Geological data retrieval method and device, electronic device and storage medium

Abstract

The invention relates to a geological data retrieval method and device, an electronic device and a storage medium, and the geological data retrieval method comprises the steps: converting a to-be-queried geological text into a to-be-queried geological text feature vector; the similarity between the to-be-queried geological text feature vector and each feature vector in a preset multi-modal feature vector pool is calculated, and the preset multi-modal feature vector pool is obtained by mapping geological text data, geological image data and sensor data collected in advance to a unified semantic space through a trained cross-modal model; determining a modal weight of the similarity according to a preset modal weight adjustment rule, and calculating a comprehensive similarity according to the modal weight and the similarity; obtaining a corresponding initial geological query result according to the comprehensive similarity; and sorting the initial geological query results according to the comprehensive similarity to obtain a target query result of the to-be-queried geological text. Through the geological data retrieval method and device, the geological data retrieval accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of geological data retrieval, and particularly to a geological data retrieval method, device, electronic device, and storage medium. Background Art

[0002] In geological research and applications, the management and retrieval of geological data face many challenges. Geological data is usually scattered in different institutions and departments, and its storage formats are diverse, including text files, image materials, database tables, maps, etc. This dispersion and format non-uniformity bring great inconvenience to geological workers. In addition, there are significant differences in the expression of geological terms in different data sources. For example, terms such as "fault" and "lithology" may have different descriptions in different data sources, which further increases the complexity of data retrieval and analysis. When conducting research, geological workers often need to retrieve multiple different types of data across time, space, and types, such as text descriptions, spatial data, and numerical data, etc. This not only takes time and effort but also easily leads to information omission or misunderstanding.

[0003] Currently, geological data is mainly retrieved through the CLIP model and graph neural network. The CLIP model learns to map text and images to a shared semantic space, and the graph neural network represents geological elements as graph nodes and uses edge relationships to represent the associations between different geological elements to capture the complex relationships between geological data. However, the existing CLIP model is not deeply adapted to the geological field. The semantic complexity of geological terms (such as "gneiss" and "thrust fault") is much higher than that in the general field, resulting in insufficient semantic alignment accuracy. Model fine-tuning requires a large amount of labeled data, but the cost of labeling geological data is high and scarce.

[0004] Moreover, the CLIP model needs to process high-dimensional image features (such as 1024-dimensional vectors) and text features (such as 768-dimensional vectors) simultaneously. The matching of high-dimensional feature spaces requires large-scale matrix operations to achieve modality alignment, with high computational complexity. When dealing with a large amount of data, the video memory occupancy surges. If relying on distributed training and model parallel technology, it will lead to cross-device communication overhead, further exacerbating the efficiency loss. The scale of geological data is huge (for example, existing geological picture libraries contain TB-level image data), and the inference speed of existing models cannot meet the real-time retrieval requirements. The hardware cost is high, and ordinary GPUs are difficult to support large-scale vector retrieval (such as a 1-billion-level vector library).

[0005] Regarding the problem of inaccurate geological data retrieval in the related art, no effective solution has been proposed yet. Summary of the Invention

[0006] In this embodiment, a geological data retrieval method, device, electronic device, and storage medium are provided to solve the problem of inaccurate geological data retrieval in the related art.

[0007] In a first aspect, in this embodiment, a geological data retrieval method is provided, including:

[0008] Obtain the geological text to be queried;

[0009] Convert the geological text to be queried into a feature vector to obtain the feature vector of the geological text to be queried;

[0010] Calculate the similarity between the feature vector of the geological text to be queried and each feature vector in the preset multi-modal feature vector pool, where the preset multi-modal feature vector pool is obtained by mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model;

[0011] Determine the modal weight of the similarity according to the preset modal weight adjustment rule, and calculate the comprehensive similarity of the feature vector of the geological text to be queried according to the modal weight and the similarity;

[0012] Obtain the corresponding initial geological query result according to the comprehensive similarity;

[0013] Sort the initial geological query result according to the comprehensive similarity to obtain the target query result of the geological text to be queried.

[0014] In some embodiments, before converting the geological text to be queried into a feature vector, it further includes:

[0015] Obtain the original text to be queried, perform denoising and word segmentation processing on the original text to be queried to obtain the geological text to be queried.

[0016] In some embodiments, the training process of the trained cross-modal model includes:

[0017] Clean the geological text training data and label the corresponding geological terms to obtain the cleaned geological text training data;

[0018] Extract the attribute relationship between the geological terms and construct a triple, and construct a geological knowledge graph based on the triple;

[0019] Train the initial cross-modal model based on the cleaned geological text training data to obtain the initially trained cross-modal model;

[0020] Taking the loss function of the geological knowledge graph as a regularization term, performing regularization constraints on the initially trained cross-modal model, and obtaining the trained cross-modal model.

[0021] In some of these embodiments, determining the modal weights of the similarity according to a preset modal weight adjustment rule includes:

[0022] Detecting keywords in the geological text to be queried;

[0023] Determining the modal weights corresponding to the keywords according to the keywords.

