A method, device, equipment and medium for generating a mineral prediction workflow
By using intent recognition and large model generation technologies, the mineral prediction workflow is automatically generated, solving the problem of relying on human experience in existing technologies. This achieves an efficient and standardized mineral prediction workflow, which is suitable for large-scale and multi-scale mineral prediction tasks under complex geological conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG LAB
- Filing Date
- 2026-06-24
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies rely on human experience in mineral prediction, resulting in low efficiency and high costs. There is a lack of automated methods for a complete mineral prediction workflow, making it difficult to meet the needs of efficient large-scale, multi-scale mineral prediction tasks under complex geological conditions.
By obtaining the user's input description of the mineral prediction task, the generation strategy is determined using intent recognition technology. The initial mineral prediction workflow is generated by combining a template generation strategy or an operator recall generation strategy with a large model, and then verified to ensure the standardization and executability of the workflow.
It achieves efficient workflow generation without manual processing, supports template generation and operator recall, improves workflow generation efficiency and quality, and meets the needs of large-scale, multi-scale mineral prediction tasks under complex geological conditions.
Smart Images
Figure CN122431654A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of mineral prediction technology, and in particular to a method, apparatus, equipment and medium for generating a mineral prediction workflow. Background Technology
[0002] Mineral resource prediction is a core link in mineral exploration breakthroughs. Traditional methods rely heavily on the human experience of geological experts and require the integration of heterogeneous data from multiple sources, including geology, geophysics, and geochemistry. This results in problems such as low work efficiency, significant subjective interference, and high costs.
[0003] Existing automated assistance methods are mostly limited to single data processing or single algorithm calls, lacking a approach for a complete mineral prediction workflow. The mineral prediction process typically involves multiple stages, such as data preprocessing, extraction of mineralized features, inference of geological patterns, and delineation of prospecting target areas. The coordination and connection between these stages and the construction of the workflow still rely heavily on manual operation, resulting in low overall efficiency and making it difficult to meet the needs of efficient large-scale, multi-scale mineral prediction tasks under complex geological conditions.
[0004] Therefore, how to efficiently obtain high-quality mineral forecasting workflows is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] In view of this, one aspect of this application provides a method for generating a mineral prediction workflow, the method comprising: Obtain the task description of the mineral prediction task input by the user; The task description is subjected to intent recognition to determine the generation strategy for the mineral prediction workflow; wherein, the generation strategy includes a template generation strategy and an operator recall generation strategy; the template generation strategy is a strategy of generating a workflow based on a workflow template pre-stored in an expert knowledge base; the operator recall generation strategy is a strategy of recalling the target tools of the mineral prediction task from the expert knowledge base and combining them to form a workflow. Based on the generation strategy, an initial mineral prediction workflow is generated by specifying a large model; The initial mineral prediction workflow is validated to obtain the target mineral prediction workflow.
[0006] Optionally, the template generation strategy includes a complete template generation strategy and a single template generation strategy; The workflow templates include complete workflow templates and single workflow templates; The expert knowledge base includes a tool library, a complete template library, and a single template library.
[0007] Optionally, the step of performing intent recognition on the task description to determine the generation strategy for the mineral prediction workflow includes: The task description is parsed using natural language to obtain the parsing results; Based on the analysis results, determine whether the user's intent is to obtain a complete mineral prediction workflow; If so, then the generation strategy is determined to be the complete template generation strategy; If not, proceed with the following steps: The task description is scored for similarity with the templates in the single template library; and the highest similarity value is selected. Determine whether the highest similarity value is greater than the similarity threshold; If it is greater than, then the generation strategy is determined to be the single template generation strategy; If the difference is small, then the generation strategy is determined to be the operator recall generation strategy.
[0008] Optionally, when the generation strategy is the single template generation strategy; the step of generating an initial mineral prediction workflow by specifying a large model according to the generation strategy includes: Based on the highest similarity value, the target single workflow template is retrieved from the single template library; Based on the task description and the target single workflow template, generate a process-oriented process description; The initial mineral prediction workflow is generated by taking the target single workflow template, the tool library, and the process description as inputs to the specified large model.
[0009] Optionally, when the generation strategy is the complete template generation strategy; the step of generating an initial mineral prediction workflow by specifying a large model according to the generation strategy includes: Extract key information from the task description; the key information includes at least the prediction method. According to the prediction method, the target complete workflow template is loaded from the expert knowledge base; By using the key information, the target complete workflow template is populated to obtain the global workflow; Determine whether there are any missing local workflows in the global workflow; If it exists, the single template generation strategy is invoked to generate the local workflow; and the global workflow and the local workflow are merged to obtain the initial mineral prediction workflow. If it does not exist, the global workflow is used as the input to the specified large model to generate the initial mineral prediction workflow.
[0010] Optionally, when the generation strategy is the operator recall generation strategy; the step of generating an initial mineral prediction workflow by specifying a large model according to the generation strategy includes: The relevance between the task description and the tools in the tool library is scored; and initial tools with a relevance value greater than the relevance threshold are selected. Based on the correlation value, the initial tools are sorted in descending order; and according to the sorting result, the first specified number of tools are selected as the target tools. Based on the target tool, the initial mineral prediction workflow is formed by combining the specified large model according to preset rules.
