Early warning method for inclined shaft TBM (Tunnel Boring Machine) deslagging system based on cloud side-end cooperation

Abstract

The invention provides an inclined shaft TBM (Tunnel Boring Machine) deslagging system early warning method based on cloud edge-end cooperation, which comprises the following steps of: acquiring an image of slag stone in an inclined shaft TBM slag chute through an acquisition terminal, and transmitting sample data of the image of the slag stone to an edge layer; the edge layer receives and preprocesses the slag stone image sample data, and transmits the processed image data to the cloud layer; dividing sample data into a training set and a test set according to a certain proportion for training an image recognition model, and updating the recognition model to an edge layer; the marginal layer identifies and classifies slag stone accumulation through the updated identification model and transmits signals to the terminal; and the terminal scouring equipment correspondingly scours the slag stones according to the transmission signals. According to the invention, the particle movement condition of the slag chute can be displayed in real time, the rock slag accumulation condition of the slag chute can be efficiently monitored, and the easy accumulation area can be scoured, so that the scour is more purposeful, the resource consumption is reduced, and the construction efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent monitoring of hard rock tunnel shield machines, and in particular relates to an early warning method for an inclined shaft TBM slag discharge system based on cloud-edge-end collaboration. Background Art

[0002] In recent years, the construction of subway tunnels and mountain tunnels has grown rapidly. After the localization of shield machines, they have been widely used in subway tunnels, mountain tunnels and other projects. Among them, TBM is mainly used for hard rock tunnel excavation. During the excavation process, most TBMs can only work in the horizontal direction, while the high-angle inclined shaft TBM can cope with larger slopes. Its structure enables it to excavate stably on larger slopes. At present, the high-angle inclined shaft TBM is mainly used for tunnel excavation in water conservancy projects, such as reservoirs, hydropower stations and other projects. The use of high-angle inclined shaft TBM can significantly improve the efficiency of tunnel excavation and reduce the impact on the environment. It plays an important role in modern tunnel engineering and provides a more efficient and reliable solution for the implementation of engineering projects. The high-angle inclined shaft TBM cuts the rock and soil materials into small particles through the cutting head. The slag cut off can rely on its own gravity to pass through the slag chute due to the influence of the inclination angle, and is transported to the horizontal conveyor belt at the bottom, and finally transported out of the tunnel by the transport vehicle. Due to the shape and viscosity of the cut slag, the slag may accumulate in the slag chute and cannot be discharged in time. Once too much slag accumulates in the slag chute, it will have a serious impact on the slag chute, which may cause the slag chute to collapse and cause a safety accident. It will also affect the normal progress of TBM excavation work and have a serious impact on construction efficiency and project progress. How to realize intelligent monitoring of the slag discharge system and timely warn and deal with the accumulation of slag has an important impact on construction safety and project progress. Summary of the invention

[0003] In response to the problem that slag accumulates in the slag chute and cannot be discharged in time, affecting the construction progress, the present invention proposes an inclined shaft TBM slag discharge system early warning method based on cloud-edge-end collaboration, and realizes the inclined shaft TBM slag discharge system early warning based on cloud-edge-end collaboration through collaborative cooperation between different levels; at the same time, a slag discharge system including a terminal, an edge layer and a cloud layer is constructed. Through the cloud-edge-end architecture, the system can respond quickly locally, process images and control devices in real time, reduce delays, and at the same time improve the scalability and long-term performance of the system through the powerful computing power and storage resources of the cloud.

[0004] In order to achieve the above object, the technical solution of the present invention is achieved as follows:

[0005] An early warning method for inclined shaft TBM deslagging system based on cloud-edge-end collaboration builds a deslagging system including terminal, edge layer and cloud layer. Through the cloud-edge-end architecture, the deslagging system responds quickly locally, processes images and controls devices in real time, and reduces delays. The steps are as follows:

[0006] S1: According to the working conditions of the TBM project, terminal equipment is built, mainly including acquisition terminal, flushing terminal and display terminal;

[0007] S2: The edge layer receives the image data set obtained by the acquisition terminal and preprocesses the image data set;

[0008] S3: The cloud layer receives edge layer data and uses a deep learning model to train the preprocessed images to obtain an image recognition model;

[0009] S4: The edge layer updates the image recognition model, recognizes the real-time image to be processed, and feeds back the recognition information to the terminal;

[0010] S5: The terminal receives the identification information and makes flushing decisions and warnings based on the identification information.

