Rock-soil layer intelligent analysis method and device based on drilling data

Abstract

The embodiment of the invention discloses a rock-soil layer intelligent analysis method and device based on drilling data. The method comprises the following steps: acquiring to-be-diagnosed drilling data, and constructing structured data corresponding to the drilling data; and based on the structured data, outputting a rock-soil layer analysis result by an analysis model deployed in the vLLM engine. According to the embodiment of the invention, training can be carried out through the lightweight large model formed by injecting the LoRA adapter into the attention mechanism and the feedforward network, and compared with all-parameter fine tuning, the computing resource consumption is remarkably reduced, the training speed is improved, and rapid completion of model adaptation for a specific work area of a drilling stratum on a consumption-level GPU becomes possible. Meanwhile, the model is deployed in a vLLM engine, so that the throughput of online reasoning service is improved by several times, the requirement of a drilling site for data analysis is met, and the actual application effect in a complex stratum mode is improved.

Description

Technical Field

[0001] The embodiments in this specification pertain to the field of borehole geological engineering diagnosis, and specifically relate to a method and apparatus for intelligent analysis of soil and rock layers based on borehole data. Background Technology

[0002] As exploration and development extend into deeper and more complex formations, accurate identification of drilled formations and assessment of downhole conditions are crucial for ensuring borehole safety and improving drilling efficiency. Currently, there are solutions using general-purpose large models with full parameter fine-tuning for borehole data diagnosis. However, because these solutions simply fine-tune general-purpose large models used in other fields, they result in enormous computational resource requirements, slow model iteration speeds, and long training cycles. Furthermore, the cost of such lengthy and resource-intensive training is prohibitive for most borehole operations or small to medium-sized technical service companies. Secondly, general-purpose inference frameworks perform poorly in terms of inference speed (throughput) and latency when handling long sequences and high-concurrency requests. In real-time borehole decision-making scenarios, such response delays of several seconds or even longer are unacceptable. In summary, existing solutions based on full parameter fine-tuning of general-purpose large models are not ideal in actual borehole production applications. Summary of the Invention

[0003] The embodiments of this disclosure provide a method and apparatus for intelligent analysis of soil and rock layers based on borehole data, which aims to solve one or more of the above-mentioned problems and other potential problems.

[0004] According to a first aspect of this disclosure, a method for intelligent analysis of soil and rock layers based on borehole data is provided. The method includes acquiring borehole data to be diagnosed and constructing structured data corresponding to the borehole data. Based on the structured data, an analysis model deployed in a vLLM engine outputs soil and rock layer analysis results, which are used to characterize borehole event categories. The analysis model is trained in a training environment with an Unsloth acceleration framework. The analysis model has a trainable LoRA adapter injected into the attention mechanism of the Transformer layer and the feedforward network.

[0005] According to a second aspect of this disclosure, a smart geotechnical analysis device based on borehole data is provided. The device includes a data acquisition module configured to acquire borehole data to be diagnosed and construct structured data corresponding to the borehole data; and a model prediction module configured to output geotechnical analysis results based on the structured data and an analysis model deployed in a vLLM engine. The geotechnical analysis results are used to characterize the stratigraphic state of the geotechnical layers. The analysis model is trained in a training environment with an Unsloth acceleration framework. The analysis model has a trainable LoRA adapter injected into the attention mechanism of the Transformer layer and the feedforward network.

[0006] According to a third aspect of this disclosure, an electronic device is provided, including one or more processors and a memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform a method provided according to a first scheme.

[0007] According to a fourth aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method provided according to the first aspect.

[0008] The solution provided in this specification, through training a lightweight large model (i.e., an analysis model) composed of an attention mechanism and a LoRA adapter injected into a feedforward network, significantly reduces computational resource consumption and improves training speed compared to full parameter fine-tuning. This makes it possible to quickly adapt the model to specific work areas of the drilled formation on consumer-grade GPUs. Simultaneously, deploying the model on the vLLM engine increases the throughput of the online inference service several times over, meeting the data analysis requirements of the drilling site and improving the practical application effect of the solution in complex geological patterns. Attached Figure Description

[0009] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0010] Figure 1 A flowchart illustrating an intelligent analysis method for soil and rock layers based on borehole data, according to some embodiments of this disclosure, is shown.