[0024] In some of these embodiments, the process of constructing the preset multi-modal feature vector pool includes:

[0025] Obtaining geological text data, geological image data, and sensor data;

[0026] Extracting the feature vector of the geological text data through the trained cross-modal model to obtain a geological text feature vector;

[0027] Extracting the feature vector of the geological image data through the trained cross-modal model to obtain a geological image feature vector;

[0028] Extracting the feature vector of the sensor data through a preset neural network model to obtain a geological parameter feature vector;

[0029] Mapping the geological text feature vector, the geological image feature vector, and the geological parameter feature vector to a unified semantic space to obtain the multi-modal feature vector pool.

[0030] In some of these embodiments, the target query result includes the unique identifier of the geological parameter and the location information of the geological parameter.

[0031] In some of these embodiments, after obtaining the target query result of the geological text to be queried, it further includes:

[0032] Generating an interactive map and / or an analysis report according to the target query result.

[0033] In a second aspect, in this embodiment, a geological data retrieval device is provided, including: an acquisition module, a feature vector conversion module, a first calculation module, a second calculation module, and an output module, where

[0034] The acquisition module is used to acquire the geological text to be queried;

[0035] The feature vector conversion module is used to convert the geological text to be queried into a feature vector to obtain a geological text feature vector to be queried;

[0036] The first calculation module is configured to calculate the similarity between the geological text feature vector to be queried and each feature vector in a preset multi-modal feature vector pool, where the preset multi-modal feature vector pool is obtained by mapping pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model;

[0037] The second calculation module is configured to determine the modal weight of the similarity according to a preset modal weight adjustment rule, and calculate the comprehensive similarity of the geological text feature vector to be queried according to the modal weight and the similarity;

[0038] The output module is configured to obtain a corresponding initial geological query result according to the comprehensive similarity; sort the initial geological query result according to the comprehensive similarity to obtain the target query result of the geological text to be queried.

[0039] In a third aspect, an electronic device is provided in this embodiment, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the geological data retrieval method described in the first aspect above is implemented.

[0040] In a fourth aspect, a storage medium is provided in this embodiment, on which a computer program is stored. When the program is executed by a processor, the geological data retrieval method described in the first aspect above is implemented.

[0041] Compared with the related art, in the geological data retrieval method provided in this embodiment, by obtaining the geological text to be queried; converting the geological text to be queried into a feature vector to obtain a geological text feature vector to be queried; calculating the similarity between the geological text feature vector to be queried and each feature vector in a preset multi-modal feature vector pool, where the preset multi-modal feature vector pool is obtained by mapping pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model; determining the modal weight of the similarity according to a preset modal weight adjustment rule, and calculating the comprehensive similarity of the geological text feature vector to be queried according to the modal weight and the similarity; obtaining a corresponding initial geological query result according to the comprehensive similarity; sorting the initial geological query result according to the comprehensive similarity to obtain the target query result of the geological text to be queried, the problem of inaccurate geological data retrieval is solved, and the accuracy of geological data retrieval is improved.

[0042] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0044] Figure 1 It is a hardware structure block diagram of the terminal of the geological data retrieval method of this embodiment.

[0045] Figure 2 It is a flowchart of the geological data retrieval method of this embodiment.

[0046] Figure 3 It is a cross-modal model training flowchart of the geological data retrieval method in this embodiment.

[0047] Figure 4 It is a flowchart of another geological data retrieval method of this embodiment.

[0048] Figure 5 It is a structure block diagram of the geological data retrieval device of this embodiment. Detailed implementation manners

[0049] To understand the purpose, technical solution and advantages of the present application more clearly, the present application will be described and explained below with reference to the accompanying drawings and embodiments.

[0050] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meanings understood by those with ordinary skills in the technical field to which this application belongs. In this application, words such as "a", "an", "one kind", "the", "these", etc. do not indicate a limitation in quantity, and they can be singular or plural. The terms "including", "comprising", "having" and any variants thereof involved in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product or device including a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent in these processes, methods, products or devices. The words such as "connected", "linked", "coupled", etc. involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "plurality" involved in this application means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships can exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, and B exists alone. Usually, the character " / " indicates that the objects associated before and after are an "or" relationship. The terms "first", "second", "third", etc. involved in this application only distinguish similar objects and do not represent a specific order for the objects.

[0051] The method embodiment provided in this embodiment can be executed on a terminal, a computer or a similar computing device. For example, when running on a terminal, Figure 1 is a block diagram of the hardware structure of the terminal of the geological data retrieval method in this embodiment. As Figure 1 shown, the terminal may include one or more ( Figure 1 only one is shown in the figure) processors 102 and a memory 104 for storing data. Among them, the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA. The above terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above terminal. For example, the terminal may further include more or fewer components than those shown in Figure 1 the figure, or have a different configuration from that shown in Figure 1 the figure.