[0011] Optionally, the step of validating the initial mineral prediction workflow to obtain the target mineral prediction workflow includes: Illegal nodes are removed from the initial mineral prediction workflow; illegal nodes include nodes that are illegally connected between data nodes and between algorithm nodes; data nodes are used to represent data objects, and algorithm nodes are used to represent data processing operations. Name verification is performed on each of the algorithm nodes; and erroneous names are corrected to names that exist in the algorithm library using a semantic fuzzy matching algorithm. For nodes in the initial mineral prediction workflow that include multiple inputs and / or multiple outputs, the input and output formats are validated.
[0012] Another aspect of this application provides an apparatus for generating a mineral prediction workflow, the apparatus comprising: The task description acquisition module is used to acquire the task description of the mineral prediction task input by the user. A generation strategy determination module is used to perform intent recognition on the task description to determine the generation strategy for the mineral prediction workflow; wherein, the generation strategy includes a template generation strategy and an operator recall generation strategy; the template generation strategy is a strategy of generating a workflow based on a workflow template pre-stored in an expert knowledge base; the operator recall generation strategy is a strategy of recalling the target tools for the mineral prediction task from the expert knowledge base and combining them to form a workflow. The workflow generation module is used to generate an initial mineral prediction workflow by specifying a large model according to the generation strategy. The workflow verification module is used to verify the initial mineral prediction workflow to obtain the target mineral prediction workflow.
[0013] Another aspect of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the computer program to implement the steps of the method for generating the mineral prediction workflow.
[0014] Another aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for generating the mineral prediction workflow.
[0015] The method, apparatus, equipment, and medium for generating a mineral prediction workflow provided in this application offer the following advantages: It automatically determines the generation strategy through intent recognition and generates the workflow using a large model based on the strategy, eliminating the need for manual processing at any stage, reducing reliance on expert experience, and improving workflow generation efficiency. Simultaneously, it supports multiple strategies, including template generation and operator recall, enabling the rapid generation of high-quality workflows using pre-stored workflow templates and the rapid generation of workflows for new tasks, overcoming the limitations of single data processing or single algorithm invocation. Furthermore, it validates the initially generated workflow to ensure its standardization and executability, improving workflow quality and thus meeting the needs of large-scale, multi-scale mineral prediction tasks under complex geological conditions. Attached Figure Description
[0016] Figure 1 A flowchart illustrating a method for generating a mineral prediction workflow provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a mineral prediction workflow generation system provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the principle of determining a generation strategy provided in an embodiment of this application. Figure 4 This is a schematic diagram of an initial mineral prediction workflow provided in an embodiment of this application; Figure 5 A schematic diagram illustrating the principle of generating an initial mineral prediction workflow provided in an embodiment of this application; Figure 6 A schematic diagram of an initial mineral prediction workflow provided for another embodiment of this application; Figure 6 (a) is Figure 6 An enlarged schematic diagram of the first part; Figure 6 (b) is Figure 6 An enlarged schematic diagram of the second part; Figure 6 (c) is Figure 6 An enlarged schematic diagram of the third part; Figure 6 (d) is Figure 6 An enlarged schematic diagram of the fourth part; Figure 6 (e) is Figure 6Enlarged schematic diagram of the fifth section; Figure 6 (f) is Figure 6 An enlarged schematic diagram of the sixth section; Figure 6 (g) is Figure 6 An enlarged schematic diagram of the seventh section; Figure 7 A schematic diagram of a device for generating a mineral prediction workflow provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.
[0017] The attached diagram is labeled as follows: 20 is the expert knowledge base, 21 is the generative reasoning module, 200 is the tool library, 201 is the complete template library, 202 is the single template library, 210 is the natural language processing function, 211 is the large model, 70 is the task description acquisition module, 71 is the generation strategy determination module, 72 is the workflow generation module, 73 is the workflow verification module, 80 is the memory, 81 is the processor, 82 is the display screen, 83 is the input / output interface, 84 is the communication interface, 85 is the power supply, 86 is the communication bus, 801 is the computer program, 802 is the operating system, and 803 is the data. Detailed Implementation
[0018] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.
[0019] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."
[0020] Figure 1 This is a flowchart illustrating a method for generating a mineral prediction workflow provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes: S10: Obtain the task description of the mineral prediction task input by the user; First, it should be noted that the generation method provided in this application embodiment can be applied to servers, electronic devices, or cloud computing platforms to automatically generate an executable workflow based on user-inputted mineral prediction tasks. Furthermore, the executing entity of this method can be any computer device deployed with an expert knowledge base 20 and a Large Language Model (LLM).
[0021] Figure 2 This is a schematic diagram of a mineral prediction workflow generation system provided in an embodiment of this application. In an optional embodiment, the main body executing the generation method provided in this application is... Figure 2 The system shown includes an expert knowledge base 20 and a generative reasoning module 21. The following will combine... Figure 2 Please provide a detailed explanation.
[0022] In a specific embodiment, the user can... Figure 2 The system's interactive interface allows users to input a task description regarding their mineral prediction needs. This interface may include, but is not limited to, a command line, a web page, or an API interface. The user-input task description can be natural language text or spoken language, which will be converted into descriptive text. This application does not limit the format of the input task description. For ease of understanding, examples of input task descriptions for mineral prediction are provided below.
[0023] For example, the task description in the natural language text entered by the user is "Based on the rock mass strategy, use the 'Geological Map of the West Qinling Region' data to obtain a raster map of the influence range of favorable rock masses for mineralization in the West Qinling Region", or "Use the fuzzy weight of evidence method to make a complete process prediction of the distribution of Au element minerals in the Jiaodong working area".