[0011] Figure 2 A schematic diagram of the structure of an intelligent rock and soil layer analysis device based on borehole data, according to some embodiments of this disclosure, is shown.

[0012] Figure 3 A schematic block diagram of an electronic device according to some embodiments of the present disclosure is shown. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] The terms “comprising” and “having”, and any variations thereof, in this specification, claims, and the foregoing drawings are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. Depending on the context, the word “if” as it applies herein may be interpreted as “when”, “in response to determination”, or “in response to detection”.

[0015] Figure 1 A flowchart illustrating a method 100 for intelligent analysis of soil and rock layers based on borehole data, according to some embodiments of this disclosure, is shown. Method 100 can be executed, for example, by a terminal, which may include, but is not limited to, mobile phones, tablets, desktop computers, servers, etc. Figure 1 As shown, in method 100, step 102 can obtain the borehole data to be diagnosed and construct the structured data corresponding to the borehole data.

[0016] In this embodiment, after drilling begins, the drilling parameter system can acquire drilling data in real time for diagnostic purposes. This drilling data can include time, hole depth, drilling speed, rotational torque, water pressure, air pressure, impact frequency, shock wave waveform, thrust, rotational speed, slag return status, and data acquisition status. Time can be a precise timestamp of the data record, used to synchronize all parameters, analyze process dynamics, and trace back events. Hole depth can be the current vertical or oblique depth of the drill bit from the surface, used to locate the formation. Drilling speed can be the depth drilled per unit time, used to reflect the drillability of the formation. Rotational torque can be the torsional torque required to overcome formation resistance when the drill rod rotates. Water pressure can be the mud circulation pressure at the drill pump outlet, or the water supply pressure of the hydrodynamic impactor. Air pressure can be the pressure of the flushing air used for slag removal, or the air supply pressure of the pneumatic impactor. Hydraulic pressure can be the oil supply pressure of the hydraulic power source driving the hydraulic impactor. Thrust can be the axial force applied to the drill bit. Impact frequency can be the number of impacts per unit time generated by a hydrodynamic, pneumatic, or hydraulic impactor, or it can be the frequency of shock waves generated during the interaction (collision, friction, and compression) between other drilling tools and the rock and soil strata. The shock wave waveform can be the shape of the shock wave generated by the interaction between the aforementioned impactor or drilling tool and the rock and soil strata, as sensed by a sensor. Rotation speed can be the rotation speed of the drill rod or top drive. Backfill status can be the particle state of the backfill fluid or airflow from the borehole (including particle size, particle shape, lithology, and composition). Acquisition status can be a flag indicating data quality or system operating status, used to indicate whether the current recorded data is valid, whether the sensor is functioning properly, and whether the system is in a stable acquisition mode. After acquiring borehole data, it can be structured to obtain structured data that subsequent analysis models can recognize and process. Structured conversion methods can include converting sensor readings into structured text descriptions.

[0017] In method 100, step 104 can be based on structured data, with the analysis model deployed in the vLLM engine outputting geotechnical analysis results. The geotechnical analysis results are used to characterize the borehole event category. The analysis model is trained in a training environment with the Unsloth acceleration framework set up. The analysis model has a trainable LoRA adapter injected into the attention mechanism of the Transformer layer and the feedforward network.

