A neural network training method for underground coal mine fire control based on vibration data

Abstract

The present application discloses a training method for a neural network of underground coal mine fire extinguishing control based on vibration data, which comprises: obtaining a training vibration signal, wherein the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment; converting the training vibration signal into a training vibration vector; encoding the training vibration vector using a deep neural network as an encoder to obtain a training vibration feature vector; constructing a Gaussian distribution based on the training vibration feature vector and randomly extracting a plurality of random feature vectors from the Gaussian distribution; calculating a distance function between the training vibration feature vector and the plurality of random feature vectors; obtaining a plurality of constructed vibration vectors by back propagation of gradients by minimizing the distance function; and training the deep neural network using the constructed vibration vector and the training vibration vector as an enhanced training data set.

Description

Technical Field

[0001] The present invention relates to the field of deep learning and neural network technology, and more specifically, to a neural network training method for underground coal mine fire extinguishing control based on vibration data, an underground coal mine fire extinguishing control method based on a deep neural network, a neural network training system for underground coal mine fire extinguishing control based on vibration data, an underground coal mine fire extinguishing control system and electronic equipment based on a deep neural network. Background Art

[0002] When the coal seam is far from the surface, people usually choose to dig tunnels underground to mine coal, which is generally called underground coal mining. In the mining process of underground coal mines, since it is difficult to completely extract the gas underground, when the gas accumulates to a certain amount, it will cause an explosion. In the event of a small-scale explosion, if the flames generated by the explosion cannot be effectively extinguished, the spread of the fire is likely to cause subsequent safety accidents. Therefore, automatic fire extinguishing devices are installed in more and more underground coal mines.

[0003] At present, the automatic fire extinguishing devices in underground coal mines usually achieve the purpose of extinguishing fire by spraying flame extinguishing powder outwards. These automatic fire extinguishing devices can be activated when they sense flames or smoke. However, in actual applications, it is expected that the activation time of the automatic fire extinguishing device can be further advanced. Even if the advance amount is only a small amount, it will be of great help to improve safety.

[0004] Therefore, an optimized technical solution for fire extinguishing control in underground coal mines is expected.

[0005] At present, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also shown a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields.

[0006] The development of deep learning and neural networks has provided new solutions and plans for fire extinguishing control in underground coal mines. Summary of the invention

[0007] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a neural network training method for underground coal mine fire extinguishing control based on vibration data, an underground coal mine fire extinguishing control method based on a deep neural network, a neural network training system for underground coal mine fire extinguishing control based on vibration data, an underground coal mine fire extinguishing control system and electronic equipment based on a deep neural network, which detect underground gas explosions based on the corresponding vibration of the mine wall. In particular, in the training process of the neural network for underground coal mine fire extinguishing control, the obtained training feature vectors are enhanced in the feature space and the enhanced sample data are obtained through the back propagation of the deep neural network, and then the model is trained in a way that the enhanced sample data and the original training feature vectors are combined, thereby enhancing the robustness of the model.

[0008] According to one aspect of the present application, a method for training a neural network for underground coal mine fire extinguishing control based on vibration data is provided, which comprises:

[0009] Acquire a training vibration signal, wherein the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment;

[0010] converting the training vibration signal into a training vibration vector;

[0011] encoding the training vibration vector using a deep neural network as an encoder to obtain a training vibration feature vector;

[0012] constructing a Gaussian distribution based on the training vibration feature vector and randomly extracting a plurality of random feature vectors from the Gaussian distribution;

[0013] calculating a distance function between the training vibration feature vector and the plurality of random feature vectors;

[0014] Obtaining a plurality of constructed vibration vectors by back propagation of gradients by minimizing the distance function;

[0015] The deep neural network is trained using the constructed vibration vector and the training vibration vector as enhanced training data sets.

[0016] In the above-mentioned neural network training method for underground coal mine fire extinguishing control based on vibration data, the training vibration signal is converted into a training vibration vector, including: converting the training vibration signal into a training vibration vector by vibration vectorization.

[0017] In the above-mentioned neural network training method for underground coal mine fire extinguishing control based on vibration data, calculating the distance function between the training vibration feature vector and the multiple random feature vectors includes: calculating the L1 distance function between the training vibration feature vector and the multiple random feature vectors.

[0018] In the above-mentioned neural network training method for underground coal mine fire extinguishing control based on vibration data, calculating the distance function between the training vibration feature vector and the multiple random feature vectors includes: calculating the L2 distance function between the training vibration feature vector and the multiple random feature vectors.

