A surrounding rock displacement real-time observation device for urban underground space engineering

Abstract

The application provides a surrounding rock displacement real-time observation device for urban underground space engineering and relates to the technical field of surrounding rock monitoring.The device comprises a laser source and an electric conversion unit, and an optical fiber is attached to the surface of surrounding rock; a signal stabilizer comprises a signal amplifier, a BPF filter, a signal rectifier and an integrator which are electrically connected in sequence; the signal amplifiers in the two signal processing links are opposite to each other; and a background management system obtains an observation result by processing and receiving surrounding rock displacement data through a neural network. The device solves the problems of the prior art, i.e., the sensor needs to punch holes to observe surrounding rock, and the signal is susceptible to interference, by combining an optical fiber sensor, a signal stabilizer, a collector and a background management system with a deep learning function. The device does not need to punch holes and has strong anti-interference capability. The device can not only monitor continuously for 24 hours, but also can further improve the accuracy of data discrimination as the monitoring time is longer and the data volume is larger.

Description

Technical Field

[0001] This invention relates to the field of surrounding rock monitoring technology, and more specifically, to a real-time observation device for surrounding rock displacement in urban underground space engineering. Background Technology

[0002] Observing surrounding rock displacement typically refers to the long-term or short-term deformation monitoring and recording of underground engineering projects or surface rock masses to understand the movement patterns, stability, and changing trends of the surrounding rock. This information is crucial for assessing and controlling geological hazard risks such as groundwater inrush, ground subsidence, and seismic activity, as well as ensuring the safety and stability of underground engineering projects. Specifically, observing surrounding rock displacement can be used in the following ways: Geological hazard risk assessment: By observing surrounding rock displacement, the deformation of the surrounding rock around underground engineering projects can be obtained, allowing for the prediction and assessment of potential geological hazard risks. Engineering safety assessment: Surrounding rock displacement monitoring can reflect the real-time operational status of underground engineering projects and changes in the surrounding environment, helping to assess the safety and stability of underground engineering projects. Building safety assessment: Observing the displacement of surface rock masses can monitor potential surface movements and seismic activity, enabling timely measures to ensure the safety of buildings. Groundwater management: Changes in surrounding rock displacement can reflect changes in groundwater levels, which is of great value for the scientific and rational management and utilization of groundwater resources. In conclusion, surrounding rock displacement monitoring is an important geological exploration technology with broad application prospects in many fields.

[0003] Although various sensors have continuously improved with the development of science and technology, including inductively modulated frequency displacement sensors, distance measuring sensors, and sliding barometers, and have achieved good results, and these sensors have been applied to surrounding rock displacement monitoring devices, existing surrounding rock displacement monitoring devices still have the following problems. Firstly, when observing surrounding rock displacement, these sensors require drilling into the surrounding rock, which can easily damage it. Secondly, the observation process requires long-term continuous observation, during which the signal is easily interfered with. Current technologies lack adequate design in this regard, resulting in a high probability of interference. Signal interference from the observed surrounding rock can adversely affect the observation results. Furthermore, existing technologies require manual verification of the device's output signals to determine if surrounding rock displacement exists. Although the equipment can operate 24 hours a day, human intervention is insufficient for 24-hour real-time verification. Moreover, human fatigue during work can reduce concentration, potentially affecting the judgment of data and signals. This can lead to data errors and, in severe cases, safety accidents. Therefore, a real-time surrounding rock displacement monitoring device for urban underground space engineering that can at least partially solve the above problems is needed. Summary of the Invention

[0004] To overcome or at least partially solve the above problems, embodiments of the present invention provide a real-time monitoring device for surrounding rock displacement in urban underground space engineering. By combining fiber optic sensors, signal stabilizers, and data acquisition devices with a background management system that has deep learning capabilities, the device solves the problems of existing technologies where sensors require drilling to observe the surrounding rock and the signals are easily interfered with. It eliminates the need for drilling, has strong anti-interference capabilities, and can monitor continuously for 24 hours. Furthermore, as the monitoring time increases and the data volume grows, the accuracy of data identification can be further improved.

[0005] The embodiments of the present invention are implemented as follows:

[0006] This application provides a real-time monitoring device for surrounding rock displacement in urban underground space engineering, comprising:

[0007] The fiber optic sensor includes a laser source for providing a single-wavelength laser, an optical fiber for transmitting the light, and a photoelectric conversion unit for receiving the single-wavelength laser and converting it into a waveform signal, wherein the optical fiber is attached to at least a portion of the surface of the surrounding rock to be observed.

[0008] A signal stabilizer, electrically connected to the photoelectric conversion unit, is used to process and stabilize the waveform signal. It has two signal processing links, which are combined into a single signal output via a signal coupler. Each signal processing link includes, in sequence, a signal amplifier, a BPF filter, a signal rectifier, and an integrator. The signal amplifiers in the two signal processing links are out of phase. The output of one integrator is electrically connected to one input of the signal coupler, and the output of the other integrator is electrically connected to the other input of the signal coupler.

[0009] The signal acquisition unit is electrically connected to the output terminal of the signal coupler and the data transmission device, respectively, and is used to receive the displacement signal output by the signal stabilizer and process the displacement signal into surrounding rock displacement data;

[0010] The background management system is electrically connected to the data transmission device. The background management system includes a neural network model, and processes and analyzes the surrounding rock displacement data received from the data transmission device through the neural network model to obtain observation results.

[0011] In some embodiments of the present invention, the signal amplifier includes: a first operational amplifier, wherein its inverting input terminal is electrically connected to a first resistor as a signal input terminal and is electrically connected to its output terminal through a second resistor, and its non-inverting input terminal is grounded; and a second operational amplifier, wherein its non-inverting input terminal is electrically connected to a third resistor as a signal input terminal and is electrically connected to its output terminal through a fourth resistor, and its inverting input terminal is grounded.

[0012] In some embodiments of the present invention, the BPF filter includes: a third operational amplifier and a fourth operational amplifier connected in series;

[0013] The third operational amplifier has a first capacitor connected in series at its non-inverting input terminal and grounded through a sixth capacitor, and its inverting input terminal grounded through a fifth resistor and electrically connected to its output terminal through a seventh resistor.

[0014] The fourth operational amplifier has its non-inverting input terminal electrically connected to the output terminal of the third operational amplifier through a ninth resistor and grounded through a second capacitor; its inverting input terminal grounded through an eighth resistor and electrically connected to its output terminal through a tenth resistor; and its output terminal grounded through an eleventh resistor.