[0024] S11: Perform intent recognition on the task description to determine the generation strategy for the mineral prediction workflow; wherein, the generation strategy includes a template generation strategy and an operator recall generation strategy; the template generation strategy is a strategy to generate a workflow based on a workflow template pre-stored in the expert knowledge base 20; the operator recall generation strategy is a strategy to recall the target tools for the mineral prediction task from the expert knowledge base 20 and combine them to form a workflow. After obtaining the task description of the mineral prediction task, through Figure 2 The system's natural language processing function 210, as shown, performs natural language processing on the task description to identify user intent and determine the appropriate method for generating the workflow. The template generation strategy is suitable for scenarios where user needs highly match existing workflow templates, enabling rapid generation of standard processes. The operator recall generation strategy, on the other hand, is suitable for scenarios where no existing templates exist, requiring dynamic recall tools to combine them into new processes.
[0025] Therefore, it can be understood that in specific embodiments, Figure 2 The expert knowledge base 20 shown includes a large number of workflow templates, so that when it is determined that the user's intention is to generate a workflow based on an existing template, the corresponding template can be quickly retrieved from the knowledge base to generate a standard process.
[0026] S12: Generate an initial mineral prediction workflow by specifying a large model based on the generation strategy; like Figure 2 As shown, the generation inference module 21 in the mineral prediction workflow generation system deploys one or more large models 211, which have the ability to understand task descriptions, combine tools, and generate workflows that conform to format standards.
[0027] Therefore, according to the above generation strategy, when the user's requirement is to generate a workflow based on an existing template, the generation inference module 21 retrieves the corresponding workflow template from the expert knowledge base 20 and performs inference based on the deployed specified large model 211 to generate an initial mineral prediction workflow. When the user's requirement does not have a ready-made workflow template in the expert knowledge base 20, the specified large model 211 is used to understand the task description, combine tools, and generate an initial mineral prediction workflow that conforms to the format standard.
[0028] The initial mineral prediction workflow includes node name declaration, node type format declaration, node type declaration (including start node, data type node, algorithm type node, and end node), and node connection declaration.
[0029] It should be noted that the large model 211 deployed in the generative inference module 21 may include, but is not limited to, the GPT (Generative Pre-trained Transformer) series, the BERT model, large models based on the Transformer architecture, LLaMA (Large Language Model Meta AI), and Tongyi Qianwen. This application does not limit the large model.
[0030] Of course, in an optional embodiment, in order to ensure the generation accuracy of the mineral prediction workflow, the large model 211 can be a vertical domain large model trained with a large amount of mineral domain expertise, and this application does not limit this.
[0031] In addition, it should be noted that for the expert knowledge base 20, existing expert knowledge can be imported into the expert knowledge base module through experts. Existing expert knowledge includes, but is not limited to, all tools that are allowed to be called by the workflow and their corresponding tool usage instructions, existing workflow templates, tool name verification templates, etc.
[0032] S13: Verify the initial mineral prediction workflow to obtain the target mineral prediction workflow.
[0033] Furthermore, considering that the generated initial mineral prediction workflow may contain issues such as illegal nodes, incorrect tool names, or mismatched input / output formats, as an optional embodiment, the initial mineral prediction workflow is validated to improve its quality. This validation includes, but is not limited to, removing illegal nodes, correcting incorrect tool names, and adjusting input / output formats. This application does not limit the scope of the validation process.
[0034] Therefore, the mineral prediction workflow generation method provided in this application automatically determines the generation strategy through intent recognition and generates the workflow using a large model 211 based on the generation strategy. This eliminates the need for manual processing at any stage, reducing reliance on expert experience and improving workflow generation efficiency. Simultaneously, it supports multiple strategies, including template generation and operator recall. This allows for the rapid generation of high-quality workflows using pre-stored workflow templates, as well as the rapid generation of workflows for new tasks, overcoming the limitations of single data processing or single algorithm invocation. Furthermore, the initially generated workflow is validated to ensure its standardization and executability, improving workflow quality and thus meeting the needs of efficient large-scale, multi-scale mineral prediction tasks under complex geological conditions.
[0035] In one optional embodiment, the template generation strategy includes a complete template generation strategy and a single template generation strategy; the workflow template includes a complete workflow template and a single workflow template; such as Figure 2 As shown, the expert knowledge base 20 includes a tool library 200, a complete template library 201, and a single template library 202.
[0036] In one alternative embodiment, the expert knowledge base 20 can be built using the open-source RAGFlow architecture, which can be specifically divided into three parts, such as... Figure 2 The sub-libraries shown include tool library 200, complete template library 201, and single template library 202. Among them, tool library 200 is used to store all tools that can be called by the workflow and their usage instructions.
[0037] The single template library 202 is used to store reusable single workflow templates, such as partial processes like "element singularity analysis" and "structural line direction extraction". The complete template library 201 is used to store complete mineral exploration process templates, such as "complete prediction process using the evidence right method".
[0038] Therefore, the template generation strategies in the above embodiments can be categorized into a complete template generation strategy and a single template generation strategy. The complete template generation strategy generates a workflow based on a pre-stored complete workflow template in the complete template library 201. The single template generation strategy generates a workflow based on a pre-stored single workflow template in the single template library 202.
[0039] In a specific implementation, when the user intends to obtain a complete workflow covering the entire chain from data preprocessing to target area delineation, a complete template generation strategy is adopted; when the user only needs to obtain a certain partial process, a single template generation strategy is adopted.