[0019] In the above-mentioned neural network training method for underground coal mine fire extinguishing control based on vibration data, the constructed vibration vector and the training vibration vector are used as enhanced training data sets to train the deep neural network, including: passing the constructed vibration vector and the training vibration vector through the deep neural network to obtain an enhanced vibration feature vector; passing the enhanced feature vector through a classifier to obtain a classification loss function; and updating the classifier based on the classification loss function.

[0020] In the above-mentioned neural network training method for underground coal mine fire extinguishing control based on vibration data, the enhanced feature vector is passed through a classifier to obtain a classification loss function, including: passing the enhanced feature vector through a fully connected layer to obtain a classification feature vector; and inputting the classification feature vector into a Softmax classification function to obtain the classification loss function.

[0021] According to another aspect of the present application, a method for controlling underground coal mine fire extinguishing based on a deep neural network is provided, which includes:

[0022] Acquire a vibration signal of a mine wall to be detected;

[0023] Inputting the vibration signal into a deep neural network and a Softmax function trained according to the neural network training method for underground coal mine fire extinguishing control based on vibration data as described above, the output of the Softmax function is a first probability representing turning on the automatic fire extinguishing device and a second probability representing not turning on the automatic fire extinguishing device; and

[0024] Whether to activate an automatic fire extinguishing device is determined based on the first probability and the second probability.

[0025] According to another aspect of the present application, a neural network training system for underground coal mine fire extinguishing control based on vibration data is provided, which includes:

[0026] A training vibration signal acquisition unit, used to acquire a training vibration signal, wherein the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment;

[0027] a training vibration signal conversion unit, configured to convert the training vibration signal obtained by the training vibration signal acquisition unit into a training vibration vector;

[0028] a training vibration feature vector generating unit, configured to encode the training vibration vector obtained by the training vibration signal converting unit by using a deep neural network as an encoder to obtain a training vibration feature vector;

[0029] a random feature vector generating unit, configured to construct a Gaussian distribution based on the training vibration feature vector obtained by the training vibration feature vector generating unit and randomly extract a plurality of random feature vectors from the Gaussian distribution;

[0030] a distance function calculation unit, configured to calculate a distance function between the training vibration feature vector obtained by the training vibration feature vector generation unit and the plurality of random feature vectors obtained by the random feature vector generation unit;

[0031] a construction vibration vector generating unit, configured to obtain a plurality of construction vibration vectors by back propagation of gradients by minimizing the distance function obtained by the distance function calculating unit; and

[0032] A deep neural network training unit is used to train the deep neural network by using the construction vibration vector obtained by the construction vibration vector generating unit and the training vibration vector obtained by the training vibration signal converting unit as an enhanced training data set.

[0033] In the above-mentioned neural network training system for underground coal mine fire extinguishing control based on vibration data, the training vibration signal conversion unit is further used to: convert the training vibration signal into a training vibration vector through vibration vectorization.

[0034] In the above-mentioned neural network training method for underground coal mine fire extinguishing control based on vibration data, the distance function calculation unit is further used to: calculate the L1 distance function between the training vibration feature vector and the multiple random feature vectors.

[0035] In the above-mentioned neural network training system for underground coal mine fire extinguishing control based on vibration data, the distance function calculation unit is further used to: calculate the L2 distance function between the training vibration feature vector and the multiple random feature vectors.

[0036] In the above-mentioned neural network training system for underground coal mine fire extinguishing control based on vibration data, the deep neural network training unit includes: an enhanced vibration feature vector generation subunit, which is used to pass the constructed vibration vector and the training vibration vector through a deep neural network to obtain an enhanced vibration feature vector; a classification loss function generation subunit, which is used to pass the enhanced feature vector obtained by the enhanced vibration feature vector generation subunit through a classifier to obtain a classification loss function; and a classifier updating subunit, which is used to update the classifier based on the classification loss function obtained by the classification loss function generation subunit.

[0037] In the above-mentioned neural network training system for underground coal mine fire extinguishing control based on vibration data, the classification loss function generating subunit is further used to: pass the enhanced feature vector through a fully connected layer to obtain a classification feature vector; and input the classification feature vector into a Softmax classification function to obtain the classification loss function.

[0038] According to another aspect of the present application, a deep neural network-based underground coal mine fire extinguishing control system is provided, which includes:

[0039] A vibration signal acquisition unit, used to acquire a vibration signal of a mine wall to be detected;

[0040] a classification unit, configured to input the vibration signal obtained by the vibration signal acquisition unit into a deep neural network and a Softmax function trained according to the neural network training method for underground coal mine fire extinguishing control based on vibration data as described above, wherein the output of the Softmax function is a first probability representing turning on the automatic fire extinguishing device and a second probability representing not turning on the automatic fire extinguishing device; and

[0041] A control result generating unit is used to determine whether to start the automatic fire extinguishing device based on the first probability and the second probability obtained by the classification unit.