[0015] In some embodiments of the present invention, the signal rectifier includes: a fifth operational amplifier and a sixth operational amplifier;

[0016] The fifth and sixth operational amplifiers have their non-inverting input terminals grounded, and their inverting input terminals are respectively connected to the twelfth and thirteenth resistors. The end where the twelfth and thirteenth resistors are connected is the input terminal of the signal rectifier.

[0017] The inverting input terminal of the fifth operational amplifier is electrically connected to the first terminal of the fourteenth resistor, the second terminal of the fourteenth resistor is electrically connected to the first terminal of the fifteenth resistor, and the second terminal of the fifteenth resistor is electrically connected to the inverting input terminal of the sixth operational amplifier; the inverting input terminal of the sixth operational amplifier is electrically connected to its output terminal through the sixteenth resistor, and is electrically connected to one end of the third capacitor through the seventeenth resistor, serving as the output terminal of the signal rectifier, wherein the other end of the third capacitor is grounded;

[0018] The first diode has its anode electrically connected to the output terminal of the fifth operational amplifier and its cathode electrically connected to the inverting input terminal of the fifth operational amplifier.

[0019] The second diode has its anode connected to the point where the fourteenth and fifteenth resistors are connected, and its cathode connected to the output of the fifth operational amplifier.

[0020] In some embodiments of the present invention, the integrator includes: a seventh operational amplifier, the non-inverting input of which is grounded through a nineteenth resistor, the inverting input of which is connected to an eighteenth resistor as the input of the integrator, and a fourth capacitor electrically connected through a twentieth resistor. The fourth capacitor is electrically connected to the output of the seventh operational amplifier and a first terminal of the twenty-first resistor. The second terminal of the twenty-first resistor is electrically connected to one end of a fifth capacitor as the output of the integrator. The other end of the fifth capacitor is grounded.

[0021] The signal coupler has two input terminals for electrical connection to the output terminal of the integrator; and one output terminal for electrical connection to the input terminal of the signal acquisition unit.

[0022] In some embodiments of the present invention, the signal coupler includes: a set of PNP transistors and a set of NPN transistors; the bases of the PNP transistors and NPN transistors are electrically connected to a 22nd resistor and a 23rd resistor, respectively, serving as two input terminals of the signal coupler, and their bases are also electrically connected to a 24th resistor and a 26th resistor, respectively, for pulling up the input signal; the emitter of the PNP transistor is electrically connected to a 25th resistor and one end of a 6th capacitor for pulling up the output signal, and the other end of the 6th capacitor is electrically connected to the anode of a 3rd diode; the emitter of the NPN transistor is electrically connected to one end of a 29th resistor and one end of a 7th capacitor, and the other end of the 7th capacitor is electrically connected to the anode of a 4th diode; the collector of the PNP transistor is grounded through a 28th resistor, and the other end of the 29th resistor is grounded; the collector of the NPN transistor is connected to a DC power supply terminal through a 27th resistor.

[0023] The cathodes of the third diode and the fourth diode are electrically connected and serve as the output terminal of the signal coupler, which is electrically connected to the input terminal of the signal acquisition device.

[0024] In some embodiments of the present invention, the neural network model includes:

[0025] The preprocessing module is used to preprocess the surrounding rock displacement data, and the preprocessing includes noise reduction and normalization.

[0026] The feature extraction module is used to extract features from the preprocessed surrounding rock displacement data through a neural network to obtain feature data, and to use the feature data as training set data and validation set data for model training.

[0027] The model training module is used to train the neural network model using the training set data, and optimize the weights and model structure to obtain a trained model.

[0028] The model validation module is used to input the validation set data into the trained model, test and validate the model, and adjust and optimize the model to obtain an optimized model.

[0029] The model deployment module is used to deploy optimized models into specific applications to form a backend management system with deep learning capabilities.

[0030] In some embodiments of the present invention, the preprocessing module includes:

[0031] The noise reduction submodule is used to remove noise from the surrounding rock displacement data through an autoencoder network model. Specifically, the surrounding rock displacement data is input into the autoencoder network model, and the autoencoder network model reduces the surrounding rock displacement data to a low-dimensional representation to obtain noise-free surrounding rock displacement data.

[0032] The normalization submodule is used to convert the noise-free surrounding rock displacement data into a standard normal distribution with a mean of 0 and a variance of 1.

[0033] In some embodiments of the present invention, the feature extraction module includes:

[0034] The data extraction submodule is used to divide and label the preprocessed surrounding rock displacement data to obtain training set data and validation set data with feature annotations.

[0035] The model design submodule is used to design a model based on the type of surrounding rock displacement data and determine the neural network model used for model training.

[0036] In some embodiments of the present invention, the model training module includes:

[0037] The training submodule is used to input the prepared training set data into the neural network model, output the result through the loss function, and optimize the training output through the backpropagation algorithm and the loss function to approximate the result data.

[0038] The parameter tuning submodule is used to update and adjust the parameters of each layer of the neural network model using the backpropagation algorithm, the approximate result data, the gradient of the loss function, and the training set data, to obtain a trained model.

[0039] In some embodiments of the present invention, the model verification module includes:

[0040] The testing submodule is used to test and evaluate the neural network model using the training set data, and obtain the evaluation results.

[0041] The optimization submodule is used to optimize and adjust the neural network model based on the evaluation results to obtain an optimized neural network model.

[0042] In some embodiments of the present invention, the loss function includes a squared error loss function and an absolute error loss function for regression tasks, and a binary cross-entropy loss function and a logarithmic loss function for classification tasks.

[0043] In some embodiments of the present invention, the parameter tuning submodule further includes a hidden layer transfer function and a processing function; wherein, the hidden layer transfer function is used to calculate the hidden state vector at the current time from the hidden state vector at the previous time step and the input vector at the current time step, and output it to the processing functions at the next time step and other layers; the processing function is used to convert the hidden state vector into an output vector.

