DAS data intelligent identification method, device, system and program

Abstract

The invention discloses a DAS data intelligent identification method, device, system and program, and provides a DAS data real-time intelligent identification method in the technical field of distributed optical fiber sensing. An environment optical cable sensing signal is converted into a DAS data sample in real time, the sample is intelligently identified and classified through a specific deep neural network model, and a classification result serves as an early warning basis to help a user to judge and decide external abnormal events. And the engineering application value of the DAS equipment is improved. The embodiment of the invention provides a CNN + Transform deep neural network model architecture. The architecture has the advantages of ideal characteristics of a convolutional neural network (CNN), such as shift, zoom and distortion invariance, and the advantages of a Transform network, such as dynamic attention, a global receptive field and higher generalization performance at the same time. The problems that an existing intelligent recognition model is low in accuracy and poor in generalization are solved. The embodiment of the invention also discloses a DAS intelligent identification system and equipment, and can solve the problems that the current DAS equipment is high in false alarm rate, slow in system response and low in intelligence. And the engineering use experience of the user is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed optical fiber sensing, and specifically to a real-time processing method for DAS data, the implementation of a deep learning model and a DAS intelligent identification and early warning system. Background Art

[0002] Distributed acoustic vibration sensing systems have been applied to perimeter security, oil and gas pipeline safety warning, train operation safety, geological disaster monitoring and other fields due to their advantages such as high sensitivity, electromagnetic interference resistance, long monitoring distance, and low cost.

[0003] However, when most distributed optical fiber acoustic vibration sensing devices (DAS) on the current market are implemented in engineering applications, the environmental data collected at the device end is too large, reaching an operation processing volume of more than 250 M bytes per second, resulting in an excessive computing power burden on the computer. Therefore, when adding a data recognition algorithm function to the device, the system response is slow, increasing the risk of device downtime and not meeting the requirements of real-time monitoring and early warning in engineering applications.

[0004] Regarding the recognition of DAS environmental data, due to the limitations of DAS device performance and the sensing optical fiber itself, there are problems such as large noise and unclear signal characteristics in the obtained environmental data. At the same time, the event sample data belongs to one-dimensional time series data and has a frequency drift problem itself. The data features obtained directly using traditional signal feature extraction algorithms (expert knowledge) cannot truly reflect the essential features of DAS data, resulting in model failure. This greatly reduces the application value of distributed optical fiber acoustic vibration sensing devices.

[0005] There is a currently publicly disclosed intelligent DAS data recognition method. Patent application number CN 114510960 A discloses using a Vison Transfomer network for pattern recognition in a distributed optical fiber sensing system. This method requires a large amount of data preprocessing calculations, and the time-frequency map data input to the model is greatly affected by noise, resulting in inaccurate model training samples. At the same time, performing signal processing in the way of image vision requires extremely high computational overhead. In engineering applications, it will cause problems such as excessive system construction costs, slow judgment response, and low discrimination accuracy. Now, an intelligent DAS data recognition method, device, system, and program are designed. The 1D DAS data is directly imported into the designed cnn + transformer network without converting the data into a time-frequency map, avoiding the problem of data feature loss caused by excessive data conversion. It fully utilizes the advantages of the CNN network's data space feature scaling invariance, the global receptive field of the transformer network, and the dynamic attention mechanism. It enhances the data feature mining ability and improves the model generalization performance. Correspondingly, this design discloses the composition of the device and the DAS system. The environmental DAS data obtained by real-time collection by the device is identified through a computer device and an intelligent model, and the recognition result is used as the decision-making basis in engineering applications, improving the engineering application value and user experience of the DAS system. Summary of the Invention

[0006] In view of this, it is necessary to provide an intelligent DAS data recognition method suitable for engineering applications to solve the problems of low accuracy, poor generalization, and slow system response in current DAS data intelligent recognition. Provide a method for realizing a DAS intelligent early warning system of "data real-time perception + model recognition + early warning", enhancing the engineering application value of DAS devices and improving the engineering application experience of users.

