Mine deformation monitoring and early warning method and system based on microwave sensing

Abstract

The present invention proposes a method and system for mine deformation monitoring and early warning based on microwave sensing. A microwave radar based on microwave sensing technology is used to obtain continuous-wave microwave signals at different parts of the mine. The absolute displacement amounts at different parts of the mine are obtained by inverting the signals using the interference phase modulation information, thereby providing data support for subsequent analysis. After a period of data collection, a dynamic standard mine deformation database is established, and the existing data is used to train the WGAN model to learn the model. Thereafter, the prediction result data of the WGAN model is used. According to the statistical characteristics of the prediction data and the mathematical properties of the normal distribution (σ represents the standard deviation), when it exceeds ±3σ, the monitoring frequency corresponding to the predicted time is increased. When the monitoring result also exceeds ±3σ, early warning is carried out, so that potential risks can be timely managed, ensuring the safe production of the mine and providing a new method for dynamic monitoring and early warning of mine deformation.

Description

Technical Field

[0001] The present invention belongs to the technical field of monitoring and early warning, and in particular relates to a method and system for monitoring and early warning of mine deformation based on microwave perception. Background Art

[0002] 80% of the ores in my country's metallurgical mines come from open-pit mining. With the continuous increase in mining depth, most open-pit mines in my country have entered the deep pit mining mode, resulting in the continuous increase in the height and angle of the pit slope, and the decrease in slope stability. For deep pit open-pit mines, the stability of high and steep slopes is an important factor affecting mine safety production. The occurrence of open-pit slope sliding disasters appears to be sudden, but the process is gradually brewing and developing. There will be a long-term accumulation of slow deformation before the disaster occurs. Therefore, taking effective measures to monitor the stability of high and steep slopes in real time and timely manage potential dangers are important guarantees for mine safety production.

[0003] Under the current situation of frequent sudden geological environmental damage in mining areas, there are few technologies for fast and accurate monitoring of mine deformation over a large area. Traditional mine deformation monitoring mostly uses triangulation, leveling and GPS measurement methods, which have the disadvantages of large workload, high cost, difficulty in preserving measurement points and inability to achieve detailed measurement over a large area. The widely used synthetic aperture radar interferometry (INSAR) technology fails seriously when interpreting surface deformation information in mining areas, has many errors and the accuracy of the results cannot be guaranteed. It is urgent to use new technologies to supplement mine deformation monitoring technology.

[0004] The dynamic response monitoring of structures is mainly reflected in the monitoring of vibration responses under external load excitation. Generally speaking, vibration measurement technologies and methods can be divided into two categories: contact and non-contact. Contact vibration measurement sensors, such as accelerometers, need to be in contact with the object being measured. On the one hand, they will bring additional mass to the structure, and on the other hand, the installation of the sensor is often time-consuming and laborious. Typical non-contact vibration measurement technologies include laser Doppler vibrometers and vision-based vibration measurement technologies. Laser-based vibration measurement technologies and methods are often used for micro-amplitude and high-frequency vibration measurements of small structures or surfaces, and have high requirements for the measurement environment. Vision-based vibration measurement methods can achieve full-field measurement, but video measurement methods often require more complex calibration operations, are greatly affected by light and imaging quality, and are limited by the depth of field of the camera, so there are certain limitations in large-scale vibration measurement.

[0005] Relying solely on monitoring equipment may misjudge some warning situations, but the generative adversarial neural network (GAN) can be trained to better predict mine deformation variables, play a reference role in the monitoring system warning, and achieve true safety and convenience.

[0006] In GAN training, compared with KL divergence and JS divergence, the greatest advantage of the Wasserstein distance is that even if there is no overlap or very little overlap between distributions, it can still calculate the distance between these two distributions, and there is no problem of gradient disappearance, which causes the generative model to stop learning, thus greatly improving the prediction accuracy.

[0007] As a new non-contact vibration and deformation monitoring technology based on microwave sensing, it has the advantages of high measurement accuracy, large measurement range, low power consumption, low cost, and strong environmental adaptability.