[0044] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0045] A waveform signal is generated by detecting displacement using an optical fiber sensor, wherein the optical fiber is attached to at least a portion of the surrounding rock surface to be observed. A signal stabilizer is electrically connected to the photoelectric conversion unit to process and stabilize the waveform signal. The stabilizer includes two signal processing links, which are combined into a single signal output via a signal coupler. Each signal processing link includes, in sequence, a signal amplifier, a BPF filter, a signal rectifier, and an integrator. The signal amplifiers in the two signal processing links are out of phase, and the output of one integrator is electrically connected to one input of the signal coupler, while the output of the other integrator is electrically connected to the other input of the signal coupler. A signal acquisition unit is electrically connected to the output of the signal coupler and a data transmission device to receive the displacement signal output by the signal stabilizer and process it into surrounding rock displacement data. A background management system is electrically connected to the data transmission device. The background management system includes a neural network model, which processes and analyzes the surrounding rock displacement data received from the data transmission device to obtain observation results. By combining fiber optic sensors, signal stabilizers, and data acquisition devices with a back-end management system that has deep learning capabilities, the problem of existing sensors requiring drilling to observe the surrounding rock and being susceptible to signal interference has been solved. This new technology eliminates the need for drilling, has strong anti-interference capabilities, and can monitor continuously for 24 hours. Furthermore, as the monitoring time increases and the amount of data grows, the accuracy of data identification can be further improved. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the application structure of a real-time monitoring device for surrounding rock displacement in urban underground space engineering, provided by an embodiment of the present invention.

[0048] Figure 2 This is a schematic diagram of the circuit structure of a signal coupler provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of a portion of the circuit structure in a signal stabilizer provided in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram comparing the normal waveform and the waveform when displacement occurs output by the fiber optic sensor according to an embodiment of the present invention.

[0051] Figure 5 This is a schematic diagram of the waveform output by the integrating circuit when the normal waveform is not disturbed, according to an embodiment of the present invention.

[0052] Figure 6 This is a schematic diagram of the waveforms before and after signal merging provided in an embodiment of the present invention;

[0053] Figure 7 This is a schematic diagram of the module structure of a neural network model provided in an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram of a hybrid convolutional recurrent neural network structure provided in an embodiment of the present invention;

[0055] Figure 9 This is a schematic diagram of the input structure of a combined convolutional neural network and a recurrent neural network provided in an embodiment of the present invention.

[0056] Icons: 1-Fiber optic sensor, 2-Signal stabilizer, 21-Signal amplifier, 22-BPF filter, 23-Signal rectifier, 24-Integrator, 25-Signal coupler, 3-Signal acquisition unit, 4-Data transmission device, 5-Back-end management system, 51-Preprocessing module, 52-Feature extraction module, 53-Model training module, 54-Model validation module, 55-Model deployment module. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0058] It should be noted that, in any embodiment of this application, the neural network is a computational model that simulates the human nervous system. It consists of a large number of interconnected processing units (or "neurons") and can learn to recognize data patterns, classify, and predict. It includes, but is not limited to, computational models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Convolutional LSTM Networks (ConvLSTMs), and can also be novel computational models formed by combining existing models. LSTM (Long Short-Term Memory) is a type of recurrent neural network, an improved version of RNN.

[0059] Example 1

[0060] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the various embodiments and features described below can be combined with each other.

[0061] Please refer to Figure 1 This invention provides a real-time monitoring device for surrounding rock displacement in urban underground space engineering, comprising: a fiber optic sensor 1, including a laser source 11 for providing single-wavelength laser light, an optical fiber 12 for transmitting light, and a photoelectric conversion unit 13 for receiving single-wavelength laser light and converting it into a waveform signal, wherein the optical fiber is attached to at least a portion of the surface of the surrounding rock to be observed; and a signal stabilizer 2, electrically connected to the photoelectric conversion unit 13, for processing and stabilizing the waveform signal, wherein it has two signal processing links, each of which includes, in sequence, a signal amplifier 21, a BPF filter 22, a signal rectifier 23, and an integrator 24. 4; In this system, the signal amplifiers 21 in the two signal processing links are out of phase, and the output of one integrator 24 is electrically connected to one input of the signal coupler 25, while the output of the other integrator 24 is electrically connected to the other input of the signal coupler 25; The signal acquisition unit 3 is electrically connected to the output of the signal coupler 25 and the data transmission device 4, and is used to receive the displacement signal output by the signal stabilizer 2 and process the displacement signal into surrounding rock displacement data; The background management system 5 is connected to the data transmission device 4. The background management system 5 includes a neural network model, and processes and analyzes the surrounding rock displacement data received from the data transmission device 4 through the neural network model to obtain the observation results.

[0062] In the above embodiment, the fiber optic sensor 1 solves the problem that other sensors used to measure surrounding rock displacement, such as inductive frequency-modulated displacement sensors or displacement transmission rods, require drilling holes in the surrounding rock, which damages the surface structure of the surrounding rock. The fiber optic sensor 11 attaches its optical fiber 12 to the surface of the surrounding rock to be observed. A single-wavelength laser is emitted from one end of the fiber via a laser source 11, and the optical signal is received and converted into a waveform signal by a photoelectric conversion unit 13 at the other end. The displacement of the surrounding rock is observed by observing the changes in the waveform, avoiding damage to the surrounding rock caused by sensor installation. The signal stabilizer 2 splits the signal output from the photoelectric conversion unit 13 into two paths. One path and the other path are amplified in opposite directions after entering an amplifier, ensuring that the two signal waveforms are completely identical. In the opposite state, when the light is interfered with, the interference on both signals is the same. After a series of processing steps, the signals reach the signal coupler 25 for coupling, and one signal is inverted and then combined into another signal. Since the inverted signal is the same as the other signal, while the phase of the interference signal is opposite, the interference signal can be directly eliminated after merging. At the same time, the signal level can be amplified to about twice that before coupling. The signal is then collected and converted into a digital signal by the acquisition device and sent to the background management system through the data transmission device 4. The signal acquisition device 3, combined with the background management system 5 with deep learning function, can not only monitor continuously for 24 hours, but also further improve the accuracy of data identification as the monitoring time increases and the data volume increases.

[0063] It should be noted that the principle of fiber optic sensor 1 is to analyze the optical characteristics of the light signal in the optical fiber 12 during transmission within the optical cable, thereby achieving the measurement of physical quantities such as displacement, strain, and temperature. Specifically, fiber optic sensor 1 involves winding and fixing a section of optical fiber 12 to the surface of the structure being measured, such as the surface of the surrounding rock. When the surrounding rock is subjected to force and undergoes displacement, its surface also deforms. Since the attached optical fiber 12 has the same shape as the surrounding rock surface, the light signal in the optical fiber 12 will undergo a slow phase shift, causing a change in the intensity or frequency of the light signal. This change can be converted into an electrical signal by a photoelectric conversion device (photoelectric conversion unit 13), and then processed by a signal acquisition and processing system to achieve the measurement of physical quantities such as displacement, strain, and temperature. (See reference...) Figure 4 As shown, when no displacement occurs, such as Figure 4 The waveform above, when displacement occurs, such as... Figure 4 The waveform below.