[0040] Therefore, by dividing templates into complete templates and single templates, user needs can be matched more accurately, redundant or insufficient workflows can be avoided, and the efficiency and quality of workflow generation can be improved.
[0041] Figure 3 This is a schematic diagram illustrating the principle of determining a generation strategy provided in an embodiment of this application. Based on the above embodiments, as an optional embodiment, intent recognition is performed on the task description to determine the generation strategy for the mineral prediction workflow, including: The task description is parsed using natural language to obtain the parsing results; Based on the analysis results, determine whether the user's intent is to obtain a complete mineral prediction workflow; If so, then the generation strategy is determined to be the complete template generation strategy; If not, proceed with the following steps: The task description is scored for similarity with templates in a single template library; and the highest similarity value is selected. Determine whether the highest similarity value is greater than the similarity threshold; If it is greater than, then the generation strategy is determined to be a single template generation strategy; If the difference is small, then the generation strategy is determined to be the operator recall generation strategy.
[0042] like Figure 3 As shown, in a specific embodiment, by Figure 2 The natural language processing function in the reasoning module 21 performs natural language parsing on the task description (i.e., the user's question about the required mineral prediction process) to obtain the parsing result. Specifically, the natural language processing method can be keyword extraction, such as detecting keywords like "complete process" or "full process prediction", or it can be an intent recognition classifier based on the large model 211. This application does not limit this approach.
[0043] Further, see Figure 3Based on the parsing results, determine whether the user's intent is to obtain a complete mineral prediction workflow. If the determination result is "yes", then the generation strategy is determined to be the complete template generation strategy, and the process ends.
[0044] If the judgment result is "no", it is necessary to further confirm whether the generation strategy is a single template generation strategy or an operator recall generation strategy. Specifically, such as... Figure 3 As shown, a similarity score is calculated between the task description and all templates in the single template library 202 of the expert knowledge base 20, resulting in a similarity value. The similarity value characterizes the degree of similarity between the task description and each template in the single template library 202. A higher similarity value indicates a higher degree of similarity, signifying a greater match between the corresponding single template and the template used to generate the current workflow.
[0045] Furthermore, based on the calculated similarity values, the highest similarity value is selected. Then, the highest similarity value is compared with a pre-set similarity threshold. If the highest similarity value is greater than the similarity threshold, the generation strategy is determined to be a single template generation strategy; if it is not greater, the generation strategy is determined to be an operator recall generation strategy, thus completing the process of determining the generation strategy.
[0046] It is worth noting that the similarity threshold can be set during the parameter setting stage, for example, it can be set to 0.3. The similarity threshold can be adjusted according to actual business needs, and this application does not impose any restrictions on it. Furthermore, regarding... Figure 2 The parameter setting phase of the system also includes setting parameters for the expert knowledge base 20, large model inference parameters, tool recall parameters, and workflow format standards.
[0047] Based on the above embodiments, when the generation strategy is a single template generation strategy, as an optional embodiment, according to the generation strategy, an initial mineral prediction workflow is generated by specifying a large model, including: Based on the highest similarity value, retrieve the target single workflow template from a single template library; Generate a process-oriented description based on the task description and the single-goal workflow template; The initial mineral prediction workflow is generated by taking a single target workflow template, toolkit, and process description as input to a specified large model.
[0048] Specifically, in a specific embodiment, when the generation strategy label is "single template generation", the target single workflow template corresponding to the highest similarity value is retrieved from the single template library 202 based on the highest similarity value in the above embodiment. Figure 2 The system shown can generate a process-oriented, structured process description based on the user's task description and a single target workflow template through the large model 211.
[0049] For example, a task description can be transformed into a flowchart describing the following three steps: Step 1: Central Log-Ratio Transform. The input is the vector point file "deposits-reprojection". The process performed is to use the Central Log-Ratio (CLR) transformation function to eliminate the closure effect of the data. The output is the generated vector point file "Geochemical Transformation".
[0050] Step 2: Inverse Distance Weighted Interpolation. The input is the vector point file "Geochemical Transformation". The process involves selecting the W elements from the "Geochemical Transformation" as the object and performing interpolation using the Inverse Distance Weighted Interpolation (IDW) method. Specific parameter settings include a distance power of 2, a smoothing factor of 0, and a search radius of 2000. The resulting output is a raster file named "w element interpolation".
[0051] Step 3: Local Singularity Analysis. The input is the raster file "w element interpolation". The process performed is to use the Local Singularity Analysis (LSA) function, where the minimum window radius is set to 3, the maximum window radius to 15, and the window growth step size to 2. The output is the analysis result for generating the raster file "w element (LSA)".
[0052] Furthermore, the target single workflow template, tool library, and the generated process description are input together into the large model 211 in the generation inference module 21. The large model 211 then generates a complete single workflow according to the specified workflow output format and the input / output specifications of the calling tools.
[0053] Figure 4 This is a schematic diagram of an initial mineral prediction workflow provided in an embodiment of this application. In an optional embodiment, when the task description of the mineral prediction task input by the user is "based on the rock mass strategy, use the 'Geological Map of the West Qinling Region' data to obtain a raster map of the influence range of favorable mineralized rock masses in the West Qinling Region", and the generation strategy is a single template generation strategy, an initial mineral prediction workflow including only one input and one output can be generated according to the above method.