[0008] Through the above analysis, the problems and defects existing in the prior art are as follows:

[0009] (1) The existing monitoring methods have a large workload, high cost, or the accuracy cannot be guaranteed;

[0010] (2) The installation of traditional contact sensors is time-consuming and laborious, while non-contact sensors rely on laser or vision-based technologies, which have high environmental requirements, are greatly affected by light or imaging quality, and have great limitations;

[0011] (3) Early warning monitoring can only judge whether to give an early warning through real-time data, and false alarms may occur when the monitoring system has problems.

[0012] The difficulties in solving the above problems and defects are as follows: 1) Propose a monitoring method with low cost, low power consumption, high accuracy, and large measurement range; 2) Propose a non-contact monitoring technology with strong environmental adaptability; 3) Propose a monitoring technology that uses the results obtained through the prediction function as early warning parameters.

[0013] The significance of solving the above problems and defects is as follows: Solving the above problems can provide new monitoring means for mine deformation monitoring, so as to be able to timely control potential dangers and ensure the safe production of mines. Summary of the Invention

[0014] In view of this, the present invention proposes a mine deformation monitoring and early warning method and system based on microwave sensing.

[0015] The technical solution of the present invention is realized as follows:

[0016] On the one hand, the present invention provides a mine deformation monitoring and early warning method based on microwave sensing, including the following steps,

[0017] S1, obtaining a continuous wave microwave signal;

[0018] S2, analyzing and processing the obtained continuous wave microwave signal;

[0019] S3, establishing a dynamic standard mine deformation amount database;

[0020] S4. Issue a warning when the deformation amount exceeds a certain range.

[0021] Based on the above technical solutions, preferably, in step S1, continuous wave microwave signals are obtained by arranging microwave radars in the mine for monitoring.

[0022] In step S2, the microwave signals are inversed based on the interference phase modulation information to obtain the deformation amount of the mine.

[0023] In step S3, a dynamic standard mine deformation amount database is established, and a WGAN model is trained for prediction.

[0024] In step S4, when it exceeds a certain range, increase the monitoring frequency corresponding to the predicted time, and issue a warning when the monitoring result also exceeds this range.

[0025] Based on the above technical solutions, preferably, in step S1, continuous wave microwave signals of different parts of the mine are obtained by using microwave radars.

[0026] In step S2, the microwave signals are inversed by using the interference phase modulation information to obtain the absolute displacement amounts of different parts of the mine.

[0027] In step S3, the obtained absolute displacement amounts are analyzed, a dynamic standard mine deformation table is established, and a WGAN model is trained with this, and the displacement amounts are predicted by using the trained WGAN model.

[0028] In step S4, if the prediction result data of the WGAN model exceeds ±3σ, increase the monitoring frequency corresponding to the predicted time, and issue a warning when the measured result also exceeds ±3σ.

[0029] Based on the above technical solutions, preferably, in step S2, the inversion of the microwave signals by using the interference phase modulation information specifically includes:

[0030] The baseband I / Q signal can be expressed as:

[0031]

[0032]

[0033] In the formula, DC I , DC Q are the DC offset amounts of the I / Q channels respectively, A I and A Q are the signal amplification factors of the I / Q channels respectively, x(t) is the time-domain vibration displacement, λ is the carrier wavelength, θ = 4πR0 / λ + θ0 is the constant displacement, where R0 is the distance between the target and the radar antenna, θ0 is the displacement change generated by the reflection surface, and △φ(t) is the remaining phase noise.

[0034] The change in the vibration displacement of the target at adjacent moments can be calculated as follows:

[0035]

[0036] In the formula, Δφ(t) is the change in the interference phase of the line at time t, and t is the time variable;

[0037] During the i-th frequency sweep period,

[0038]

[0039] In the formula, f0 is the initial carrier frequency, f c is the center carrier frequency, λ0 and λ c are the wavelengths corresponding to the initial and center carrier frequencies respectively, c is the electromagnetic wave propagation speed, and x(iT) are the initial phase of the difference signal and the target vibration displacement during the i-th frequency sweep period respectively.

[0040] Based on the above technical solution, preferably, in step S3, the process of training the WGAN model includes:

[0041] The training objective function is:

[0042]

[0043] Among them, G and D represent the generator and the discriminator respectively; the input of G is the random variable z; the input of D is the mixed data, including the real data labeled as 1 and the generated data labeled as 0; G(z) represents the sample generated by G; when the mixed data is input into D, D makes a binary classification discrimination on the data. If D discriminates that the current data is real data P data , it outputs closer to 1; if it discriminates as generated data P g , it outputs closer to 0; where X represents the real data; represents the min-max game value function.