[0011] Preferably, the acquisition terminal is to capture images of the accumulation of slag in the slag chute by using an image acquisition device;

[0012] The display terminal displays the slag and rock images collected in real time by the collection terminal according to a certain image ratio;

[0013] The flushing terminal receives the execution signal given by the edge layer and performs flushing operations of different degrees on the slag accumulation area.

[0014] Preferably, the step S1 specifically includes the following steps:

[0015] S1.1: Statistical analysis of particle size distribution in the slag chute;

[0016] S1.2: Collect statistics on the movement of rock slag particles, including rock slag velocity and mass;

[0017] S1.3: By setting the rock slag velocity threshold and mass threshold, the area where the rock slag velocity is less than the threshold or the rock slag mass is greater than the threshold is calibrated as the rock slag easy accumulation area;

[0018] S1.4: Install the image acquisition terminal in the area where the slag is likely to accumulate, marked on the slag chute, and install it above the slag chute through a bracket, and display the acquired image through the display terminal;

[0019] S1.5: Align the image acquisition terminal with the area of ​​the flushing terminal to achieve flushing of the corresponding area.

[0020] Preferably, the method for preprocessing the image data set in step S2 includes filtering noise reduction, image enhancement and image filtering; specifically includes the following steps:

[0021] S2.1: receiving the original image data set acquired by the terminal device;

[0022] S2.2: Use image filter to remove noise from the image;

[0023] S2.3: Compare the RGB color model and the HSV color model and perform dimensionality reduction on the image;

[0024] S2.4: Perform background segmentation on the image and use the optimal threshold to segment the image to obtain a binary image.

[0025] Preferably, the method of using a deep learning model to train the preprocessed image in step S3 includes:

[0026] S3.1: The preprocessed image samples are aggregated and divided into a training set and a prediction set according to a certain ratio;

[0027] S3.2: A deep learning network model is used, which is mainly divided into the input layer, the trunk part, the neck and the output layer, to extract features from the image samples and capture various features in the image;

[0028] S3.3: The model is iteratively trained using the training set, and the trained model is tested using the test set;

[0029] S3.4: Establish an accuracy evaluation formula, adjust the weight of each channel, suppress useless or less effective features, and train the model until the accuracy evaluation index fluctuates less and tends to be stable.

[0030] Preferably, the edge layer is used to realize real-time image recognition and stacking level judgment, model reasoning, and local data caching. At the same time, the edge layer stores temporary image data and processing results, quickly responds to terminal requests, and avoids frequent data requests from the cloud layer.

[0031] Preferably, the cloud layer centrally manages and stores large amounts of data, trains complex deep learning models, and performs tasks that require large computing resources.

[0032] Preferably, the collaborative work of the cloud layer, edge layer and terminal includes:

[0033] The cloud layer provides overall data analysis and model training support, while the edge layer implements more complex real-time reasoning tasks. After optimization and updating of the cloud layer, the edge layer can perform new reasoning tasks.

[0034] The edge layer uploads data to the cloud layer through the Internet or a dedicated network for large-scale storage, long-term analysis, and model training; the cloud layer trains the model and updates the image recognition model of the edge layer;

[0035] The terminal device communicates with the edge computing device through the communication protocol; the image data is transmitted from the terminal device to the edge layer for real-time processing, the edge layer performs preliminary analysis and recognition, and feeds back the results to the terminal device or the cloud layer.

[0036] A computer device is used to execute an inclined shaft TBM slag discharge system early warning method based on cloud-edge-end collaboration, including a processor, a memory and a network interface. The processor is connected to the memory, and the memory stores instructions; the processor is used to execute the instructions stored in the memory; at the same time, the network interface connects the computer device and other electronic devices.

[0037] Beneficial effects of the present invention:

[0038] 1) The present invention realizes image acquisition, image display and flushing at the terminal by building a cloud-edge-end architecture, and realizes real-time image recognition and cloud-based training recognition model at the edge layer, thereby achieving real-time monitoring of the accumulation of slag chutes and timely unblocking. Compared with the traditional method of continuous flushing with water, the present invention can efficiently monitor the accumulation of slag in the slag chutes, and the flushing is more purposeful, reducing resource consumption and improving construction efficiency.