[0007] In the first aspect of the embodiment of the present invention, an intelligent DAS data recognition method is disclosed, including: receiving DAS data, model recognition, and realizing an intelligent early warning system.

[0008] Adopting this technical solution can combine the real-time collection of DAS data and the intelligent model recognition and analysis of the data to realize real-time intelligent judgment of the occurrence of external environmental events for users to use, improving the engineering application value and application experience of DAS devices.

[0009] Preferably, the receiving of DAS data is mainly the acoustic vibration data of each position point sensed along the environmental optical cable uploaded by the device, which should include data time, position point, and optical cable distance information.

[0010] Preferably, the received data further includes data processing. The backward Rayleigh scattering signal reflecting the environmental conditions obtained through an external optical cable is encoded according to information. The encoded information includes the environmental point position, time, and data length. After encoding, the data can be transmitted to the receiving device in real time through the network, saved locally, or directly called by some model functions.

[0011] Adopting this technical solution can improve the migration efficiency and real-time performance of DAS data, so as to enhance the convenience of equipment replacement for users in the implementation of specific engineering projects.

[0012] Preferably, in possible implementation manners, the model recognition method includes a series of steps of model creation and training. Specifically, the following steps should be included:

[0013] S1: Establishment of DAS data samples;

[0014] S2: Normalization of DAS data;

[0015] S3: Creation of a deep learning model;

[0016] S4: Model training and use;

[0017] Preferably, in some implementation manners, in the above S1 step, the receiving device obtains DAS data containing information encoding, which is called the first data. Data parsing is performed to generate DAS data of a standard length, and sample labels are added for use in model creation. A standard DAS data sample set is established, which is called the second data, including a training set, a test set, and a validation set.

[0018] Preferably, in some implementation manners, in the above S2 step, data standardization is performed before model creation. The calculation formula of the processing method is:

[0019]

[0020] In the above formula, x′ k is the standardization of the k samples, μ is the data mean, and σ is the data standard deviation.

[0021] Preferably, in some implementation manners, in the above S3 step, the model creation is to select the CNN+Transformer architecture.

[0022] Preferably, the CNN+Transformer network architecture is to first extract the shallow features of the data through the CNN layer after standardizing the DAS data model, and then use the Transformer-encoder layer to extract the global features of the data. After multiple feature extractions and feature normalizations, the classification results are output by the Dense layer.

[0023] Preferably, the Transformer-encoder layer is characterized in that the multi-head attention mechanism uses the eigenvalues output by the CNN layer as the Q, K, and V vectors, and calculates the scaled dot-product attention of Q, K, and V to obtain the global feature output.

[0024] Preferably, the scaled dot-product attention is characterized in that its calculation formula is:

[0025]

[0026] f out (f score , V) = f score ·V

[0027] Preferably, in some implementation manners, the CNN+Transformer network in the above S3 should include multiple CNN layers and Transformer-encoder layers to achieve multiple extractions of complex data features and enhance the model's data feature extraction ability.

[0028] Adopting this model solution introduces the ideal characteristics of the convolutional neural network (CNN) - shift, scale, and distortion invariance, and at the same time introduces the advantages of the Transformer network - dynamic attention, global receptive field, and stronger generalization performance. It has strong data essential feature mining ability and high intelligent recognition accuracy.

[0029] Preferably, in some implementation manners, for the model training and use in the above S4 step, the model with an identification accuracy greater than 95% on the data test set is saved, and the optimal model is selected as the final intelligent recognition model. The recognition result is the third data.

[0030] Preferably, in possible implementation manners, the intelligent early warning method is to convert the third data into a user-receivable form, such as visualization on the Internet of Things platform, early warning data reports, etc., to help users analyze and judge in a timely manner and take early warning measures.

[0031] Adopting this technical solution converts the external DAS first data into the second data for model training, uses the CNN+Transformer model architecture, and mines the hidden essential features of the second data to improve the recognition ability of DAS data.