[0064] Example 2

[0065] Reference Figures 1 to 3As shown, a real-time rock displacement monitoring device for urban underground space engineering includes an optical fiber sensor 1, a laser source 11 for providing single-wavelength laser light, an optical fiber 12 for transmitting the light, and a photoelectric conversion unit 13 for receiving the single-wavelength laser light and converting it into a waveform signal. The optical fiber is attached to at least a portion of the surface of the rock to be observed. A signal stabilizer 2, electrically connected to the photoelectric conversion unit 13, processes and stabilizes the waveform signal. It contains two signal processing links, each including a signal amplifier 21, a BPF filter 22, a signal rectifier 23, and an integrator 24, which are sequentially electrically connected. The signal amplifiers 21 in the two signal processing links are out of phase, and the output of one integrator 24 is electrically connected to one input of the signal coupler 25, while the output of the other integrator 24 is electrically connected to the other input of the signal coupler 25. The signal acquisition unit 3 is electrically connected to the output of the signal coupler 25 and the data transmission device 4, and is used to receive the displacement signal output by the signal stabilizer 2 and process the displacement signal into surrounding rock displacement data. The background management system 5 is connected to the data transmission device 4. The background management system 5 includes a neural network model, and processes and analyzes the surrounding rock displacement data received from the data transmission device 4 through the neural network model to obtain the observation results.

[0066] The signal amplifier 21 includes: a first operational amplifier U1a, whose inverting input terminal is electrically connected to a first resistor R1 as one signal input terminal of the signal amplifier 21, and the inverting input terminal is electrically connected to its output terminal through a second resistor R2, and its non-inverting input terminal is grounded to GND; and a second operational amplifier U1b, whose non-inverting input terminal is electrically connected to a third resistor R3 as the other signal input terminal of the signal amplifier 21, and the non-inverting input terminal is electrically connected to its output terminal through a fourth resistor R4, and its inverting input terminal is grounded to GND. By setting two identical operational amplifiers and only swapping one of their input terminals, amplification is achieved while ensuring that the output signals of both are exactly inverted, eliminating the need for an inverter. Operational amplifiers also have the advantages of simple structure, easy implementation, and low cost.

[0067] In a preferred embodiment, the BPF filter 22 includes: a third operational amplifier U2a and a fourth operational amplifier U2b connected in series; the third operational amplifier U2a has a first capacitor C1 connected in series at its non-inverting input terminal and grounded to GND through a sixth capacitor C6, its inverting input terminal grounded to GND through a fifth resistor R5, and its output terminal electrically connected to its output terminal through a seventh resistor R7; the fourth operational amplifier U2b has its non-inverting input terminal electrically connected to the output terminal of the third operational amplifier U2a through a ninth resistor R9 and grounded to GND through a second capacitor C2, its inverting input terminal grounded to GND through an eighth resistor R8, and its output terminal electrically connected to its output terminal through a tenth resistor R10, and its output terminal also grounded to GND through an eleventh resistor R11.

[0068] Reference Figure 3 As shown, the BPF filter 22 includes a high-pass filter composed of the third operational amplifier U2a and corresponding components, and a low-pass filter composed of the fourth operational amplifier U2b and corresponding components. The signal is coupled to the non-inverting input of the third operational amplifier U2a by the first capacitor C1, and its output is connected to its inverting input via the seventh resistor R7 as a negative feedback signal. The output of the third operational amplifier U2a is connected to the non-inverting input of the fourth operational amplifier U2b via the ninth resistor R9. A second capacitor C2 is also electrically connected to the non-inverting input of the fourth operational amplifier U2b, and its output is electrically connected to its inverting input via the tenth resistor R10 as negative feedback. The third operational amplifier U2a and the fourth operational amplifier U2b can preferably be IC741.

[0069] Among them, the BPF filter 22 mentioned above is a band-pass filter, which is a device that allows waves of a specific frequency band to pass through while blocking other frequency bands. It forms a second-order filter by passing through a high-pass filter and a low-pass filter. Second-order filters are superior to first-order filters in terms of filtering effect, filtering slope, phase shift characteristics, and stability.

[0070] In a preferred embodiment, the signal rectifier 23 includes: a fifth operational amplifier U3a and a sixth operational amplifier U3b; the non-inverting input terminals of the fifth operational amplifier U3a and the sixth operational amplifier U3b are grounded to GND, and their inverting input terminals are respectively connected to a twelfth resistor R12 and a thirteenth resistor R13, wherein the end where the twelfth resistor R12 and the thirteenth resistor R13 are connected is the input terminal of the signal rectifier 23; the inverting input terminal of the fifth operational amplifier U3a is electrically connected to the first terminal of the fourteenth resistor R14, the second terminal of the fourteenth resistor R14 is electrically connected to the first terminal of the fifteenth resistor R15, and the second terminal of the fifteenth resistor R15 is electrically connected to... The inverting input terminal of the sixth operational amplifier U3b; the inverting input terminal of the sixth operational amplifier U3b is electrically connected to its output terminal through the sixteenth resistor R15, and is electrically connected to one end of the third capacitor C3 through the seventeenth resistor R17, serving as the output terminal of the signal rectifier 23, wherein the other end of the third capacitor C3 is grounded to GND; the anode of the first diode D1 is electrically connected to the output terminal of the fifth operational amplifier U3a, and its cathode is electrically connected to the inverting input terminal of the fifth operational amplifier U3a; the anode of the second diode D2 is electrically connected to the position where the fourteenth resistor R14 and the fifteenth resistor R15 are connected, and its cathode is electrically connected to the output terminal of the fifth operational amplifier U3a.

