GIS workflow optimization method based on dynamic experience learning

Abstract

The invention discloses a GIS (Geographic Information System) workflow optimization method based on dynamic experience learning, which comprises the following steps of: firstly, receiving a natural language GIS task request input by a user, analyzing the natural language GIS task request into a structured task description by utilizing a large language model, and generating candidate GIS workflows in combination with a static GIS tool; searching historical workflow execution experience matched with the candidate GIS workflow from a dynamically maintained workflow experience knowledge base, and optimizing the candidate GIS workflow under the assistance of the large language model based on the searched historical workflow execution experience to obtain optimized GIS workflow; and finally, executing the optimized GIS workflow, capturing feedback information in the execution process, and updating the feedback information into the workflow experience knowledge base. According to the method, the adaptability to agriculture and forestry data diversity and task complexity is enhanced, the operation threshold of complex agriculture and forestry GIS tasks is reduced, and the user trust is enhanced.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence and geographic information science and technology, and in particular relates to a GIS workflow optimization method based on dynamic experience learning. Background Technology

[0002] Geographic Information Systems (GIS), as a core tool for processing and analyzing spatial data, are becoming increasingly powerful, but this also brings with it increased operational complexity. Its applications in agriculture and forestry, such as crop growth monitoring, forest resource surveys, pest and disease early warning, and disaster assessment, are particularly prominent. In recent years, the rise of Large Language Models (LLM) has brought hope for simplifying GIS operations and enabling interactive spatial analysis using natural language.

[0003] Existing research has proposed embedding LLM into existing GIS platforms (such as QGIS) to enable spatial analysis workflows driven by natural language. These systems typically include modules for query optimization, RAG-based (RAG) tool retrieval, data understanding, code generation, and real-time error debugging. They can transform user requests into executable sequences of GIS operations and automate processes to a certain extent.

[0004] However, despite the progress made in automation by the aforementioned systems, they still have limitations in handling the deep complexities of agroforestry geospatial analysis tasks. These limitations mainly stem from the fact that LLMs themselves may lack in-depth GIS domain knowledge and agroforestry professional practical experience. For example, the system may struggle to anticipate potential hidden problems in specific combinations of agroforestry data (such as remote sensing imagery and ground survey data from different time periods) or operational sequences (such as complex model chains for crop yield estimation based on multi-source data). These problems may include distorted vegetation indices due to ineffective atmospheric correction of image data, geometric calculation errors due to coordinate system mismatch, or subtle requirements of specific tools on input data formats such as forest compartment zoning Shapefile files. Even if syntax errors can be resolved through on-the-spot debugging, it may be difficult to fundamentally optimize the overall efficiency and robustness of the workflow.

[0005] Furthermore, current automation relies primarily on understanding the static documentation of tools and basic error-correction logic, lacking a mechanism that allows agents to learn from the successes and failures of each task execution and use these lessons to guide future task planning, especially in agricultural and forestry applications that require a high degree of domain knowledge and practical experience.

[0006] Therefore, improving its intelligence level and enabling it to better assist in complex agricultural and forestry geospatial analysis remains a key challenge in this research field. Summary of the Invention

[0007] The purpose of this application is to provide a GIS workflow optimization method based on dynamic experience learning, overcoming the shortcomings of existing technologies in GIS workflows for agricultural and forestry tasks. To achieve the above objectives, the technical solution of this application is as follows: A GIS workflow optimization method based on dynamic experience learning includes: Step 1: Receive the natural language GIS task request input by the user, parse it into a structured task description using a large language model, and generate candidate GIS workflows using static GIS tools. Step 2: Retrieve historical workflow execution experience that matches the candidate GIS workflow from the dynamically maintained workflow experience knowledge base; Step 3: Based on the retrieved historical workflow execution experience, and with the assistance of a large language model, optimize the candidate GIS workflow to obtain the optimized GIS workflow; Step 4: Execute the optimized GIS workflow and capture feedback information during the execution process, updating it to the workflow experience knowledge base.

