Distributed sound wave sensing monitoring device and method for sea wave impact force

Abstract

The invention discloses a distributed sound wave sensing monitoring device and method for sea wave impact force. The device comprises a vibration sensing optical fiber, an anchoring unit, a distributed optical fiber sound wave sensing demodulator, a vibration signal processing unit and a sea wave impact force inversion unit. The method comprises the following steps: fixing the vibration sensing optical fiber on a coast impacted by sea waves through the anchoring unit; the distributed optical fiber sound wave sensing demodulator transmits a monitored sea wave vibration signal to the vibration signal processing unit for preprocessing; the vibration signal processing unit transmits the processed optical fiber vibration amplitude signal to the sea wave impact force inversion unit; obtaining an optical fiber vibration amplitude signal and sea wave impact force training data set through a calibration experiment, and constructing an attention convolutional neural network model for training; and the sea wave impact force inversion unit establishes a corresponding relation between the optical fiber vibration amplitude signal and the sea wave impact force by using an attention convolutional neural network model to obtain the real impact force of the coast sea wave. According to the invention, accurate monitoring of the coastal wave impact force is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of ocean wave monitoring, and in particular to a distributed acoustic wave sensing monitoring device and method for ocean wave impact force. Background Art

[0002] Sea waves are the propagation of the undulating shape of the sea surface. They are a kind of fluctuation formed by water particles leaving the equilibrium position, vibrating periodically, and propagating in a certain direction. The vibration of water particles generates kinetic energy, and the undulation of sea waves generates potential energy, which contains huge energy. When sea waves reach the offshore coast, shore waves will be formed. Under the action of shore waves, the coast will be eroded or even collapsed, causing the coastline to retreat. The collapse and retreat of the coast seriously affect the safety of life and property of coastal people and the safety of navigation. The impact of sea waves on the coast is closely related to the impact force of sea waves. Therefore, the monitoring of sea wave impact force is of great significance to the protection of coastal coasts and marine ecological environment.

[0003] Traditional wave monitoring devices include wave buoys, hydraulic wave meters and remote sensing wave meters.

[0004] A wave buoy is an observation instrument that drifts with the ocean current or observes the sea surface amplitude, wave period, wave direction, and wave spectrum at a fixed point. It indirectly calculates the wave energy by combining the observed data with wave theory. However, this method can only obtain the wave energy, but not the impact of the wave on the coast.

[0005] The hydraulic wave meter is an instrument that uses the wave analysis method of linear wave theory to measure wave parameters. The device has high accuracy in monitoring long-period waves and is mainly used for shallow water wave monitoring. Like the wave buoy, this wave measurement method cannot monitor the impact force of waves.

[0006] Remote sensing wave meters include laser wave meters and radar wave meters. Laser wave meters have high ranging accuracy, can continuously record wave change information and realize all-weather monitoring, but are greatly affected by high temperature, high salinity, high humidity and typhoons and other disastrous weather. Radar wave meters can measure many ocean elements, have high measurement accuracy and high spatial resolution, but they occupy a large area, are difficult to move, are expensive, and are easily damaged by typhoons and other disastrous weather. Although the above two methods can obtain a variety of information about waves and can also obtain wave energy distribution through certain calculations, neither of them can monitor the impact force of waves.

[0007] How to conduct distributed real-time monitoring of wave impact force is a technical problem that needs to be solved urgently. Summary of the invention

[0008] Purpose of the invention: The purpose of the present invention is to propose a distributed acoustic wave sensing monitoring device and method for wave impact force; obtain neural network model training data through calibration tests conducted on simulated wave impact indoor coastal models, and design and train the designed attention convolutional neural network model, i.e., the wave impact force judgment model. The vibration sensing optical fiber is laid on the coastal rock wall, and the distributed optical fiber acoustic wave sensing method is used to collect the optical fiber vibration amplitude signal of the wave impacting the coast in real time. After preprocessing the optical fiber vibration amplitude signal, the optical fiber vibration amplitude signal is transmitted to the wave impact force inversion unit, and the trained wave impact force judgment model is used to judge the impact force caused by the waves.

[0009] Technical solution: To achieve the above-mentioned purpose, the distributed acoustic wave sensing monitoring device of the wave impact force of the present invention comprises a vibration sensing optical fiber, an anchoring unit, a distributed optical fiber acoustic wave sensing demodulator, a vibration signal processing unit and a wave impact force inversion unit; the vibration sensing optical fiber is fixed to the coast surface through the anchoring unit; at the same time, the vibration sensing optical fiber is connected to the distributed optical fiber acoustic wave sensing demodulator through an optical fiber lead; the distributed optical fiber acoustic wave sensing demodulator collects the optical fiber vibration amplitude signal and transmits it to the vibration signal processing unit; the vibration signal processing unit pre-processes the optical fiber vibration amplitude signal to improve the signal-to-noise ratio, and transmits the collected vibration amplitude signal to the wave impact force inversion unit; the wave impact force inversion unit obtains the real-time monitoring result of the wave impact force through the wave impact force judgment model.