[0018] In this embodiment, a pre-trained analysis model is used. This model can be based on the Qwen3-4B-2507-Thinking model, which demonstrates a balanced performance in instruction compliance and reasoning. The model structure is also adjusted by injecting trainable LoRA adapters into the attention mechanism and feedforward network of the Transformer layer. This adds a low-rank parameter branch to the model without altering the original weights of the base model, generating a simplified thought chain. This allows the model to quickly learn the domain features of borehole multimodal time-series signals with minimal additional computation, while maintaining the stability of the model's existing language understanding and sequence modeling capabilities. Thus, the model does not update all its parameters during training, eliminating the need for costly full-parameter fine-tuning, and can make correct judgments regarding high-frequency and noisy geological events such as sudden advances and stuck drill bits. Furthermore, the model is not trained in an unoptimized general training environment, but in a training environment with the Unsloth acceleration framework enabled. Through the Unsloth acceleration framework's built-in fusion kernel optimization and efficient memory management technology, operator fusion and contiguous memory allocation can be automatically performed in the background while keeping the model's original knowledge unchanged. This reduces the training task that originally required tens of GB of GPU memory to only about 10 GB, and allows for rapid convergence within a few hours. The result is a lightweight model weight file that incorporates drilling domain knowledge, enabling the model to quickly learn the domain features of drilling data with extremely low computational overhead.

[0019] In practical applications, the trained analysis model is deployed in the vLLM inference engine for uniformity. This leverages the vLLM engine's PagedAttention memory management mechanism, which efficiently manages key-value caches for different inference requests within GPU memory, similar to how an operating system manages virtual memory through paging. This avoids memory fragmentation caused by varying sequence lengths. By efficiently handling the key-value cache generated during model inference, memory fragmentation and waste are significantly reduced. This allows for throughput far exceeding traditional frameworks and extremely low inference latency when facing continuous streaming requests from borehole data, ensuring real-time output of analysis results even in complex geological conditions. Ultimately, the analysis model can output corresponding geotechnical analysis results based on the input structured data. As an example, the results of the geotechnical analysis can be expressed as follows: "[2025 / 08 / 18 14:22:37-2025 / 08 / 18 14:22:39] a sudden advance occurred, [2025 / 08 / 18 14:22:55-2025 / 08 / 18 14:22:57] a sudden advance occurred, [2025 / 08 / 18 14:23:17-2025 / 08 / 18 14:23:19] a sudden advance occurred."

[0020] In one possible implementation, structured data corresponding to the borehole data is constructed, including:

[0021] After preprocessing the borehole data, the borehole data at each time step is spliced ​​into text description data according to a preset template format, and the text description data is then converted into structured data.

[0022] In this embodiment, after acquiring the borehole data, it is first preprocessed. Preprocessing may include, for example, aligning timestamps of data from different physical sources within a fixed time window (e.g., 100 seconds) and filtering out spike noise caused by sensor jitter using a sliding window averaging method. Next, multiple sensor readings from each time step in the borehole data are dynamically concatenated according to a pre-set template format to obtain a structured text description data. This text description data is then serialized to obtain structured data, for example, a JSONL format file.

[0023] As an example, an instance constructed using a preset template format can be represented as "Time, Hole Depth / cm, Drilling Speed ​​cm / s, Rotational Torque / Nm, Water Pressure / bar, Thrust / kN, Rotational Speed / rpm, Acquisition Status;\n 2025-08-18 14:21:40, 1563, 1.7, 500, 10.0, 10, 80, 1;\n 2025-08-18 14:21:41, 1564, 1.7, 500, 10.0, 10, 80, 1;\n 2025-08-18 14:21:42, 1565, 2.0, 500, 10.0, 10, 80, 1,....".

[0024] In one possible implementation, the method further includes:

[0025] Based on historical diagnostic data, a training set is determined, which includes structured data samples and soil and rock layer analysis results samples.

[0026] Based on structured data samples, the initial model generates predicted geotechnical layer analysis results. The initial model is injected with LoRA adapters in the Query projection layer and Value projection layer of the attention mechanism, as well as the upper projection layer and lower projection layer of the feedforward network.

[0027] Using the soil and rock layer analysis results as a monitoring signal, the initial model is trained for at least one round to obtain the analysis model.