[0054] Figure 5 This is a schematic diagram illustrating the principle of generating an initial mineral prediction workflow provided in an embodiment of this application. In another optional embodiment, when the generation strategy is a complete template generation strategy; according to the generation strategy, the initial mineral prediction workflow is generated by specifying a large model, including: Extract key information from the task description; the key information should include at least the prediction method. Based on the prediction method, load the target complete workflow template from the expert knowledge base; By filling in the target complete workflow template with key information, the global workflow is obtained; Determine if any local workflows are missing from the global workflow; If it exists, invoke the single template generation strategy to generate a local workflow; then merge the global workflow and the local workflow to obtain the initial mineral prediction workflow; If it does not exist, use the global workflow as input to the specified large model to generate the initial mineral prediction workflow.
[0055] In a specific embodiment, such as Figure 5 As shown, key information is extracted from the task description. This key information includes at least the prediction method, and also includes, but is not limited to, geographical location information, target mineral type for mineral exploration prediction, prediction range, prior information, prediction elements, tag files, and initial data. For ease of understanding, examples will be provided below.
[0056] For example, the user-input task description is "Using the fuzzy evidence weight method, based on the Jiaodong geological map, geochemical exploration, and aeromagnetic data... to conduct a complete process prediction of the distribution of Au ore in the Jiaodong working area." Furthermore... Figure 2 The system shown extracts the prediction method as "fuzzy evidence weight method" through the large model 211, and extracts geographical location information (Jiaodong working area), target mineral type (Au), prediction element list (aeromagnetic anomaly grid, favorable rock mass influence range grid, etc.), prior information (mineral occurrence influence range), initial data (geological map, geochemical data), etc.
[0057] After obtaining the prediction method, further, such as Figure 5 As shown, based on the prediction method, a complete target workflow template is loaded from the expert knowledge base 20. For example, the fuzzy evidence weight method corresponds to a complete mineral exploration process template, which includes multiple stages such as data preprocessing, feature extraction, vector to raster conversion, and prediction probability raster estimation.
[0058] Using the key information extracted above, the target complete workflow template is populated to obtain the global workflow. It should be noted that, in specific embodiments, populating includes, but is not limited to, filling specific data names, parameter values, etc., into the corresponding nodes of the template.
[0059] See Figure 5 To further improve the quality of the mineral prediction workflow, after obtaining the global workflow, it is determined whether there are any missing local workloads. For example, some complete templates may only define the main framework, while a certain sub-step (such as the extraction of abnormal isovalue regions of specific elements) does not provide a specific implementation method. In this case, it is necessary to call a single template generation strategy (such as a single-element abnormal graph generation strategy) to complete it.
[0060] In one optional embodiment, if local gaps exist in the global workflow, a single template generation strategy is invoked to generate the local workflow. The global workflow and the local workflow are then merged to obtain the initial mineral prediction workflow. It is worth noting that in the specific embodiment of merging the global and local workflows, duplicate data nodes in different local workflows need to be merged to avoid redundancy.
[0061] In another alternative embodiment, if there are no local missing features, the global workflow is used as input to the specified large model 211 to generate an initial mineral prediction workflow.
[0062] It should be noted that, in one optional implementation, when determining whether a local workflow is missing, the system can traverse each node of the global workflow and check if any node is marked as "to be implemented" or lacks the corresponding algorithm tool, thereby determining whether a local workflow is missing. This application does not limit the specific method for determining the missing local workflow.
[0063] Figure 6 This is a schematic diagram of an initial mineral prediction workflow provided in another embodiment of this application. Figure 6 (a) is Figure 6 An enlarged schematic diagram of the first part. Figure 6 (b) is Figure 6 An enlarged schematic diagram of the second part of the image. Figure 6 (c) is Figure 6 An enlarged schematic diagram of the third part. Figure 6 (d) is Figure 6 Enlarged schematic diagram of the fourth part. Figure 6 (e) is Figure 6 An enlarged schematic diagram of the fifth part. Figure 6 (f) is Figure 6 Enlarged schematic diagram of the sixth part of the diagram. Figure 6 (g) is Figure 6 An enlarged schematic diagram of the seventh section. For ease of understanding, the following will combine... Figure 6 as well as Figure 6 (a) to Figure 6 (g) Provide an example.
[0064] For example, in one optional embodiment, the user-input task description is: "Using the fuzzy evidence weight method, based on the Jiaodong geological map, geochemical and aeromagnetic data, with the 'Jiaodong working area' data as the prediction range and the 'Jiaodong area mineral occurrence influence range' data as the prior information, and using the data of 'Jiaodong area aeromagnetic anomaly raster,' 'Jiaodong area_favorable rock mass influence range graded raster,' 'Jiaodong area_mineral-controlling fault influence range binary raster,' 'Jiaodong area Au (gold) element anomaly graded raster_traditional,' 'Jiaodong area Ag (silver) element anomaly graded raster_traditional,' 'Jiaodong area Hg (mercury) element anomaly graded raster_traditional,' 'Jiaodong area As (arsenic) element anomaly graded raster_traditional,' and 'Jiaodong area Sb (antimony) element anomaly graded raster_traditional' as prediction elements, a complete process prediction of the distribution of Au ore in the Jiaodong working area is performed." After determining the current generation strategy as a complete template generation strategy using the above method, a complete template generation strategy can be generated. Figure 6 The initial mineral prediction workflow is shown.