[0039] 2) The present invention can display the movement of slag chute particles in real time, can efficiently monitor the accumulation of slag in the slag chute, and can flush the areas prone to accumulation, making the flushing more purposeful, reducing resource consumption and improving construction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A schematic diagram of the installation of terminal equipment of the slag discharge system constructed in the present invention.

[0042] Figure 2 This is the main flow chart of the early warning method proposed in the present invention.

[0043] Figure 3 This is a detailed flow chart of the present invention using a deep learning model for training.

[0044] Figure 4A schematic diagram of the framework of the slag discharge system constructed in the present invention.

[0045] Figure 5 Schematic diagram of computer equipment for the slag removal system constructed for the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The embodiment of the present invention provides an early warning method for the TBM slag discharge system in an inclined shaft based on cloud-edge-end collaboration, constructing a slag discharge system including a terminal, an edge layer, and a cloud layer. Through the cloud-edge-end architecture, the slag discharge system can respond quickly locally, process images and control devices in real time, reduce delays, and improve the scalability and long-term performance of the system through the powerful computing power and storage resources of the cloud. Among them, the terminal is used for data acquisition, image display, and flushing control; the edge layer is used for real-time image recognition and accumulation level judgment, model reasoning, and local data caching; the cloud layer is used for big data storage, deep learning training and optimization, and model management and update.

[0048] According to the working conditions of the TBM project, this embodiment provides a terminal device construction of an early warning method for a TBM slag removal system in a inclined shaft based on cloud-edge-end collaboration, such as Figure 1 As shown, the terminal can be built in multiple projects at the same time and is not limited by the number of projects; and the terminal equipment mainly includes a collection terminal, a flushing terminal and a display terminal. The specific steps of implementing the terminal equipment of the present invention are as follows:

[0049] S101: Counting the particle size distribution in the TBM slag chute 3 of the inclined shaft;

[0050] S102: Collecting statistics on the movement of slag particles, including slag velocity and mass;

[0051] S103: by setting a rock slag velocity threshold and a mass threshold, etc., calibrating an area where the rock slag velocity and mass are less than the threshold as an area where rock slag is prone to accumulate;

[0052] S104: installing the image acquisition terminal 5 in the area where slag is likely to accumulate, which is marked in the slag chute of the inclined shaft TBM, and installing it above the slag chute through the bracket 2, and displaying the acquired image through the display terminal;

[0053] S105: Match the image acquisition terminal 5 with the area of ​​the flushing terminal 6 to achieve flushing of the corresponding area.

[0054] There is no limit on the installation location of all terminal devices, but it is guaranteed that the installation location can effectively perform tasks in the project area. Image acquisition terminals are not limited to cameras, scanners, infrared cameras and other image acquisition devices; image display terminals are not limited to the resolution, type and size of the display screen; flushing terminals include a variety of flushing types, including fixed flushing, adjustable flushing, rotating flushing, automatic rotating flushing, fan-shaped flushing, cone flushing, multi-hole flushing, etc.

[0055] The flushing is set to three levels but not limited to three levels, including heavy accumulation, moderate accumulation and light accumulation. The flushing terminal performs different flushing tasks according to different signals.

[0056] like Figure 2 The figure is a flow chart of the early warning method for the inclined shaft TBM slag discharge system based on cloud-edge-end collaboration provided in this embodiment. The TBM slag discharge early warning method is applied to the inclined shaft TBM slag discharge system based on cloud-edge-end collaboration. The early warning method for the inclined shaft TBM slag discharge system based on cloud-edge-end collaboration includes:

[0057] S201: collecting slag images in the slag chute of the TBM in the inclined shaft through the collection terminal, and transmitting the slag image sample data to the edge layer;

[0058] S202: The edge layer receives and pre-processes the slag rock image sample data, and transmits the processed image data to the cloud layer;

[0059] S203: Divide the sample data into a training set and a test set according to a certain ratio for training the image recognition model, and update the recognition model to the edge layer;

[0060] S204: The edge layer identifies and classifies the slag accumulation through the updated recognition model and transmits the signal to the terminal;

[0061] S205: The terminal flushing equipment flushes the slag stone accordingly according to the transmission signal.