[0028] In this embodiment, based on historical diagnostic data, structured data samples obtained from historical diagnostic tasks and soil and rock layer analysis result samples labeled after manual diagnosis based on the structured data samples can be acquired. A training set for the model is then constructed based on these structured data samples and soil and rock layer analysis result samples. During the initial model training process using the training set, the model's generator can generate predicted soil and rock layer analysis results based on the structured data samples. The generator loss is obtained by comparing the soil and rock layer analysis result samples with the predicted soil and rock layer analysis results. This generator loss is then used to assess the loss of the predicted soil and rock layer analysis results. The loss assessment can utilize a comparison loss function (e.g., a labeled smoothed cross-entropy loss function). The generator loss can be a relatively large value, and it can then be backpropagated to the generator to guide the optimization of its parameters, achieving one round of supervised training. This training process can be iteratively executed round after round until the generator can generate more accurate predicted soil and rock layer analysis results, i.e., until the loss value calculated by the loss function is smaller. After training, the analysis model can output the soil and rock layer analysis results.

[0029] In one possible implementation, the method further includes:

[0030] The trained analysis model is evaluated using the evalscope evaluation framework;

[0031] In response to the evaluation results indicating that the overall accuracy is higher than the first threshold and the F1 score of each key subclass is higher than the second threshold, the analysis model is deployed to the vLLM engine.

[0032] In this embodiment, another problem with existing model evaluation methods is that they are too general and lack fine-grained diagnosis of key but rare events in the field of geological engineering. Therefore, for the trained analysis model, the evalscope evaluation framework is also used to evaluate the model. The evaluation script of the evalscope framework not only calculates the accuracy of the model on the overall test set, but also automatically classifies the predicted results and true labels into key subcategories such as "stuck drill bit," "breakthrough," and "sudden drop" according to a predefined label mapping table, and generates an independent performance report for each subcategory, including precision, recall, and F1 score (i.e., the model's performance score on the selected task), thereby focusing on the model's performance on "long-tail events" with few training samples but high engineering value. Only when the overall accuracy is higher than the first threshold and the F1 score of each key subcategory is higher than the second threshold, is the analysis model considered to meet the requirements of geological diagnosis at the borehole site and can be deployed to the vLLM engine for use. The first and second thresholds can be set according to actual needs, and generally the first threshold can be slightly higher than the second threshold. Key subcategories can include various strata (such as sandstone, mudstone, and carbonate rocks) and various drilling events (such as stuck drill, breakout, and sudden drop).

[0033] As an example, evalScope first loads a standardized dataset from a built-in or user-specified benchmark. These datasets typically contain questions, options, context, reference answers, and corresponding metadata. Next, its Dataset Adapter transforms each data point into a prompt format required by the model; for example, multiple-choice questions are constructed as a uniform template of "question + options," and math problems are constructed as a "question + solution requirements" structure. Then, evalScope calls the Model Adapter to send input to the model in a consistent inference mode (such as greedy, temperature=0, or chain-of-thought control mode) and collects the model's raw output. The system then calls the corresponding AnswerParser based on the specific task type to parse the free text generated by the model into the final evaluable answer (such as extracting A / B / C / D, extracting the last number, removing the inference chain and retaining only the conclusion, normalizing units, etc.). The parsed prediction is compared with the standard answer, and accuracy, exact match (EM), and token-level are calculated according to the task definition. The evaluation criteria include match, numerical error, step accuracy, and macro / micro metrics. For complex tasks such as mathematical reasoning or multi-step problem-solving, evalScope also utilizes its judge module to automatically score using another large model, addressing the issue of unstructured model answers preventing direct comparison. Finally, the system compiles all results, error causes, model outputs, parsing processes, and overall statistical metrics into a structured evaluation report, and can generate visualizations, inter-model comparison charts, or leaderboard summaries as needed. The entire process strictly maintains the same prompt, inference parameters, and parsing logic for each model, ensuring fairness and reproducibility in the evaluation.

[0034] In one possible implementation, the trained analysis model is evaluated based on the evalscope evaluation framework, including:

[0035] Construct a test set, which includes test data that was not used in the training and covers both typical and rare operating conditions;

[0036] The trained analytics model is evaluated using the evalscope evaluation framework based on the test set.

[0037] In this embodiment, in order to better evaluate the model, a test set will be specially constructed. The test data included in the test set will not only be the data that was not used in training, but will also cover typical and rare working conditions, so as to better evaluate each key sub-category.