[0044] First, fix G and train D:

[0045] The best case of D is:

[0046] Subsequently, fix D and train G:

[0047]

[0048] If and only if P data = P g At this time, The generator can generate samples with a distribution different from that of the real data.

[0049] Based on the above technical solutions, preferably, in step S3, the Wasserstein distance calculation formula is as follows:

[0050]

[0051] W(p1,p2) is the Wasserstein distance, ∏(P1,P2) is the set of all possible joint distributions combined by the P1 and P2 distributions. For each possible joint distribution γ, a sample x and y can be sampled from it as (x,y)~γ, and the pair of samples ||x - y|| can be calculated, so as to calculate the expected value of the sample pair distance under the joint distribution γ

[0052] Rewritten according to Lipschitz continuity:

[0053]

[0054] Among them, W(p r ,p θ ) is the equivalent Wasserstein distance obtained according to the Kantorovich-Rubinstein duality. For the continuous function f, L represents the norm, K is the Lipschitz constant of the function, and a restriction is imposed on f, requiring that there exists a constant such that for any two elements x1,x2 in the domain, |f(x1) - f(x2)| ≤ K|x1 - x2|, thereby restricting the maximum local variation amplitude of the continuous function.

[0055] Based on the above technical solutions, preferably, in step S3, finally two loss functions of the WGAN are obtained:

[0056] WGAN generator loss function:

[0057] WGAN discriminator loss function:

[0058] Among them, f w is the discriminator network with the last layer being a non-linear activation layer constructed.

[0059] Based on the above technical solutions, preferably, in step S3, an LSTM is used as the time series generator and a CNN is used as the discriminator in the WGAN model. Among them, the LSTM is as follows: First, determine which information to discard from the node state, which is determined by the "forget gate layer"; Next, determine which new information to store in the cell node state, which is achieved through the "input gate layer" and the tanh layer; Then multiply the old state by f t , and then add Ct ; Finally, determine the output content, set the unit state to tanh and multiply it by the sigmoid gate output.

[0060] Based on the above technical solution, preferably, a long short-term memory network (LSTM) is selected as the generator in the WGAN. The storage unit of the LSTM consists of a forget gate, an input gate, and an output gate. The three thresholds are composed of the sigmoid activation function and pointwise multiplication. The gates of the LSTM network are represented as follows:

[0061] i t = σ(w ri s t-1 + w ri x t + w ci c t-1 + b i )

[0062] f t = σ(w rf s t-1 + w xf x t + w cf c t-1 + b f )

[0063] c t = f t × c t-1 + i t × σ(w rc s t-1 + w xc x t + b c )

[0064] o t = σ(w ro s t-1 + w xo x t + w co c t-1 + b o )

[0065] s t = o t × tanh(c t )

[0066] Where x t represents the input at time t, s t-1 represents the output of the previous moment, c t-1 represents the hidden state of the previous moment. The value of the current time LSTM memory cell is c t , and the output value is s t. Where i t , o t , f t are the values of the input gate, output gate, and forget gate at time t, w is the weight, b is the bias term, and σ is the activation function respectively.

[0067] In a second aspect, the present invention provides a mine deformation monitoring and early warning system based on microwave sensing. Using the method described in the first aspect of the present invention, it includes,

[0068] A signal generation module that generates a modulated wave voltage signal through a signal generator and then controls the signal through a signal conditioning circuit;

[0069] A microwave transmitting and receiving module that transmits and receives continuous wave microwave signals through a microwave radar;

[0070] A data acquisition module that acquires radar baseband signals through a data acquisition card;

[0071] A data processing module that processes the signals to calculate the absolute displacement and trains the WGAN model for data prediction;

[0072] An early warning module that determines whether to give an early warning by comparing with standard mine data.