[0008] Preferably, after step 1, the method further includes: Perform a preliminary internal logical consistency check on the candidate GIS workflow. If the check fails, return to re-execute step 1.

[0009] Preferably, the workflow experience knowledge base stores historically executed GIS workflows and their complete contextual information, including: task profiles, workflow details, execution results and performance, solutions, or user correction records.

[0010] Preferably, the optimization of candidate GIS workflows based on retrieved historical workflow execution experience, with the assistance of a large language model, to obtain an optimized GIS workflow includes: Utilize large language models to analyze retrieved historical workflow execution experience; Based on the analysis results, key operations were pre-set for candidate GIS workflows, tool parameters were optimized, and workflow paths were reconstructed. The candidate GIS workflows are sorted and selected to obtain the optimized GIS workflow.

[0011] Preferably, the step of capturing feedback information during the execution process and updating it in the workflow experience knowledge base includes: Receive feedback information, clean and denoise the feedback information, and use a large language model to assist in information extraction and abstraction; The refined and abstracted information is formatted according to the predefined item structure of the workflow experience knowledge base; Write the formatted data into the workflow experience knowledge base.

[0012] Preferably, the GIS workflow optimization method based on dynamic experience learning further includes: Before or during the execution of the optimized GIS workflow, review and correct any syntax and basic logic errors in the optimized GIS workflow.

[0013] The GIS workflow optimization method based on dynamic experience learning proposed in this application has at least the following beneficial effects: Significantly enhance the "agricultural and forestry practice wisdom" and problem-predicting capabilities of GIS task automation agents: By building and dynamically maintaining a workflow experience knowledge base and utilizing an experience-driven analysis and optimization engine, the agent can learn from the successes and failures of historical executions, proactively identify and avoid potential operational risks (such as common errors in specific remote sensing image data types or crop growth stages, inefficient vegetation index calculation parameter settings, etc.), rather than simply relying on understanding the static documentation of the tools or reactive debugging of errors that have occurred.

[0014] Achieving proactive optimization and continuous self-improvement of agricultural and forestry GIS workflows: Instead of simply generating a "usable" workflow, it proactively selects or constructs agricultural and forestry analysis workflow solutions with expected higher success rates, better execution efficiency, or higher result quality based on insights driven by historical data. As the experience knowledge base continues to grow, task planning and decision-making capabilities can continuously evolve, demonstrating true "self-growth" characteristics.

[0015] Further lowering the operational threshold for complex agricultural and forestry GIS tasks and enhancing user trust: Through smarter and more robust workflow generation, even users lacking deep GIS expertise or agricultural and forestry backgrounds can successfully complete more complex spatial analysis tasks (such as regional crop yield prediction based on multi-source data) with the help of agents. By drawing on historical experience (especially successful user modification examples) for optimization, its behavior is more in line with agricultural and forestry GIS practices, thus making it easier to gain user recognition and trust.

[0016] It enhances the adaptability to the diversity of agricultural and forestry data and the complexity of tasks: The solutions accumulated in the experience knowledge base for different crop types, forest stand structures, remote sensing data source characteristics, agricultural and forestry task types (such as drought monitoring and soil nutrient assessment) and error scenarios enable the agent to have richer coping strategies and stronger adaptability when facing new, non-standardized or edge situations, thereby improving the generalization ability and practical value of the entire system. Attached Figure Description

[0017] Figure 1 This is a flowchart of the GIS workflow optimization method based on dynamic experience learning proposed in this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] Example 1, as Figure 1 As shown, a GIS workflow optimization method based on dynamic experience learning is provided, including: Step S1: Receive the natural language GIS task request input by the user, parse it into a structured task description using a large language model, and generate candidate GIS workflows by combining static GIS tools.