[0038] In one possible implementation, the method further includes:

[0039] In response to the evaluation results indicating that the recall rate of the target event corresponding to any rare operating condition is lower than the third threshold, iterative optimization suggestions are generated based on the target event analysis model.

[0040] In this embodiment, if the recall rate of the target event corresponding to a rare working condition is low, it means that the model has missed a large number of samples that should have been correctly identified or answered in the task of that key subcategory. Therefore, corresponding iterative optimization suggestions can be generated so that when users iterate and update the model in the future, they can focus on using the training data corresponding to the target event to complete the data supplementation and provide accurate guidance for model iteration.

[0041] Figure 2 Schematic diagrams of the structure of a smart geotechnical analysis device 200 based on borehole data, according to some embodiments of this disclosure, are shown. The various embodiments in this specification are described in a progressive manner; similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments. Figure 2 As shown, the device 200 includes a data acquisition module 201, configured to acquire borehole data to be diagnosed and construct structured data corresponding to the borehole data; and a model prediction module 202, configured to output soil and rock layer analysis results based on the structured data and the analysis model deployed in the vLLM engine. The soil and rock layer analysis results are used to characterize the stratigraphic state of the soil and rock layers. The analysis model is trained in a training environment with the Unsloth acceleration framework. The analysis model has a trainable LoRA adapter injected into the attention mechanism of the Transformer layer and the feedforward network.

[0042] In one possible implementation, the drilling data includes time, hole depth, drilling speed, rotational torque, water pressure, air pressure, impact frequency, shock wave waveform, propulsion force, rotational speed, slag return status, and acquisition status.

[0043] In one possible implementation, the data acquisition module 201 is further configured to preprocess the borehole data, then concatenate the borehole data at each time step into text description data according to a preset template format, and convert the text description data into structured data.

[0044] In one possible implementation, the device further includes a model training module configured to determine a training set based on historical diagnostic data, the training set including structured data samples and soil and rock layer analysis result samples; based on the structured data samples, an initial model generates predicted soil and rock layer analysis results, the initial model having LoRA adapters injected into the Query projection layer and Value projection layer of the attention mechanism, as well as the upper projection layer and lower projection layer of the feedforward network; using the soil and rock layer analysis result samples as supervision signals, the initial model is trained for at least one round to obtain the analysis model.

[0045] In one possible implementation, the device further includes a model evaluation module configured to evaluate the trained analysis model based on the evalscope evaluation framework; in response to the evaluation results indicating that the overall accuracy is higher than a first threshold and the F1 score of each key sub-category is higher than a second threshold, the analysis model is deployed to the vLLM engine.

[0046] In one possible implementation, the model evaluation module is further configured to construct a test set, which includes test data that was not used in training and covers typical and rare operating conditions; based on the test set, the trained analysis model is evaluated by the evalscope evaluation framework.

[0047] In one possible implementation, the model evaluation module is further configured to generate iterative optimization suggestions for the analysis model based on the target event in response to the evaluation result indicating that the recall rate of the target event corresponding to any rare operating condition is lower than a third threshold.

[0048] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this specification is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0049] Figure 3 A block diagram of an electronic device 300 that can implement various embodiments of the present disclosure is shown. For example... Figure 3 As shown, the electronic device 300 includes a processor 310, a disk drive 320, an input / output interface 330, a network interface 340, and a memory 350. The processor 310, disk drive 320, input / output interface 330, network interface 340, and memory 350 can communicate with each other via a communication bus 360.

[0050] The processor 310 can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits to execute relevant programs in order to implement the technical solution provided in this application.

[0051] The memory 350 can be implemented in the form of ROM (Read Only Memory), RAM (Read Access Memory), static memory, dynamic storage devices, etc. The memory 350 can store the operating system 351 used to control the operation of the electronic device 300, and the basic input / output system (BIOS) 352 used to control the low-level operations of the electronic device 300. Additionally, it can store a web browser 353, a data storage management system 354, etc. In summary, when the technical solution provided in this application is implemented through software or firmware, the relevant program code is stored in the memory 350 and is called and executed by the processor 310.