[0052] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the geological data retrieval method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implements the above method. The memory 104 may include a high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.

[0053] The transmission device 106 is used to receive or send data via a network. The above network includes a wireless network provided by a communication provider of the terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one instance, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0054] In this embodiment, a geological data retrieval method is provided. Figure 2 is a flowchart of the geological data retrieval method of this embodiment, as Figure 2 shown, and this process includes the following steps:

[0055] Step S201, obtain the geological text to be queried; convert the geological text to be queried into a feature vector to obtain the feature vector of the geological text to be queried.

[0056] Specifically, obtain the geological text to be queried input by the user. Among them, the geological text to be queried input by the user can be natural language. For example: "Is there a thrust fault in a certain area?" Use a pre-trained language model (such as BERT, Word2Vec) to convert the geological text to be queried into a feature vector with a fixed dimension, and output the feature vector of the text to be queried.

[0057] Step S202, calculate the similarity between the feature vector of the geological text to be queried and each feature vector in the preset multi-modal feature vector pool, where the preset multi-modal feature vector pool is obtained by mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model.

[0058] Specifically, geological data includes multiple modalities, such as text (reports, descriptions), images (remote sensing, borehole histograms), numerical values (seismic wave velocities, lithological parameters), etc. There are significant differences in semantic expressions among these different modalities of data, making it difficult to directly associate and uniformly process them. For example, there is a lack of a unified semantic framework between the "fault" in the text and the fault line features in the image, and between the numerical data (such as borehole lithological stratification) and the spatial data (such as geological maps). Moreover, geological data comes from a wide range of sources and has diverse formats, including PDF reports, GeoTIFF images, database tables, Shapefile vector maps, etc., with inconsistent data structures, coordinate systems, accuracies, etc. Also, existing cross-modal models (such as CLIP) are not deeply adapted to the geological field, resulting in insufficient semantic alignment accuracy, and the semantic complexity of geological terms (such as "gneiss" and "thrust fault") is much higher than that in the general field. Based on this, in this embodiment, the cross-modal model is improved by training the cross-modal model with geological terms to obtain a trained cross-modal model, and then mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through the trained cross-modal model, unifying the dimensions of the geological text data, geological image data, and sensor data to obtain a multi-modal feature vector pool, where the sensor data is data such as borehole parameters collected by sensors. By constructing a multi-modal feature pool, the feature vectors of each modality are fused to achieve semantic alignment of the data of each modality and improve the accuracy of retrieval. Calculate the similarity between the geological text feature vector to be queried and the feature vectors of each modality in the multi-modal feature vector pool, where the calculation of the similarity can be the calculation of cosine similarity.

[0059] Step S203, determine the modality weights of the similarity according to the preset modality weight adjustment rule, and calculate the comprehensive similarity of the geological text feature vector to be queried according to the modality weights and the similarity.

[0060] Specifically, corresponding weight ratios are set in advance for the similarities of each modality. Exemplarily, if the geological text to be queried contains the word "image", the modality weight of the similarity related to the image modality is set to 0.6, and if the geological text to be queried contains "borehole", the modality weight of the similarity related to the numerical modality is set to 0.4. According to the focus of the geological parameter query in the geological text to be queried and the pre-set modality weight ratio allocation rule, allocate the modality weights to the calculated similarities, and calculate the comprehensive similarity according to the allocated weights.

[0061] Step S204, obtain the corresponding initial geological query result according to the comprehensive similarity; sort the initial geological query result according to the comprehensive similarity to obtain the target query result of the geological text to be queried.

[0062] Specifically, each comprehensive similarity corresponds to corresponding geological parameters. The geological parameters corresponding to the comprehensive similarity are sorted from high to low according to the level of the comprehensive similarity, and the target query result of the geological text to be queried is obtained.

[0063] Through the above steps S201 to S204, the geological text to be queried is obtained; the geological text to be queried is converted into a feature vector to obtain the feature vector of the geological text to be queried; the similarity between the feature vector of the geological text to be queried and each feature vector in the preset multi-modal feature vector pool is calculated, where the preset multi-modal feature vector pool is obtained by mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model; the modal weight of the similarity is determined according to the preset modal weight adjustment rule, and the comprehensive similarity of the feature vector of the geological text to be queried is calculated according to the modal weight and the similarity; the corresponding initial geological query result is obtained according to the comprehensive similarity; the initial geological query result is sorted according to the comprehensive similarity to obtain the target query result of the geological text to be queried. Compared with the prior art in which the geological data is retrieved through the CLIP model and the graph neural network, in this embodiment, the cross-modal model is trained by combining geological terms to obtain a trained cross-modal model. Then, the pre-collected geological text data, geological image data, and sensor data are mapped to a unified semantic space through the trained cross-modal model to obtain a multi-modal feature vector pool. By calculating the similarity between the feature vector of the geological text to be queried and each feature vector in the multi-modal feature vector pool, the similarity corresponding to each modal feature vector is obtained. Finally, the modal weight of each similarity is determined according to the preset modal weight adjustment rule, and the comprehensive similarity is calculated according to the modal weight. The corresponding geological parameter, that is, the target query result, is obtained according to the comprehensive similarity. By training the cross-modal model by combining geological parameters, and then unifying the feature vectors of each modality through the trained cross-modal model, when retrieving the geological text to be queried, the accuracy of the retrieval is improved by comparing the similarity with the unified multi-modal feature vector pool, and then the modal weight of each modal similarity is adjusted to further improve the accuracy of the retrieval.