[0065] like Figure 6 As shown, the initial mineral prediction workflow generated based on the complete template generation strategy includes multiple individual workflows; that is, the complete mineral prediction workflow consists of multiple individual workflows. To clearly illustrate the complete mineral prediction workflow exemplified above, [the following is a list of workflows]. Figure 6 Multiple enlarged details are shown in the diagram. Figure 6 (a) to Figure 6 (g).
[0066] It should be noted that, for Figure 6 (c) Figure 6 (d) and Figure 6 (e) To clearly illustrate the workload involved, the placement of nodes in the workflow has been adjusted, but the connection relationships and order between the nodes remain the same. Figure 6 Consistent.
[0067] Therefore, the generation method provided in this application first loads a complete template and fills in key information, then detects whether there are any local omissions, and calls a single template generation strategy according to the omissions, and finally merges to generate a complete workflow, thereby realizing the fully automated construction of complex mineral prediction tasks and ensuring the integrity, reliability and coherence of the workflow.
[0068] In another alternative embodiment, when the generation strategy is an operator recall generation strategy; based on the generation strategy, an initial mineral prediction workflow is generated by specifying a large model, including: The relevance scores between the task description and the tools in the tool library are evaluated, and initial tools with relevance scores greater than the relevance threshold are selected. Based on the relevance score, the initial tools are sorted in descending order; and according to the sorting result, the first specified number of tools are selected as target tools. Based on the target tool, an initial mineral prediction workflow is formed by combining a specified large model according to preset rules.
[0069] In a specific embodiment, the relevance score is calculated between the task description and the tools in the tool library 200 to obtain a relevance value. The relevance value is used to characterize the degree of relevance between the task description and the various tools in the tool library 200. The higher the relevance value, the higher the degree of relevance, that is, the relevance value and the degree of relevance are positively correlated.
[0070] Based on the relevance score, initial tools with relevance scores greater than the relevance threshold are selected. Specifically, Figure 2 The system shown uses the large model 211 to create a decision chain for each tool in the tool library 200, giving the correlation value between the tool and the user task description (e.g., a value between 0 and 1), and using the tools with correlation values greater than a preset correlation threshold (e.g., 0.7) as the initial toolset.
[0071] Furthermore, based on the relevance values, the initial tools are sorted in descending order, and the top specified number of tools are selected as target tools according to the sorting results. This specified number can be determined by the "allowed tool recall quantity" parameter in the pre-set tool recall parameters, for example, it can be set to 20. In other words, under the operator recall generation strategy, the tool recall parameters, including the tool recall quantity and the preset relevance threshold, are crucial to the quality of the final generated workflow.
[0072] Based on the target tools, an initial mineral prediction workflow is formed by combining the specified large model 211 according to preset rules. Specifically, the large model 211 connects the selected target tools in an orderly manner according to the data flow, based on the semantic logic of the task description. For example, in an optional embodiment, reprojection is performed first, followed by buffer analysis, then masking, and finally resampling, thereby generating a complete operator chain workflow.
[0073] It should be noted that, in specific embodiments, when screening initial tools, the system can not only screen by the relevance threshold, but also make weighted adjustments based on the tool's historical call frequency and user preferences. This application does not limit this.
[0074] In addition, it should be noted that, in an optional embodiment, when selecting a specified number of tools after sorting in descending order, if the number of tools with the same relevance score exceeds the remaining quota, they can be randomly selected or all can be retained.
[0075] Therefore, the generation method provided in this application, by combining an expert knowledge base with large-scale model reasoning, achieves intelligent classification of user-input questions and automatic generation of processes.
[0076] In one optional embodiment, the initial mineral prediction workflow is validated to obtain the target mineral prediction workflow, including: Illegal nodes are removed from the initial mineral prediction workflow. Illegal nodes include nodes that make illegal connections between data nodes and between algorithm nodes. Data nodes are used to represent data objects, and algorithm nodes are used to represent data processing operations. Name verification is performed on each algorithm node; and erroneous names are corrected to names that exist in the algorithm library using a semantic fuzzy matching algorithm. For nodes in the initial mineral prediction workflow that include multiple inputs and / or multiple outputs, perform input / output format validation.
[0077] In a specific embodiment, the initial mineral prediction workflow is parsed, and all nodes and the connections between them are checked. If an illegal connection exists, such as "data node D1 → data node D2 (e.g., D*** → |data|D*** in mermaid format)" or "algorithm node A1 → algorithm node A2 (e.g., A*** → |analysis|A*** in mermaid format)," then that connection edge and its associated isolated nodes are deleted. In fact, in this specific embodiment, legal connections are an alternating pattern of "data node → algorithm node → data node."
[0078] It should be noted that data nodes are used to represent data objects, such as raster data and vector data, while algorithm nodes are used to represent data processing operations, such as resampling and buffer analysis. In an optional embodiment, the system can use a script (such as Python) to traverse the directed edges of the workflow, check the node types at both ends of the edge, and automatically delete illegal edges (i.e., illegal connections).
[0079] In another optional implementation, the names of all algorithm nodes in the workflow are extracted and compared with the algorithm names existing in the tool's knowledge base. For mismatched names, a semantic fuzzy matching algorithm (such as similarity calculation based on edit distance or word vectors) is used to find the closest correct name and automatically correct it. If no match can be found, the user can be prompted for intervention.
[0080] In another alternative embodiment, for algorithms with multiple inputs or multiple outputs (such as the "mask" algorithm which requires two input data nodes), the number and data format of their input and output edges are checked to ensure they meet the algorithm requirements. If not, the large model 211 automatically adjusts the connections or adds necessary transformation nodes.
