A method and system for intelligent manufacturing of seismic exploration equipment based on a large language model, a terminal, and a storage medium

Abstract

This invention relates to the field of data processing technology and discloses a method, system, terminal, and storage medium for intelligent manufacturing of seismic exploration equipment based on a large language model. The method includes: acquiring multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment and processing the data to obtain target multi-source seismic data; constructing a knowledge graph to obtain a marine geological knowledge graph; determining the target large language model and using retrieval enhancement technology to perform autonomous learning processing based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model; acquiring current seismic profile data, performing autonomous decision-making processing, outputting seismic exploration parameter configuration suggestions, and adjusting the initial seismic exploration parameters of the seismic exploration equipment; and optimizing the seismic exploration parameter adjustment model based on actual exploration data. This invention significantly improves the imaging resolution, penetration capability, and data acquisition efficiency of seismic exploration equipment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, terminal, and computer-readable storage medium for intelligent manufacturing of seismic exploration equipment based on a large language model. Background Technology

[0002] Marine geological exploration is a crucial foundation for engineering activities such as offshore oil and gas resource development, submarine wind farm layout, submarine pipeline laying, and geological hazard monitoring. With the continuous expansion of marine engineering projects, higher demands are being placed on the accuracy, efficiency, and intelligence of geological exploration. Among these, marine seismic exploration, as a mainstream geological structure imaging method, has been widely applied in many fields.

[0003] However, existing marine seismic exploration systems generally use source and receiver arrays with fixed parameter configurations. Key parameter ranges such as source frequency, power, and receiver spacing are preset before the equipment leaves the factory. In sea areas with relatively simple geological conditions, basic exploration targets can be achieved. However, in complex geological environments such as wind power exploration, submarine landslides, and shallow gas-rich areas, conventional parameter combinations are difficult to achieve high-quality imaging. Imaging resolution and penetration depth often cannot be balanced, affecting the accuracy of geological target identification and exploration efficiency.

[0004] Therefore, there is an urgent need for an intelligent manufacturing method and system with self-learning and parameter self-adjustment capabilities to meet the pressing need for high-resolution seismic imaging in variable marine environments and to enable automatic configuration and precise operation of seismic exploration equipment in complex geological scenarios. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent manufacturing method, system, terminal, and computer-readable storage medium for seismic exploration equipment based on a large language model. This aims to address the problem that existing marine seismic exploration systems generally use source and receiver arrays with fixed parameter configurations, lack adaptive adjustment mechanisms, and are unable to meet the requirements for refined imaging in complex geological environments.

[0006] To achieve the above objectives, the present invention provides an intelligent manufacturing method for seismic exploration equipment based on a large language model, the method comprising the following steps:

[0007] Acquire multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment, and process the multi-source seismic data to obtain target multi-source seismic data;

[0008] A knowledge graph is constructed based on the target multi-source seismic data to obtain a marine geological knowledge graph;

[0009] A target large language model is determined, and a retrieval enhancement technique is used to autonomously learn the target large language model based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model.

[0010] The system acquires current seismic profile data, performs autonomous decision-making based on the current seismic profile data through the seismic exploration parameter adjustment model, outputs seismic exploration parameter configuration suggestions, and adjusts the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment.

[0011] Acquire actual exploration data and use a reinforcement learning strategy to optimize the seismic exploration parameter adjustment model mounted on the target seismic exploration equipment based on the actual exploration data.

[0012] Optionally, in the intelligent manufacturing method for seismic exploration equipment based on a large language model, the data processing includes preprocessing and reprocessing.

[0013] The process of acquiring multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment and processing the multi-source seismic data to obtain target multi-source seismic data specifically includes:

[0014] Determine the initial seismic exploration parameters of the seismic exploration equipment, acquire reflected wave signals based on the initial seismic exploration parameters, and obtain the exploration environment corresponding to the reflected wave signals. The initial seismic exploration parameters include the initial source frequency and the initial receiver array deployment method.

[0015] Acquire existing seismic exploration data, and construct multi-source seismic data based on the existing seismic exploration data, the initial seismic exploration parameters, the reflected wave signal, and the exploration environment;

[0016] The multi-source seismic data is preprocessed and reprocessed to obtain the target multi-source seismic data;

[0017] The preprocessing includes cleaning, standardization, anomaly detection, and error correction.

[0018] The reprocessing includes standardization, format conversion, noise reduction and completion, and accuracy verification.

[0019] Optionally, the intelligent manufacturing method for seismic exploration equipment based on a large language model, wherein the step of constructing a knowledge graph based on the target multi-source seismic data to obtain a marine geological knowledge graph specifically includes:

[0020] Determine map nodes, wherein the map nodes include stratigraphic units, reflectance features, sedimentary types, equipment parameters, and imaging effects;

[0021] A knowledge graph is constructed from the target multi-source seismic data using a triplet model based on the graph nodes, resulting in a marine geological knowledge graph. The triplet in the triplet model includes source parameters, geological features, and exploration results.

[0022] Optionally, in the intelligent manufacturing method for seismic exploration equipment based on a large language model, determining the target large language model specifically includes:

[0023] Obtain preset geological environment and preset seismic source parameters, and encode the preset geological environment and preset seismic source parameters to obtain an encoded sequence;

[0024] A preset large language model is determined, and the preset large language model is trained according to the encoding sequence to obtain an initial large language model;

[0025] The initial large language model was fine-tuned using LoRA technology to obtain the target large language model.

[0026] Optionally, the intelligent manufacturing method for seismic exploration equipment based on a large language model, wherein the step of using retrieval enhancement technology to train the target large language model based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model, further includes:

[0027] The vector data in the marine geological knowledge graph is obtained, and the similarity of the vector data is calculated using a hierarchical clustering algorithm to obtain the similarity results;

[0028] The vector data is categorized based on the similarity results to obtain visualized data clusters.

[0029] Add tag metadata to each data sample in the visualized data cluster, wherein the tag metadata includes the region, data type, exploration target, stratigraphic age, acquisition method, and source batch.

[0030] Optionally, the intelligent manufacturing method for seismic exploration equipment based on a large language model, wherein the step of using retrieval enhancement technology to train the target large language model based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model specifically includes:

[0031] Obtain the user query statement and input the user query statement into the target large language model;

[0032] The target large language model is used to perform keyword parsing and weight calculation on the user query statement to obtain initial candidate text;

[0033] The initial candidate texts are subjected to high-dimensional vectorization encoding, semantic similarity calculation, and reordering to obtain semantic matching data.

[0034] Based on the tag metadata in the marine geological knowledge graph, the semantic matching data is targeted for retrieval processing, and the retrieval results are output. The target large language model is trained and the seismic exploration parameter adjustment model is obtained.

[0035] Optionally, the intelligent manufacturing method for seismic exploration equipment based on a large language model, wherein acquiring current seismic profile data, performing autonomous decision-making processing based on the current seismic profile data through the seismic exploration parameter adjustment model, outputting seismic exploration parameter configuration suggestions, and adjusting the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment, specifically includes:

[0036] Acquire the current seismic profile data and input the current seismic profile data into the seismic exploration parameter adjustment model;

[0037] The current seismic profile data is analyzed and processed using the seismic exploration parameter adjustment model to obtain the geological environment and seismic exploration targets.