[0010] The distributed acoustic wave sensing monitoring method of the wave impact force of the present invention comprises the following steps:

[0011] Step (1), conduct a calibration test on an indoor coastal model: paste the vibration sensing optical fiber on the wave impact force judgment model, paste the force sensor on the vibration sensing optical fiber, and use a wave maker to generate waves with different impact forces. Collect the vibration amplitude data on the vibration sensing optical fiber through a distributed optical fiber acoustic wave sensor demodulator, and simultaneously record the impact force monitored by the force sensor; divide the vibration amplitude data of the vibration sensing optical fiber at the force sensor during each wave impact into a training set, a validation set, and a test set;

[0012] Step (2), based on the data collected from the calibration test and the training set, validation set and test set of step (1), an attention convolutional neural network model, i.e., a wave impact force judgment model, is designed and trained, thereby obtaining the functional relationship between the optical fiber vibration amplitude signal and the wave impact force, and judging the magnitude of the wave impact force; the process is:

[0013] In step (2.1), since the optical fiber vibration amplitude signal has time dependence, the wave impact force judgment model adopts a one-dimensional convolution layer to extract the time characteristics in the signal; that is, the optical fiber vibration amplitude signal is processed by a one-dimensional convolution layer to capture the time dependence characteristics of the signal.

[0014] The specific process is that the one-dimensional convolution layer calculates the weighted sum in each window by sliding the convolution kernel along the length of the optical fiber vibration amplitude signal, extracts the time series characteristics of the optical fiber vibration amplitude signal, and thus improves the accuracy of the wave impact force judgment model for judging the wave impact force.

[0015] In step (2.2), after the one-dimensional convolution layer, the wave impact force judgment model adds an attention mechanism module, which determines the influence weight of the amplitude data in the optical fiber amplitude signal on the impact force through pooling convolution.

[0016] In step (2.3), after the attention mechanism module and the one-dimensional convolutional layer extract the amplitude time series features, the fully connected layer uses the output of the previous layer as input, uses weights and biases for conversion, and outputs one or more predicted values ​​to judge the final size of the wave impact force.

[0017] Step (2.4), in the process of training the wave impact force judgment model, the mean square error MSE is used as the loss function to measure the deviation between the predicted value and the true value. By optimizing the loss function, the prediction accuracy of the model is improved. Due to the impact of waves, the optical fiber laid on the limestone coast model is slightly deformed, which will change the phase information of the internal optical signal transmission, thereby changing the optical fiber vibration amplitude signal. After model training, the functional relationship between the optical fiber vibration amplitude signal and the wave impact force is as follows:

[0018]

[0019] Where: F is the impact force; A is the fiber vibration amplitude signal; λ is the wavelength of the laser pulse; the phase change Δφ; n is the refractive index of the fiber; ξ is the elastic-optical coefficient; L g is the gauge length; the pulse interval is ΔT; and f is the model function relationship for predicting the impact force of waves.

[0020] Step (3), fix the vibration sensing optical fiber to the rocky coast impacted by the waves through an anchoring unit, and check the integrity of the optical fiber line.

[0021] Step (4), connecting the vibration sensing optical fiber to the distributed optical fiber acoustic wave sensor demodulator through an optical fiber lead, setting the same sampling parameters as the calibration test, and collecting the optical fiber vibration amplitude signal generated by the wave impact vibration sensing optical fiber.

[0022] Step (5), transmitting the optical fiber vibration amplitude data collected by the distributed optical fiber acoustic wave sensor demodulator to the vibration signal processing unit, preprocessing the optical fiber vibration amplitude data to improve the data signal-to-noise ratio.

[0023] In step (6), the vibration signal processing unit transmits the preprocessed optical fiber vibration amplitude data to the wave impact force inversion unit. In the wave impact force inversion unit, the preprocessed optical fiber vibration amplitude data is first accurately divided into multiple time window data of fixed length. Then, each time window data is separately sent to the wave impact force judgment model. The one-dimensional convolution layer of the wave impact force judgment model first extracts the data features of each time window, and automatically captures the key time features through multiple filters; then, the attention mechanism module adjusts the weights of the extracted data features, and gives priority to the amplitude time series features that are key to predicting the wave impact force. Finally, the fully connected layer combines the attention-weighted features to form a comprehensive feature vector to predict the impact force of the waves in the time window data; the fully connected layer outputs a quantized impact force value, which reflects the actual force of the waves on the optical fiber sensor in the specific time window data.

[0024] In step (1), waves with different impact forces are generated indoors using a wave maker, and the coastal model equipped with vibration sensing optical fibers and force sensors is impacted to obtain optical fiber vibration amplitude data and force sensor data. The impact force monitored by the force sensor is used as a label, and the optical fiber vibration amplitude data at the position of the force sensor during the wave impact period is used as a data set. The data set is divided into a training set, a validation set, and a test set for the neural network model.

[0025] In step (2.4), the wave impact force judgment model uses the mean square error as the loss function to measure the deviation between the predicted value and the true value, and optimizes the loss function; at the same time, the wave impact force judgment model uses the Adam optimizer to update parameters. The optimizer combines momentum and adaptive learning rate adjustment to increase the training speed; it effectively accelerates the training process and helps avoid falling into the local minimum, thereby ensuring faster and more stable convergence performance.

[0026] In step (3), after the vibration sensing optical fiber is laid, the integrity of the optical fiber line is ensured, and the optical fiber is fixed with an anchoring unit to ensure the stability of the optical fiber position, so as to ensure that the optical fiber is stably laid on the coast and has the ability to resist seawater corrosion.