[0081] Therefore, the method for generating a mineral prediction workflow provided in this application, through a multi-verification mechanism, effectively corrects structural errors, naming errors, and input-output mismatches in the large model 211 generation workflow, thereby improving the quality of the workflow.
[0082] In the above embodiments, the method for generating a mineral prediction workflow has been described in detail. This application also provides an embodiment of a device for generating a mineral prediction workflow.
[0083] Figure 7 This is a schematic diagram of the structure of a mineral prediction workflow generation device provided in an embodiment of this application, as shown below. Figure 7 As shown, the device includes: The task description acquisition module 70 is used to acquire the task description of the mineral prediction task input by the user. The generation strategy determination module 71 is used to perform intent recognition on the task description in order to determine the generation strategy of the mineral prediction workflow. The generation strategy includes a template generation strategy and an operator recall generation strategy. The template generation strategy is a strategy to generate a workflow based on a workflow template pre-stored in the expert knowledge base. The operator recall generation strategy is a strategy to recall the target tools of the mineral prediction task from the expert knowledge base and combine them to form a workflow. Workflow generation module 72 is used to generate an initial mineral prediction workflow based on a specified large model according to the generation strategy. The workflow verification module 73 is used to verify the initial mineral prediction workflow to obtain the target mineral prediction workflow.
[0084] The template generation strategy includes a complete template generation strategy and a single template generation strategy; the workflow template includes a complete workflow template and a single workflow template; and the expert knowledge base includes a tool library, a complete template library, and a single template library.
[0085] Furthermore, the mineral prediction workflow generation apparatus provided in this application embodiment also includes: The parsing module is used to perform natural language parsing on the task description and obtain the parsing results; The intent determination module is used to determine whether the user's intent is to obtain a complete mineral prediction workflow based on the parsing results; if so, the generation strategy is determined to be the complete template generation strategy; otherwise, the target processing module is called. The target processing module is used to score the similarity between the task description and the templates in the single template library; and to select the highest similarity value; to determine whether the highest similarity value is greater than the similarity threshold; if it is greater, the generation strategy is determined to be the single template generation strategy; if it is not greater, the generation strategy is determined to be the operator recall generation strategy.
[0086] When the generation strategy is a single template generation strategy, the process description generation module, the target template retrieval module, and the first workflow generation sub-module are invoked. The target template retrieval module is used to retrieve a target single workflow template from a single template library based on the highest similarity value. The process description generation module is used to generate process-oriented process descriptions based on task descriptions and target single workflow templates. The first workflow generation submodule is used to generate an initial mineral prediction workflow by taking the target single workflow template, tool library and process description as input to the specified large model.
[0087] When the generation strategy is a complete template generation strategy, the key information extraction module, template loading module, filling module, and missing information handling module are invoked. The key information extraction module is used to extract key information from the task description; the key information includes at least the prediction method. The template loading module is used to load the target complete workflow template from the expert knowledge base according to the prediction method; The populate module is used to populate the target complete workflow template with key information to obtain the global workflow; The missing workflow handling module is used to determine whether there are missing local workflows in the global workflow. If there are, a single template generation strategy is invoked to generate the local workflow. The global workflow and the local workflow are then merged to obtain the initial mineral prediction workflow. If there are no missing local workflows, the global workflow is used as the input of a specified large model to generate the initial mineral prediction workflow.
[0088] When the generation strategy is the operator recall generation strategy, the relevance scoring module, the sorting module, and the second workflow generation submodule are invoked. The relevance scoring module is used to score the relevance between the task description and the tools in the tool library; and to filter out the initial tools whose relevance values are greater than the relevance threshold. The sorting module is used to sort the initial tools in descending order based on their relevance values; and to select the first specified number of tools as target tools based on the sorting results. The second workflow generation submodule is used to form an initial mineral prediction workflow by combining a specified large model according to preset rules based on the target tool.
[0089] The node removal module is used to remove illegal nodes in the initial mineral prediction workflow. Illegal nodes include nodes that form illegal connections between data nodes and between algorithm nodes. Data nodes are used to represent data objects, and algorithm nodes are used to represent data processing operations. The name verification module is used to verify the names of each algorithm node; and corrects erroneous names to names that exist in the algorithm library through a semantic fuzzy matching algorithm. The format verification module is used to verify the input and output formats of nodes in the initial mineral prediction workflow that include multiple inputs and / or multiple outputs.
[0090] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 8 As shown, the electronic device includes: a memory 80 for storing computer programs; The processor 81 is configured to execute a computer program to implement the steps of the method for generating a mineral prediction workflow as described in the above embodiments.
[0091] The electronic devices provided in this embodiment may include, but are not limited to, laptops or desktop computers.
[0092] The processor 81 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 81 may be implemented using at least one of the following hardware forms: Digital Signal Processor (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 81 may also include a main processor and a coprocessor. The main processor, also known as the Central Processing Unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 81 may integrate a Graphics Processing Unit (GPU), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 81 may also include an Artificial Intelligence (AI) processor, which is used to handle computational operations related to machine learning.
[0093] The memory 80 may include one or more computer-readable storage media, which may be non-transitory. The memory 80 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory 80 is used to store at least the following computer program 801, which, after being loaded and executed by the processor 81, is capable of implementing the relevant steps of the mineral prediction workflow generation method disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 80 may also include an operating system 802 and data 803, etc., and the storage method may be temporary storage or permanent storage. The operating system 802 may include Windows, Unix, Linux, etc. The data 803 may include, but is not limited to, the relevant data involved in the mineral prediction workflow generation method.