[0064] In some of these embodiments, before converting the geological text to be queried into a feature vector, it further includes:

[0065] The original text to be queried is obtained, and the original text to be queried is denoised and segmented to obtain the geological text to be queried.

[0066] Specifically, natural language processing libraries (such as NLTK, spaCy) are used to remove noise such as stop words, punctuation marks, and HTML tags from the original text to be queried, improve the text quality, and enhance the accuracy of subsequent semantic understanding. A tokenization tool (such as NLTK, spaCy) is used to tokenize the original text to be queried. Geological texts contain a large number of professional terms and concepts. Tokenization can accurately divide these terms. For example, rock names such as "granite" and "shale", and geological structure terms such as "fold structure" and "fault" are obtained to get the geological text to be queried.

[0067] In another embodiment, Figure 3 is the cross-modal model training flowchart of the geological data retrieval method in this embodiment, as Figure 3 shown. The training process of the trained cross-modal model includes:

[0068] Step S301, clean the geological text training data and label the corresponding geological terms to obtain the cleaned geological text training data.

[0069] Specifically, text data is crawled from network resources (such as geological report libraries) through a distributed crawler, or structured data, remote sensing data (images) are extracted from a database through an ETL collection tool, or image data is obtained through a satellite image API, or real-time numerical data is obtained from a sensor network through a sensor interface. Further, PDF / Word documents are crawled from a geological report library (such as USGS), and the text content is extracted through OCR (Optical Character Recognition) to obtain text data; the satellite image API (such as Sentinel-2) is called to download the target area image to obtain image data; lithology, depth and other parameters are extracted from the borehole database to obtain numerical data. The text data, image data, and numerical data constitute the geological text training data. The geological text training data is preprocessed. Among them, natural language processing libraries (such as NLTK, spaCy) are used to remove noise (such as HTML tags), the text data after removing noise is tokenized and geological terms (such as "fault") are labeled, and the output is: the cleaned text data; image processing libraries (such as GDAL, OpenCV) are used to perform radiometric correction and geometric correction on remote sensing images, and the output is the corrected image data; geographic information system (GIS) tools (such as QGIS, ArcGIS) are used to map the borehole data to a unified coordinate system (such as WGS84), and the output is the standardized numerical data. And the processed data is stored in a distributed file system (such as HDFS).

[0070] Step S302, extract the attribute relationship between geological terms, construct triples, and construct a geological knowledge graph based on the triples.

[0071] Specifically, extract geological terms (such as "intrusive rock") and their relationships (such as "belongs to - magmatic rock"), construct triples (entity - relationship - entity: intrusive rock - belongs to - magmatic rock) based on the geological terms and their relationships, and then use a knowledge graph construction tool to construct a knowledge graph and store it in a graph database (such as Neo4j). Among them, the geological domain knowledge graph is a technical framework that integrates multi-source heterogeneous data in a structured and semantic way for geological knowledge and endows the computer with the ability of understanding and reasoning. The geological knowledge is formally expressed and organized in the form of a directed graph, and the geological knowledge is constructed into a semantic network that can be understood, calculated, and reasoned by the computer. The core of the geological knowledge graph is to represent geological entities (such as strata, rock masses, faults, etc.), attributes (such as lithology, thickness, dip angle, etc.), and relationships (such as spatial position, genetic connection, etc.) in the form of triples, and construct a knowledge network through these triples. When the knowledge graph needs to be updated, the data is processed by the acquisition system and the server cluster to update the knowledge graph.

[0072] Step S303: Train the initial cross-modal model based on the cleaned geological text training data to obtain the initially trained cross-modal model.

[0073] Specifically, fine-tune the CLIP model using geological domain data, freeze the image encoder, and only train the text encoder. The image encoder is a pre-trained convolutional neural network (such as ResNet), which may already have good generalization ability in geological domain data. Therefore, the image encoder is selected to be frozen to avoid overfitting and save computing resources at the same time. The text encoder (usually a Transformer architecture) needs to be fine-tuned for the text data in the geological domain to better understand the semantics of geological descriptions. Therefore, in this embodiment, the initial cross-modal model is trained with geological text training data to obtain the initially trained cross-modal model.

[0074] Step S304: Use the loss function of the geological knowledge graph as a regularization term to perform regularization constraints on the initially trained cross-modal model to obtain the trained cross-modal model.

