performed, generating a text report through DAG state iteration. Furthermore, memory triples can be extracted by extracting and storing the corrected records formed during the regeneration process into a memory card network to form new memory cards.

[0073] The simulation result evaluation report generation method provided in this application embodiment firstly acquires heterogeneous data and processes the heterogeneous data to obtain vector data; wherein, the heterogeneous data includes indicator files and scenario files; then, a drawing logic summary is generated based on the vector data, visualization instruction code is generated based on the drawing logic summary, and the visualization instruction code is run in an isolated execution environment to generate a target chart; finally, the vector data and the structured chart description corresponding to the target chart are input into a large language model to obtain qualitative evaluation text corresponding to the target chart and its underlying data, and further a complete evaluation report is generated, the report content including background, situation, effectiveness analysis, and improvement suggestions.

[0074] This application achieves strict semantic and numerical alignment of data, charts, and text by dual-track vectorization of indicator files and scenario files into numerical summary features and background knowledge vectors. Combined with isolated environment chart self-repair generation, DAG-driven text-image alignment generation, and memory-enhanced bias optimization, it significantly improves the accuracy, interpretability, maintainability, and adaptability of the evaluation report. The text-image mutual verification includes forward structural verification from data to charts, semantic constraint generation from charts to text, and reverse consistency verification from text to underlying data.

[0075] Please see Figure 3 and 4 , Figure 3 This is one of the structural schematic diagrams of a simulation result evaluation report generation device provided in an embodiment of this application. Figure 4 This is a second schematic diagram of a device for generating a simulation result evaluation report, provided in an embodiment of this application. Figure 3 As shown, the generating apparatus 200 includes: The data processing module 301 is used to acquire heterogeneous data and process the heterogeneous data to obtain vector data; wherein the heterogeneous data includes indicator files and scenario files; wherein the vector data includes numerical summary features and background knowledge vectors. Chart generation module 302 is used to generate a drawing logic summary based on the vector data, generate visualization instruction code based on the drawing logic summary, and run the visualization instruction code in an isolated execution environment to generate the target chart; The evaluation text generation module 303 is used to input the vector data and the structured chart description corresponding to the target chart into the large language model to obtain the qualitative evaluation text corresponding to the target chart.

[0076] Furthermore, the numerical summary features are used for chart generation and numerical verification, and the background knowledge vector is used for semantic retrieval and contextual constraints; when the data processing module 301 processes the heterogeneous data to obtain vector data, the data processing module 301 is also used for: After converting the index file from spreadsheet format to structured data format, it is divided into data segments corresponding to multiple time steps according to the time sequence of the simulation run. Then, a set of key points reflecting the changing trend and statistical features characterizing the distribution characteristics are extracted from the data fragments to obtain the numerical summary features; The hypothetical file is segmented into semantic units, and the hypothetical file is mapped to a vector space through a text embedding function to obtain the background knowledge vector; The numerical summary features and the background knowledge vector together constitute the vector data.

[0077] Furthermore, when the chart generation module 302 is used to run the visualization instruction code in an isolated execution environment to generate the target chart, the chart generation module 302 is also used to: The initial script is generated by calling a preset code template based on the drawing logic summary. The anchor agent runs the initial script in the isolated execution environment and monitors the execution status, structure information and chart rendering effect of the initial script in real time. When the process fails or the chart does not meet the visual structure constraints, the reasoning agent combines the error message with the visual structure constraints to iteratively rewrite the script until a target chart that meets the requirements is generated.

[0078] Furthermore, the qualitative assessment text is generated through a report generation process organized using a directed acyclic graph structure; In the directed acyclic graph structure, each node represents an evaluation subtask, and the nodes are connected by directed edges to indicate the order of evaluation subtasks. The generation of each node is performed by a reasoning agent, and the execution of the reasoning agent is based on the content of the node's predecessor node, the background information matched from the knowledge vector base, and the historical similar reasoning patterns retrieved from the continuously evolving memory card network.

[0079] Furthermore, the generation device 300 also includes a modification module 304, which, after obtaining the qualitative evaluation text, is used to: In response to a user's modification operation on any segment of the qualitative evaluation text, the modified content is obtained, or the child nodes of the directed acyclic graph are automatically subjected to reverse self-modification checks to obtain the modified content. Perform causal logic verification on the modified content to determine whether it conflicts with the underlying data subset; If no conflict is determined, then based on the modified content and the topological connection relationship of the directed acyclic graph structure, all successor nodes that depend on the modified node are automatically identified, triggering the regeneration of the successor nodes.

[0080] Furthermore, the generation device 300 also includes an optimization module 305, which, after obtaining the qualitative evaluation text, is used to: In response to the automatic optimization instruction for the qualitative evaluation text, the anchor agent performs reverse numerical consistency verification on the underlying data corresponding to the qualitative evaluation text. When a discrepancy is found between the text description and the data facts, a historical logical pattern matching the type of the discrepancy is retrieved from the memory card network, and this logical pattern is input as a hard constraint signal into the reasoning agent to drive the reasoning agent to regenerate the discrepancy paragraph. The correction records generated during the regeneration process are extracted and stored in the memory card network to form new memory cards.

[0081] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 As shown, the electronic device 500 includes a processor 510, a memory 520, and a bus 550.