[0062] The edge layer preprocesses the image sample data. The image preprocessing includes filtering and noise reduction, color model selection and grayscale, and background removal, but is not limited to these links. The main purpose of image preprocessing is to reduce the amount of calculation, remove interference information in the image, and improve the detectability of image-related information. The specific steps are as follows:

[0063] S2.1: Use image enhancement technology to increase the actual working condition image data set;

[0064] S2.2: Use an image filter to remove noise in the image; for example, use a median filter to remove salt and pepper noise in the image and smooth the image by calculating the median of the neighborhood around the pixel.

[0065] S2.3: Compare the RGB color model and the HSV color model, where the formula for converting the RGB color model to the HSV color model is:

[0066]

[0067] v=max

[0068] Among them, max represents the maximum of the r, g, and b components of the pixel point, and min represents the minimum of the r, g, and b components of the pixel point; the image is processed by dimensionality reduction, that is, the image is grayed to reduce the amount of calculation. The RGB format image is divided into R component image, G component image, and B component image by using the three-part method, and the HSV format image is divided into H component image, S component image, and V component image; the differences of the six color component images are compared, and the component image with obvious distinction between the slag and the background, such as the V component image, is selected to facilitate subsequent image processing;

[0069] S2.4: Perform background segmentation on the image; for example, use the Otsu algorithm to segment the image foreground and background, calculate the image histogram, and normalize it. The normalization formula is:

[0070]

[0071] In the formula, X is the gray value of the image, Xmin is the minimum value of the data, and Xmax is the maximum value of the data;

[0072] Traverse all possible thresholds and calculate the inter-class variance corresponding to each threshold. The calculation formula is:

[0073] σ 1 =ω 1 ×ω 2 ×(μ 1 -μ 2 ) 2

[0074] In the formula, ω 1 represents the probability of the prospect appearing, ω 2 represents the probability of background appearance, μ 1 is the average gray value of the foreground, μ 2 is the average gray value of the background;

[0075] The threshold t that maximizes the inter-class variance is selected as the optimal threshold, and the image is segmented using the optimal threshold to obtain a binary image.

[0076] The cloud layer receives preprocessed sample data, such as Figure 3 Use the deep learning model to learn and train the processed images. For example, take the YOLOv8 network model as an example and use the preprocessed image dataset as a training sample. The specific steps are as follows:

[0077] S301: Summarize the preprocessed image samples and divide them into a training set and a prediction set according to a ratio of 7:3;

[0078] S302: The YOLOv8 deep learning network model used is divided into an input layer, a trunk, a neck, and an output layer to extract features from image samples and capture various features in the image;

[0079] S303: the model is iteratively trained by the training samples, and the trained model is tested by the test set;

[0080] S304: YOLOv8 accuracy evaluation formula:

[0081] mAP=(AP_1+AP_2+...+AP_n) / N

[0082] mAP is a comprehensive performance evaluation indicator that combines the accuracy and recall rate in target detection, where AP_1, AP_2, ..., AP_n represent the average precision of the model in each category.

[0083] S305: Add a squeeze and excitation (SE) module to adjust the weights of each channel, suppress useless or less effective features, and train the model until the mAP fluctuates less and tends to be stable.

[0084] According to the recognition model, a classification standard is further established. For mild accumulation: the proportion of slag and stone is <30%; for moderate accumulation: 30%≤slag and stone proportion≤70%; for severe accumulation: the proportion of slag and stone is >70%. At the same time, the confidence of the classification results can be evaluated. The edge layer classifies and recognizes the images collected by the terminal through training and updating the recognition model, sends different signals to the terminal, and the flushing terminal executes different flushing results. Further, for the flushing terminal, when the degree of slag and stone accumulation reaches moderate or severe, an audible and visual alarm is triggered. Adjust the parameters of the flushing equipment according to the degree of accumulation: mild accumulation: short-term low-pressure flushing; moderate accumulation: medium-time and pressure flushing; severe accumulation: long-term high-pressure flushing. Further monitor the operating status of the flushing equipment in real time and record the operating data.

[0085] like Figure 4 As shown in the figure, the TBM slag discharge early warning method is applied to the inclined shaft TBM slag discharge system based on cloud-edge-end collaboration. The inclined shaft TBM slag discharge system early warning method based on cloud-edge-end collaboration includes:

[0086] For the image acquisition terminal, image data of the slag and rock area is collected through physical equipment to provide image and depth information.