[0038] Keyword extraction and knowledge graph matching are performed on the geological environment and the seismic exploration target to obtain seismic exploration parameter configuration suggestions. The initial seismic exploration parameters are then adjusted according to the seismic exploration parameter configuration suggestions, which include the target source frequency and the target receiver array deployment method.

[0039] Furthermore, to achieve the above objectives, the present invention also provides an intelligent manufacturing system for seismic exploration equipment based on a large language model, wherein the intelligent manufacturing system for seismic exploration equipment based on a large language model includes:

[0040] The multi-source seismic data acquisition module is used to acquire multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment, and to process the multi-source seismic data to obtain target multi-source seismic data.

[0041] The knowledge graph construction module is used to construct a knowledge graph based on the target multi-source seismic data to obtain a marine geological knowledge graph.

[0042] The autonomous learning module is used to determine the target large language model and to use retrieval enhancement technology to perform autonomous learning processing on the target large language model based on the marine geological knowledge graph to obtain the seismic exploration parameter adjustment model.

[0043] The autonomous decision-making module is used to acquire current seismic profile data, perform autonomous decision-making processing based on the current seismic profile data through the seismic exploration parameter adjustment model, output seismic exploration parameter configuration suggestions, and adjust the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment.

[0044] An optimization feedback module is used to acquire actual exploration data and employ a reinforcement learning strategy to optimize the seismic exploration parameter adjustment model mounted on the target seismic exploration equipment based on the actual exploration data.

[0045] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a large language model-based intelligent manufacturing program for seismic exploration equipment stored in the memory and executable on the processor. When the large language model-based intelligent manufacturing program for seismic exploration equipment is executed by the processor, it implements the steps of the large language model-based intelligent manufacturing method for seismic exploration equipment as described above.

[0046] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a smart manufacturing program for seismic exploration equipment based on a large language model, and when the smart manufacturing program for seismic exploration equipment based on a large language model is executed by a processor, it implements the steps of the smart manufacturing method for seismic exploration equipment based on a large language model as described above.

[0047] In this invention, multi-source seismic data corresponding to the initial seismic exploration parameters of a seismic exploration equipment are acquired, and the multi-source seismic data is processed to obtain target multi-source seismic data. A knowledge graph is constructed based on the target multi-source seismic data to obtain a marine geological knowledge graph. A target large-scale language model is determined, and retrieval enhancement technology is used to autonomously learn the target large-scale language model based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model. Current seismic profile data is acquired, and the seismic exploration parameter adjustment model performs autonomous decision-making based on the current seismic profile data to output seismic exploration parameter configuration suggestions. The initial seismic exploration parameters of the seismic exploration equipment are adjusted according to the seismic exploration parameter configuration suggestions to obtain a target seismic exploration equipment. Actual exploration data is acquired, and a reinforcement learning strategy is used to optimize the seismic exploration parameter adjustment model mounted on the target seismic exploration equipment based on the actual exploration data. This invention constructs a knowledge graph and uses retrieval enhancement technology to combine the knowledge graph with a large language model for autonomous learning, thereby constructing a seismic exploration parameter adjustment model. The seismic exploration parameter adjustment model can make autonomous decisions based on actual geological conditions, thereby automatically adjusting the exploration parameters of marine seismic exploration, significantly improving seismic imaging resolution, data acquisition efficiency, and automation level. Attached Figure Description

[0048] Figure 1 This is a comparison chart of traditional penetration depth and resolution;

[0049] Figure 2 This is the first flowchart of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention;

[0050] Figure 3 This is the second flowchart of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention;

[0051] Figure 4 This is a schematic diagram of the marine geological intelligent manufacturing system architecture of the intelligent manufacturing method for seismic exploration equipment based on a large language model, as described in this invention.

[0052] Figure 5 This is a third flowchart of a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model of the present invention;

[0053] Figure 6 This is a fourth flowchart of a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model of the present invention;

[0054] Figure 7 This is a fifth flowchart of a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model of the present invention;

[0055] Figure 8 This is the sixth flowchart of a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model of the present invention;

[0056] Figure 9 This is the seventh flowchart of a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model of the present invention;

[0057] Figure 10 This is a schematic diagram of the hardware and software architecture and data acquisition module composition of the intelligent manufacturing system of the preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention.

[0058] Figure 11 This is a schematic diagram of the knowledge graph relationship network for optimizing seismic exploration parameters, representing a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention.

[0059] Figure 12 This is a schematic diagram of the intelligent seismic data processing and decision optimization process of a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention.

[0060] Figure 13 This is a schematic diagram of the retrieval and recommendation process of the RAG (Retrieval Enhanced Generation) system, a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention.

[0061] Figure 14 This is a schematic diagram of a multi-level semantic retrieval system, representing a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention.

[0062] Figure 15 This is a schematic diagram of the model fine-tuning process of a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model according to the present invention.

[0063] Figure 16 This is a schematic diagram of the strategy iteration and fine-tuning process of the optimization feedback module in a preferred embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model of the present invention.

[0064] Figure 17 This is a structural diagram of a preferred embodiment of the intelligent manufacturing system for seismic exploration equipment based on a large language model according to the present invention;

[0065] Figure 18 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0067] like Figure 1 As shown, while high-frequency seismic sources offer good vertical resolution and are suitable for imaging shallow details, their signal attenuation is rapid and their penetration depth is limited, making it difficult to detect deep structures. Low-frequency seismic sources, on the other hand, possess strong penetration capabilities and are suitable for mid-to-deep exploration, but their resolution is lower, easily missing shallow geological information. In actual exploration, it is often necessary to simultaneously image both shallow and deep structures, but existing marine seismic exploration systems lack the ability to automatically adjust source and receiver parameters according to different exploration targets, resulting in a trade-off between imaging quality and efficiency.

[0068] Furthermore, the seabed geological environment varies significantly, with different regions exhibiting diverse sediment types, stratigraphic structures, and porosity characteristics. Currently, seismic exploration operations are typically categorized into four types: conventional, high-resolution, ultra-high-resolution, and shallow seismic profiling. Each type employs fixed configurations of source power, frequency, and receiver spacing, lacking the ability to adaptively optimize parameters for specific geological conditions. This results in the ineffective identification of some crucial geological anomalies.

[0069] In summary, existing marine seismic exploration technologies suffer from problems such as fixed source and receiver parameters, limited imaging capabilities, poor environmental adaptability, and a lack of intelligent adjustment mechanisms, making it difficult to meet the technical requirements of next-generation marine resource development and deep-sea complex environment exploration. Therefore, there is an urgent need for an intelligent seismic exploration system that can adaptively adjust and optimize source parameters based on geological environmental characteristics, in order to improve imaging resolution, extend exploration depth, and enhance overall operational efficiency.

[0070] To address the aforementioned issues, this invention proposes an intelligent manufacturing method and system for seismic exploration equipment based on a large language model. This method is particularly suitable for seismic exploration tasks under complex geological conditions (enabling high-precision seismic data acquisition and analysis), improving the resolution and data acquisition efficiency of marine seismic exploration, and enabling precise detection and imaging of seabed geological structures. This invention integrates autonomous learning, autonomous decision-making, and continuous optimization capabilities, enabling intelligent adjustment of source frequency, energy, and receiver deployment parameters. This significantly improves seismic imaging resolution, data acquisition efficiency, and automation levels, making it suitable for various operational needs, including shallow high-resolution stratigraphic exploration and mid-level structural imaging.