[0032] By adopting this technical solution, an intelligent processing data stream of "real-time perception + intelligent judgment" for DAS data is realized. It can realize the real-time analysis and judgment of external environmental events and improve the engineering application value of DAS devices.

[0033] Further, in the second aspect of the embodiments of the present invention, a device for realizing the transmission and calculation of the first data, the second data, and the third data is disclosed. It includes a device for collecting DAS data by the first device and a device for calculating the second device data.

[0034] Preferably, in a possible implementation manner, the first device DAS data acquisition device includes a distributed fiber optic acoustic vibration demodulation optical module, a data acquisition module, a high-speed data calculation module, and an external sensing optical cable.

[0035] Preferably, the above-mentioned distributed fiber optic acoustic vibration demodulation optical module is based on an optical demodulation system. The data acquisition module completes the acquisition of the acoustic vibration data of the environmental optical cable, and the data calculation module processes and generates the first data.

[0036] Through this technical solution, the fiber optic Rayleigh scattering signal caused by the external environmental acoustic vibration is converted into computable acoustic vibration data, and the data is encoded and added with the optical cable position point and time information to generate the first data.

[0037] Preferably, in a possible implementation manner, the second device includes a high-performance CPU and a GPU processor, enabling the conversion and transmission among the first data, the second data, and the third data to be executed.

[0038] Adopting this technical solution, with a processor with strong computing power, big data operations can be processed at high speed, improving the processing speed of converting the first data into the second data and converting the second data into the third data for user use, and enhancing the real-time monitoring and response ability of the system.

[0039] Further, in the third aspect of the embodiments of the present invention, a computer program product is provided to enable the device to execute the processing and conversion among the first data, the second data, and the third data. When the computer program product runs on a computer, the computer is enabled to execute the processing and transmission methods of the first data, the second data, and the third data.

[0040] Further, in the fourth aspect of the embodiments of the present invention, a DAS intelligent early warning system device is disclosed. It has the behavioral functions in the method provided in the first aspect above. Specifically, it includes real-time acquisition of environmental acoustic vibration signals based on the external optical cable at the bottom layer, intelligent identification of the middle layer model, and output and early warning of data analysis results at the application layer. The functions are implemented by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions.

[0041] Adopting this technical solution solves the problem that traditional DAS devices must rely on expert knowledge and experience to judge external events, and has the disadvantage of a high false alarm rate for event early warnings. The real-time identification method of the intelligent identification model is added, improving the engineering application value of the DAS system.

[0042] In summary, compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] 1. The DAS data intelligent recognition method, device, system and program realize the combination of real-time acquisition of DAS data and identification and analysis of the data intelligent model, realize real-time intelligent recognition of the occurrence of external environmental events for users to use, and enhance the engineering application value and application experience of DAS devices.

[0044] 2. The intelligent recognition method adopts the CNN+Transformer neural network architecture. By extracting shallow data features through the CNN layer and deep data extraction through the Transformer layer, the model's ability to extract hidden data features is enhanced, and the recognition ability and generalization of the model are improved.

[0045] 3. The data processing involved in this method does not rely on expert knowledge analysis, which improves the computer processing speed and the accuracy of system early warning.

[0046] It can be understood that the computer program product described in the third aspect and the DAS intelligent early warning system device described in the fourth aspect provided above correspond to the methods in the first aspect or the second aspect. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a schematic diagram of the DAS intelligent early warning system architecture provided by an embodiment of the present application;

[0048] Figure 2 is a schematic diagram of the intelligent recognition method provided by an embodiment of the present application;

[0049] Figure 3 is a schematic diagram of the CNN+Transformer model structure provided by an embodiment of the present application;

[0050] Figure 4 is a schematic diagram of the first data structure provided by an embodiment of the present application;

[0051] Figure 5 is a schematic diagram of the second data structure provided by an embodiment of the present application;

[0052] Figure 6 is a schematic diagram of the third data structure provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] It should be noted that in this application, "at least one" means one or more, and "a plurality" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims, and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0054] For ease of understanding, some explanations of concepts related to the embodiments of this application are exemplarily given for reference.