[0073] The mine deformation monitoring and early warning method and system based on microwave sensing of the present invention have the following beneficial effects compared with the prior art:

[0074] (1) Use a microwave radar based on microwave sensing technology to obtain continuous wave microwave signals from different parts of the mine, and use the interference phase modulation information to invert the signals to obtain the absolute displacement of different parts of the mine, thereby providing data support for subsequent analysis;

[0075] (2) Establish a dynamic standard mine deformation database through data collection for a period of time, and use the existing data to train the Wasserstein GAN (abbreviated as WGAN) deep learning model;

[0076] (3) Subsequently, use the prediction result data of the WGAN model. According to the statistical characteristics of the prediction data and the mathematical properties of the normal distribution (σ represents the standard deviation), when it exceeds ±3σ, increase the monitoring frequency corresponding to the predicted time, and give an early warning when the monitoring result also exceeds ±3σ, so as to be able to timely control potential dangers, ensure the safe production of the mine, and also provide a new method for the dynamic monitoring and early warning of mine deformation Description of the Drawings

[0077] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0078] Figure 1 It is a schematic structural diagram of the mine deformation monitoring and early warning system based on microwave sensing of the present invention;

[0079] In the figure, 1. Signal generation module; 2. Microwave transmitting and receiving module; 3. Data acquisition module; 4. Data processing module; 5. Early warning module;

[0080] Figure 2 It is a flowchart of the mine deformation monitoring and early warning method based on microwave sensing of the present invention;

[0081] Figure 3 It is an LSTM flowchart;

[0082] Figure 4 It is the WGAN training process. Specific embodiments

[0083] The following will combine 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 a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0084] As Figure 2 shown, the mine deformation monitoring and early warning method based on microwave sensing of the present invention includes the following steps S1 to S4.

[0085] S1. Obtain a continuous wave microwave signal. Specifically, a continuous wave microwave signal can be obtained by monitoring a mine with a microwave radar. More specifically, continuous wave microwave signals of different parts of the mine can be obtained by using a microwave radar.

[0086] S2. Analyze and process the obtained continuous wave microwave signal. Specifically, the microwave signal is inverted based on the interference phase modulation information to obtain the mine deformation amount. More specifically, the absolute displacement amounts of different parts of the mine are obtained by inverting the microwave signal using the interference phase modulation information.

[0087] In step S2, the inversion of the microwave signal using the interference phase modulation information specifically includes:

[0088] The baseband I / Q signal can be expressed as:

[0089]

[0090]

[0091] where DC I and DC Q are the DC offsets of the I / Q channels respectively, A I and A Q are the signal amplification factors of the I / Q channels respectively, x(t) is the vibration displacement in the time domain, λ is the carrier wavelength, θ = 4πR0 / λ + θ0 is the constant displacement, where R0 is the distance between the target and the radar antenna, θ0 is the displacement change generated by the reflecting surface, and △φ(t) is the residual phase noise;

[0092] The vibration displacement change of the target at adjacent times can be calculated as:

[0093]

[0094] where is the interference phase change amount at the line time, and t is the time variable.

[0095] Within the i-th sweep period time,

[0096]

[0097] where f0 is the initial carrier frequency, f c is the center carrier frequency, λ0 and λ c are the wavelengths corresponding to the initial and center carrier frequencies respectively, c is the electromagnetic wave propagation speed, and x(iT) are the initial phase of the difference signal and the target vibration displacement within the i-th sweep period time respectively.

[0098] S3. Establish a dynamic standard mine deformation database. Specifically, establish a dynamic standard mine deformation database and train a WGAN model for prediction. More specifically, analyze the obtained absolute displacement amount, establish a dynamic standard mine deformation table, and use this to train the WGAN model, and use the trained WGAN model to predict the displacement amount.

[0099] In step S3, the process of training the WGAN model includes:

[0100] As Figure 4 shown, use G and D to represent the generator and discriminator respectively; the input of G is the random variable z; the input of D is the mixed data, including the real data labeled as 1 and the generated data labeled as 0; G(z) represents the sample generated by G; when the mixed data is input into D, D makes a binary classification discrimination on the data. If D discriminates that the current data is real data Pdata , the output is closer to 1; if it is determined to be the generated data P g , the output is closer to 0; in order to make the data generated by G deceive D as much as possible, it is necessary to train the generative model and the discriminative model against each other to continuously improve the capabilities of both models. Its training objective function is:

[0101]

[0102] Among them, X represents the real data; represents the minimax game value function;

[0103] Train D to maximize the objective function to ensure that the data can be correctly discriminated as real as possible; at the same time, train G to minimize log(1 - D(G(z))).