[0071] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 2 The first embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model in this invention includes:

[0072] The basic scheme of the present invention:

[0073] Step S10: Obtain multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment, and process the multi-source seismic data to obtain target multi-source seismic data.

[0074] This invention acquires reflected wave signal data collected in real time by seismic exploration equipment during operation, and also acquires data resources from different survey lines, historical operation areas and publicly available scientific research institutions. These data are integrated into multi-source seismic data, which is then preprocessed and further processed before being stored in the database to obtain the target multi-source seismic data.

[0075] Step S20: Construct a knowledge graph based on the target multi-source seismic data to obtain a marine geological knowledge graph.

[0076] To enhance the semantic understanding and associative reasoning capabilities of data, this invention constructs a marine geological knowledge graph based on target multi-source seismic data. The graph reveals the multidimensional logical relationships between data through a graph structure. This graph integrates historical exploration records, scientific research data, and expert annotation results, enabling deep coupling between exploration parameters and geological background, and providing knowledge support for parameter recommendation and intelligent reasoning.

[0077] Step S30: Determine the target large language model, and use retrieval enhancement technology to perform autonomous learning processing on the target large language model based on the marine geological knowledge graph to obtain the seismic exploration parameter adjustment model.

[0078] This invention combines a large language model, a marine geological knowledge graph, and retrieval enhancement technology to perform autonomous learning, thereby constructing a seismic exploration parameter adjustment model. This seismic exploration parameter adjustment model has the ability to identify geological scenes, generate parameter suggestions, and make dynamic corrections, enabling efficient and accurate analysis of user queries and seismic profile data.

[0079] Step S40: Obtain the current seismic profile data, perform autonomous decision-making processing based on the current seismic profile data through the seismic exploration parameter adjustment model, output seismic exploration parameter configuration suggestions, and adjust the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment.

[0080] When the present invention receives real-time acquired seismic profile data, it automatically analyzes the waveform features, reflection structures and stratigraphic interface information through the seismic exploration parameter adjustment model, and performs semantic matching in combination with existing knowledge graphs to determine the geological environment type of the current working area, such as soft sedimentary layer, hard bedrock, landslide or fault zone, thereby achieving accurate output of seismic exploration parameter configuration suggestions (automatically updating source frequency, power and receiver array configuration to generate final seismic exploration equipment parameters).

[0081] Step S50: Obtain actual exploration data and use reinforcement learning strategy to optimize the seismic exploration parameter adjustment model carried in the target seismic exploration equipment based on the actual exploration data.

[0082] This invention is based on an optimized feedback mechanism. It inputs the imaging results obtained during the actual exploration process, the data transmitted back by the equipment, and the user evaluation results into the system. It uses a reinforcement learning strategy to continuously adjust and optimize the seismic exploration parameter adjustment model, thereby improving the accuracy and stability of parameter recommendations and realizing closed-loop adaptive optimization.

[0083] In summary, this invention can intelligently optimize and recommend source frequency, power, and receiver array configuration based on existing seismic profile data, significantly improving the imaging resolution, penetration capability, and data acquisition efficiency of the equipment. This invention possesses continuous optimization capabilities, allowing for dynamic fine-tuning of parameter strategies based on feedback results, making it suitable for the intelligent manufacturing of high-precision seismic exploration equipment for complex seabed geological structures.

[0084] Further technical solutions of the present invention:

[0085] Please see Figure 3 The second embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model in this invention includes:

[0086] S101. Determine the initial seismic exploration parameters of the seismic exploration equipment, collect reflected wave signals according to the initial seismic exploration parameters, and obtain the exploration environment corresponding to the reflected wave signals. The initial seismic exploration parameters include the initial source frequency and the initial receiving array deployment method.

[0087] like Figure 4 As shown, the present invention includes four core modules: a data acquisition module, an autonomous learning module, an autonomous decision-making module, and an optimization feedback module. These modules work together to form an exploration system that is adaptively optimized for dynamic geological environments.

[0088] For the data acquisition module, this invention employs a multi-frequency seismic source and a variable-spacing receiver array to achieve dynamic adjustment of the source frequency within the range of 300Hz to 3000Hz, and supports flexible configuration of the receiver hydrophone array spacing between 0.5m and 2m to meet the needs of seismic exploration at different depths and resolutions. This module can acquire reflected wave signals from seafloor strata in real time and generate high-precision raw seismic profiles, providing a data foundation for subsequent geological interpretation and parameter optimization.

[0089] S102. Obtain existing seismic exploration data, and construct multi-source seismic data based on the existing seismic exploration data, the initial seismic exploration parameters, the reflected wave signal, and the exploration environment;

[0090] In addition to the function of physical acquisition and measurement data, this invention also integrates a structured storage mechanism and semantic parsing framework for instrument parameters, geological structure and unstructured data, supporting the fusion of seismic instrument configuration data (such as source type, receiver deployment method and sampling accuracy) and geological exploration data (such as seismic profiles, well logging data and fault distribution) into the database, thereby building a unified data management system.

[0091] S103. Preprocess and reprocess the multi-source seismic data to obtain the target multi-source seismic data;

[0092] S104, wherein the preprocessing includes cleaning, standardization, anomaly detection, and error correction.

[0093] S105, The reprocessing includes standardization processing, format conversion processing, noise reduction and completion processing, and accuracy verification processing.

[0094] Please see Figure 5 The third embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model in this invention includes:

[0095] S201. Determine the map nodes, wherein the map nodes include stratigraphic units, reflection characteristics, sedimentary types, equipment parameters, and imaging effects;

[0096] S202. Using a triplet model, a knowledge graph is constructed from the target multi-source seismic data based on the graph nodes to obtain a marine geological knowledge graph. The triplet in the triplet model includes source parameters, geological features, and exploration results.

[0097] This invention employs a knowledge graph construction method to semantically model core elements in the seabed geological environment, such as sedimentation, structure, lithological assemblage, and oil and gas reservoir indicators, and constructs a seismic data-driven geological relationship network, thereby enhancing the modeling and reasoning capabilities for complex geological backgrounds in target areas.

[0098] This invention incorporates a self-learning module to intelligently learn and understand historical data, measured exploration data, expert experience, and literature. This module integrates retrieval enhancement generation technology and knowledge graph construction technology to form a unified data integration and semantic enhancement framework. First, this invention standardizes key information such as seismic source parameters, receiver configuration, and exploration area characteristics, then stores it in vector form to achieve efficient retrieval based on semantics and similarity. The knowledge graph constructs a network of relationships between geological elements, using "exploration target - geological unit - instrument parameters - imaging effect" as the core logical chain, thereby enhancing the system's ability to understand complex marine geological conditions.