[0042] According to another aspect of the present application, an electronic device is provided, comprising: a processor; and a memory, in which computer program instructions are stored, and when the computer program instructions are executed by the processor, the processor executes the neural network training method for underground coal mine fire extinguishing control based on vibration data as described above, or the underground coal mine fire extinguishing control method based on deep neural network.

[0043] According to yet another aspect of the present application, a computer-readable medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the neural network training method for underground coal mine fire extinguishing control based on vibration data as described above, or the underground coal mine fire extinguishing control method based on a deep neural network.

[0044] Compared with the prior art, the present application provides a neural network training method for underground coal mine fire extinguishing control based on vibration data, an underground coal mine fire extinguishing control method based on a deep neural network, a neural network training system for underground coal mine fire extinguishing control based on vibration data, an underground coal mine fire extinguishing control system and electronic equipment based on a deep neural network, which detects underground gas explosions based on the corresponding vibration of the mine wall. In particular, in the training process of the neural network for underground coal mine fire extinguishing control, the obtained training feature vectors are enhanced in the feature space and the enhanced sample data are obtained through the back propagation of the deep neural network, and then the model is trained in a way that the enhanced sample data and the original training feature vectors are combined, thereby enhancing the robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0046] Figure 1 The diagram illustrates an application scenario diagram of a neural network training method for underground coal mine fire extinguishing control based on vibration data and an underground coal mine fire extinguishing control method based on a deep neural network according to an embodiment of the present application;

[0047] Figure 2 A flowchart of a neural network training method for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application is illustrated;

[0048] Figure 3 A schematic diagram of the system architecture of a neural network training method for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application is illustrated;

[0049] Figure 4 A flowchart of a neural network training method for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application is illustrated, in which the constructed vibration vector and the training vibration vector are used as enhanced training data sets to train the deep neural network;

[0050] Figure 5 A flowchart of a method for controlling underground coal mine fire extinguishing based on a deep neural network according to an embodiment of the present application is illustrated;

[0051] Figure 6 A block diagram of a neural network training system for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application is illustrated.

[0052] Figure 7 The diagram illustrates a block diagram of a deep neural network training unit in a neural network training system for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application.

[0053] Figure 8 The figure illustrates a block diagram of an underground coal mine fire extinguishing control system based on a deep neural network according to an embodiment of the present application.

[0054] Fig. 9 A block diagram of an electronic device according to an embodiment of the present application is illustrated. DETAILED DESCRIPTION

[0055] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0056] Scenario Overview

[0057] As mentioned above, more and more underground coal mines are equipped with automatic fire extinguishing devices, which usually achieve the purpose of extinguishing fire by spraying flame extinguishing powder. These automatic fire extinguishing devices can be activated when they sense flames or smoke. However, in actual applications, it is expected that the activation time of the automatic fire extinguishing device can be further advanced. Even if the advance is only a small amount, it will be of great help to improve safety.

[0058] Therefore, the inventor of the present application considers detecting underground gas explosions by using vibration data obtained by a vibration sensor, because once an underground gas explosion occurs, the corresponding vibration of the mine wall can be easily detected, and even if the explosion location is a certain distance from the sensor, such vibration can be detected immediately. In this way, the safety can be improved by activating the automatic fire extinguishing device on the flame propagation path in advance.

[0059] However, in practice, since most vibration data are obtained through simulation in an experimental environment, there may be differences between them and the real data in the field environment. Moreover, this difference is directly related to the complex factors in the geological conditions of the area where the mine is located, and it is difficult to make a qualitative analysis. As a result, when deep neural networks and classifiers are used to classify the collected vibration data, the model is not robust enough.

[0060] The inventor of this application also found that if the traditional data enhancement method is used, that is, the effect is not good by enhancing the collected vibration data. The applicant found that this is because when mapping the sample data from the sample data to the feature space, the randomness added to the sample data cannot be well retained after mapping to the feature space, and thus cannot play a significant role in improving the robustness of the model.

[0061] In response to the above problems, the inventors of the present application adopt a method of enhancing the obtained feature vector in the feature space, and further obtain enhanced sample data through back propagation of a deep neural network, and train the model in a manner of combining the enhanced sample data and the feature vector, thereby enhancing the robustness of the model. Specifically, for each vibration data used for training, a feature vector is first input into a deep neural network, a Gaussian distribution is constructed based on the feature vector, and multiple random feature vectors are randomly extracted from the Gaussian distribution, and multiple updated vibration data are obtained through back propagation of the gradient by minimizing the distance between the feature vector and the multiple reference feature vectors, and the model is trained with the multiple updated vibration data and the original vibration data.