[0027] In step (4), ensure that the vibration sensing optical fiber, optical fiber lead and distributed optical fiber acoustic wave sensor demodulator are connected, and the sampling parameters related to the channel spacing and sampling frequency of the distributed optical fiber acoustic wave sensor demodulator are designed according to the parameters of the distributed optical fiber acoustic wave sensor demodulator used in the indoor calibration test.

[0028] In step (5), the vibration signal processing unit performs preprocessing operations such as pinching out, de-linear trending, and filtering on the collected optical fiber vibration amplitude signal to improve the signal-to-noise ratio of the collected signal.

[0029] In step (6), the fully connected layer forms a comprehensive feature vector using the attention-weighted features to predict the impact force of the waves in the time window data, and outputs a quantified impact force value through a set threshold or calculation model.

[0030] In step (6), when the wave impact force inversion unit processes the optical fiber vibration amplitude signal, the time length of the time window data is set to half of the typical wave cycle, and each time window data overlaps with the next time window data by half of the time length, so that each wave event appears completely in at least one time window data. Specifically, the time length of the time window data is set to half of the typical wave cycle to ensure that each wave event is fully displayed in at least one window, even if the start or end part of the event is located at the boundary of two time window data. In addition, each time window data overlaps with the subsequent time window data by half of the time, reducing the data loss and boundary effect caused by time window switching, which not only improves the continuity and integrity of the data, but also enhances the ability of the wave impact force judgment model to capture the characteristics of each wave event.

[0031] In step (6), the vibration amplitude data contained in each window is first standardized before being input into the wave impact force judgment model to eliminate possible deviations caused by different weather conditions, thereby making a more accurate judgment on the model.

[0032] In step (6), in the process of judging the impact force of waves, the wave impact force inversion unit first inputs the standardized time window data into the wave impact force judgment model for judgment. The model processes the preprocessed data window through a multi-layer attention convolution layer, learns the key amplitude time series features in the vibration signal, and focuses on the signal part that has the greatest influence on the judgment of the impact force. The fully connected layer forms a comprehensive feature vector with the weighted features of attention, and maps it into a judgment value using linear and nonlinear transformations, that is, the judgment value is the impact force of the waves.

[0033] Working principle: The present invention utilizes the characteristics of distributed fiber optic acoustic wave sensing method to achieve long-distance and large-scale monitoring, and conducts long-distance and large-scale monitoring of the impact force of waves in coastal areas. The distributed fiber optic acoustic wave sensing technology has a high spatial sampling rate and a wide frequency response range, so it is easy to achieve spatial intensive sampling; at the same time, it ensures sufficient data volume, which is convenient for analyzing the distribution of wave impact force in each time period. The distributed fiber optic acoustic wave sensing technology also has the advantages of anti-electromagnetic interference, high durability, and real-time response, which has great advantages for monitoring the impact force of waves along the coast.

[0034] The distributed fiber optic acoustic wave sensing technology used in the present invention has a test frequency of 0.1Hz to 40kHz. By capturing the vibration signal of the entire process of wave impact and exploring the relationship between the distributed fiber optic acoustic wave sensing vibration signal and the impact force, combined with a convolutional neural network, distributed real-time monitoring of the wave impact force is achieved.

[0035] The distributed acoustic wave sensing monitoring device and method for the wave impact force of the present invention uses an anchoring unit to fix a vibration sensing optical fiber to the surface of the coast impacted by the wave; the vibration sensing optical fiber is connected to a distributed optical fiber acoustic wave sensing demodulator through an optical fiber lead; the other end of the distributed optical fiber acoustic wave sensing demodulator is connected to a vibration signal processing unit; the distributed optical fiber acoustic wave sensing demodulator transmits the monitored wave vibration signal to the vibration signal processing unit for preprocessing to improve the signal-to-noise ratio; the vibration signal processing unit transmits the optical fiber vibration amplitude signal obtained after processing to the wave impact force inversion unit; and a wave maker is used to perform an indoor calibration experiment to obtain the optical fiber vibration amplitude signal and the wave impact The wave impact force inversion unit uses the trained attention convolutional neural network model to establish the corresponding relationship between the optical fiber vibration amplitude signal and the wave impact force. The vibration signal processing unit transmits the pre-processed optical fiber vibration amplitude signal to the wave impact force inversion unit. The wave impact force inversion unit first segments the optical fiber vibration amplitude data into time series, and then imports the segmented optical fiber vibration amplitude signal into the trained wave impact force judgment model to obtain the corresponding wave impact force monitoring result, thereby obtaining the real impact force of coastal waves in real time and realizing distributed real-time monitoring of coastal wave impact force.

[0036] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0037] (1) The present invention utilizes a convolutional neural network combined with an attention mechanism to learn key amplitude time series features through a wave impact force judgment model, and accurately associates the relationship between the signal features in each time window data and the actual wave impact force. This not only improves the accuracy of the wave impact force judgment model in processing dynamic and complex wave data, but also enhances the accuracy of the prediction, thereby achieving accurate monitoring of the wave impact force along the coast.

[0038] (2) The present invention adopts the attention mechanism to prioritize the amplitude timing information that is most critical for judgment, thereby enhancing the nonlinear processing capability of the wave impact force judgment model and improving the accuracy and efficiency of judging the magnitude of the wave impact force.