[0099] In practical applications, the self-learning module can automatically parse various geological exploration data, including but not limited to seismic profiles, sedimentary environment descriptions, fault markings, well location measurements, and multi-source geophysical results. Geological exploration data is mainly divided into two categories: seismic instrument parameters and geological exploration results. Seismic instrument parameters cover technical indicators such as source type, operating frequency, pulse energy, waveform characteristics, receiver array configuration, receiver spacing, deployment method, dynamic range, and sensitivity. These parameters are standardized, stored as high-dimensional vectors, and quickly matched with the geological characteristics of specific areas to intelligently recommend the optimal source scheme and receiver array layout. Geological exploration results include seismic profiles, well logging data, sedimentary environment characteristics, laboratory analysis, exploration reports, well location information, stratigraphic division, sedimentary physical parameters, and historical exploration lines. All data undergoes noise removal, format standardization, and accuracy verification before being stored to ensure information integrity and consistency. Subsequently, based on vector retrieval technology, this invention enables rapid comparison and reuse of similar cases in the current exploration area, providing highly relevant data support for parameter recommendation and strategy formulation.

[0100] Please see Figure 6 The fourth embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model in this invention includes:

[0101] S301. Obtain a preset geological environment and preset seismic source parameters, and encode the preset geological environment and preset seismic source parameters to obtain an encoding sequence;

[0102] S302. Determine a preset large language model, and train the preset large language model according to the encoding sequence to obtain an initial large language model.

[0103] S303. The initial large language model is fine-tuned using LoRA (Low-Rank Adaptation, a parameter adjustment method for deep learning models) to obtain the target large language model.

[0104] The autonomous decision-making module in this invention is the core of its intelligence. Relying on a large language model and a deep neural network model, it possesses multiple functions, including seismic data analysis, exploration strategy reasoning, and parameter generation. Through a dialogue mode, it inputs exploration requirements and obtains instrument parameters. This module receives real-time input and background data provided by the data acquisition module and the autonomous learning module, automatically analyzes the geological characteristics of the exploration target area, and combines historical best practices with model predictions to generate optimal suggestions for source type, power, frequency combination, receiver spacing, array structure, and other parameters.

[0105] Furthermore, this invention possesses adaptive control capabilities. During actual exploration, if the imaging quality of the input measured data is poor, the signal-to-noise ratio is low, or the structure is unclear, the autonomous decision-making module can automatically correct the parameter settings and provide new suggestions based on the feedback mechanism. For example, in shallow water soft sedimentary environments, the system tends to recommend high-frequency, low-energy seismic sources to improve near-surface resolution; while in deep water hard strata areas, it automatically selects low-frequency, high-energy seismic sources to obtain deep target imaging.

[0106] Please see Figure 7 The fifth embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model in this invention includes:

[0107] S304. Obtain vector data from the marine geological knowledge graph, and use a hierarchical clustering algorithm to calculate the similarity of the vector data to obtain the similarity result;

[0108] S305. Based on the similarity results, the vector data is categorized to obtain a visual data cluster;

[0109] S306. Add tag metadata to each data sample in the visualized data cluster, wherein the tag metadata includes the region, data type, exploration target, stratigraphic age, acquisition method, and source batch.

[0110] Please see Figure 8 The sixth embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model in this invention includes:

[0111] S307. Obtain the user query statement and input the user query statement into the target large language model;

[0112] S308. The user query statement is parsed and weighted using the target large language model to obtain initial candidate text;

[0113] S309. Perform high-dimensional vectorization encoding, semantic similarity calculation, and reordering on the initial candidate texts to obtain semantic matching data.

[0114] S310. Based on the tag metadata in the marine geological knowledge graph, perform targeted retrieval processing on the semantic matching data, output the retrieval results, and complete the training of the target large language model to obtain the seismic exploration parameter adjustment model.

[0115] The optimization feedback module in this invention is based on reinforcement learning and closed-loop control strategies. It continuously receives exploration results, user evaluations, and measured data transmitted from the equipment during actual operations, and automatically iterates and updates the model. This module uses a reward and penalty mechanism to evaluate the contribution of different parameter configurations to the imaging results and mission objectives, and dynamically optimizes the model parameters accordingly.

[0116] After each exploration task, the system automatically compares the deviation between predicted parameters and measured data, analyzes the sources of imaging errors, and records them in the model training feedback set. Through multiple rounds of optimization, the system has gradually formed a parameter configuration strategy library applicable to different geological conditions. The reinforcement learning framework ensures that the model still has the ability to quickly adapt and make judgments when facing new geological conditions or anomalous areas.

[0117] Furthermore, the optimized feedback module supports a human-machine collaborative feedback mechanism. Geological experts can annotate key anomalies and correct recommended parameter results through a visual interface, thereby updating the knowledge graph and parameter biases, and gradually achieving human-machine joint training. This mechanism not only improves the model's interpretability and expert acceptance but also enhances the model's reliability and transparency in real-world production scenarios. This invention is equipped with a 3D seismic imaging visualization module, which can display in real time the impact of seismic source layout, profile response, stratigraphic distribution, and parameter variations on imaging results. It supports interactive adjustments by users, significantly improving the efficiency of exploration strategy decision-making.

[0118] Please see Figure 9 The seventh embodiment of the intelligent manufacturing method for seismic exploration equipment based on a large language model in this invention includes:

[0119] S401. Obtain the current seismic profile data and input the current seismic profile data into the seismic exploration parameter adjustment model;

[0120] S402. The current seismic profile data is analyzed and processed using the seismic exploration parameter adjustment model to obtain the geological environment and seismic exploration targets.

[0121] Based on the aforementioned large language model, this invention analyzes the waveform features and reflection structure information in seismic profile data to identify geological environment types and corresponding seismic exploration targets.

[0122] S403. Perform keyword extraction and knowledge graph matching on the geological environment and the seismic exploration target to obtain seismic exploration parameter configuration suggestions, and adjust the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment. The seismic exploration parameter configuration suggestions include the target source frequency and the target receiving array deployment method.

[0123] This invention extracts keywords from the geological environment and exploration targets using a large language model, and combines this with a marine geological knowledge graph to complete semantic matching and reasoning, generating seismic exploration parameter adjustment results. Based on these results, the initial parameters of the seismic exploration equipment are intelligently adjusted to obtain the target seismic exploration equipment configuration.

[0124] Example application:

[0125] The intelligent manufacturing method and system for seismic exploration equipment based on a large language model described in this invention are designed for complex marine geological environments. The system includes a data acquisition module, an autonomous learning module, an autonomous decision-making module, and an optimization feedback module, forming a closed-loop intelligent optimization system for exploration strategies.

[0126] like Figure 10 As shown, the hardware platform of this invention includes a multi-frequency seismic source, a variable-spacing receiving array, high-speed data communication, and a shipboard power supply system. The seismic source system (corresponding to the multi-frequency seismic source) encompasses Chirp (a linear frequency modulated signal, a signal whose frequency changes linearly with time), Boomer (a type of marine seismic exploration source device, called a Boomer source), and Sparker devices (electric spark sources), allowing for multi-band selection based on the depth and structural complexity of the geological target. The receiving system (corresponding to the variable-spacing receiving array) adopts an array layout, achieving wide-area coverage and fine sampling of seismic signals through optimization of array spacing and arrangement. The data transmission module (corresponding to high-speed data communication) supports high-speed, low-latency transmission, ensuring data integrity and real-time performance in harsh marine environments. The shipboard power supply system adapts to long-cycle operation requirements, providing continuous and stable power support.