[0052] Input / output interface 330 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0053] Network interface 340 is used to connect a communication module (not shown in the figure) to enable communication and interaction between the device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0054] Bus 360 includes a pathway for transmitting information between various components of the device, such as processor 310, disk drive 320, input / output interface 330, network interface 340, and memory 350.

[0055] It should be noted that although the above-described device only shows the processor 310, disk drive 320, input / output interface 330, network interface 340, memory 350, bus 360, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the method of this application, and does not necessarily include all the components shown in the figures.

[0056] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0057] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. Furthermore, although operations are depicted in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0058] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for intelligent analysis of rock-soil layers based on borehole data, characterized in that, The method comprises: obtaining drilling data to be diagnosed, and constructing structured data corresponding to the drilling data; based on the structured data, an analysis model deployed in a vLLM engine outputs a geotechnical layer analysis result, the geotechnical layer analysis result is used to represent a drilling event category, the analysis model is trained in a training environment provided with an Unsloth acceleration framework, and the analysis model injects a trainable LoRA adapter into the attention mechanism of the Transformer layer and the feedforward network. 2.The intelligent analysis method of rock-soil layer based on drilling data according to claim 1, characterized in that, The drilling data includes time, hole depth, drilling speed, rotary torque, water pressure, air pressure, impact frequency, impact wave form, thrust force, rotation speed, backwash state and collection state. 3.The intelligent analysis method of rock-soil layer based on drilling data according to claim 1, characterized in that, The construction of the structured data corresponding to the drilling data comprises: After preprocessing the drilling data, the drilling data at each time step is spliced into text description data according to a preset template format, and the text description data is converted into structured data. 4.The intelligent analysis method of rock-soil layer based on drilling data according to claim 1, characterized in that, The method further comprises: based on historical diagnosis data, a training set is determined, the training set includes structured data samples and geotechnical layer analysis result samples; based on the structured data samples, an initial model generates predicted geotechnical layer analysis results, the initial model injects LoRA adapters into the Query projection layer and the Value projection layer of the attention mechanism and the upper projection layer and the lower projection layer of the feedforward network, respectively; using the geotechnical layer analysis result samples as a supervision signal, the initial model is trained for at least one round to obtain an analysis model. 5.The intelligent analysis method of rock-soil layer based on drilling data according to claim 1, characterized in that, The method further comprises: based on the evalscope evaluation framework, the trained analysis model is evaluated; in response to the evaluation result representing that the overall accuracy is higher than a first threshold and the F1 score of each key subcategory is higher than a second threshold, the analysis model is deployed to the vLLM engine. 6.The intelligent analysis method of rock-soil layer based on drilling data according to claim 5, characterized in that, The evaluation of the trained analysis model based on the evalscope evaluation framework comprises: constructing a test set, the test set includes test data that does not participate in training and covers typical working conditions and rare working conditions; based on the test set, the trained analysis model is evaluated by the evalscope evaluation framework. 7.The intelligent analysis method of rock-soil layer based on drilling data according to claim 5, characterized in that, The method further comprises: in response to the evaluation result representing that the recall rate of any rare working condition corresponding to a target event is lower than a third threshold, an iterative optimization suggestion for the analysis model is generated based on the target event.

8. A rock-soil layer intelligent analysis device based on borehole data, characterized by, The device comprises: a data acquisition module configured to obtain drilling data to be diagnosed, and construct structured data corresponding to the drilling data; a model prediction module configured to output a geotechnical layer analysis result based on the structured data, the geotechnical layer analysis result is used to represent the formation state of the geotechnical layer, the analysis model is trained in a training environment provided with an Unsloth acceleration framework, and the analysis model injects a trainable LoRA adapter into the attention mechanism of the Transformer layer and the feedforward network.

9. An electronic device, comprising: comprises: one or more processors, and A memory associated with the one or more processors, the memory for storing program instructions that, when read and executed by the one or more processors, perform the steps of the method for intelligent analysis of rock-soil layers based on drilling data according to any one of claims 1-7.

10. Computer program product, characterized in that, A computer program that, when executed by a processor, implements the method for intelligent analysis of rock-soil layers based on drilling data according to any one of claims 1-7.

Images