[0055] Please refer to Figure 1-6 , the present invention provides a technical solution: a DAS data intelligent recognition method, device, system, and program, including real-time acquisition and reception of DAS data. The received data is real-time recognized and classified through a designed CNN+Transformer neural network, and the classification result is used as a warning basis for display on the device side or by engineering personnel.

[0056] Working principle: When using this DAS data intelligent recognition method, device, system, and program, first, through the DAS device, the environmental optical cable is used as a sensor to real-time collect environmental data (DAS data). The data is transmitted to a high-performance computing device through the device for neural network intelligent recognition and classification. The classification result is processed into a certain data format for users to use (such as viewing, data reports, etc.) according to a computer program, and is displayed on the device side to help users make decisions. The engineering application of the DAS intelligent warning system is realized.

[0057] The above is the working process of this DAS data intelligent recognition method, device, system, and program.

[0058] Next, in combination with Figure 1 An exemplary introduction is made to the architecture schematic diagram of the DAS intelligent warning system provided by the embodiments of the present invention.

[0059] The DAS intelligent recognition system includes a first electronic device 01, a second electronic device 02, and a third electronic device 03. The first electronic device is composed of an external environment optical cable and a DAS optical module. The second electronic device refers to high-performance computers, graphics workstations, servers, etc. Its characteristic is that the device has a high-performance CPU, GPU, and a large-capacity storage medium. The third device refers to terminal alarm devices such as terminal displays, mobile phones, and tablets. For the convenience of describing data transmission between multiple devices below, it is specifically assumed that 001 is the signal port of the external optical cable and the DAS optical module in the first device, 100 is the data transmission end of the first device. 101 is the data receiving end of the second device, 200 is the data sending end of the second device. 201 is the data receiving end of the third device.

[0060] In some embodiments, the port 001 of the first device is connected to an external optical cable. The optical cable serves as an environmental acoustic vibration sensor, and the monitoring distance can reach dozens of kilometers. The DAS optical module of the first device is a Ф-OTDR optical system, which converts the backward Rayleigh scattering optical signal in the optical cable into an environmental acoustic vibration signal. The first device injects optical pulses with a certain pulse width into the environmental optical cable at a repetition frequency of 2 - 5KHz per second. Through the Ф-OTDR signal demodulation system, 1-second environmental vibration data at each position point along the optical cable is obtained in the memory of the first device, and the data length is 4 - 10K bytes. Information such as each data position point, time, data length, and data frame header is written at the front end of the data for encoding. After encoding, the first data is generated. The first data is sent from the first device 100 as the data sending end, and the data stored in the first device can be sent to the second device 02 through a wired network or a wireless network.

[0061] In some embodiments, the second device completes data transmission matching with the first device. The 101 port of the second device serves as the data receiving end and continuously receives the first data. The second device first converts the first data into the second data required by the model. The second data is called a data sample, which includes a training set, a test set, and a validation set.

[0062] The following will refer to the schematic Figure 2 Introduce the data processing method in the second device.

[0063] The 101 of the second device serves as the data receiving end of the device, obtains the data transmitted from the first device, and is finally converted into the third data for the next process of the system according to the shown data processing flow.

[0064] Specifically, S1 is to convert the first data format into the second data - the model sample. The sample content includes data and labels.

[0065] The step S2 is to process the second data in the data normalization manner. Adopting this technical method is beneficial to model training, avoids problems such as gradient disappearance and gradient explosion of the model, and enhances the generalization of the model.

[0066] The step S3 is model training. This execution process usually requires the second device to include a GPU device and a large-capacity storage medium to support the efficient completion of model training.

[0067] The step S4 is model saving and using. The specific processing method is to save the model with a test set greater than 95%. The saved model is directly imported into the first data of the receiver 101 of the second device in the next step, and it can convert the first data into the third data for user use in real time.