[0075] Specifically, in this embodiment, the contrastive learning loss (InfoNCE) is used as the loss function. The contrastive learning loss is used to maximize the similarity between positive sample pairs (such as matching images and texts), while minimizing the similarity between negative sample pairs (such as non-matching images and texts), thereby improving the accuracy of model recognition. To further enhance the adaptability of the model in the geological field, a domain knowledge regularization term is introduced in this embodiment. In this embodiment, the loss function of the knowledge graph in the geological field is introduced as the regularization term to provide additional constraints for the text encoder and ensure that the knowledge learned by the model is consistent with the domain knowledge. Thus, a trained cross-modal model is obtained. Among them, a domain knowledge regularization term is added on the basis of CLIP to constrain the model to learn the semantic consistency of geological terms. The specific formula is as follows:

[0076] ;

[0077] where λ is a hyperparameter, KnowledgeLoss is based on the triple constraints of the knowledge graph, and InfoNCE is the loss function of the geological knowledge graph.

[0078] In some embodiments, the modal weights of the similarity are determined according to a preset modal weight adjustment rule, including:

[0079] Detect the keywords in the geological text to be queried; determine the modal weights corresponding to the keywords according to the keywords.

[0080] Specifically, in the above step S203, setting the modal weights according to the preset modal weight adjustment rule is mainly determined according to the keywords in the geological text to be queried. The modal weights of the similarity of the matched text, image or numerical value are adjusted according to whether the preset keywords are included in the text to be queried. For example, detect whether a specific keyword, such as "image" or "drilling hole", is included in the query. If the query includes "image", increase the modal weight of the similarity corresponding to the image modality (such as setting 0.6); if the query includes "drilling hole", increase the modal weight of the similarity corresponding to the numerical modality (such as setting 0.4). Among them, the specific keywords and modal weights can be set according to the actual situation, and no specific limitation is made in this embodiment.

[0081] In another embodiment, the construction process of the preset multi-modal feature vector pool includes:

[0082] Obtain geological text data, geological image data, and sensor data; extract the feature vectors of the geological text data through the trained cross-modal model to obtain geological text feature vectors; extract the feature vectors of the geological image data through the trained cross-modal model to obtain geological image feature vectors; extract the feature vectors of the sensor data through the preset neural network model to obtain geological parameter feature vectors; map the geological text feature vectors, geological image feature vectors, and geological parameter feature vectors to a unified semantic space to obtain a multi-modal feature vector pool.

[0083] Specifically, after obtaining the geological text data, geological image data, and sensor data, where the sensor data is the geological parameter data actually collected by the sensor, the feature vectors of each data are extracted according to the trained cross-modal model in the above embodiment. Input the geological text description, and use the fine-tuned CLIP text encoder to encode the geological text description, and output a feature vector with a fixed dimension to obtain the geological text feature vector, for example, output a 768-dimensional vector. Input the geological image data, and use the frozen CLIP image encoder to encode the geological image data, and output a feature vector with a fixed dimension to obtain the geological image feature vector, for example, output a 1024-dimensional vector. Input the sensor data, and map the sensor data to a 512-dimensional vector through the MLP feedforward neural network to obtain the geological parameter feature vector. Among them, the sensor data can be drilling parameters, which are geological information obtained through drilling engineering and are used to reveal the structure, properties, and distribution of underground rock and soil layers. Physical properties of rock and soil layers: such as color, particle size, hardness, density, etc. Chemical composition of rock and soil layers: mineral composition, element content, etc. obtained through sampling analysis. Stratigraphic structure: such as the thickness, dip angle, and fault location of rock layers. Groundwater conditions: such as groundwater level, water quality, water pressure, etc. Engineering geological parameters: such as the bearing capacity, shear strength, and compression modulus of rock and soil.

[0084] Map the geological text feature vectors, geological image feature vectors, and geological parameter feature vectors to a unified semantic space so that data of different modalities can be analyzed and processed under the same framework. Integrate the geological text feature vectors, geological image feature vectors, and geological parameter feature vectors into a multi-modal feature vector. The specific integration method can adopt simple splicing, weighted average, or more complex fusion strategies. Exemplary: Assume that the geological text feature vector is v t (768-dimensional), the geological image feature vector is v s (1024-dimensional), and the numerical feature is v n (512-dimensional). They can be spliced into a 2304-dimensional vector, v = [v t , v s , v n .

[0085] Use a vector database such as Milvus to store multimodal feature vectors. Milvus is an open-source vector database that supports efficient vector retrieval and similarity search. The feature vectors can be stored in Milvus and retrieved and queried quickly through the API provided by Milvus, thus obtaining a multimodal feature vector pool.

[0086] In some of these embodiments, the target query result includes the unique identifier of the geological parameter and the location information of the geological parameter.