[0127] During the data acquisition phase, this invention is deployed on an oceanographic survey vessel to acquire seismic data by selecting optimal seismic sources (such as Sparker, Boomer, Chirp) and underwater receiving arrays (such as towed hydrophone arrays) that support variable spacing. Its frequency adjustment capability within the 300Hz to 3000Hz range allows for the selection of seismic source devices and automatic adjustment of transmission frequency and energy output based on the target layer depth and geological structure complexity. The receiver array supports flexible spacing configurations within the range of 0.5m to 2.0m to adapt to different resolution and penetration depth requirements.

[0128] like Figure 10As shown, on the software platform, the data acquisition and control unit generates an initial acquisition parameter combination scheme, including the seismic source and receiving array, based on the geological model of the task area, historical exploration data, and current operation feedback, and applies it to the control program. During operation, the system acquires reflected wave signals in real time and completes preliminary imaging processing. Simultaneously, it stores the acquired parameters, environmental information, and raw waveform data to provide a basis for subsequent model evaluation and parameter optimization.

[0129] The data management unit establishes a unified multi-source data management system, standardizing and integrating data resources from different survey lines, historical work areas, and publicly available research institutions. The data covers seismic profiles, well logs, sediment experimental results, and source configuration parameters. The software platform connects to subsequent self-learning and optimization feedback modules through database expansion, constructing a structured index system to improve retrieval efficiency and enhance subsequent self-learning capabilities.

[0130] The data preprocessing unit is used to clean, standardize, detect anomalies, and correct errors in the acquired data. For example, it automatically identifies and removes abnormal waveforms and low-quality data; it uses a format converter to standardize different data formats (such as SEG-Y, CSV, LAS, and other seismic data formats), improving data universality and model adaptability. Historical data, after comparison and correction, is used as input for model pre-training, significantly reducing sample bias and improving prediction accuracy.

[0131] To enhance the semantic understanding and associative reasoning capabilities of data, this invention introduces a seismic exploration knowledge graph construction mechanism. The knowledge graph uses instrument parameters, seismic waveform characteristics, geological environmental parameters, and imaging effects as core nodes, revealing multidimensional logical relationships between data through a graph structure. This graph (i.e., the marine geological knowledge graph in this invention) integrates historical exploration records, scientific research data, and expert annotation results, enabling deep coupling between exploration parameters and geological background, and providing knowledge support for parameter recommendation and intelligent reasoning.

[0132] For example, in a certain exploration area, this invention identifies a region dominated by interbedded sand and mud layers based on a knowledge graph. The seismic source signal exhibits significant low-frequency attenuation in the high-porosity sandstone region. This invention uses a semantic matching mechanism to link reflection features with coarse sand labeling information and adjusts the seismic source frequency to 300Hz to enhance the imaging effect of the target layer's reflection signal, thereby improving reservoir identification accuracy. As another example, in areas with dense fault zones, this invention automatically triggers a frequency up-adjustment mechanism based on the fault-fluid channel coupling model in the knowledge graph and optimizes the receiver deployment density to enhance the detailed resolution of faults and the ability to identify inter-strata relationships.

[0133] like Figure 11As shown, to enhance knowledge reasoning capabilities, this invention constructs a knowledge index (i.e., the marine geological knowledge graph in this invention), establishing a network of connections between "source parameters—geological characteristics—exploration results" based on a ternary model. Graph nodes include stratigraphic units, reflection characteristics, sedimentary types, equipment parameters, and imaging effects, which are refined through expert annotation and system learning to form a knowledge structure that supports reasoning. This graph not only supports earthquake anomaly identification, stratigraphic comparison analysis, and parameter optimization recommendations, but also enables pattern transfer under different geological scenarios, improving system adaptability.

[0134] As the semantic hub of this invention, the knowledge graph connects data, models, and application scenarios, supporting transfer learning across different geological environments and enabling cross-regional and cross-project knowledge reuse and strategy transfer. For example, in offshore wind power surveys, the source deployment strategy for shallow-water deltaic muddy sediments can be applied to deep-water marginal basins with similar sedimentary properties and reflection structure characteristics in oil and gas development projects through similarity reasoning from the knowledge graph, thereby improving the system's adaptability and intelligent parameter recommendation capabilities in new work areas.

[0135] like Figure 12 As shown, during the self-learning phase, this invention constructs a knowledge-driven structure based on localized data. By integrating a vector database and a semantic indexing engine, it performs unified encoding and in-depth processing on multi-source heterogeneous data related to marine seismic exploration. The data covers historical seismic profile images, well logging curves, sediment physical parameters, exploration reports, source equipment parameters, etc. All data undergoes standardization, format conversion, noise reduction and completion, and accuracy verification before being stored in the database. Figure 12 Data preprocessing is performed to ensure that the data quality meets the requirements of model training and semantic reasoning.

[0136] This invention employs Retrieval-Augmented Generation (RAG) technology to achieve deep semantic modeling and intelligent retrieval of unstructured data (such as...). Figure 12 and Figure 13 (As shown). First, at the base layer, an embedded semantic encoder (such as BERT, Bidirectional Encoder Representations from Transformers, a pre-trained language representation model; SBERT, Sentence-BERT, a type of BERT model) is used to vectorize textual data. Combined with similarity calculation in the high-dimensional feature space, semantic matching between complex geological texts and instrument parameter descriptions is achieved. This invention integrates a local vector index storage (such as FAISS or Milvus) to achieve intelligent knowledge retrieval completely independent of the external network environment, ensuring information security and engineering controllability.

[0137] Furthermore, at the logical layer, the data is categorized into two main types: seismic instrument parameters and geological exploration results. Seismic instrument parameters include source type, excitation frequency, pulse duration, energy level, typical penetration depth, signal resolution, and receiver configuration (see Table 1). This type of data is transformed into a high-dimensional vector after structured processing. Combined with the geological background characteristics of the exploration area, this data is used by the system to automatically recommend the optimal source configuration and receiver deployment to meet the requirements of different exploration depths and resolutions (see Table 2).

[0138] Table 1: Comparison of Parameters of Commonly Used Source Equipment in Marine Seismic Exploration

[0139]

[0140] Regional geological information data falls under the category of exploration results, encompassing profile images, sedimentary environment classifications, structural features, anomalous reflection characteristics, well locations and depths, well logging response curves, experimental analysis results, structural divisions, sedimentary physical parameters, and historical exploration survey lines, as shown in Table 2. This invention employs a regional-level partitioned index for this type of data and establishes a multi-dimensional index system based on well location or work area numbers, supporting semantic matching and inversion logic reasoning on the basis of vectorized encoding.

[0141] Table 2: Comparison of Typical Marine Geological Environment Characteristics and Applicable Scenarios for Seismic Exploration

[0142]

[0143] This invention employs a hierarchical clustering algorithm to automatically classify vector data with high similarity, forming visualized data clusters and constructing a multi-dimensional labeling system. Each data sample is accompanied by label metadata, including its region, data type, exploration target, stratigraphic age, collection method, and source batch, enabling users to filter data as needed and improving the accuracy and interpretability of semantic retrieval. Based on label filtering and similarity ranking, the vector retrieval engine outputs candidate knowledge items for subsequent autonomous decision-making modules to use (e.g., ...). Figure 13 (As shown).