[0020] Users input natural language GIS task requests through a natural language interface, such as: identify which areas of my land have poor corn growth. Then, using Large Language Modeling (LLM) and static GIS tools, one or more candidate GIS workflows are generated.

[0021] Specifically, a large language model is used to perform deep analysis on the natural language GIS task request, transforming it into a structured task description. This description clarifies the analysis objectives (such as identifying areas of poor corn growth), the objects of operation, and related constraints. The objects of operation and related constraints can be obtained from a stored database based on the task request or input by the user; they can be represented by UAV multispectral imagery and plot boundaries.

[0022] Then, the data features of the operation object are extracted. For example, by reading the metadata of the image file to obtain its band information, spatial resolution, coordinate system, etc., or by analyzing the attribute table structure of the vector data. Next, based on the aforementioned structured task description and the extracted data features, a series of related GIS tools (such as vegetation index calculation tools and reclassification tools) are retrieved from a preset GIS tool knowledge base that includes information such as functions, parameters, and calling methods.

[0023] These retrieved GIS tools are organized into a logically coherent sequence of operations, and corresponding initial execution code is generated, thus forming one or more preliminary candidate GIS workflows. This candidate workflow is a collection of tool operation sequences and their execution code, defining the initial technical path to fulfill the user's request.

[0024] Step S2: Retrieve historical workflow execution experience that matches the candidate GIS workflow from the dynamically maintained workflow experience knowledge base.

[0025] This embodiment dynamically maintains a workflow experience knowledge base. The information stored in the workflow experience knowledge base includes: agricultural and forestry task types (such as crop yield estimation, forest fire risk assessment, and pest and disease monitoring), abstract characteristics of input data (such as remote sensing data sources, image resolution, spatiotemporal distribution of ground survey data, geometric type of forest compartments or plots, coordinate system status, and data quality issues such as cloud cover), as well as user preferences or constraints (such as prioritizing fast-growing forest areas and excluding protected areas); as well as the success / failure status of workflow execution, specific error information encountered (such as errors in band calculation formulas, and deviations in terrain factor calculations caused by insufficient DEM data accuracy), the steps used for error correction, processing time, and (if available) quality assessment of output results (such as verification accuracy with real ground data and the Kappa coefficient of classification results).

[0026] The workflow experience knowledge base is used to structure and store historically executed agricultural and forestry GIS workflows and their complete contextual information. Each experience entry follows a predefined structure, which includes at least: (1) Task profile: An abstract description of the task, including the type of agricultural and forestry task (such as "forest resource inventory" and "crop disease identification"), the characteristics of the input data (such as "Landsat 8 multispectral image with cloud cover less than 10%), and the description of the user's goal.

[0027] (2) Workflow details: The specific sequence of tools and parameter configurations to be executed (e.g., “Calculate NDVI using bands 4 and 5, and extract vegetation by threshold segmentation”).

[0028] (3) Execution results and effectiveness: Record the success / failure status of the workflow, the specific error information or type encountered (such as "NDVI value calculation is abnormal, possibly due to input band error"), processing time, and quality assessment of the output results (such as "85% agreement with ground survey points").

[0029] (4) Solutions or User Correction Records: Effective remedial measures for failures or inefficiencies (such as "correcting the NDVI formula to (NIR-Red) / (NIR+Red)"), or content and effects manually adjusted by users to optimize results. This knowledge base is preferably organized using a graph database or a relational database with specific semantic indexes to support efficient similarity calculation and pattern mining.

[0030] During the retrieval process, the "task profile" is first extracted from the candidate GIS workflow to be executed. This task profile is an abstract representation of the candidate workflow context, containing information such as task type, input data characteristics, and core objectives. Then, using this task profile as query criteria, historical experience entries with high similarity are retrieved from the experience knowledge base to obtain a set of relevant historical workflow execution experiences. These experiences contain all the detailed information mentioned above (task profile, workflow details, execution results, solutions, etc.).