[0087] Specifically, in this embodiment, the output target query result includes the unique identifier of the geological parameter and the location information of the geological parameter. The unique identifier of the geological parameter is to accurately distinguish and locate each specific geological parameter in a huge geological data system, ensuring that there is no confusion during data query, management, and analysis. For example, in a geological database containing multiple ore deposits, each ore deposit's geological parameters (such as ore grade, rock type, etc.) have a unique identifier. This identifier can be a code, such as "ore deposit 001 - parameter 01", so that when querying the ore grade of ore deposit 001, the corresponding data can be quickly and accurately found through this unique identifier. The location information of the geological parameter refers to the specific location of the parameter in the geographical space. This includes geographical coordinate information such as longitude, latitude, altitude, and possibly depth information (such as a rock sample at a certain depth underground). Location information is an indispensable part of geological research and applications because geological phenomena and parameters are often closely related to geographical locations. For example, in mineral exploration, knowing the specific location of the ore deposit is a prerequisite for on-site exploration and mining. Through accurate location information, the geographical location of the ore deposit can be determined, and its relative position to other geological structures (such as faults, folds, etc.) can be evaluated, thus better planning exploration and mining plans. When a user queries geological data, the unique identifier and location information in the target query result can help quickly locate the required data.

[0088] In another embodiment, after obtaining the target query result of the geological text to be queried, it further includes: generating an interactive map and / or an analysis report according to the target query result.

[0089] Specifically, the annotation function of the map library is used to annotate the target query results on the map. Or a PDF analysis report is generated using a front-end library (such as jsPDF). The analysis report includes detailed information of the retrieval results, a map screenshot, a link to the original data, etc. The three-dimensional visualization of geological data is realized through engines such as Cesium and Unity, supporting various display forms such as cross-sectional views and stereograms. The client also provides a button or menu option. When the user clicks it, the report generation is triggered, and a PDF report containing the retrieval results and the original data link is generated and downloaded. Specifically, it can be in the form of an interactive map, an analysis report, etc. The user can also view the target query results in real time through operations such as dragging and zooming, and the interaction response time is less than 200 ms. Multi-terminal collaboration of the Web side, mobile side, and desktop side is supported, and the data synchronization delay is less than 1 second.

[0090] In this embodiment, a geological data retrieval method is also provided. Figure 4 It is a flowchart of another geological data retrieval method of this embodiment, as Figure 4 shown, and this process includes the following steps:

[0091] Step S401, obtain the original text to be queried, perform denoising and word segmentation processing on the original text to be queried, and obtain the geological text to be queried;

[0092] Step S402, convert the geological text to be queried into a feature vector to obtain the feature vector of the geological text to be queried;

[0093] Step S403, calculate the similarity between the feature vector of the geological text to be queried and each feature vector in the preset multi-modal feature vector pool, where the preset multi-modal feature vector pool is obtained by mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model;

[0094] Step S404, detect the keywords in the geological text to be queried; determine the modal weights corresponding to the keywords according to the keywords;

[0095] Step S405, calculate the comprehensive similarity of the feature vector of the geological text to be queried according to the modal weights and the similarity;

[0096] Step S406, obtain the corresponding initial geological query results according to the comprehensive similarity, and sort the initial geological query results according to the comprehensive similarity to obtain the target query results of the geological text to be queried. Among them, the target query results include the unique identifier of the geological parameter and the location information of the geological parameter;

[0097] Step S407, generate an interactive map and / or an analysis report according to the target query results.

[0098] Exemplarily:

[0099] User query: Enter "Is there a gold mine in a certain area?" at the front end.

[0100] System processing: Query parsing: Extract the text vector of "gold mine", and use a pre-trained language model (such as BERT) to convert "gold mine" into a text vector with a fixed dimension.

[0101] Knowledge graph matching: Search for terms and relationships related to "gold mine" in the knowledge graph. For example, the knowledge graph may contain triples such as "gold mine - often appears in - alteration zone", "gold mine - related to - quartz vein", etc. Match the relevant terms in the knowledge graph (such as "alteration zone").

[0102] Dynamically adjust the modality weights: Detect whether the query contains keywords related to a specific modality (such as "image", "drilling"). If "image" is mentioned in the query, increase the image modality weight; if "drilling" is mentioned, increase the numerical modality weight. Prioritize matching the alteration features in remote sensing images.

[0103] Retrieve multi-modal feature vectors related to "gold mine" from a vector database (such as Milvus). Calculate the cosine similarity between the query vector and the feature vectors, and generate a comprehensive score by combining the modality weights.

[0104] Result visualization: Highlight the potential gold mine areas on the map and provide borehole data verification. Highlight the potential gold mine areas on the map. Provide borehole data verification and display the locations and geological information of the relevant boreholes.

[0105] Output results:

[0106] Interactive map: Highlight the potential gold mine areas on the map. Provide borehole data verification and display the locations and geological information of the relevant boreholes.

[0107] Analysis report: Generate a PDF report containing the retrieval results, the links to the original data, the relevant geological terms, and the knowledge graph information.

[0108] Exemplarily: User query: Upload seismic wave velocity data and enter "Is there an active fault?"

[0109] System processing:

[0110] Data processing and vectorization: Map the uploaded seismic wave velocity data into numerical vectors, and use a pre-trained MLP (multi-layer perceptron) model to convert the wave velocity data into numerical vectors with a fixed dimension.