[0039] (3) The device and method described in the present invention use anchoring units to lay optical fibers on the surface of the coast, which reduces costs and fixes the optical fibers. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the composition of the distributed acoustic wave sensor monitoring device for the impact force of ocean waves of the present invention;

[0041] Figure 2 It is a schematic diagram of the indoor calibration test of the present invention;

[0042] Figure 3 It is a schematic diagram of the wave impact force judgment model of the present invention;

[0043] Figure 4 Schematic diagram of the attention mechanism module of the present invention;

[0044] Figure 5 Schematic diagram of the original vibration signal obtained when the ocean wave impacts the optical fiber in an embodiment of the present invention;

[0045] Figure 6 Schematic diagram of the result of the vibration signal processing unit in the embodiment of the present invention. DETAILED DESCRIPTION

[0046] The distributed acoustic wave sensing monitoring device for wave impact force of the present invention comprises a vibration sensing optical fiber 1, an anchoring unit 2, a distributed optical fiber acoustic wave sensing demodulator 3, a vibration signal processing unit 4 and a wave impact force inversion unit 5. The vibration sensing optical fiber 1 is fixed to the surface of the rocky coast through the anchoring unit 2; one end of the distributed optical fiber acoustic wave sensing demodulator 3 is connected to the vibration sensing optical fiber 1 through an optical fiber lead, and the other end is connected to the vibration signal processing unit 4; the vibration signal processing unit 4 is connected to the wave impact force inversion unit 5.

[0047] The distributed acoustic wave sensing monitoring method of the wave impact force of the present invention comprises the following steps:

[0048] Step (1), a calibration test is performed on an indoor coastal model: the vibration sensing optical fiber 1 is pasted on the coastal model, the force sensor 7 is pasted on the vibration sensing optical fiber 1, and a wave maker 6 is used to generate waves with different impact forces. The vibration amplitude signal on the vibration sensing optical fiber is collected by a distributed optical fiber acoustic wave sensor demodulator 3, and the impact force monitored by the force sensor 7 is recorded at the same time. The impact force monitored by the force sensor 7 is used as a label, and the vibration amplitude data of the vibration sensing optical fiber at the position of the force sensor during each wave impact is divided into a training set, a validation set, and a test set of the neural network model.

[0049] Step (2) designs and trains an attention convolutional neural network model, i.e., a wave impact force judgment model, based on the data collected from the calibration test and the training set, validation set, and test set. The wave impact force judgment model uses a convolutional neural network in combination with an attention mechanism to judge the magnitude of the wave impact force through the functional relationship between the vibration amplitude signal and the impact force.

[0050] Step (2.1), since the optical fiber vibration amplitude signal has time dependence, the wave impact force judgment model contains a one-dimensional convolution layer, and the one-dimensional convolution layer extracts the time series characteristics in the signal through the convolution kernel. The present invention is different from the conventional convolution network. The size and step size of each convolution kernel of the present invention are optimized to match the typical wave cycle, so as to more accurately capture the time-dependent characteristics related to the wave impact force. Such a design not only enhances the functional relationship between the optical fiber vibration amplitude signal and the wave impact force, but also improves the response speed and processing efficiency of the wave impact force judgment model.

[0051] In this step (2.1), the wave impact force judgment model includes five sequentially arranged one-dimensional convolutional layers, each of which is followed by batch normalization and ReLU activation function. This hierarchical structure helps to extract complex features in the optical fiber vibration amplitude signal while maintaining computational efficiency. In the specific implementation process, each convolutional layer uses a set of specific convolution kernels to extract features from the input vibration amplitude signal. These convolution kernels slide on the signal data and calculate the dot product in the coverage area to extract local features. Batch normalization is performed immediately after each convolutional layer, and its main function is to normalize the feature map output by the convolutional layer and standardize it to a distribution with a mean close to 0 and a variance close to 1. This step improves the stability of the wave impact force judgment model training and accelerates the convergence process. Next, the ReLU activation function is applied to the output of batch normalization, introducing nonlinear processing to enhance the ability of the wave impact force judgment model to process complex signal features. Through the above hierarchical structure design, the wave impact force judgment model deeply explores and extracts the complex features in the signal layer by layer. The feature extraction of each layer of convolution and the subsequent normalization processing, as well as the nonlinear activation work together to improve the accuracy of the wave impact force judgment model in processing vibration signals.

[0052] Step (2.2), adding an attention mechanism module to the wave impact force judgment model, the attention mechanism module includes a channel attention mechanism and a spatial attention mechanism. The attention mechanism module not only analyzes the importance of each channel in the optical fiber vibration amplitude signal, but also determines the contribution of each optical fiber vibration amplitude signal in space, so that the wave impact force judgment model focuses on the amplitude time series signal that is most critical to the wave impact force judgment, thereby improving the accuracy of the judgment.