[0104] Specifically, first fix G and train D:

[0105] Through training, the larger the value of this formula, the better. The best case for D is:

[0106]

[0107] The proof process is as follows:

[0108] Since V is continuous, it can be written in integral form to represent the expectation:

[0109]

[0110] Make some transformations:

[0111]

[0112] Continue the calculation:

[0113] = ∫ x p data (x)log(D(x))dx + ∫ x p z (G -1 (x))log(1 - D(x))(G -1 )'(x)dx

[0114] = ∫ x p data (x)log(D(x))dx + ∫ x p g (x)log(1 - D(x))dx

[0115] = ∫ x p data (x)log(D(x)) + pg \(\int(x)\log(1 - D(x))dx\)

[0116] Then maximize \(V(D, G)\):

[0117]

[0118] Subsequently, fix \(D\) and train \(G\):

[0119]

[0120] If and only if \(P\) data \( = P\) g at this time, the generator can generate samples with a distribution different from the true data.

[0121] For the generative adversarial network model, the best state is when \(p\) data and \(p\) g are close enough, or even \(p\) data \( = p\) g , but in fact, the learning capabilities of the discriminator and the generator are inconsistent. The discriminator is more capable than the generator, which makes it impossible for the generator to obtain useful information from the discriminator's feedback to optimize the quality of the generated samples. It is possible that there is no overlap between \(p\) data and \(p\) g , or the overlap is so small that it can be ignored, and the gradient of the generator disappears, resulting in the loss not decreasing. For example:

[0122] ① \(p\) g (x) = 0 and \(p\) data (x) = 0, and the JS divergence is meaningless

[0123] ②

[0124] ③

[0125] ④ \(p\) g (x) ≠ 0 and \(p\) data (x) ≠ 0. If the overlapping part can be ignored, it is equivalent to the first case, and the Js divergence is meaningless. In fact, the probability that the overlapping part between \(p\) data and \(p\) g can be ignored is almost 1. Because the generator usually outputs a high-dimensional (such as 28×28) sample from a low-dimensional (such as 100-dimensional) through a neural network. When the parameters in the generative model are fixed, the generated distribution \(p\) g Although it seems to be a sample in a high-dimensional space, in fact, the changes it can produce have been limited by the input dimension. At the same time, due to the mapping transformation between layers of the neural network, there is a situation of data dimensionality reduction, resulting in \(p\) gThe substantial dimension can only be lower than the input dimension. However, the true distribution p data is a high-dimensional distribution. Therefore, p g almost impossible to overlap with p data If there is an overlap, the dimension of this part must be lower than that of p g , so the overlapping part can be ignored.

[0126] From the above four cases, regardless of whether p data and p g are far apart or close, as long as there is no overlapping part or the overlapping part is negligible between the two distributions, the loss of the generator is a constant -log(2). Then there will be a vanishing gradient problem, and the generator cannot update the model parameters.

[0127] Therefore, the distance metric is crucial in the generative model. The Wasserstein distance calculation method can provide meaningful gradients. The Wasserstein distance calculation formula is:

[0128]

[0129] W(p1,p2) is the Wasserstein distance, ∏(P1,P2) is the set of all possible joint distributions of the combined P1 and P2 distributions. For each possible joint distribution γ, a sample x and y can be sampled from it (x,y)~γ, and the distance between this pair of samples ||x - y|| can be calculated. Thus, the expected value of the sample pair distance under the joint distribution γ can be calculated.

[0130] According to the Lipschitz continuity, it is rewritten as:

[0131]

[0132] Among them, W(p r ,p θ ) is the equivalent Wasserstein distance obtained according to the Kantorovich-Rubinstein duality. For the continuous function f, L represents the norm, and K is the Lipschitz constant of the function. A restriction is imposed on f, requiring that there exists a constant such that for any two elements x1,x2 in the domain, |f(x1)-f(x2)|≤K|x1 - x2|, thereby restricting the maximum local variation amplitude of the continuous function. Finally, the two loss functions of WGAN are obtained:

[0133] WGAN generator loss function:

[0134] WGAN discriminator loss function:

[0135] Among them, f w is a discriminator network with the last layer constructed as a non-linear activation layer.