[0087] For the image display terminal, the slag rock image collected in real time by the physical device is displayed according to a certain image ratio. The image zoom, move and area selection functions are provided to facilitate observation.

[0088] For the flushing terminal, it receives the execution signal given by the edge layer and performs flushing operations of different degrees on the slag accumulation area.

[0089] For the edge layer, temporary image data and processing results are stored to quickly respond to device-side requests and avoid frequent data requests from the cloud. It is mainly used to achieve real-time image recognition and stacking level judgment, model reasoning, and local data caching.

[0090] For the cloud layer, efficient storage services (such as cloud databases, object storage, and distributed storage systems) are provided based on the cloud to store historical image data, training data sets, model weights, log information, etc. The cloud can perform large-scale data analysis and processing, such as slag accumulation trend prediction, historical data query, and system performance optimization.

[0091] The collaborative work between the cloud layer, edge layer, and terminals includes:

[0092] The cloud layer provides overall data analysis and model training support, while the edge layer implements more complex real-time reasoning tasks. After optimization and updating of the cloud layer, the edge layer can perform new reasoning tasks.

[0093] The edge layer uploads data to the cloud layer through the Internet or a dedicated network for large-scale storage, long-term analysis, and model training; the cloud layer trains the model and updates the image recognition model of the edge layer;

[0094] The terminal device communicates with the edge computing device through the communication protocol; the image data is transmitted from the terminal device to the edge layer for real-time processing, the edge layer performs preliminary analysis and recognition, and feeds back the results to the terminal device or the cloud layer.

[0095] like Figure 5 As shown, the present invention provides a computer device, including: at least one processor 51; and a memory 52 communicatively connected to the at least one processor 51; wherein the memory 52 stores instructions executable by the at least one processor 51, and the instructions are executed by the at least one processor 51, so that the at least one processor 51 can execute the above-mentioned inclined shaft TBM slag discharge system early warning method based on cloud-edge-end collaboration, and at the same time, the network interface 53 connects the computer device and other electronic devices.

[0096] Processor 51 is mainly used for image preprocessing: performing filtering, noise reduction, graying, background removal and other operations to improve image quality. Image training: Use a deep learning framework (such as TensorFlow, PyTorch) to train the collected image data. Establish a classification model and image recognition based on the degree of slag accumulation: Use the trained model to perform real-time recognition on the collected images and output the accumulation level. At the same time, the processor types include central processing units (CPUs), graphics processing units (GPUs) and artificial intelligence accelerators (such as TPUs, FPGAs), etc. At the same time, it is necessary to ensure that the data transmission bandwidth between the processor and the memory is high enough and optimize the program algorithm to reduce the load on the processor and improve real-time performance.

[0097] The memory 52 is used to store computer programs (system operating procedures, algorithm models, parameter configurations, etc.); store the collected slag image data and intermediate processing results; and save the trained models and classification results for subsequent calls. The types and functions of the storage include random access memory (RAM): used to temporarily store running programs and real-time collected image data; cache intermediate results in the calculation process to improve processing efficiency. Read-only memory (ROM): stores fixed initialization programs and firmware to ensure that the system can start and run, and non-volatile memory: including hard disk (HDD), solid-state drive (SSD) or embedded flash memory (eMMC); used for long-term storage of historical image data, classification results and training models.

[0098] The network interface 53 may include a wireless network interface or a wired network interface. The network interface 53 is generally used to establish a communication connection between the computer device and other electronic devices.

[0099] Through the above solution, the terminal device can communicate with the edge computing device through the local area network (LAN), Wi-Fi, Bluetooth or ZigBee. Image data is transmitted from the terminal device to the edge layer for real-time processing. The edge layer performs preliminary analysis and recognition and feeds the results back to the terminal device or cloud layer. The edge layer can upload data to the cloud layer through the Internet or a dedicated network for large-scale storage, long-term analysis and model training. The cloud layer trains the model and can regularly update the recognition model of the edge layer through OTA. The cloud layer provides overall data analysis and model training support, while the edge layer undertakes more complex real-time reasoning tasks. After cloud optimization and updating, the edge layer can perform new reasoning tasks, thereby improving the real-time and accuracy of the system.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration, characterized in that: Build a slag removal system including the terminal, edge layer, and cloud layer. Through the cloud-edge-end architecture, the slag removal system responds quickly locally, processes images and controls devices in real time, and reduces latency. The steps are as follows: S1: According to the working conditions of the TBM project, terminal equipment is built, mainly including acquisition terminal, flushing terminal and display terminal; S2: The edge layer receives the image data set obtained by the acquisition terminal and preprocesses the image data set; S3: The cloud layer receives edge layer data and uses a deep learning model to train the preprocessed images to obtain an image recognition model; S4: The edge layer updates the image recognition model, recognizes the real-time image to be processed, and feeds back the recognition information to the terminal; S5: The terminal receives the identification information and makes flushing decisions and warnings based on the identification information.