[0053] In this step (2.2), the wave impact force model adds an attention mechanism module after the third and fifth convolutional layers. The attention mechanism module includes a channel attention mechanism module and a spatial attention mechanism module. These two attention mechanisms work together to optimize the processing of each data point in the optical fiber vibration amplitude signal by the wave impact force judgment model. The channel attention mechanism module first uses global average pooling and maximum pooling operations to extract the statistical information of each channel, and then learns the weight coefficient of each channel through a fully connected layer, thereby strengthening the response to important feature channels. The spatial attention mechanism module focuses on the spatial distribution of the signal and identifies important feature areas in space by performing pooling and convolution operations on each feature map. The attention mechanism module further improves the ability of the wave impact force judgment model to distinguish complex wave impact forces by focusing on the areas that contribute greatly to the prediction results in space, thereby enabling the attention mechanism module to learn and emphasize the most critical influence weights on the impact force judgment in the optical fiber amplitude signal, thereby enabling the wave impact force judgment model to more accurately predict different impact forces when processing wave signals with complex dynamic characteristics, thereby improving the accuracy of the prediction and the practicality of the wave impact force judgment model.

[0054] In step (2.3), after the attention mechanism module and the one-dimensional convolutional layer extract the amplitude time series features, the wave impact force judgment model uses a fully connected layer to take the output of the previous layer as input, converts it using weights and biases, and outputs one or more predicted values ​​to judge the final wave impact force. The fully connected layer not only includes linear nodes, but also introduces nonlinear processing units such as ReLU or Leaky ReLU to enhance the nonlinear expression ability of the wave impact force judgment model. In addition, the number of nodes and the hierarchical structure in the fully connected layer are optimized to ensure the integration and mapping of the amplitude time series features to the final impact force judgment value.

[0055] In step (2.4), in the process of training the wave impact force judgment model, the mean square error (MSE) is selected as the loss function because the loss function intuitively quantifies the error between the impact force judgment value and the actual value, and because a regularization term is added during the optimization of the wave impact force judgment model to prevent overfitting. At the same time, the Adam optimizer is used for adjustment, and the momentum and adaptive learning rate parameters are set according to the characteristics of the wave data to accelerate the convergence speed and improve the stability and prediction accuracy of the wave impact force judgment model under different sea conditions.

[0056] Step (3), laying the vibration sensing optical fiber 1 through the anchoring unit 2 on the coast hit by the waves, and checking the integrity of the optical fiber line;

[0057] Step (4), connecting the vibration sensing optical fiber 1 to the distributed optical fiber acoustic wave sensor demodulator 3 through an optical fiber lead, checking the smoothness of optical fiber signal transmission, setting the same sampling parameters as the calibration test, and collecting the optical fiber vibration amplitude signal generated by the wave impact vibration sensing optical fiber;

[0058] Step (5), transmitting the optical fiber vibration amplitude signal data collected by the distributed optical fiber acoustic wave sensor demodulator 3 to the vibration signal processing unit 4, and preprocessing the optical fiber vibration amplitude data to improve the data signal-to-noise ratio.

[0059] Step (6), the vibration signal processing unit 4 transmits the pre-processed optical fiber vibration amplitude data to the wave impact force inversion unit 5, and the wave impact force inversion unit 5 divides the pre-processed continuous optical fiber vibration amplitude data into a plurality of fixed-length time window data, so that the wave impact force judgment model can capture the time-varying characteristics caused by the wave impact, thereby improving the timeliness and accuracy of monitoring. Each time window data is then input into the trained wave impact force judgment model. The wave impact force judgment model uses the internal one-dimensional convolution layer, attention mechanism module and full connection layer to comprehensively analyze the vibration characteristics of each time window data, thereby judging the size of the wave impact force at the corresponding moment.

[0060] In step 1, the vibration sensing optical fiber 1 is horizontally arranged on the coastal model, and the force sensor 7 is evenly pasted along the vibration sensing optical fiber. A wave maker is used to create waves with different impact forces, and the optical fiber vibration amplitude data and force sensor data obtained by impacting the coastal model are obtained. The impact force monitored by the force sensor is used as a label, and the optical fiber vibration amplitude data at the position of the force sensor during the wave impact time period is used as a data set. The data set is divided into a training set, a validation set, and a test set of the neural network model in a ratio of 8:1:1.

[0061] In step 2, due to the impact of waves, the optical fiber laid on the limestone coast model is slightly deformed, which will change the phase information of the internal optical signal transmission, thereby changing the optical fiber vibration amplitude signal. After training the wave impact force judgment model, the functional relationship between the optical fiber vibration amplitude signal and the wave impact force is obtained:

[0062]

[0063] Where: F is the impact force; A is the fiber vibration amplitude signal; λ is the wavelength of the laser pulse; the phase change Δφ; n is the refractive index of the fiber; ξ is the elastic-optical coefficient; L g is the gauge length; the pulse interval is ΔT; and f is the model function relationship.

[0064] In step 2.1, the wave impact force judgment model contains five one-dimensional convolutional layers, each followed by batch normalization and ReLU activation function. This hierarchical structure facilitates the gradual extraction of complex features in the signal while maintaining computational efficiency. Convolutional layers are not directly connected, but are transitioned through batch normalization and activation function layers, which facilitates stability and nonlinear enhancement during the learning process of the wave impact force model.

[0065] In step 2.2, the wave impact force judgment model adds an attention mechanism module after the third and fifth convolutional layers respectively. The attention mechanism module includes a channel attention mechanism module and a spatial attention mechanism module. The attention mechanism module determines the influence weight of each amplitude data in the optical fiber amplitude signal on the impact force judgment through pooling convolution operations, which is crucial to distinguish complex waves with different impact forces.