[0111] Knowledge graph matching: Combine the fault distribution information in the geological map and extract the geological terms and location information related to faults.

[0112] Dynamic adjustment of modal weights: Since the query involves spatial distribution information (such as fault locations), the spatial modal weight is dynamically increased.

[0113] Calculation of fault activity probability: Calculate the similarity between the uploaded wave velocity data vector and the known fault distribution feature vector. Use machine learning models (such as random forest, support vector machine) to combine wave velocity data and fault location information to calculate the fault activity probability.

[0114] Result visualization: Generate a heat map of fault activity risk to warn high-risk areas.

[0115] Output result: A heat map of fault activity risk is generated on the map. High-risk areas are shown in red, and low-risk areas are shown in blue. Users can intuitively see potential active fault areas.

[0116] Warning information: Provide detailed information on high-risk areas, including fault names, locations, and activity probabilities.

[0117] Through the above steps S401 to S407, by collecting geological data (text, images, numerical values, spatial data) from multi-source heterogeneous data sources, cleaning and standardizing the database, and mapping geological text data, geological image data, and sensor data to a unified semantic space based on a trained cross-modal model. Then, according to the keywords in the text data queried by the user, dynamically adjust the modal weights of the similarities of the matched modalities, and achieve cross-modal retrieval. Overlay the retrieval results with spatial data to generate a geological element distribution map or specific analysis results. Compared with the prior art that retrieves geological data through the CLIP model and graph neural network, in this embodiment, the cross-modal model is trained with geological professional terms, which improves the accuracy of the cross-modal model in extracting geological data features. Then, by setting a multi-modal feature vector pool with the trained cross-modal model, mapping multi-modal data to a unified semantic space realizes multi-format compatibility, supports automatic parsing of more than 20 geological data formats such as Shapefile, DXF, GeoTIFF, 3DS, etc., reduces manual intervention, and lowers the data integration cost. It bridges the cross-modal semantic gap in the geological field, improves the applicability and retrieval accuracy. Further combining the modal weight calculation to obtain a comprehensive retrieval result further improves the accuracy of geological data retrieval.

[0118] In this embodiment, a geological data retrieval device is also provided. This device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated here. The following terms such as "module", "unit", "sub-unit", etc. can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0119] Figure 5 is the structural block diagram of the geological data retrieval device of this embodiment. As Figure 5 shown, the device 50 includes: an acquisition module 51, a feature vector conversion module 52, a first calculation module 53, a second calculation module 54, and an output module 55. Among them,

[0120] The acquisition module 51 is used to acquire the geological text to be queried;

[0121] The feature vector conversion module 52 is used to convert the geological text to be queried into a feature vector to obtain the feature vector of the geological text to be queried;

[0122] The first calculation module 53 is used to calculate the similarity between the feature vector of the geological text to be queried and each feature vector in the preset multi-modal feature vector pool. Among them, the preset multi-modal feature vector pool is obtained by mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model;

[0123] The second calculation module 54 is used to determine the modal weight of the similarity according to the preset modal weight adjustment rule, and calculate the comprehensive similarity of the feature vector of the geological text to be queried according to the modal weight and the similarity;

[0124] The output module 55 is used to obtain the corresponding initial geological query result according to the comprehensive similarity; sort the initial geological query result according to the comprehensive similarity to obtain the target query result of the geological text to be queried.

[0125] It should be noted that the above-mentioned each module can be a functional module or a program module, and can be implemented either by software or by hardware. For the modules implemented by hardware, the above-mentioned each module can be located in the same processor; or the above-mentioned each module can also be located in different processors in any combination form.

[0126] In this embodiment, an electronic device is also provided, including a memory and a processor. A computer program is stored in the memory, and the processor is set to run the computer program to execute the steps in any one of the above method embodiments.

[0127] Optionally, the above-mentioned electronic device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above-mentioned processor, and the input / output device is connected to the above-mentioned processor.

[0128] Optionally, in this embodiment, the above-mentioned processor may be set to execute the following steps through a computer program:

[0129] S1. Acquire the geological text to be queried;

[0130] S2. Convert the geological text to be queried into a feature vector to obtain the feature vector of the geological text to be queried;

[0131] S3. Calculate the similarity between the feature vector of the geological text to be queried and each feature vector in the preset multi-modal feature vector pool, where the preset multi-modal feature vector pool is obtained by mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model;

[0132] S4. Determine the modal weight of the similarity according to the preset modal weight adjustment rule, and calculate the comprehensive similarity of the feature vector of the geological text to be queried according to the modal weight and the similarity;

[0133] S5. Obtain the corresponding initial geological query result according to the comprehensive similarity; sort the initial geological query result according to the comprehensive similarity to obtain the target query result of the geological text to be queried.

[0134] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and alternative embodiments, and will not be elaborated in this embodiment.