[0136] A WGAN model that uses an LSTM as the time series generator and a CNN as the discriminator. Among them, as Figure 3 , the LSTM is as follows: First, determine which information to discard from the node state, which is decided by the "forget gate layer"; Next, determine which new information to store in the cell node state, which is achieved through the "input gate layer" and the tanh layer; Then multiply the old state by f t , then add C t , and finally decide the output content, place the cell state through tanh and multiply it by the sigmoid gate output.

[0137] Based on the above technical solutions, preferably, in the WGAN, a long short-term memory network (LSTM) is selected as the generator. The storage unit of the LSTM consists of a forget gate, an input gate, and an output gate. The three thresholds are composed of the sigmoid activation function and pointwise multiplication. The gates of the LSTM network are represented as follows:

[0138] i t = σ(w ri s t-1 + w ri x t + w ci c t-1 + b i )

[0139] f t = σ(w rf s t-1 + w xf x t + w cf c t-1 + b f )

[0140] c t = f t × c t-1 + i t × σ(w rc s t-1 + w xc x t + b c )

[0141] o t = σ(w ro s t-1 + w xo x t + w coc t-1 +b o )

[0142] s t = o t × tanh(c t )

[0143] where x t represents the input at time t, s t-1 represents the output of the previous time step, c t-1 represents the hidden state of the previous time step, and the value of the current time LSTM memory cell is c t , and the output value is s t . Where i t , o t , f t are the values of the input gate, output gate, and forget gate at time t respectively, w is the weight, b is the bias term, and σ is the activation function.

[0144] S4. Issue a warning when the deformation amount exceeds a certain range.

[0145] The following describes the mine deformation monitoring and warning system based on microwave sensing of the present invention. Using the method of the present invention, such as Figure 1 , which includes,

[0146] A signal generation module that generates a modulated wave voltage signal through a signal generator and then controls the signal by a signal conditioning circuit;

[0147] A microwave transmitting and receiving module that transmits and receives continuous wave microwave signals through a microwave radar;

[0148] A data acquisition module that acquires the radar baseband signal through a data acquisition card;

[0149] A data processing module that processes the signal to calculate the absolute displacement amount and trains the WGAN model for data prediction;

[0150] A warning module that determines whether to issue a warning by comparing with the standard mine data.

[0151] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A mine deformation monitoring and early warning method based on microwave sensing, characterized in that: including the following steps, S1. By deploying microwave radars in the mine for monitoring, continuous wave microwave signals at different parts of the mine are obtained; S2. Using the interference phase modulation information to invert the microwave signals to obtain the absolute displacement amounts at different parts of the mine; S3. In step S3, analyze the obtained absolute displacement amounts, establish a dynamic standard mine deformation table, and use this to train the WGAN model, and use the trained WGAN model to predict the displacement amounts; S4. If the data of the prediction result of the WGAN model exceeds ±3σ, increase the monitoring frequency corresponding to the predicted time, and issue a warning when the measured result also exceeds ±3σ.

2. The method for monitoring and warning of mine deformation based on microwave sensing according to claim 1, wherein: In step S2, using the interference phase modulation information to invert the microwave signals specifically includes: The baseband I / Q signal can be expressed as: where DC I and DC Q are the DC offsets of the I / Q channels respectively, A I and A Q are the signal amplification factors of the I / Q channels respectively, x(t) is the vibration displacement in the time domain, λ is the carrier wavelength, θ = 4πR0 / λ + θ0 is the constant displacement, where R0 is the distance between the target and the radar antenna, θ0 is the displacement change generated by the reflecting surface, and Δφ(t) is the residual phase noise; The change in the vibration displacement of the target at adjacent times can be calculated as: where is the interference phase change of the line time, and t is the time variable; Within the i-th sweep period time, where f0 is the initial carrier frequency, f c is the center carrier frequency, λ0 and λ c are the wavelengths corresponding to the initial and center carrier frequencies respectively, c is the electromagnetic wave propagation speed, and x(iT) are the initial phase and the target vibration displacement of the chirp signal within the time of the i-th frequency sweep period respectively.