2. The early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The acquisition terminal uses an image acquisition device to capture images of the accumulation of slag in the slag chute; The display terminal displays the slag and rock images collected in real time by the collection terminal according to a certain image ratio; The flushing terminal receives the execution signal given by the edge layer and performs flushing operations of different degrees on the slag accumulation area.

3. The early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The step S1 specifically includes the following steps: S1.1: Statistical analysis of particle size distribution in the slag chute; S1.2: Collect statistics on the movement of rock slag particles, including rock slag velocity and mass; S1.3: By setting the rock slag velocity threshold and mass threshold, the area where the rock slag velocity is less than the threshold or the rock slag mass is greater than the threshold is calibrated as the rock slag easy accumulation area; S1.4: Install the image acquisition terminal in the area where the slag is likely to accumulate, marked on the slag chute, and install it above the slag chute through a bracket, and display the acquired image through the display terminal; S1.5: Align the image acquisition terminal with the area of ​​the flushing terminal to achieve flushing of the corresponding area.

4. The early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The method for preprocessing the image data set in step S2 includes filtering noise reduction, image enhancement and image filtering; specifically includes the following steps: S2.1: receiving the original image data set acquired by the terminal device; S2.2: Use image filter to remove noise from the image; S2.3: Compare the RGB color model and the HSV color model and perform dimensionality reduction on the image; S2.4: Perform background segmentation on the image and use the optimal threshold to segment the image to obtain a binary image.

5. The early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The method for training the preprocessed image using a deep learning model in step S3 includes: S3.1: The preprocessed image samples are aggregated and divided into a training set and a prediction set according to a certain ratio; S3.2: A deep learning network model is used, which is mainly divided into the input layer, the trunk part, the neck and the output layer, to extract features from the image samples and capture various features in the image; S3.3: The model is iteratively trained using the training set, and the trained model is tested using the test set; S3.4: Establish an accuracy evaluation formula, adjust the weight of each channel, suppress useless or less effective features, and train the model until the accuracy evaluation index fluctuates less and tends to be stable.

6. The early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The edge layer is used to realize real-time image recognition and stacking level judgment, model reasoning, and local data caching. At the same time, the edge layer stores temporary image data and processing results, quickly responds to terminal requests, and avoids frequent data requests from the cloud layer.

7. The early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The cloud layer centrally manages and stores large amounts of data, trains complex deep learning models, and performs tasks that require large computing resources.

8. The early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to claim 1 is characterized in that: The collaborative work between the cloud layer, edge layer, and terminals includes: The cloud layer provides overall data analysis and model training support, while the edge layer implements more complex real-time reasoning tasks. After optimization and updating of the cloud layer, the edge layer can perform new reasoning tasks. The edge layer uploads data to the cloud layer through the Internet or a dedicated network for large-scale storage, long-term analysis, and model training; the cloud layer trains the model and updates the image recognition model of the edge layer; The terminal device communicates with the edge computing device through the communication protocol; the image data is transmitted from the terminal device to the edge layer for real-time processing, the edge layer performs preliminary analysis and recognition, and feeds back the results to the terminal device or the cloud layer.

9. A computer device for executing the early warning method for inclined shaft TBM slag removal system based on cloud-edge-end collaboration according to any one of claims 1 to 8, characterized in that: The device comprises a processor (51), a memory (52) and a network interface (53), wherein the processor (51) is connected to the memory (52), and the memory (52) stores instructions; the processor (51) is used to execute the instructions stored in the memory (52); and the network interface (53) is connected to computer devices and other electronic devices.

Images