[0066] In step 3, after the vibration sensing optical fiber is laid, the integrity of the optical fiber line must be ensured, and the optical fiber must be fixed with an anchor unit to ensure the stability of the optical fiber position. It must also be ensured that the optical fiber is stably laid on the coast and has the ability to resist seawater corrosion.

[0067] In step 4, it should be ensured that the optical fiber is tightly connected to the distributed optical fiber acoustic wave sensor demodulator, and the sampling parameters related to the track spacing and sampling frequency of the distributed optical fiber acoustic wave sensor demodulator are determined according to the parameters of the distributed optical fiber acoustic wave sensor demodulator used in the indoor calibration test.

[0068] In step 5, the vibration signal processing unit performs pinch-out, de-linear trend, and filtering preprocessing operations on the collected optical fiber vibration amplitude signal to improve the signal-to-noise ratio of the collected signal.

[0069] In step 6, the length of the time window data is set to half of the typical wave cycle. Each window overlaps with the next window for half the time length, so that each wave event appears completely in at least one window, even if its start or end is at the junction of two windows, thereby ensuring the continuity and integrity of the data and avoiding information loss caused by data cutting at the window boundary.

[0070] In step 6, the data contained in each time window is first standardized before being input into the wave impact force judgment model to eliminate possible deviations caused by different weather conditions.

[0071] In step 6, the standardized time window data is input into the wave impact force judgment model for judgment. The wave impact force judgment model processes the preprocessed data window through a multi-layer attention convolution layer to learn the key amplitude time series features in the vibration signal. The amplitude time series features ensure that the wave impact force judgment model focuses on the signal part that has the greatest influence on the impact force judgment through the attention mechanism. Then, the fully connected layer integrates these features and uses linear and nonlinear transformations to map them to a specific judgment value, that is, the impact force of the waves. In this process, the wave impact force judgment model accurately associates and judges the wave impact force in each time window data, improving the accuracy and reliability of the model judgment.

[0072] Example

[0073] like Figure 1 As shown, the distributed acoustic wave sensing monitoring device for the wave impact force of the present invention comprises a vibration sensing optical fiber 1, an anchoring unit 2, a distributed optical fiber acoustic wave sensing demodulator 3, a vibration signal processing unit 4 and a wave impact force inversion unit 5. The vibration sensing optical fiber 1 is fixed to the coastline through the anchoring unit 2; one end of the distributed optical fiber acoustic wave sensing demodulator 3 is connected to the vibration sensing optical fiber 1, and the other end is connected to the vibration signal processing unit 4. The vibration signal processing unit 4 is connected to the wave impact force inversion unit 5. The wave maker 6 is a push-plate wave maker in the form of a servo motor driving a ball screw, and the maximum amplitude of the waves produced is 20 cm, and various waveforms are produced.

[0074] The distributed acoustic wave sensing monitoring method and system of the present invention are used to monitor the wave impact force of a certain coast in real time. The selected coast is a limestone coast, which is 2km long and 50m high. The distance between the sea level and the coastline is 30m. According to the characteristics of the limestone, the anchoring unit selects a cross anchor.

[0075] The distributed acoustic wave sensing monitoring method of the wave impact force of the present invention comprises the following steps:

[0076] Step (1), using Figure 2 The indoor calibration test device shown in the figure is used for the calibration test. Limestone is selected to make the coastal model. The vibration sensing optical fiber is horizontally arranged on the coastal model. Four force sensors 7 are evenly pasted along the vibration sensing optical fiber. The sampling frequency of the distributed optical fiber demodulator is set to 100Hz. Waves with different impact forces are generated by the wave maker 6 to impact the limestone coastal model with the vibration sensing optical fiber fixed. Since the impact of the waves in the impact process is between a few seconds and thirty seconds, the time when the wave maker emits the wave is taken as the starting time of each effective optical fiber vibration amplitude signal. The optical fiber vibration amplitude signal monitored in the calibration test is sliced ​​for 30 seconds. Therefore, each segment of data has 3000 characteristic optical fiber vibration amplitude data, and a total of 5000 segments of calibration test data are collected.

[0077] At the same time, according to the correspondence between the optical fiber vibration amplitude data and the impact force monitored by the force sensor, each segment of optical fiber vibration amplitude data is labeled with the impact force monitored by the force sensor, and the corresponding feature optical fiber data is divided into training set, validation set, and test set. The data volume ratio is 8:1:1, that is, 4000 segments of training set data, 500 segments of validation set and 500 segments of test set respectively.

[0078] In step (2), the impact causes the optical fiber installed on the limestone coast model to be slightly deformed, which in turn changes the phase information of the internal optical signal transmission, thereby changing the optical fiber vibration amplitude signal. After model training, the functional relationship between the optical fiber vibration amplitude signal and the wave impact force is obtained:

[0079]

[0080] Where: F is the impact force; A is the fiber amplitude; λ is the wavelength of the laser pulse; the phase change Δφ; n is the refractive index of the fiber; ξ is the elastic-optical coefficient; L g is the gauge length; the pulse interval is ΔT; and f is the model function relationship.

[0081] It can be concluded from the above that based on the data collected from the calibration test, an attention convolutional neural network model, namely the wave impact force judgment model, is designed and trained to obtain the functional relationship between the optical fiber vibration amplitude signal and the wave impact force.