[0082] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 is running, the processor 510 and the memory 520 communicate via the bus 550. When the machine-readable instructions are executed by the processor 510, they can perform the operations described above. Figure 1 The steps of generating the simulation result evaluation report in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0083] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of generating the simulation result evaluation report in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0084] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0086] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0087] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0088] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating simulation result evaluation reports based on graphics and text, characterized in that, The method for generating the simulation result evaluation report includes: Heterogeneous data is acquired and processed to obtain vector data; wherein the heterogeneous data includes indicator files and scenario files; wherein the vector data includes numerical summary features and background knowledge vectors; A drawing logic summary is generated based on the vector data, visualization instruction code is generated based on the drawing logic summary, and the visualization instruction code is run in an isolated execution environment to generate the target chart. The vector data and the structured chart description corresponding to the target chart are input into the large language model to obtain the qualitative evaluation text corresponding to the target chart.

2. The method for generating simulation result evaluation reports according to claim 1, characterized in that, The numerical summary features are used for chart generation and numerical verification, and the background knowledge vectors are used for semantic retrieval and contextual constraints; the processing of the heterogeneous data to obtain vector data includes: After converting the index file from spreadsheet format to structured data format, it is divided into data segments corresponding to multiple time steps according to the time sequence of the simulation run. Then, a set of key points reflecting the changing trend and statistical features characterizing the distribution characteristics are extracted from the data fragments to obtain the numerical summary features; The hypothetical file is segmented into semantic units, and the hypothetical file is mapped to a vector space through a text embedding function to obtain the background knowledge vector; The numerical summary features and the background knowledge vector together constitute the vector data.

3. The method for generating simulation result evaluation reports according to claim 1, characterized in that, The step of running the visualization instruction code in an isolated execution environment to generate the target chart includes: The initial script is generated by calling a preset code template based on the drawing logic summary. The anchor agent runs the initial script in the isolated execution environment and monitors the execution status, structure information and chart rendering effect of the initial script in real time. When the process fails or the chart does not meet the visual structure constraints, the reasoning agent combines the error message with the visual structure constraints to iteratively rewrite the script until a target chart that meets the requirements is generated.

4. The method for generating simulation result evaluation reports according to claim 1, characterized in that, The qualitative assessment text is generated through a report generation process organized using a directed acyclic graph structure. In the directed acyclic graph structure, each node represents an evaluation subtask, and the nodes are connected by directed edges to indicate the order of evaluation subtasks. The generation of each node is performed by a reasoning agent, and the execution of the reasoning agent is based on the content of the node's predecessor node, the background information matched from the knowledge vector base, and the historical similar reasoning patterns retrieved from the continuously evolving memory card network.

5. The method for generating a simulation result evaluation report according to claim 4, characterized in that, After obtaining the qualitative evaluation text, the method for generating the simulation result evaluation report further includes: In response to a user's modification operation on any segment of the qualitative evaluation text, the modified content is obtained, or the child nodes of the directed acyclic graph are automatically subjected to a reverse self-modification check to obtain the modified content. Perform causal logic verification on the modified content to determine whether it conflicts with the underlying data subset; If no conflict is determined, then based on the modified content and the topological connection relationship of the directed acyclic graph structure, all successor nodes that depend on the modified node are automatically identified, triggering the regeneration of the successor nodes.

6. The method for generating a simulation result evaluation report according to claim 1, characterized in that, After obtaining the qualitative evaluation text, the method for generating the simulation result evaluation report further includes: In response to the automatic optimization instruction for the qualitative evaluation text, the anchor agent performs reverse numerical consistency verification on the underlying data corresponding to the qualitative evaluation text. When a discrepancy is found between the text description and the data facts, a historical logical pattern matching the type of the discrepancy is retrieved from the memory card network, and this logical pattern is input as a hard constraint signal into the reasoning agent to drive the reasoning agent to regenerate the discrepancy paragraph. The correction records generated during the regeneration process are extracted and stored in the memory card network to form new memory cards.

7. A simulation result evaluation report generation device based on graphics and text, characterized in that, The simulation result evaluation report generation device includes: A data processing module is used to acquire heterogeneous data and process the heterogeneous data to obtain vector data; wherein the heterogeneous data includes indicator files and scenario files; wherein the vector data includes numerical summary features and background knowledge vectors; The chart generation module is used to generate a drawing logic summary based on the vector data, generate visualization instruction code based on the drawing logic summary, and run the visualization instruction code in an isolated execution environment to generate the target chart. The evaluation text generation module is used to input the vector data and the structured chart description corresponding to the target chart into the large language model to obtain the qualitative evaluation text corresponding to the target chart.

8. The simulation result evaluation report generation device according to claim 7, characterized in that, The numerical summary features are used for chart generation and numerical verification, and the background knowledge vectors are used for semantic retrieval and contextual constraints; when the data processing module processes the heterogeneous data to obtain vector data, the data processing module is also used for: After converting the index file from spreadsheet format to structured data format, it is divided into data segments corresponding to multiple time steps according to the time sequence of the simulation run. Then, a set of key points reflecting the changing trend and statistical features characterizing the distribution characteristics are extracted from the data fragments to obtain the numerical summary features; The hypothetical file is segmented into semantic units, and the hypothetical file is mapped to a vector space through a text embedding function to obtain the background knowledge vector; The numerical summary features and the background knowledge vector together constitute the vector data.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the graph-based simulation result evaluation report generation method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method for generating a simulation result evaluation report based on graphics and text as described in any one of claims 1 to 6.

Images