[0144] To improve the retrieval efficiency and semantic matching accuracy of geological exploration data, this invention, based on data vectorization storage and the construction of a multi-dimensional tag system, further introduces a multi-level semantic retrieval strategy (such as...) at the strategy layer. Figure 14 (As shown). This strategy integrates traditional keyword matching mechanisms (BM25 algorithm is an algorithm for information retrieval that calculates the degree of matching between a document and a query condition based on the similarity between the document and the query condition) with semantic vector ranking methods (such as SBERT / BERT), realizing a full-process retrieval from keyword triggering to semantic understanding, from coarse screening to fine-tuning, and can effectively cope with multi-source heterogeneous and complex data environments.

[0145] The specific implementation steps of the multi-level semantic retrieval strategy are as follows:

[0146] The first-level retrieval is based on the BM25 inverted index mechanism. The system performs keyword parsing and weight calculation on the user's input query to quickly locate semantically relevant initial candidate text. This layer focuses on coverage speed and keyword accuracy, and is suitable for query requests with clear structure and concentrated terminology, such as engineering retrieval tasks like "fault zone + earthquake source," and can complete text filtering in milliseconds.

[0147] The second-level retrieval focuses on improving semantic accuracy. The system uses a semantic embedding model (such as Sentence-BERT) to perform high-dimensional vectorization encoding of candidate data and leverages vector databases (such as FAISS, Milvus, or Qdrant) to calculate and re-rank semantic similarity. At this stage, the system can effectively identify sentence variations and contextual logical relationships, making it particularly suitable for analysis scenarios with inconsistent expressions, overlapping terminology, or non-standardized input. For example, when a user asks about "landslide source configuration strategies," the system can link to materials related to "imaging optimization of unstable sedimentary bodies," achieving precise semantic matching.

[0148] The third-layer retrieval enables tag-assisted fine-grained filtering. The system attaches multi-dimensional tag metadata to each data sample, covering elements such as the region, data type, exploration target, stratigraphic age, acquisition method, and data source. Users can combine and limit the search based on tags to further narrow the search scope and enhance professional matching. For example, when searching for data in the categories of "deep water area + landslide body + high-frequency seismic source + grid layout," the system can achieve targeted retrieval based on specific engineering needs while preserving semantic matching, improving the controllability and interpretability of the results.

[0149] All data processing is performed in the local operating environment, and the system has the capability for local data maintenance and dynamic updates. New data can be entered into the system manually or imported in batches. After standardization processing, vector encoding and map linking are automatically completed. The system supports version control and update tracking mechanisms to ensure data traceability and version consistency. At the same time, it can suggest archiving for data that has not been used for a long time, realizing the structural cleanup and space optimization of data resources.

[0150] During the autonomous decision-making phase, this invention integrates a Large Language Model (LLM), as well as the marine geological knowledge graph and retrieval enhancement generation system from the previous two phases (such as...). Figure 12As shown, the system possesses the ability to identify geological scenarios and generate parameter recommendations. Upon receiving real-time acquired seismic profile data, the system automatically analyzes the waveform features, reflection structures, and stratigraphic interfaces, performing semantic matching based on existing knowledge graphs to determine the geological environment type of the current work area, such as soft sedimentary layers, hard bedrock, landslides, or fault zones. After identifying the geological environment, the system automatically generates optimal source type, operating frequency, and pulse interval recommendations, along with receiver array deployment suggestions. During parameter generation, the system also references successful cases from similar geological scenarios, using RAG matching to find the best parameter combinations. These recommendations not only consider historical performance but also incorporate real-time analysis of current signal-to-noise ratio, waveform continuity, and other indicators, ensuring that the recommended solutions are highly consistent with current geological conditions. The system supports a closed-loop decision-making model of "operation, analysis, and optimization simultaneously," dynamically adjusting exploration strategies during data acquisition to improve operational efficiency and imaging accuracy.

[0151] The autonomous decision-making capability of this invention relies on the following key mechanisms: First, the LLM can simultaneously receive structured and unstructured data inputs, including historical seismic profiles, well logging data, sedimentary environment descriptions, and equipment parameter logs; second, this invention has a built-in geological knowledge graph module that can analyze the correlation between geological body characteristics and equipment configuration, identify similar historical scenarios through triple matching, and retrieve parameter configuration schemes from past successful cases; subsequently, the model generates seismic source configuration suggestions based on similarity ranking and semantic reasoning, such as recommending the use of seismic source equipment such as electric spark, Boomer, Sparker, or air gun, as well as the corresponding frequency range, pulse interval, and energy level; finally, the system supports user feedback mechanisms and reinforcement learning methods, enabling the parameter recommendation strategy to self-correct and optimize during continuous use.

[0152] For example, in shallow water soft sediment environments, this invention identifies seismic wave characteristics with continuous but rapidly attenuating reflection interfaces. Furthermore, by combining historical knowledge graphs, it discovers that the Boomer source provides better resolution in similar environments. The system then recommends this source, setting the frequency in the 1500Hz–3000Hz range, with a pulse interval of 0.1 seconds, and suggests a receiver spacing of 1 meter using a linear array layout to obtain clear images of shallow structures.

[0153] In another case, facing a deep-water work area with frequent fault activity and complex landslide structures, the system automatically read unconformities, dipping bedding, and reflection interference signals from the seismic profile and compared them with historical fault zone exploration data, identifying insufficient penetration capability of mid-to-high frequency signals. The system then recommended switching to an electric spark source, setting the operating frequency to 600Hz–1000Hz, with a pulse interval of 0.5 seconds, and suggested that the receiver array be deployed in a fan shape with the spacing reduced to 0.5 meters to enhance the detection capability of fault boundaries.

[0154] Taking a locally fine-tuned LLM (or LLaMA) model as an example to implement this decision-making capability, this invention employs LoRA technology to achieve model adaptation by training only a small number of learnable weights without modifying the basic model parameters. First, the system loads the basic LLaMA model and its Tokenizer (a core component used to convert raw text into structured data that the model can process) and initializes the model structure. During the loading phase, the undefined pad_token (padding token) in the Tokenizer is set to eos_token (sequence end token) to avoid alignment errors during batch input. Subsequently, the system injects a LoRA-based trainable subsystem, fine-tuning only key sub-modules to effectively control computational resource consumption.

[0155] like Figure 15 As shown, the model training data is generated using structured templates. The content revolves around typical geological environments (such as shallow-water soft sedimentary layers, fault zones, and landslide areas) and recommended source parameters (such as type, frequency, and receiver deployment) to form prompt and completion fields, which are then encoded as sequential inputs using a tokenizer. During training, the system constructs a DataLoader to batch process the training data and initializes an optimizer (such as AdamW). Training follows a standard loop of forward propagation, backpropagation, and parameter updates. After each training round, the system evaluates the loss value to optimize model performance. After training, the model and its tokenizer are saved and deployed in the actual inference system.

[0156] During the deployment and inference process, the system parses the user's input question into geological environment and exploration targets. Through keyword extraction and knowledge graph matching mechanisms, it determines the geological scene category to which the question belongs and retrieves the recommended parameter table to return the corresponding seismic source and receiver configuration suggestions. If a valid scene cannot be identified, the system will prompt the user to supplement information to improve matching accuracy.