[0082] Step (2.1), since the optical fiber vibration amplitude signal has time dependence, the wave impact force judgment model is designed to use five one-dimensional convolutional layers to extract the time features in the signal, and each layer is followed by batch normalization and ReLU activation function. This hierarchical structure facilitates the extraction of complex features in the signal while maintaining computational efficiency. The convolutional layers are not directly connected, but are transitioned through batch normalization and activation function layers, which facilitates the stability and nonlinear enhancement of the wave impact force judgment model learning process.

[0083] In step (2.2), the wave impact force judgment model adds attention mechanism modules after the third and fifth convolutional layers, respectively. The attention mechanism modules include channel attention mechanism modules and spatial attention mechanism modules. The design uses the attention mechanism modules to enhance the network's perception of important features in the input signal. The wave impact force judgment model focuses on the most informative part of the optical fiber vibration amplitude data, improving the accuracy of the model's judgment of the wave impact force.

[0084] In step (2.3), after the attention mechanism module and the one-dimensional convolutional layer extract the amplitude time series features, the wave impact force judgment model uses the fully connected layer to further integrate these features. The fully connected layer synthesizes the local features obtained by the wave impact force judgment model to make the final impact force judgment.

[0085] In step (2.4), the wave impact force judgment model training process uses the mean square error (MSE) as the loss function, which intuitively quantifies the error between the impact force judgment value and the actual value and is suitable for continuous value judgment. At the same time, the Adam optimizer is used, which combines the advantages of momentum and adaptive learning rate to improve the training speed of the wave impact force judgment model.

[0086] In step (3), the vibration sensing optical fiber 1 is fixedly laid at a position 0.5 m above the sea level using an anchoring unit 2 to ensure that the waves impact the optical fiber and generate vibration amplitude data. The total length of the optical fiber is 1.5 km. After the laying is completed, the integrity of the optical fiber line needs to be ensured.

[0087] Step (4), after the vibration sensing optical fiber 1 is connected to the distributed optical fiber acoustic wave sensor demodulator 3 through the optical fiber lead, the smoothness of the optical fiber channel is checked, the sampling frequency of the distributed optical fiber acoustic wave sensor demodulator 3 is set to 100 Hz, the channel spacing is 5m, the number of channels is 600, and the optical fiber vibration amplitude data caused by the slight deformation of the optical fiber caused by the waves hitting the coast is collected in real time.

[0088] Step (5), the distributed optical fiber acoustic wave sensor demodulator 3 transmits the optical fiber vibration amplitude signal to the vibration signal processing unit 4, and the extracted raw data is as follows: Figure 5 In the vibration signal processing unit 4, the optical fiber vibration amplitude signal is extinguished to remove the linear trend and improve the signal-to-noise ratio. Since the calibration test found that the sea wave signal in the 1-50 Hz frequency band is more significant, the sea wave signal is subjected to a 1-50 Hz bandpass filter. Figure 6 It is the result of the 30s optical fiber vibration amplitude signal preprocessed by the vibration signal processing unit 4.

[0089] Step (6), the vibration signal processing unit 4 transmits the pre-processed data to the wave impact force inversion unit 5. In order to accurately judge the wave impact force in each 30-second window and associate an accurate time tag, the wave impact force inversion unit 5 first performs a highly refined time series segmentation on the 24-hour optical fiber vibration amplitude signal. Each data window is fixed to 30 seconds. By designing a time window with a 15-second overlap (i.e., the start time of each window is 15 seconds later than the previous window), the continuity and integrity of the wave event are ensured to be retained in at least one complete window. The time window data is standardized to eliminate errors caused by temperature and humidity on site. The standardized time window data is input into the wave impact force judgment model for judgment. The wave impact force judgment model processes the pre-processed data window through a multi-layer attention convolution layer to learn the key amplitude time series features in the vibration signal. These amplitude time series features are emphasized through the attention mechanism to ensure that the wave impact force judgment model focuses on the signal part that has the greatest influence on the judgment of the impact force. Then, the fully connected layer integrates these features and maps them to a specific judgment value, i.e., the impact force of the wave, using linear and nonlinear transformations. Each judgment value is accompanied by a time tag, which directly corresponds to the actual time of its data window. In this process, the wave impact force judgment model accurately associates and judges the wave impact force in each time window data, improving the accuracy and reliability of the judgment.

[0090] According to the distributed optical fiber facilities deployed on site, the distributed optical fiber acoustic wave sensing technology is used to monitor the wave impact in real time and obtain the optical fiber vibration amplitude signal of the wave impact. In the indoor calibration test, the attention convolutional neural network model is used for training to realize the correspondence between the optical fiber vibration amplitude data and the wave impact force. The wave impact force inversion unit then divides the optical fiber vibration amplitude data into time series and imports it into the wave impact force judgment model. The wave impact force judgment model judges the wave impact force at each moment based on the correspondence between the optical fiber vibration amplitude data obtained through training and the wave impact force, thereby realizing intelligent real-time monitoring of the coastal wave impact force.