[0068] In some possible implementation manners, the third device is the visualization terminal of the system. It includes a display terminal and a server. Its function is to display the on-site environmental conditions in real time and issue early warnings for environmental dangerous events. Specifically, as shown in Figure 01, the port 201 of the third device serves as a data receiver to obtain the third data uploaded by the second device 02. The data includes information such as data type, time, and environmental location. The third device 03 displays the third data to the user in real time on the display device through a program, or adds a corresponding early warning function according to user requirements.

[0069] Refer to Figure 3 As shown, the embodiment of the present invention application discloses an intelligent recognition model, which is applied to the second device 02 to realize the conversion of the second data into the third data. It includes a data input layer, a CNN data feature extraction layer, a Transformer-encoder layer, a Dense layer, and an output layer.

[0070] In some embodiments, the samples generated from the second data enter the input layer, and the data samples are normalized, filtered, etc. to enhance the sample features and reduce the influence of data noise on the model. The shown CNN layer is a convolutional neural network with a convolutional kernel of 7, a channel number of 64, and a stride of 2. It completes the shallow extraction and local extraction of signal features.

[0071] The Transformer-encoder layer includes a multi-head attention layer, a normalization layer, a feed-forward neural network layer, and a normalization layer.

[0072] The shown class_token is a vector of all zeros, and the dimension size is equal to the size of the recognition label. In the specific implementation manner, the class_token is added at the beginning of the input sequence as the label for the final classification task.

[0073] In some implementation manners, the Q, K, and V values of the shown Transformer-encoder layer are the feature vectors of the upper CNN layer as the initial values for calculation, used to calculate the single-point attention and generate an attention high-level feature output.

[0074] In some embodiments, the multi-head attention calculation mechanism is based on the SDPA scoring function, and its formula is expressed as:

[0075]

[0076] The multi-head attention output function can be expressed by the formula:

[0077] f out (f score , V) = f score ·V

[0078] In some embodiments, the number of the shown Transformer-encoder layers is 3 to obtain more hierarchical complex data features and achieve the optimal recognition effect.

[0079] In some embodiments, the shown Dense layer is a linear multi-variable neural network to complete the induction of the upper-layer data features.

[0080] In some embodiments, the shown output layer refers to the softmax function to obtain the classification label result with the highest probability.

[0081] The embodiments of the present invention application disclose a possible implementation effect. 5103 second data samples are tested, and the accuracy rate is 97.02%. For the recognition of multi-type acoustic vibration events, the confusion matrix of the model prediction results is shown in the following table:

[0082]

[0083] Referring to Figure 4 , the embodiments of the present invention application provide an encoding method for the first data, which is used for data transmission between the first device 01 and the second device 02 and data migration between data devices.

[0084] Referring to Figure 5 As shown, in some possible embodiments, the embodiments of the present application provide a conversion format for the second data, which is used for model creation and training of the second device 02.

[0085] Referring to Figure 6 As shown, in some possible embodiments, the embodiments of the present application provide a third data presentation format, which is used for the display of the system intelligent recognition results in the third device. It is convenient for users to view the environmental event prediction results in a timely manner.

[0086] This embodiment also provides a computer program product. When the computer program product runs on a computer, it enables the computer to execute the above-related steps to implement the data processing method in the above embodiments.

[0087] The first device 01, the second device 02, and the third device 03 provided in this embodiment constitute a set of DAS data real-time intelligent recognition systems. The computer storage medium, computer program product, or chip therein are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be elaborated here.

[0088] Through the description of the above embodiments, those skilled in the art can clearly understand that for the convenience and brevity of description, only the above division of each functional module is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0089] In several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device device embodiments described above are only illustrative. For example, the division of device units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple device units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point, the coupling or direct coupling or communication connection between the devices shown or discussed can be through some interfaces. The indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0090] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered by the protection scope of this application.

Claims

1. A DAS data intelligent identification method, comprising: Method for receiving DAS data in real time, model identification method, intelligent early warning method. The received DAS data is mainly the acoustic vibration data of each position point obtained by sensing along the environmental optical cable uploaded by the device, which includes data time, optical cable position point, and optical cable distance information.