[0157] This invention also supports real-time transmission of source response and received waveform quality data from the receiving system. Simultaneously, it calculates indicators such as signal-to-noise ratio, propagation delay, and reflection integrity, and compares these with model predictions. If anomalies are detected, such as signal weakening, poor coupling, or blurred imaging, the system will immediately adjust the source frequency, receiver spacing, or array structure, and update the recommended parameter scheme, achieving data-driven automatic parameter tuning.

[0158] During the optimization feedback phase, the system introduces a reinforcement learning (RL) mechanism to construct a parameter tuning strategy. Based on the seismic imaging performance and signal-to-noise ratio of each exploration task, the source and receiver configurations (such as...) are adjusted. Figure 12 and Figure 16 (As shown). The system automatically compares the differences between the imaging results and the expected model, and updates its internal strategy value function by evaluating indicators such as seismic profile clarity, key layer identification accuracy, and signal continuity, thereby optimizing the next round of parameter recommendations. If insufficient image resolution is detected in a soft sedimentary environment, the system will prioritize increasing the proportion of high-frequency wave components; if insufficient penetration depth is detected, the low-frequency source power will be appropriately increased and the receiver array spacing will be widened to improve imaging depth.

[0159] User feedback is also a core element of this invention. The system supports recording modifications made by geological interpreters to prediction results, especially in areas such as determining key stratigraphic positions or seismic reflection characteristics. Each user modification is automatically recorded as a feedback sample and incorporated into the system's fine-tuning process. For example, when a user manually adjusts the system's recommended source frequency or receiver spacing, the system will analyze and correct the recommendation logic based on the geological characteristics of the current exploration area. This mechanism can gradually reduce prediction errors, enabling the system to continuously optimize with use.

[0160] The system supports continuous iterative optimization based on new samples or user operations. If a user is not satisfied with a certain exploration suggestion, their modification behavior can be automatically parsed as a new training sample, added to the training set after review, triggering the fine-tuning process, forming a human-machine collaborative enhancement closed loop. This feedback-driven mechanism continuously strengthens the model's understanding and adaptability to complex geological scenarios in actual operation.

[0161] like Figure 16 The diagram illustrates the user feedback and self-optimization process. The system first receives actual exploration data and compares it with the geological model. Based on the comparison results, the system automatically adjusts the source parameters and receiver deployment to form a preliminary parameter recommendation scheme. If the user provides no feedback, the system uses the existing recommendation strategy; if the user adjusts the recommendation, the system enters the feedback optimization process. During this process, the user corrects the source and receiver configurations through human-computer interaction, and the system performs parameter fine-tuning accordingly, revising the original recommendation strategy. All feedback data is compiled into training samples and added to the sample library for incremental fine-tuning of the large language model. The system retrains the model based on the updated samples, generating a more accurate recommendation strategy, and incorporates the new strategy into the next round of recommendation, achieving continuous learning and capability evolution. This closed-loop process combines user behavior feedback, reinforcement learning, and fine-tuning mechanisms, effectively improving the model's adaptability in complex geological environments and the accuracy of parameter recommendations.

[0162] To further enhance its intelligence, the system integrates expert annotations and historical corrections into its parameter optimization strategy, adaptively adjusting the recommended range through a parameter offset attribution mechanism. For example, if the system detects that a user has repeatedly lowered the recommended seismic source frequency from 800Hz to 600Hz in shallow gas reservoir areas, its strategy will automatically prioritize frequencies near 600Hz in similar scenarios and fine-tune it based on the actual seismic signal acquisition results. If the noise in the device's returned signal is too high or the stratigraphic level is unclear, the system can also automatically adjust the receiver deployment density, array structure, or seismic source spacing to ensure continuous optimization of data acquisition quality.

[0163] The beneficial effects of this invention are:

[0164] 1. The intelligent manufacturing system provided by this invention has significant technical advantages and application value in complex marine geological exploration scenarios. Firstly, by introducing an autonomous decision-making mechanism and parameter adaptive optimization technology, the system can intelligently match the optimal source frequency and receiver array deployment according to the actual geological environment. Compared to traditional fixed-parameter acquisition schemes, this invention can still achieve high-resolution imaging in thick sand layers, significantly improving the identification accuracy of shallow to mid-layer geological targets.

[0165] 2. This invention relies on a large language model and reinforcement learning framework, enabling the system to continuously learn from measured exploration data and automatically generate parameter recommendation strategies. During operation, the system dynamically adjusts key parameters such as source power, frequency range, and receiver spacing based on real-time feedback, achieving closed-loop control of "optimization while data acquisition," significantly reducing the intensity of manual intervention and trial-and-error costs.

[0166] 3. This invention integrates structured database and knowledge graph construction technologies to make explicit and structured the implicit relationships between exploration parameters, geological structural features, survey line layout and imaging effects in a semantic network manner, thereby enhancing the semantic understanding and logical reasoning ability of seismic data.

[0167] 4. This invention possesses high geological adaptability and scenario mobility, making it suitable for various exploration tasks, including deep-water oil and gas resource surveys, submarine landslide risk assessments, submarine pipeline site selection, and offshore wind farm foundation surveys. The system maintains high-quality data acquisition capabilities in different environments, effectively reducing invalid survey lines and repetitive work, and significantly improving operational efficiency and economic benefits.

[0168] Furthermore, this invention possesses continuous learning and autonomous evolution capabilities, with built-in model fine-tuning and strategy optimization mechanisms. It can automatically absorb feedback data after each exploration task and continuously optimize parameter recommendation logic. Through a closed-loop process of "self-learning—adaptation—self-optimization," the system's intelligence continuously increases with usage, exhibiting good lifecycle scalability and system evolution potential.

[0169] Furthermore, such as Figure 17 As shown, based on the above-mentioned intelligent manufacturing method for seismic exploration equipment based on a large language model, this invention also provides an intelligent manufacturing system for seismic exploration equipment based on a large language model, wherein the intelligent manufacturing system for seismic exploration equipment based on a large language model includes:

[0170] The multi-source seismic data acquisition module 51 is used to acquire multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment, and to process the multi-source seismic data to obtain target multi-source seismic data.

[0171] Knowledge graph construction module 52 is used to construct a knowledge graph based on the target multi-source seismic data to obtain a marine geological knowledge graph;

[0172] The autonomous learning module 53 is used to determine the target large language model and use retrieval enhancement technology to perform autonomous learning processing on the target large language model based on the marine geological knowledge graph to obtain the seismic exploration parameter adjustment model.

[0173] The autonomous decision-making module 54 is used to acquire the current seismic profile data, perform autonomous decision-making processing based on the current seismic profile data through the seismic exploration parameter adjustment model, output seismic exploration parameter configuration suggestions, and adjust the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment.

[0174] The optimization feedback module 55 is used to acquire actual exploration data and use a reinforcement learning strategy to optimize the seismic exploration parameter adjustment model carried in the target seismic exploration equipment based on the actual exploration data.

[0175] Furthermore, such as Figure 18 As shown, based on the above-mentioned intelligent manufacturing method for seismic exploration equipment based on a large language model, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 18 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0176] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard drive or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard drive, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a seismic exploration equipment intelligent manufacturing program 40 based on a large language model, which can be executed by the processor 10 to implement the seismic exploration equipment intelligent manufacturing method based on a large language model in this application.

[0177] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the intelligent manufacturing method for seismic exploration equipment based on a large language model.