Claims

1. A distributed acoustic wave sensing monitoring device for wave impact force, characterized in that: The invention comprises a vibration sensing optical fiber (1), an anchoring unit (2), a distributed optical fiber acoustic wave sensor demodulator (3), a vibration signal processing unit (4) and a sea wave impact force inversion unit (5); the vibration sensing optical fiber (1) is fixed to the surface of the coast via the anchoring unit (2); the vibration sensing optical fiber (1) is connected to the distributed optical fiber acoustic wave sensor demodulator (3) via an optical fiber lead; the distributed optical fiber acoustic wave sensor demodulator (3) transmits the collected vibration amplitude signal of the vibration sensing optical fiber to the sea wave impact force inversion unit (5) via the vibration signal processing unit (4).

2. A distributed acoustic wave sensing monitoring method for wave impact force, characterized in that: The monitoring method is implemented by the distributed acoustic wave sensor monitoring device for wave impact force according to claim 1, and comprises the following steps: Step (1), laying out the vibration sensing optical fiber (1) on an indoor coastal model for a calibration test, pasting a force sensor (7) on the vibration sensing optical fiber (1), and generating waves with different impact forces through a wave maker (6); collecting vibration amplitude data on the vibration sensing optical fiber through a distributed optical fiber acoustic wave sensor demodulator (3), and recording the impact force monitored by the force sensor; dividing the vibration amplitude data of the vibration sensing optical fiber at the force sensor during each wave impact into a training set, a validation set, and a test set; Step (2), based on the training set, validation set and test set, design and train an attention convolutional neural network model, i.e., a wave impact force judgment model, to obtain the functional relationship between the optical fiber vibration amplitude signal and the wave impact force, and judge the magnitude of the wave impact force; the process is: Step (2.1), the wave impact force judgment model uses a one-dimensional convolution layer to calculate the weighted sum in each window along the length of the optical fiber vibration amplitude signal by sliding the convolution kernel to extract the time series characteristics of the optical fiber vibration amplitude signal; Step (2.2), adding an attention mechanism module to the wave impact force judgment model, and the attention mechanism module determines the influence weight of the amplitude data in the optical fiber amplitude signal on the impact force through pooling convolution; Step (2.3), after the attention mechanism module and the one-dimensional convolutional layer extract the amplitude time series features, the fully connected layer uses the output of the previous layer as input, uses weights and biases to transform, and outputs one or more predicted values ​​to judge the final wave impact force; Step (2.4), the wave impact force judgment model uses the mean square error as the loss function to measure the deviation between the predicted value and the true value, and optimizes the loss function; Among them, the functional relationship between the optical fiber vibration amplitude signal and the wave impact force is as follows: Where: F is the impact force of the waves; A is the fiber amplitude; λ is the wavelength of the laser pulse; the phase change Δφ; n is the refractive index of the fiber; ξ is the elastic-optical coefficient; L g is the gauge length; the pulse interval is ΔT; f is the model function relationship; Step (3), fixing the vibration sensing optical fiber (1) to the coast hit by the waves through the anchoring unit (2); Step (4), connecting the vibration sensing optical fiber (1) to the distributed optical fiber acoustic wave sensor demodulator (3) through an optical fiber lead, setting the same sampling parameters as those in the calibration test, and collecting the optical fiber vibration amplitude signal generated by the wave impact vibration sensing optical fiber; Step (5), transmitting the optical fiber vibration amplitude signal collected by the distributed optical fiber acoustic wave sensor demodulator (3) to the vibration signal processing unit (4), and preprocessing the optical fiber vibration amplitude data to improve the data signal-to-noise ratio; Step (6), the vibration signal processing unit (4) transmits the pre-processed optical fiber vibration amplitude data to the wave impact force inversion unit (5), and the pre-processed optical fiber vibration amplitude signal data is divided into a plurality of time window data of fixed length; the time window data is input into the wave impact force judgment model; the one-dimensional convolution layer of the wave impact force judgment model first extracts the data features of each time window, and then the attention mechanism module adjusts the weights of the data features, and the fully connected layer forms a comprehensive feature vector with the features weighted by attention to predict the impact force of the waves in the time window data, and outputs a quantized impact force value.

3. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 2 is characterized in that: In step (1), the magnitude of the impact force monitored by the force sensor is used as a label, and the optical fiber vibration amplitude data at the force sensor during the wave impact period is used as a data set, and the data set is divided into a training set, a validation set, and a test set.

4. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 2 is characterized in that: In step (2.4), the wave impact force judgment model uses the mean square error as the loss function to measure the deviation between the predicted value and the true value, and optimizes the loss function; at the same time, the wave impact force judgment model uses the Adam optimizer to update parameters.

5. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 2 is characterized in that: In step (5), the vibration signal processing unit (4) performs pinch-out, linear trend removal and filtering preprocessing on the collected optical fiber vibration amplitude signal.

6. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 2 is characterized in that: In step (6), the fully connected layer forms a comprehensive feature vector using the attention-weighted features to predict the impact force of the waves in the time window data, and outputs a quantified impact force value through a set threshold or calculation model.

7. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 2 is characterized in that: In step (6), the length of the time window data is half of the wave period.

8. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 2 is characterized in that: In step (6), each time window data overlaps with the subsequent time window data by half of its time length.

9. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 2 is characterized in that: In step (6), each time window data is standardized before being input into the wave impact force judgment model.

10. The distributed acoustic wave sensing monitoring method of ocean wave impact force according to claim 1, characterized in that: In step (6), the fully connected layer forms a comprehensive feature vector with the attention-weighted features, and maps it into a judgment value using linear and nonlinear transformations, where the judgment value is the impact force of the waves.

Images