[0071] Reference Figure 3As shown, the input signal to the signal rectifier 23 is first rectified by a half-wave rectifier to generate a half-wave signal. This signal is then fed into a subsequent stage to be superimposed and inverted with the input signal, resulting in a full-wave rectified signal as the output. This signal is then filtered by a first-order filter circuit to obtain a relatively stable signal. In the circuit diagram, U3a, D1, D2, R14, and R22 constitute the half-wave rectifier section, U3b, R15, R16, and R13 constitute the superposition and inversion section, and R17 and C3 constitute the first-order filter section. The first diode D1 and the second diode D2 have a forward voltage of approximately 0.6V in the circuit, while the open-circuit gain of the integrated operational amplifier is typically in the tens of thousands. A small change in the input voltage of the operational amplifier can cause the output to follow suit, thus achieving signal rectification. The model of the integrated operational amplifier U3a in the circuit is mainly selected based on the voltage amplitude and frequency of the input signal. Here, the OPA6951D is preferred, as it supports a maximum bandwidth of 500MHz.

[0072] As a preferred embodiment, refer to Figure 2 and Figure 3 As shown, the integrator 24 includes: a seventh operational amplifier U4, whose non-inverting input is grounded to GND through a nineteenth resistor R19; its inverting input is connected to an eighteenth resistor R18 as the input of the integrator 24; and is electrically connected to a fourth capacitor C4 through a twentieth resistor R20. The fourth capacitor C4 is electrically connected to the output of the seventh operational amplifier U4 and the first end of a twenty-first resistor R21; the second end of the twenty-first resistor R21 is electrically connected to one end of a fifth capacitor C5 as the output of the integrator 24; the other end of the fifth capacitor C5 is grounded to GND; a signal coupler 25 has two inputs for electrical connection to the output of the integrator 24; and one output for electrical connection to the input of the signal acquisition unit 3. Through the integrating circuit in the integrator 24, it has advantages such as filtering and anti-interference, and can process the signal to make the waveform of the input signal smoother. Therefore, it can not only achieve low-pass filtering of the signal, but also make the waveform of the signal smoother, thereby improving the signal quality and stability.

[0073] As a preferred embodiment, refer to Figure 2As shown, the signal coupler 25 includes: a set of PNP transistors Q1 and a set of NPN transistors Q2; the bases of the PNP transistors Q1 and Q2 are electrically connected to the twenty-second resistor R22 and the twenty-third resistor R23, respectively, serving as the two input terminals of the signal coupler 25, and their bases are also electrically connected to the twenty-fourth resistor R24 ​​and the twenty-sixth resistor R26, respectively, for pulling up the input signals; the emitter of the PNP transistor Q1 is electrically connected to the twenty-fifth resistor R25 for pulling up the output signal and one end of the sixth capacitor C6, and the other end of the sixth capacitor C6 is electrically connected to the third resistor R24. The anode of diode D3; the emitter of NPN transistor Q2 is electrically connected to one end of resistor R29 (29th) and one end of capacitor C7 (7th), and the other end of capacitor C7 is electrically connected to the anode of diode D4 (4th); the collector of PNP transistor Q1 is grounded to GND through resistor R28 (28th), and the other end of resistor R29 is grounded to GND; the collector of NPN transistor Q2 is connected to the DC power supply terminal VCC through resistor R27 (27th); the cathodes of diode D3 (3rd) and diode D4 (4th) are electrically connected and serve as the output terminal of signal coupler 25, which is electrically connected to the input terminal of signal acquisition unit 3.

[0074] Combination Figures 4 to 6 As shown, interference may occur after the signal is input, or there may be no interference during the period. When there is no interference, the signal becomes a normal DC signal after integration, such as... Figure 5 In this method, the two signals are in opposite phase. When interference occurs, since the two signals originate from the same source and have the same period, only their phases are opposite, the location, direction, and amplitude of the interference are all the same. Figure 6 The top two signals, when the bottom signal is inverted, form the same state as the first signal. At this time, the phases of the interference signals in the two signals are exactly opposite. After being coupled and output by the above coupling circuit, the interference signals will cancel each other out. (Refer to...) Figure 6 The last signal waveform in the middle, and the normal signal level is superimposed, so that the signal strength is about twice that of the uncombined signal.

[0075] Example 3

[0076] Reference Figure 1 The data transmission device 4 includes a wireless transmission unit and / or a wired transmission unit, used to transmit displacement data from the acquisition device to the back-end management system 5. Wired or wireless transmission modes are typically used, and specific devices may include transmission lines, multiplexers, wireless modems, fiber optic modems, etc.

[0077] In a preferred embodiment, the fiber optic sensor 1 measures the displacement of the surrounding rock and transmits the data to the signal acquisition unit 3. The signal acquisition unit 3 collects and stores the data from the displacement sensor and performs preliminary data processing and analysis. The data transmission device 4 is mainly a communication device used to transmit the collected data to the central monitoring station or other relevant personnel for real-time monitoring and early warning. The power supply provides a stable power source for the displacement sensor, signal acquisition unit, and communication device. The computer equipment at the central monitoring station is equipped with a software system, namely the background management system 5, which controls the operation of the fiber optic sensor 1 and the signal acquisition unit 3 and performs data processing and analysis through the neural network model. This system processes and analyzes the data through its internal neural network model and finally returns the results to the central monitoring station or other relevant personnel, enabling continuous operation. The device may also include safety protection equipment to protect the displacement sensor and signal acquisition unit from external interference and damage.

[0078] Because fiber optic sensors can employ distributed sensing technology, they can monitor continuous physical quantities in real time, forming displacement information and other parameters distributed across a continuous domain. In contrast, traditional inductively modulated (IMM) sensors require a large number of single-point sensors for displacement measurement; if the sensor density is insufficient or improperly arranged, it cannot form a complete distribution of physical variables. Fiber optic sensors do not require physical contact with the surrounding rock surface; the optical fiber is simply attached to the surface, preventing damage to the rock or affecting the monitoring results during measurement. In contrast, existing IIM sensors require physical bonding to the surface of the monitored structure. Furthermore, because fiber optic sensors use non-contact sensing technology, they exhibit better repeatability over long-term use and do not damage the surface material of the measured structure. Additionally, fiber optic sensors typically only require attaching, wrapping, or coiling a single fiber optic cable along the surface of the rock mass, making installation very simple and ensuring good connectivity with the rock surface, thus facilitating maintenance and replacement. In contrast, the installation of inductive frequency modulation sensors in existing technologies requires drilling and fixing of monitoring points, which is more difficult and costly to maintain.