3. The mine deformation monitoring and early warning method based on microwave sensing according to claim 2, characterized in that: In step S3, the process of training the WGAN model includes: The training objective function is: Among them, G and D represent the generator and the discriminator respectively; the input of G is the random variable z; the input of D is the mixed data, including the real data labeled 1 and the generated data labeled 0; G(z) represents the sample generated by G; when the mixed data is input into D, D makes a binary classification discrimination on the data. If D discriminates that the current data is real data P data , it outputs closer to 1; if it discriminates as generated data P g , it outputs closer to 0; where X represents the real data; represents the min-max game value function; First, fix G and train D: The best case for D is: Subsequently, fix D and train G: If and only if P data = P g then the generator can generate samples that are different from the true data distribution.

4. The method for mine deformation monitoring and early warning based on microwave sensing according to claim 3, wherein: In step S3, the Wasserstein distance calculation formula is: W(p1, p2) is the Wasserstein distance, and Π(P1, P2) is the set of all possible joint distributions combining the P1 and P2 distributions. For each possible joint distribution γ, a sample x and y can be sampled from it as (x, y) ~ γ, and the pair of samples ||x - y|| can be calculated, so as to calculate the expected value E (x,y)～γ [‖x - y‖]; Rewritten according to Lipschitz continuity to obtain: Among them, W(p r , p θ ) is the equivalent Wasserstein distance obtained according to the Kantorovich-Rubinstein duality. L represents the norm. For the continuous function f, K is the Lipschitz constant of the function, which imposes a restriction on f, requiring that there exists a constant such that for any two elements x1 and x2 in the domain, |f(x1) - f(x2)| ≤ K|x1 - x2|, thereby restricting the maximum local variation amplitude of the continuous function.

5. The method for mine deformation monitoring and early warning based on microwave sensing according to claim 1, wherein: In step S3, finally obtain the two loss functions of the WGAN: WGAN Generator Loss Function: WGAN discriminator loss function: Among them, f w is the discriminator network with the last layer constructed as a non-linear activation layer.

6. The method for mine deformation monitoring and early warning based on microwave sensing according to claim 3, wherein: In step S3, a WGAN model using LSTM as a time series generator and CNN as a discriminator is adopted. Among them, the LSTM is as follows: First, determine which information to discard from the node state, which is decided by the "forget gate layer"; Next, determine which new information to store in the cell node state, which is achieved through the "input gate layer" and the tanh layer; Then multiply the old state by f t , and then add C t ; Finally, decide the output content, put the cell state through tanh and multiply it by the sigmoid gate output.

7. The method for mine deformation monitoring and early warning based on microwave sensing according to claim 2, wherein: In the WGAN, LSTM is selected as the generator. The storage unit of the LSTM consists of a forgetting gate, an input gate, and an output gate. The three thresholds are composed of a sigmoid activation function and element-wise multiplication. The gates of the LSTM network are expressed as follows: i t = σ(w ri s t-1 + w ri x t + w ci c t-1 + b i ) f t = σ(w rf s t-1 + w xf x t + w cf c t-1 + b f ) c t = f t × c t-1 + i t × σ(w rc s t-1 + w xc x t + b c ) o t = σ(w ro s t-1 + w xo x t + w co c t-1 + b o ) s t = o t × tanh(c t ) where x t represents the input at time t, s t-1 represents the output of the previous time step, c t-1 represents the hidden state of the previous time step, and the value of the current time LSTM memory cell is c t , and the output value is s t ; where i t , o t , f t are the values of the input gate, output gate, and forget gate at time t respectively, w is the weight, b is the bias term, and σ is the activation function.

8. A mine deformation monitoring and early warning system based on microwave sensing, characterized in that: Using the method according to claim 1, it includes, A signal generation module that generates a modulated wave voltage signal through a signal generator and then controls the signal through a signal conditioning circuit; A microwave transmitting and receiving module that transmits and receives continuous wave microwave signals through a microwave radar; A data acquisition module that acquires the radar baseband signal through a data acquisition card; A data processing module that processes the signal to calculate the absolute displacement amount, trains the WGAN model, and then performs data prediction; An early warning module that determines whether to issue a warning by comparing with the standard mine data.

Images