[0031] Step S3: Based on the retrieved historical workflow execution experience, and with the assistance of a large language model, optimize the candidate GIS workflow to obtain the optimized GIS workflow.

[0032] This step optimizes candidate GIS workflows based on the retrieved historical workflow execution experience.

[0033] In a specific embodiment, the optimization process is as follows: Step 3.1: Analyze the retrieved historical workflow execution experience using a large language model.

[0034] By leveraging the historical workflow execution experience retrieved through Large Language Modeling (LLM), we can deeply identify success patterns and failure pitfalls related to the current task. Success patterns include "For wheat rust monitoring, using the red edge index (NDRE) combined with texture features yields better results"; failure pitfalls include "Failure to normalize images from multiple periods leads to unreliable change detection results".

[0035] Step 3.2: Based on the analysis results, perform pre-set key operations, optimize tool parameters, and reconstruct workflow paths for candidate GIS workflows.

[0036] This embodiment lists some specific technical means for optimizing candidate GIS workflows based on analysis results, including: Pre-set key operations: Based on historical experience, necessary preprocessing operations are automatically inserted before the core analysis steps, such as radiometric calibration and atmospheric correction for remote sensing data, or verification, image declouding, and data interpolation for common data problems.

[0037] Optimize tool parameters: Refer to the parameter settings in historical successful cases (such as the number of training samples for supervised classification and the threshold of a specific vegetation index) to fine-tune the tool parameters in the current workflow.

[0038] Refactor workflow paths: If historical experience reveals that there is a more efficient workflow structure (such as faster, more stable, or higher quality results) to complete similar tasks, then borrow or directly adopt that structure to refactor the current workflow.

[0039] Step 3.3: Sort and select the best candidate GIS workflows to obtain the optimized GIS workflow.

[0040] Sorting and Selection: If multiple feasible optimized workflow solutions are generated through modification, they are sorted according to the comprehensive performance indicators of each solution in historical experience (such as weighted average success rate, average execution time, expected result quality, etc.), and the solution with the highest ranking is selected as the final output optimized GIS workflow.

[0041] Step S4: Execute the optimized GIS workflow, capture feedback information during the execution process, and update it to the workflow experience knowledge base.

[0042] This step optimizes the GIS workflow and comprehensively and meticulously captures various feedback information during the execution process. For example, it monitors the execution status of each step in the workflow, records performance indicators (such as time consumption and resource usage), captures any errors or warnings, and records real-time user interactions or manual corrections in specific scenarios (such as a user adjusting the health level classification threshold after viewing the NDVI results).

[0043] To achieve a closed-loop guarantee of continuous evolution, this application utilizes LLM to assist in the extraction of feedback information and the abstraction of experience (for example, summarizing general error patterns from specific error logs, such as "the range of input raster data cell values ​​does not meet the tool requirements", or summarizing parameter sensitivity rules from successful user parameter adjustments, such as "for coniferous forests, the sensitive band combination of a certain disease index is X, Y, Z"). Then, according to the predefined structure of the knowledge base, these newly learned, structured experiences are securely written into the knowledge base, or used to update and correct existing relevant experience entries in the knowledge base, thereby achieving dynamic growth and optimization of the knowledge base.

[0044] In one specific embodiment, the step of capturing feedback information during the execution process and updating it in the workflow experience knowledge base includes: Step 4.1: Receive feedback information, clean and denoise the feedback information, and use a large language model to assist in information extraction and abstraction.

[0045] For example, general error cause patterns can be summarized from specific error logs (such as "the range of input raster data cell values ​​does not meet the tool requirements"), or parameter sensitivity rules can be summarized from successful parameter adjustments by users (such as "for coniferous forests, the sensitive band combination of a certain disease index is X, Y, Z").

[0046] Step 4.2: Format the refined and abstracted information according to the predefined item structure of the workflow experience knowledge base.