[0079] As a preferred embodiment, the signal acquisition unit 3 typically employs a professional displacement data acquisition unit or demodulator. It may also include supporting facilities such as a clean power supply and a mounting bracket to provide power to the monitoring equipment and secure it; for example, a solar panel can be used for power supply, or a steel bracket can be selected for mounting.

[0080] In a preferred embodiment, after the fiber optic sensor 1 acquires a signal, the sensor and data acquisition device collect and preprocess the signal transmitted from the fiber optic cable. The sensor and data acquisition device collect sensor signal data from the object or system, and then preprocess the data, including noise reduction, filtering, and normalization. Feature extraction is then performed using a neural network model; for example, convolutional neural networks (CNNs) or recurrent neural networks (RNNs) can be used to extract features from the processed signal data. CNNs are primarily used for processing time-domain data, while RNNs are suitable for processing sequence data. Additionally, hybrid convolutional recurrent neural networks (ConvLSTM) can also be used. The feature data neural network model is then trained to optimize weights and model structure, enabling more accurate classification, prediction, or analysis of target signals. The model's performance is verified by inputting validation set data into the trained model, and the model is adjusted and optimized. Finally, the trained and optimized model is integrated into a specific application. Once integrated, when the system receives the acquired signal, it can automatically identify and judge the motion signal using the aforementioned neural network model for real-time monitoring.

[0081] It should be noted that deep learning network models, due to their autonomous learning capabilities, also undergo autonomous learning during signal processing. That is, the longer they operate, the more data they can monitor, which is equivalent to having more data for learning and training, thereby further improving accuracy.

[0082] Example 4

[0083] A real-time rock displacement monitoring device for urban underground space engineering includes: a neural network model, which can be a convolutional neural network (CNN) or a recurrent neural network (RNN), or a combination of CNN and RNN to form a new hybrid convolutional recurrent neural network (ConvLSTM), as shown in the reference. Figure 7As shown, the system includes: a preprocessing module 51 for preprocessing surrounding rock displacement data, including noise reduction and normalization; a feature extraction module 52 for extracting features from the preprocessed surrounding rock displacement data using a neural network, obtaining feature data, and using this feature data as training and validation set data for model training; a model training module 53 for training the neural network model using the training set data, optimizing the weights and model structure, and obtaining a trained model; a model validation module 54 for inputting validation set data into the trained model, testing and validating the model, and adjusting and optimizing the model to obtain an optimized model; and a model deployment module 55 for deploying the optimized model to specific applications to form a backend management system with deep learning capabilities. In this application, the fiber optic sensor 1 and signal acquisition device 3 used for monitoring surrounding rock displacement can be processed using neural networks such as convolutional neural networks (CNN) and recurrent neural networks (RNN), and improved recurrent neural networks can also be used; for example... Figure 8 The diagram shows a Long Short-Term Memory (LSTM) network; alternatively, gated recurrent unit (GRU) networks can be used, which effectively addresses the problem in this application where the data from the fiber optic sensor 1 is typically high-dimensional and dense, making it difficult for existing recurrent neural networks (RNNs) to process. To further improve the processing power of the neural network model, this application combines convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to form a network such as... Figure 9 The hybrid convolutional neural network (ConvLSTM) shown is also an effective method for processing sensor signal data. Its advantage lies in its ability to simultaneously process the spatiotemporal features of sensor signals, improving the model's prediction accuracy and reliability. For example, input data (monitoring data from fiber optic sensor 1) is used as input and processed through multiple convolutional and pooling layers to extract spatial features. Because CNNs have good spatial feature extraction capabilities, they can be used to extract features from various objects, such as images, sounds, and text. The output of the CNN is then used as input to a recurrent neural network (RNN), which is processed by recurrent layers to extract time-series features. Since RNNs have good time-series feature extraction capabilities, they can be used to process a wide range of time-series data, such as speech, text, and various sensor data. Typically, the convolutional part takes the data input from fiber optic sensor 1, extracts the spatial features of the fiber optic sensor 1 signal data, and then sends the output to the recurrent part as its input, responsible for extracting time-series features. This combines spatial and time-series features; therefore, the device of this application performs excellently in the prediction and analysis of surrounding rock displacement, improving the accuracy of prediction and analysis.

[0084] As a preferred embodiment, the preprocessing module 51 includes: a noise reduction submodule, used to remove noise from the surrounding rock displacement data through an autoencoder network model; specifically, the surrounding rock displacement data is input into the autoencoder network model to reduce the surrounding rock displacement data to a low-dimensional representation, thereby obtaining noise-free surrounding rock displacement data; and a normalization submodule, used to convert the noise-free surrounding rock displacement data into a standard normal distribution with a mean of 0 and a variance of 1; i.e., x' = (xμ) / σ, where x is a sample in the original surrounding rock displacement data; x' is the sample value obtained after normalization processing; μ is the mean of the original surrounding rock displacement data, used to represent the central tendency of the original surrounding rock displacement data; and σ is the standard deviation of the original surrounding rock displacement data, used to represent the dispersion of the original surrounding rock displacement data. μ and σ are calculated from the entire original surrounding rock displacement dataset, and the value range of x' is usually [0,1] or [1,1]. The specific value range depends on the specific normalization method, which helps to improve the efficiency and accuracy of the algorithm and prevent learning bias caused by differences in feature scales. Normalization can make the feature scales of data similar, thus making it easier to discover patterns and regularities.

[0085] It's important to note that denoising refers to removing or reducing noise in data. Noise can be random or irregular signals caused by limitations of the sensor or equipment itself, the influence of the data acquisition environment, etc. The purpose of denoising is to improve data quality, making subsequent analysis and modeling more accurate and meaningful. Denoising methods include average filtering, median filtering, and wavelet transform. Filtering, on the other hand, refers to screening and processing data according to specific requirements. Unlike denoising, filtering purposefully preserves or emphasizes certain features or information in the data. For example, images of surrounding rocks in urban underground spaces acquired through image fiber optic sensors can be blurred using Gaussian filtering to remove noise and details, or edge detection algorithms can be used to emphasize features such as contours and textures in the image. Common filtering methods include linear filtering, nonlinear filtering, and convolutional neural network filtering. An autoencoder is a neural network model that can be used to remove noise from noisy data. It can compress input data into a low-dimensional representation and regenerate noise-free output data.