[0135] In addition, in combination with the geological data retrieval method provided in the above embodiments, a storage medium can also be provided in this embodiment to implement it. A computer program is stored on the storage medium; when the computer program is executed by a processor, the geological data retrieval method in any of the above embodiments is implemented.

[0136] It should be understood that the specific embodiments described here are only used to explain this application, rather than to limit it. According to the embodiments provided in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0137] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data that have been authorized by the user or fully authorized by all parties.

[0138] Obviously, the drawings are only some examples or embodiments of the present application. For those of ordinary skill in the art, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it can be understood that although the work done during the development process here may be complex and time-consuming, for those of ordinary skill in the art, some design, manufacturing, or production changes based on the technical content disclosed in the present application are only conventional technical means and should not be regarded as insufficient disclosure of the present application.

[0139] The term "embodiment" in this application means that the specific features, structures or characteristics described in connection with the embodiments may be included in at least one embodiment of this application. The phrase appears at various positions in the specification and does not necessarily mean the same embodiment, nor does it mean being independent or alternative to other embodiments and mutually exclusive. Those of ordinary skill in the art can clearly or implicitly understand that the embodiments described in this application can be combined with other embodiments without conflict.

[0140] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., and are not limited thereto.

[0141] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of patent protection. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A geological data retrieval method, characterized in that, Including: Obtain the geological text to be queried; Convert the geological text to be queried into a feature vector to obtain a geological text feature vector to be queried; Calculate the similarity between the geological text feature vector to be queried and each feature vector in a preset multi-modal feature vector pool, where the preset multi-modal feature vector pool is obtained by mapping pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model; Determine the modal weight of the similarity according to a preset modal weight adjustment rule, and calculate the comprehensive similarity of the geological text feature vector to be queried according to the modal weight and the similarity; Obtain a corresponding initial geological query result according to the comprehensive similarity; Sort the initial geological query result according to the comprehensive similarity to obtain the target query result of the geological text to be queried.

2. The geological data retrieval method according to claim 1, characterized in that, Before converting the geological text to be queried into a feature vector, the method further includes: Obtain the original text to be queried, perform denoising and word segmentation processing on the original text to be queried to obtain the geological text to be queried.

3. The geological data retrieval method according to claim 1, wherein The training process of the trained cross-modal model includes: Clean the geological text training data and label the corresponding geological terms to obtain the cleaned geological text training data; Extract the attribute relationship between the geological terms and the geological terms, construct triples, and construct a geological knowledge graph based on the triples; Train an initial cross-modal model based on the cleaned geological text training data to obtain an initially trained cross-modal model; Use the loss function of the geological knowledge graph as a regularization term to perform regularization constraints on the initially trained cross-modal model to obtain the trained cross-modal model.

4. The geological data retrieval method according to claim 1, characterized in that Determining the modal weight of the similarity according to a preset modal weight adjustment rule includes: Detect keywords in the geological text to be queried; Determine the modal weight corresponding to the keyword according to the keyword.

5. The geological data retrieval method according to claim 1, characterized in that, The construction process of the preset multi-modal feature vector pool includes: Obtain geological text data, geological image data, and sensor data; Extract the feature vector of the geological text data through the trained cross-modal model to obtain a geological text feature vector; Extract the feature vector of the geological image data through the trained cross-modal model to obtain a geological image feature vector; Extract the feature vector of the sensor data through a preset neural network model to obtain a geological parameter feature vector; Map the geological text feature vector, the geological image feature vector, and the geological parameter feature vector to a unified semantic space to obtain the multi-modal feature vector pool.

6. The geological data retrieval method according to claim 1, wherein The target query result includes the unique identifier of the geological parameter and the location information of the geological parameter.

7. The geological data retrieval method according to claim 1, wherein, After obtaining the target query result of the geological text to be queried, the method further includes: Generate an interactive map and / or an analysis report according to the target query result.

8. A geological data retrieval device, characterized in that, Including: An acquisition module, a feature vector conversion module, a first calculation module, a second calculation module, and an output module, where The acquisition module is used to acquire the geological text to be queried; The feature vector conversion module is used to convert the to-be-query geological text into a feature vector to obtain the to-be-query geological text feature vector; The first calculation module is used to calculate the similarity between the to-be-query geological text feature vector and each feature vector in the preset multi-modal feature vector pool, wherein the preset multi-modal feature vector pool is obtained by mapping the pre-collected geological text data, geological image data, and sensor data to a unified semantic space through a trained cross-modal model; The second calculation module is used to determine the modal weight of the similarity according to the preset modal weight adjustment rule, and calculate the comprehensive similarity of the to-be-query geological text feature vector according to the modal weight and the similarity; The output module is used to obtain the corresponding initial geological query result according to the comprehensive similarity; sort the initial geological query result according to the comprehensive similarity to obtain the target query result of the to-be-query geological text.

9. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is set to run the computer program to execute the geological data retrieval method 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 computer program is executed by the processor, the steps of the geological data retrieval method according to any one of claims 1 to 7 are implemented.

Images