[0178] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface.

[0179] In one embodiment, when the processor 10 executes the intelligent manufacturing program 40 for seismic exploration equipment based on a large language model stored in the memory 20, it implements the steps of the intelligent manufacturing method for seismic exploration equipment based on a large language model as described above.

[0180] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a smart manufacturing program for seismic exploration equipment based on a large language model, and when the smart manufacturing program for seismic exploration equipment based on a large language model is executed by a processor, it implements the steps of the smart manufacturing method for seismic exploration equipment based on a large language model as described above.

[0181] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0182] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0183] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for intelligent manufacturing of seismic exploration equipment based on a large language model, characterized in that, The intelligent manufacturing method for seismic exploration equipment based on a large language model includes: Acquire multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment, and process the multi-source seismic data to obtain target multi-source seismic data; A knowledge graph is constructed based on the target multi-source seismic data to obtain a marine geological knowledge graph; A target large language model is determined, and a retrieval enhancement technique is used to autonomously learn the target large language model based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model. The step of employing retrieval enhancement technology to autonomously learn the target large language model based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model, previously included: The vector data in the marine geological knowledge graph is obtained, and the similarity of the vector data is calculated using a hierarchical clustering algorithm to obtain the similarity results; The vector data is categorized based on the similarity results to obtain visualized data clusters. Add tag metadata to each data sample in the visualized data cluster, wherein the tag metadata includes the region, data type, exploration target, stratigraphic age, acquisition method, and source batch; The process of employing retrieval enhancement technology to autonomously learn from the target large language model based on the marine geological knowledge graph to obtain a seismic exploration parameter adjustment model specifically includes: Obtain the user's query statement and input the user's query statement into the target large language model; The target large language model is used to perform keyword parsing and weight calculation on the user query statement to obtain initial candidate text; The initial candidate texts are subjected to high-dimensional vectorization encoding, semantic similarity calculation, and reordering to obtain semantic matching data. Based on the tag metadata in the marine geological knowledge graph, the semantic matching data is targeted for retrieval, and the retrieval results are output to obtain the seismic exploration parameter adjustment model. The system acquires current seismic profile data, performs autonomous decision-making based on the current seismic profile data through the seismic exploration parameter adjustment model, outputs seismic exploration parameter configuration suggestions, and adjusts the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment. Acquire actual exploration data and use reinforcement learning strategies to optimize the seismic exploration parameter adjustment model mounted on the target seismic exploration equipment based on the actual exploration data.

2. The intelligent manufacturing method for seismic exploration equipment based on a large language model according to claim 1, characterized in that, The data processing includes preprocessing and reprocessing; The process of acquiring multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment and processing the multi-source seismic data to obtain target multi-source seismic data specifically includes: Determine the initial seismic exploration parameters of the seismic exploration equipment, acquire reflected wave signals based on the initial seismic exploration parameters, and obtain the exploration environment corresponding to the reflected wave signals. The initial seismic exploration parameters include the initial source frequency and the initial receiver array deployment method. Acquire existing seismic exploration data, and construct multi-source seismic data based on the existing seismic exploration data, the initial seismic exploration parameters, the reflected wave signal, and the exploration environment; The multi-source seismic data is preprocessed and reprocessed to obtain the target multi-source seismic data; The preprocessing includes cleaning, standardization, anomaly detection, and error correction. The reprocessing includes standardization, format conversion, noise reduction and completion, and accuracy verification.

3. The intelligent manufacturing method for seismic exploration equipment based on a large language model according to claim 1, characterized in that, The construction of a knowledge graph based on the target multi-source seismic data to obtain a marine geological knowledge graph specifically includes: Determine map nodes, wherein the map nodes include stratigraphic units, reflectance features, sedimentary types, equipment parameters, and imaging effects; A knowledge graph is constructed from the target multi-source seismic data using a triplet model based on the graph nodes, resulting in a marine geological knowledge graph. The triplet in the triplet model includes source parameters, geological features, and exploration results.

4. The intelligent manufacturing method for seismic exploration equipment based on a large language model according to claim 1, characterized in that, The determination of the target large language model specifically includes: Obtain preset geological environment and preset seismic source parameters, and encode the preset geological environment and preset seismic source parameters to obtain an encoded sequence; A preset large language model is determined, and the preset large language model is trained according to the encoding sequence to obtain an initial large language model; The initial large language model was fine-tuned using LoRA technology to obtain the target large language model.

5. The intelligent manufacturing method for seismic exploration equipment based on a large language model according to claim 1, characterized in that, The process of acquiring current seismic profile data, performing autonomous decision-making based on the current seismic profile data through the seismic exploration parameter adjustment model, outputting seismic exploration parameter configuration suggestions, and adjusting the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment specifically includes: Acquire the current seismic profile data and input the current seismic profile data into the seismic exploration parameter adjustment model; The current seismic profile data is analyzed and processed using the seismic exploration parameter adjustment model to obtain the geological environment and seismic exploration targets. Keyword extraction and knowledge graph matching are performed on the geological environment and the seismic exploration target to obtain seismic exploration parameter configuration suggestions. The initial seismic exploration parameters of the seismic exploration equipment are adjusted according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment. The seismic exploration parameter configuration suggestions include the target source frequency and the target receiver array deployment method.

6. A smart manufacturing system for seismic exploration equipment based on a large language model, characterized in that, The intelligent manufacturing system for seismic exploration equipment based on a large language model is applied to the intelligent manufacturing method for seismic exploration equipment based on a large language model as described in any one of claims 1-5, wherein the intelligent manufacturing system for seismic exploration equipment based on a large language model comprises: The multi-source seismic data acquisition module is used to acquire multi-source seismic data corresponding to the initial seismic exploration parameters of the seismic exploration equipment, and to process the multi-source seismic data to obtain target multi-source seismic data. The knowledge graph construction module is used to construct a knowledge graph based on the target multi-source seismic data to obtain a marine geological knowledge graph. The autonomous learning module is used to determine the target large language model and to use retrieval enhancement technology to perform autonomous learning processing on the target large language model based on the marine geological knowledge graph to obtain the seismic exploration parameter adjustment model. The autonomous decision-making module is used to acquire current seismic profile data, perform autonomous decision-making processing based on the current seismic profile data through the seismic exploration parameter adjustment model, output seismic exploration parameter configuration suggestions, and adjust the initial seismic exploration parameters of the seismic exploration equipment according to the seismic exploration parameter configuration suggestions to obtain the target seismic exploration equipment. An optimization feedback module is used to acquire actual exploration data and employ a reinforcement learning strategy to optimize the seismic exploration parameter adjustment model mounted on the target seismic exploration equipment based on the actual exploration data.

7. A terminal, characterized in that, The terminal includes: a memory, a processor, and a large language model-based intelligent manufacturing program for seismic exploration equipment stored in the memory and executable on the processor. When the large language model-based intelligent manufacturing program for seismic exploration equipment is executed by the processor, it implements the steps of the large language model-based intelligent manufacturing method for seismic exploration equipment as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a smart manufacturing program for seismic exploration equipment based on a large language model. When the smart manufacturing program for seismic exploration equipment based on a large language model is executed by a processor, it implements the steps of the smart manufacturing method for seismic exploration equipment based on a large language model as described in any one of claims 1-5.

Images