[0086] In a preferred embodiment, the feature extraction module 52 is used to extract features from the preprocessed surrounding rock displacement data using a neural network, obtain feature data, and use this feature data as training and validation set data for model training. The feature extraction module 52 includes: a data extraction submodule, used to divide and label the preprocessed surrounding rock displacement data to obtain training and validation set data with feature labels, making it suitable for training and testing using a hybrid convolutional recurrent neural network; and a model design submodule, used to design a model based on the type of surrounding rock displacement data, determining the neural network model used for model training. For example, in this application, a hybrid convolutional recurrent neural network model can be selected, including a convolutional module, a recurrent module, and a classification (or regression) output part, determining the number of layers, size, and hyperparameters of each part; data division involves dividing the dataset into training and validation set data, etc., for monitoring and evaluation during deep learning model training. Labeling involves tagging the data, adding labels or annotations, so that the deep learning model can learn and train based on the data labels. For example, data annotation includes classification annotation, object detection annotation, semantic segmentation annotation, and named entity annotation in natural language processing.

[0087] As a preferred embodiment, refer to Figure 9 The above-mentioned merging of convolutional neural networks and recurrent neural networks results in a new input component, forming a CNNLSTM. Its front end can use the input component of a CNN to process the training data, which is then subjected to convolution operations and further processed by the LSTM to obtain the posterior probability matrix. This matrix can be used as... Figure 9 The input portion of the hybrid convolutional recurrent neural network (ConvLSTM) shown.

[0088] Example 5

[0089] The aforementioned model training module 53 is used to train the neural network model using training set data, and optimize the weights and model structure to obtain a trained model. The model training module 53 includes a training submodule, which is used to input the prepared training set data into the neural network model, output through the loss function, and optimize the training output approximate result data through the backpropagation algorithm and the loss function. The parameter tuning submodule is used to update and adjust the parameters of each layer of the neural network model using the backpropagation algorithm, through the approximate result data, the gradient of the loss function, and the training set data, to obtain a trained model.

[0090] In a preferred embodiment, the aforementioned training submodule is used to input the prepared training set data into the neural network model, output the results through a loss function, and optimize the training output using a backpropagation algorithm and the loss function to approximate the training results. The loss function includes Mean Squared Error Loss and Mean Absolute Error Loss for regression tasks, and Binary Crossentropy Loss and Log Loss for classification tasks. After processing with the loss function, further optimization can be performed using an optimization function. For training hybrid convolutional recurrent neural network models, optimization algorithms such as gradient descent, Adam, and RMSprop can be used. Optimizing the weights refers to updating the weights of each neuron during neural network training, thereby making the network's prediction results more accurate. For hybrid convolutional recurrent neural networks, loss function optimization is an essential step, helping to more accurately predict and analyze surrounding rock displacement, providing strong support for fields such as geological engineering and earthquake prediction.

[0091] In a preferred embodiment, the aforementioned parameter tuning submodule further includes a hidden layer transfer function and a processing function. The hidden layer transfer function calculates the current hidden state vector from the hidden state vector of the previous time step and the input vector of the current time step, and outputs it to the processing functions of the next time step and other layers. The main function of the processing function is to convert the hidden state vector into an output vector. Optimizing the hidden layer transfer and processing functions in the hybrid convolutional recurrent neural network model, such as convolutional layers, pooling layers, recurrent layers, fully connected layers, and activation functions, is mainly used for feature extraction and feature combination to generate the final output.

[0092] As a preferred embodiment, some examples of loss functions that can be used in the above-described hybrid convolutional recurrent neural network are listed below. For example, the binary crossentropy loss function is commonly used for binary classification problems, such as determining whether the surrounding rock has shifted. Its formula is:

[0093]

[0094] In the above formula, y i The actual value (0 or 1). The value is an estimate (between 0 and 1), where N is the total sample size. It concerns the true value y and the estimated value. The loss function value; this binary cross-entropy loss function can effectively optimize binary classification problems.

[0095] For example, the Mean Squared Error Loss function can be used for regression problems, such as predicting the displacement of surrounding rock. Its formula is:

[0096]

[0097] In the above formula, yi represents the true value. Let N be the predicted value and N be the total number of samples. The squared error loss function can optimize the prediction of continuous numerical variables. Applying this loss function to hybrid convolutional recurrent neural networks can further optimize the model and improve performance.

[0098] In this embodiment, the loss functions described above can be used in combination. For example, for a binary classification problem, both the binary cross-entropy loss function and the logarithmic loss function can be used simultaneously. The combination method involves a weighted sum of these two loss functions, i.e.:

[0099]

[0100] Here, α is the weighting coefficient, used to balance the effects of the two loss functions.

[0101] Alternatively, for regression problems, both the squared error loss function and the absolute error loss function can be used simultaneously, in the following way:

[0102]

[0103] Among them, L mae β is the absolute error loss function, and β is the weighting coefficient used to balance the effects of the two losses.

[0104] Multiple loss functions can be used in combination, and weighted balancing can be applied between different loss functions to strengthen or weaken their effects.

[0105] As a preferred embodiment, the model validation module 54 described above is used to input validation set data into the trained model, test and validate the model, and adjust and optimize the model to obtain an optimized model. It includes: a testing submodule, used to test and evaluate the neural network model using training set data to obtain evaluation results; and model testing and optimization: during the model testing and optimization phase, the hybrid convolutional recurrent neural network model needs to be optimized and adjusted based on the test results and evaluation metrics to improve the model's performance.

[0106] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A real-time monitoring device for surrounding rock displacement in urban underground space engineering, characterized in that, include: The fiber optic sensor includes a laser source for providing a single-wavelength laser, an optical fiber for transmitting the light, and a photoelectric conversion unit for receiving the single-wavelength laser and converting it into a waveform signal, wherein the optical fiber is attached to at least a portion of the surface of the surrounding rock to be observed. A signal stabilizer, electrically connected to the photoelectric conversion unit, is used to process and stabilize the waveform signal. It has two signal processing links, which are combined into a single signal output via a signal coupler. Each signal processing link includes, in sequence, a signal amplifier, a BPF filter, a signal rectifier, and an integrator. The signal amplifiers in the two signal processing links are out of phase. The output of one integrator is electrically connected to one input of the signal coupler, and the output of the other integrator is electrically connected to the other input of the signal coupler. The signal acquisition unit is electrically connected to the output terminal of the signal coupler and the data transmission device, respectively, and is used to receive the displacement signal output by the signal stabilizer and process the displacement signal into surrounding rock displacement data; The background management system is electrically connected to the data transmission device. The background management system includes a neural network model, and processes and analyzes the surrounding rock displacement data received from the data transmission device through the neural network model to obtain observation results.