2. The method of claim 1 further comprising processing data, wherein The backscattered Rayleigh signal reflecting the environmental conditions obtained through the external optical cable is encoded according to the information. The encoded information includes the location of the environmental point, time, and data length. The encoded data can be transmitted to the receiving device in real time through the network. It can be saved locally or directly called by some model functions.

3. The model recognition method as claimed in claim 1 comprises a series of model creation and training steps. It is characterized in that: The following steps should be included: S1: DAS data sample establishment; S2: DAS data normalization; S3: Deep learning model creation; S4: Model training and use.

4. As in claim 3, step S1 is that the receiving device obtains DAS data containing information coding, which is called first data. Data parsing is performed to generate DAS data of standard length, and sample labels are added for model creation. A standard DAS data sample set is established, which is called second data, including a training set, a test set, and a validation set.

5. As described in claim 3, step S2 is data standardization before model creation. The calculation formula of the processing method is: In the above formula, x′ k is the standardization of k samples, μ is the data mean, and σ is the data standard deviation.

6. As claimed in claim 3, the model creation in step S3 adopts the CNN+Transformer architecture.

7. The CNN+Transformer network architecture as described in claim 6 is to standardize the DAS data model, first extract the shallow features of the data through the CNN layer, and then use the Transformer-encoder layer to extract the global features of the data. After multiple feature extraction and feature normalization, the classification result is output by the Dense layer.

8. The Transformer-encoder layer as claimed in claim 7 is characterized in that The multi-head attention mechanism uses the feature values ​​of the CNN layer output as vectors Q, K, and V. The scaled dot product attention of Q, K, and V is calculated to obtain the global feature output.

9. The CNN+Transformer network as described in claims 6, 7, and 8 should include multiple CNN layers and Transformer-encoder layers to achieve multiple extraction of complex data features and enhance the model data feature extraction capability.

10. The model training and use in step S4 of claim 3 is to save the model with a recognition accuracy greater than 95% on the data test set, and select the best model as the final intelligent recognition model. The recognition result is the third data.

11. The intelligent early warning method as claimed in claim 1 is to convert the third data into a user-receivable form, such as visualization of the Internet of Things platform, early warning data report, etc., to help users make timely analysis and judgment and take early warning measures. It should include visualization equipment, computing equipment and program products. To implement the early warning function.

12. The second aspect of the embodiment of the present invention discloses a device for transmitting and calculating first data, second data to third data, including a first device for collecting DAS data and a second device for data calculation.

13. As described in claim 12, the first device DAS data acquisition device includes a distributed optical fiber acoustic vibration demodulation optical module, a data acquisition module, a high-speed data calculation module, and an external sensing optical cable.

14. As described in claim 12, the above-mentioned distributed optical fiber acoustic vibration demodulation optical module is based on The optical demodulation system collects the acoustic vibration data of the ambient optical cable by the data acquisition module, and the data calculation module processes and generates the first data.

15. The second device as claimed in claim 12 comprises a high-performance CPU and GPU processor, so that the conversion and transmission between the first data, the second data and the third data can be performed.

16. As described in claim 12, the embodiment of the present invention publishes the data formats of the first data, the second data, and the third data.

17. A third aspect of the embodiments of the present invention provides a computer program product, which enables the device to perform processing and conversion between the first data, the second data, and the third data. When the computer program product is run on a computer, the computer executes a method for processing and transmitting the first data, the second data, and the third data.

18. The fourth aspect of the embodiment of the present invention discloses a DAS intelligent early warning system device, which has the behavioral functions in the method provided in the first aspect above, specifically including the real-time collection function of the ambient sound and vibration signals based on the external optical cable at the bottom layer, the model intelligent recognition function, and the output and early warning of the data analysis results at the application layer. The functions are implemented by hardware executing the corresponding software. The hardware or software includes one or more modules corresponding to the above functions.

Images