[0094] In some embodiments, the electronic device may further include a display screen 82, an input / output interface 83, a communication interface 84, a power supply 85, and a communication bus 86.
[0095] Those skilled in the art will understand that Figure 8 The structures shown do not constitute a limitation on electronic devices and may include more or fewer components than those shown.
[0096] The electronic device provided in this application includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the method for generating the mineral prediction workflow in the above embodiments.
[0097] It should be noted that although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.
Claims
1. A method for generating a mineral prediction workflow, characterized in that, The method includes: Obtain the task description of the mineral prediction task input by the user; The task description is subjected to intent recognition to determine the generation strategy for the mineral prediction workflow; wherein, the generation strategy includes a template generation strategy and an operator recall generation strategy; the template generation strategy is a strategy of generating a workflow based on a workflow template pre-stored in an expert knowledge base; the operator recall generation strategy is a strategy of recalling the target tools of the mineral prediction task from the expert knowledge base and combining them to form a workflow. Based on the generation strategy, an initial mineral prediction workflow is generated by specifying a large model; The initial mineral prediction workflow is validated to obtain the target mineral prediction workflow.
2. The method for generating a mineral prediction workflow as described in claim 1, characterized in that, The template generation strategy includes a complete template generation strategy and a single template generation strategy; The workflow templates include complete workflow templates and single workflow templates; The expert knowledge base includes a tool library, a complete template library, and a single template library.
3. The method for generating a mineral prediction workflow as described in claim 2, characterized in that, The process of performing intent recognition on the task description to determine the generation strategy for the mineral prediction workflow includes: The task description is parsed using natural language to obtain the parsing results; Based on the analysis results, determine whether the user's intent is to obtain a complete mineral prediction workflow; If so, then the generation strategy is determined to be the complete template generation strategy; If not, proceed with the following steps: The task description is scored for similarity with the templates in the single template library; and the highest similarity value is selected. Determine whether the highest similarity value is greater than the similarity threshold; If it is greater than, then the generation strategy is determined to be the single template generation strategy; If the difference is small, then the generation strategy is determined to be the operator recall generation strategy.
4. The method for generating a mineral prediction workflow as described in claim 3, characterized in that, When the generation strategy is the single template generation strategy; the step of generating an initial mineral prediction workflow by specifying a large model according to the generation strategy includes: Based on the highest similarity value, the target single workflow template is retrieved from the single template library; Based on the task description and the target single workflow template, generate a process-oriented process description; The initial mineral prediction workflow is generated by taking the target single workflow template, the tool library, and the process description as inputs to the specified large model.
5. The method for generating a mineral prediction workflow as described in claim 2, characterized in that, When the generation strategy is the complete template generation strategy; the step of generating an initial mineral prediction workflow by specifying a large model according to the generation strategy includes: Extract key information from the task description; the key information includes at least the prediction method. According to the prediction method, the target complete workflow template is loaded from the expert knowledge base; By using the key information, the target complete workflow template is populated to obtain the global workflow; Determine whether there are any missing local workflows in the global workflow; If it exists, the single template generation strategy is invoked to generate the local workflow; and the global workflow and the local workflow are merged to obtain the initial mineral prediction workflow. If it does not exist, the global workflow is used as the input to the specified large model to generate the initial mineral prediction workflow.
6. The method for generating a mineral prediction workflow as described in claim 2, characterized in that, When the generation strategy is the operator recall generation strategy; the step of generating an initial mineral prediction workflow by specifying a large model according to the generation strategy includes: The relevance between the task description and the tools in the tool library is scored; and initial tools with a relevance value greater than the relevance threshold are selected. Based on the correlation value, the initial tools are sorted in descending order; and according to the sorting result, the first specified number of tools are selected as the target tools. Based on the target tool, the initial mineral prediction workflow is formed by combining the specified large model according to preset rules.
7. The method for generating a mineral prediction workflow as described in claim 1, characterized in that, The step of validating the initial mineral prediction workflow to obtain the target mineral prediction workflow includes: Illegal nodes are removed from the initial mineral prediction workflow; illegal nodes include nodes that are illegally connected between data nodes and between algorithm nodes; data nodes are used to represent data objects, and algorithm nodes are used to represent data processing operations. Name verification is performed on each of the algorithm nodes; and erroneous names are corrected to names that exist in the algorithm library using a semantic fuzzy matching algorithm. For nodes in the initial mineral prediction workflow that include multiple inputs and / or multiple outputs, the input and output formats are validated.
8. A device for generating a mineral prediction workflow, characterized in that, The device includes: The task description acquisition module is used to acquire the task description of the mineral prediction task input by the user. A generation strategy determination module is used to perform intent recognition on the task description to determine the generation strategy for the mineral prediction workflow; wherein, the generation strategy includes a template generation strategy and an operator recall generation strategy; the template generation strategy is a strategy of generating a workflow based on a workflow template pre-stored in an expert knowledge base; the operator recall generation strategy is a strategy of recalling the target tools for the mineral prediction task from the expert knowledge base and combining them to form a workflow. The workflow generation module is used to generate an initial mineral prediction workflow by specifying a large model according to the generation strategy. The workflow verification module is used to verify the initial mineral prediction workflow to obtain the target mineral prediction workflow.
9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for generating a mineral prediction workflow according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method for generating the mineral prediction workflow according to any one of claims 1 to 7.