[0047] For example, the refined and abstracted information, including new success / failure cases, solutions, user wisdom, etc., is formatted according to the predefined entry structure of the workflow experience knowledge base.

[0048] Step 4.3: Write the formatted data into the workflow experience knowledge base.

[0049] This step securely writes new or updated existing experiences into the workflow experience knowledge base and updates the knowledge base index as needed to ensure that newly learned knowledge can be effectively used in future experience retrieval processes.

[0050] Example 2, based on Example 1, receives a natural language GIS task request input by the user, parses it into a structured task description using a large language model, and generates candidate GIS workflows using static GIS tools. It also includes: Perform a preliminary internal logical consistency check on the candidate GIS workflow. If the check fails, return to re-execute step S1.

[0051] Specifically, after generating candidate GIS workflows, an initial internal logical consistency check is performed, judging the rationality of tool connections based on basic GIS operation rules (such as data type matching and spatial reference consistency). If obvious logical problems are found, the process returns to re-parse the task or re-plan the workflow.

[0052] Example 3, based on the above examples, the GIS workflow optimization method based on dynamic experience learning further includes: Before or during the execution of the optimized GIS workflow, review and correct any syntax and basic logic errors in the optimized GIS workflow.

[0053] In this embodiment, syntax and basic logic errors in the code are reviewed and corrected before and during execution.

[0054] The technical solution proposed in this application can construct a complete closed loop from task understanding, preliminary planning, experience optimization, execution feedback to experience learning, so that the ability of agricultural and forestry GIS intelligent analysis agent can be continuously improved as its "experience" accumulates.

[0055] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A GIS workflow optimization method based on dynamic experience learning, characterized in that, The GIS workflow optimization method based on dynamic experience learning includes: Step 1: Receive the natural language GIS task request input by the user, parse it into a structured task description using a large language model, and generate candidate GIS workflows using static GIS tools. Step 2: Retrieve historical workflow execution experience that matches the candidate GIS workflow from the dynamically maintained workflow experience knowledge base; Step 3: Based on the retrieved historical workflow execution experience, and with the assistance of a large language model, optimize the candidate GIS workflow to obtain the optimized GIS workflow; Step 4: Execute the optimized GIS workflow and capture feedback information during the execution process, updating it to the workflow experience knowledge base.

2. The GIS workflow optimization method based on dynamic experience learning according to claim 1, characterized in that, Following step 1, the following is also included: Perform a preliminary internal logical consistency check on the candidate GIS workflow. If the check fails, return to re-execute step 1.

3. The GIS workflow optimization method based on dynamic experience learning according to claim 1, characterized in that, The workflow experience knowledge base stores historically executed GIS workflows and their complete contextual information, including: task profiles, workflow details, execution results and performance, solutions, or user correction records.

4. The GIS workflow optimization method based on dynamic experience learning according to claim 1, characterized in that, Based on the retrieved historical workflow execution experience, and with the assistance of a large language model, the candidate GIS workflow is optimized to obtain an optimized GIS workflow, including: Utilize large language models to analyze retrieved historical workflow execution experience; Based on the analysis results, key operations were pre-set for candidate GIS workflows, tool parameters were optimized, and workflow paths were reconstructed. The candidate GIS workflows are sorted and selected to obtain the optimized GIS workflow.

5. The GIS workflow optimization method based on dynamic experience learning according to claim 1, characterized in that, The feedback information captured during the execution process is updated in the workflow experience knowledge base, including: Receive feedback information, clean and denoise the feedback information, and use a large language model to assist in information extraction and abstraction; The refined and abstracted information is formatted according to the predefined item structure of the workflow experience knowledge base; Write the formatted data into the workflow experience knowledge base.

6. The GIS workflow optimization method based on dynamic experience learning according to claim 1, characterized in that, The GIS workflow optimization method based on dynamic experience learning also includes: Before or during the execution of the optimized GIS workflow, review and correct any syntax and basic logic errors in the optimized GIS workflow.

Images