[0062] Based on this, the present application proposes a training method for a neural network of underground coal mine fire extinguishing control based on vibration data, which includes: converting the training vibration signal into a training vibration vector; encoding the training vibration vector using a deep neural network as an encoder to obtain a training vibration feature vector; constructing a Gaussian distribution based on the training vibration feature vector and randomly extracting multiple random feature vectors from the Gaussian distribution; calculating the distance function between the training vibration feature vector and the multiple random feature vectors; obtaining multiple constructed vibration vectors by back propagation of gradients by minimizing the distance function; and training the deep neural network using the constructed vibration vector and the training vibration vector as an enhanced training data set.

[0063] Based on this, the present application also proposes a method for underground coal mine fire extinguishing control based on a deep neural network, which includes: obtaining a vibration signal of the mine wall to be detected; inputting the vibration signal into a deep neural network and a Softmax function trained according to the training method of the neural network for underground coal mine fire extinguishing control based on vibration data as described above, the output of the Softmax function being a first probability indicating that the vibration signal is caused by an explosion and a second probability that the vibration signal is not caused by an explosion; and, based on the first probability and the second probability, determining a detection result of whether the vibration signal is caused by an explosion.

[0064] Figure 1The diagram illustrates an application scenario diagram of a neural network training method for underground coal mine fire extinguishing control based on vibration data and an underground coal mine fire extinguishing control method based on a deep neural network according to an embodiment of the present application.

[0065] like Figure 1 As shown, in the training phase of this application scenario, a vibration sensor (e.g., Figure 1 V) as shown in the figure collects vibration signals of simulated underground coal mine explosions as training vibration signals, and then inputs the training vibration signals into a server that is deployed with a training algorithm of a neural network for underground coal mine fire extinguishing control based on vibration data (for example, as Figure 1 S) as shown in , wherein the server is capable of training the neural network of the underground coal mine fire extinguishing control based on vibration data with the training vibration signal.

[0066] After the neural network is trained by the neural network training algorithm for underground coal mine fire extinguishing control based on vibration data as described above, whether the underground coal mine fire extinguishing control is timely can be detected based on the deep neural network.

[0067] Furthermore, if Figure 1 As shown, in the detection phase of this application scenario, a vibration sensor (e.g., Figure 1 V) as shown in the figure obtains the vibration signal of the mine wall to be detected; then, the vibration signal is input into a server deployed with a detection algorithm for underground coal mine fire extinguishing control based on a deep neural network (for example, as Figure 1 S) as shown in the figure, wherein the server is capable of processing the vibration signal based on a detection algorithm for underground coal mine fire extinguishing control based on a deep neural network to generate a detection result of whether the vibration signal is caused by an explosion.

[0068] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0069] Exemplary Methods

[0070] Figure 2 The figure shows a flow chart of a neural network training method for underground coal mine fire extinguishing control based on vibration data. Figure 2As shown, according to an embodiment of the present application, a neural network training method for underground coal mine fire extinguishing control based on vibration data includes: S110, obtaining a training vibration signal, wherein the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment; S120, converting the training vibration signal into a training vibration vector; S130, encoding the training vibration vector using a deep neural network as an encoder to obtain a training vibration feature vector; S140, constructing a Gaussian distribution based on the training vibration feature vector and randomly extracting a plurality of random feature vectors from the Gaussian distribution; S150, calculating a distance function between the training vibration feature vector and the plurality of random feature vectors; S160, obtaining a plurality of constructed vibration vectors by back propagation of gradients by minimizing the distance function; and, S170, training the deep neural network using the constructed vibration vector and the training vibration vector as enhanced training data sets.

[0071] Figure 3 The figure shows a schematic diagram of the architecture of a neural network training method for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application. Figure 3 As shown, in the network architecture of the training method, first, the training vibration vector (for example, Figure 3 Vt as shown in FIG) is input into a deep neural network (e.g., Figure 3 ) to obtain a training vibration feature vector (e.g., Figure 3 V1 as shown in the figure); then, constructing a Gaussian distribution based on the training vibration feature vector and randomly extracting a plurality of random feature vectors from the Gaussian distribution (for example, Figure 3 Then, the distance function between the training vibration feature vector and the plurality of random feature vectors is calculated (for example, Figure 3 Then, by minimizing (e.g., Figure 3 The distance function is minimized as shown in FIG. 1 to obtain multiple constructed vibration vectors (for example, as Figure 3 Vc1 to Vck as shown in FIG); then, the deep neural network is trained using the constructed vibration vector and the training vibration vector as enhanced training data sets.