2. The real-time surrounding rock displacement monitoring device according to claim 1, characterized in that, The signal amplifier includes: The first operational amplifier has its inverting input terminal electrically connected to a first resistor as a signal input terminal, and electrically connected to its output terminal through a second resistor, while its non-inverting input terminal is grounded. The second operational amplifier has its non-inverting input terminal electrically connected to a third resistor as a signal input terminal, and electrically connected to its output terminal through a fourth resistor, while its inverting input terminal is grounded.

3. The real-time surrounding rock displacement monitoring device according to claim 1, characterized in that, The BPF filter includes a third operational amplifier and a fourth operational amplifier connected in series; The third operational amplifier has a first capacitor connected in series at its non-inverting input terminal and grounded through a sixth capacitor, and its inverting input terminal grounded through a fifth resistor and electrically connected to its output terminal through a seventh resistor. The fourth operational amplifier has its non-inverting input terminal electrically connected to the output terminal of the third operational amplifier through a ninth resistor and grounded through a second capacitor; its inverting input terminal grounded through an eighth resistor and electrically connected to its output terminal through a tenth resistor; and its output terminal grounded through an eleventh resistor.

4. The real-time surrounding rock displacement monitoring device according to claim 1, characterized in that, The signal rectifier includes: a fifth operational amplifier and a sixth operational amplifier; The fifth and sixth operational amplifiers have their non-inverting input terminals grounded, and their inverting input terminals are respectively connected to the twelfth and thirteenth resistors. The end where the twelfth and thirteenth resistors are connected is the input terminal of the signal rectifier. The inverting input terminal of the fifth operational amplifier is electrically connected to the first terminal of the fourteenth resistor, the second terminal of the fourteenth resistor is electrically connected to the first terminal of the fifteenth resistor, and the second terminal of the fifteenth resistor is electrically connected to the inverting input terminal of the sixth operational amplifier; the inverting input terminal of the sixth operational amplifier is electrically connected to its output terminal through the sixteenth resistor, and is electrically connected to one end of the third capacitor through the seventeenth resistor, serving as the output terminal of the signal rectifier, wherein the other end of the third capacitor is grounded; The first diode has its anode electrically connected to the output terminal of the fifth operational amplifier and its cathode electrically connected to the inverting input terminal of the fifth operational amplifier. The second diode has its anode connected to the point where the fourteenth and fifteenth resistors are connected, and its cathode connected to the output of the fifth operational amplifier.

5. The real-time surrounding rock displacement monitoring device according to claim 1, characterized in that, The integrator includes: a seventh operational amplifier, whose non-inverting input is grounded through a nineteenth resistor, whose inverting input is connected to an eighteenth resistor as the input of the integrator, and is electrically connected to a fourth capacitor through a twentieth resistor. The fourth capacitor is electrically connected to the output of the seventh operational amplifier and the first end of the twenty-first resistor. The second end of the twenty-first resistor is electrically connected to one end of a fifth capacitor as the output of the integrator. The other end of the fifth capacitor is grounded. The signal coupler has two input terminals for electrical connection to the output terminal of the integrator; and one output terminal for electrical connection to the input terminal of the signal acquisition unit.

6. The real-time surrounding rock displacement monitoring device according to claim 1 or 5, characterized in that, The signal coupler includes: a set of PNP transistors and a set of NPN transistors; the bases of the PNP and NPN transistors are electrically connected to resistors 22 and 23, respectively, serving as the two input terminals of the signal coupler; the bases of the PNP and NPN transistors are also electrically connected to resistors 24 and 26, respectively, for pulling up the input signals; the emitter of the PNP transistor is electrically connected to resistor 25, for pulling up the output signal, and one end of capacitor 6; the other end of capacitor 6 is electrically connected to the anode of diode 3; the emitter of the NPN transistor is electrically connected to resistor 29 and capacitor 7; the other end of capacitor 7 is electrically connected to the anode of diode 4; the collector of the PNP transistor is grounded through resistor 28, and the other end of resistor 29 is grounded; the collector of the NPN transistor is connected to a DC power supply terminal through resistor 27. The cathodes of the third diode and the fourth diode are electrically connected and serve as the output terminal of the signal coupler, which is electrically connected to the input terminal of the signal acquisition device.

7. The real-time surrounding rock displacement monitoring device according to claim 1, characterized in that, The neural network model includes: The preprocessing module is used to preprocess the surrounding rock displacement data, and the preprocessing includes noise reduction and normalization. The feature extraction module is used to extract features from the preprocessed surrounding rock displacement data to obtain feature data, and to use the feature data as training set data and validation set data for model training. The model training module is used to train the neural network model using the training set data, and optimize the weights and model structure to obtain a trained model. The model validation module is used to input the validation set data into the trained model, test and validate the model, and adjust and optimize the model to obtain an optimized model. The model deployment module is used to deploy optimized models into specific applications to form a backend management system with deep learning capabilities.

8. The real-time surrounding rock displacement monitoring device according to claim 7, characterized in that, The preprocessing module includes: The noise reduction submodule is used to remove noise from the surrounding rock displacement data through an autoencoder network model. Specifically, the surrounding rock displacement data is input into the autoencoder network model, and the autoencoder network model reduces the surrounding rock displacement data to a low-dimensional representation to obtain noise-free surrounding rock displacement data. The normalization submodule is used to convert the noise-free surrounding rock displacement data into a standard normal distribution with a mean of 0 and a variance of 1.

9. The real-time surrounding rock displacement monitoring device according to claim 7, characterized in that, The model training module includes: The training submodule is used to input the prepared training set data into the neural network model, output the result through the loss function, and optimize the training output through the backpropagation algorithm and the loss function to approximate the result data. The parameter tuning submodule is used to update and adjust the parameters of each layer of the neural network model using the backpropagation algorithm, the approximate result data, the gradient of the loss function, and the training set data, to obtain a trained model.

10. The real-time surrounding rock displacement monitoring device according to claim 7, characterized in that, The model validation module includes: The testing submodule is used to test and evaluate the neural network model using the training set data, and obtain the evaluation results. The optimization submodule is used to optimize and adjust the neural network model based on the evaluation results to obtain an optimized neural network model.

Images