[0072] In step S110, a training vibration signal is obtained, and the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment. Specifically, the required training vibration signal can be collected by a vibration sensor. As mentioned above, in the technical solution of the present application, the vibration data obtained by the vibration sensor is used to detect underground gas explosions. It should be understood that once an underground gas explosion occurs, the corresponding vibration of the mine wall can be easily detected, and even if the explosion location is a certain distance from the sensor, this vibration can be detected instantly. In this way, safety can be improved by starting the automatic fire extinguishing device on the flame propagation path in advance.

[0073] In a specific example of the present application, the vibration sensor may be installed in a tunnel excavated in an underground coal mine, and the vibration signal may be collected by the vibration sensor.

[0074] In step S120, the training vibration signal is converted into a training vibration vector. Specifically, in an example of the present application, the process of converting the training vibration signal into a training vibration signal includes: converting the training vibration signal into a training vibration vector by vibration vectorization. A person of ordinary skill in the art should know that vector quantization (VQ) is an extremely important lossy data compression method based on block coding rules, which is widely used in the fields of speech coding, speech recognition and speech synthesis.

[0075] It should be understood that by converting the training vibration signal into a training vibration vector, the training vibration signal is converted into a data format that can be easily calculated by a computer, that is, a training vibration vector. Specifically, a plurality of vibration scalar data groups are formed into a vector, and then the vector is quantized as a whole in the vector space, thereby compressing the training vibration signal data without losing too much information.

[0076] In step S130, the training vibration vector is encoded using a deep neural network as an encoder to obtain a training vibration feature vector. That is, the high-dimensional features in the training vibration vector are extracted using a deep neural network. It should be understood that the training vibration vector is mapped into a feature space by encoding using a deep neural network as an encoder to obtain a training vibration feature vector.

[0077] Specifically, in the embodiment of the present application, the deep neural network is a multilayer perceptron. Those skilled in the art should know that a multilayer perceptron is a deep neural network model that can fully utilize the information of each position in the audio vector and the association information between each position to extract high-dimensional features that can express audio data. The multilayer perceptron model is a feedforward artificial neural network model that can have multiple hidden layers in addition to the input and output layers, which maps multiple input data sets to a single output data set.

[0078] In step S140, a Gaussian distribution is constructed based on the training vibration feature vector and a plurality of random feature vectors are randomly extracted from the Gaussian distribution. As mentioned above, in practice, since most of the vibration data are obtained through simulation in an experimental environment, they may differ from the real data in the field environment, and this difference is directly related to the complex factors in the geological conditions of the area where the mine is located, and it is difficult to qualitatively analyze. This results in insufficient robustness of the model when the deep neural network and classifier are used to classify the collected vibration data.

[0079] Data enhancement is a simple and direct method to improve model performance. However, if the traditional data enhancement method is used, that is, the effect is not good by enhancing the collected vibration data. The inventor of the present application found that this is because when the sample data is mapped from the sample data to the feature space, the randomness added to the sample data cannot be well preserved after being mapped to the feature space, and thus cannot play a significant role in improving the robustness of the model. Therefore, in the embodiment of the present application, a method of enhancing the obtained feature vector in the feature space is adopted.

[0080] That is, the training vibration feature vector is used to construct a Gaussian distribution, and multiple random feature vectors are randomly extracted from the Gaussian distribution. It should be understood that the multiple random feature vectors randomly extracted from the Gaussian distribution constructed from the training vibration feature vector are correlated with the training vibration feature vector, and the multiple random feature vectors randomly extracted from the Gaussian distribution are enhanced training data. Among them, Gaussian distribution is a very important probability distribution in mathematics, physics and engineering.

[0081] In step S150, a distance function between the training vibration feature vector and the plurality of random feature vectors is calculated, that is, the difference between the training vibration feature vector and the random feature vector in a specific space is calculated.

[0082] Specifically, in the embodiment of the present application, the process of calculating the distance function between the training vibration feature vector and the multiple random feature vectors includes: calculating the L1 distance function between the training vibration feature vector and the multiple random feature vectors. A person skilled in the art should understand that the L1 distance function, also known as the minimum absolute deviation (LAD), is the sum of the absolute differences between the target value and the estimated value.

[0083] By calculating the L1 distance between the training vibration feature vector and each position of the multiple random feature vectors, the feature difference between the training vibration feature vector and each position of the multiple random feature vectors can be reflected in the numerical dimension, thereby excluding the part with too large difference in the numerical dimension in constructing the vibration vector.

[0084] It is worth mentioning that in other examples of the present application, the distance function between the training vibration feature vector and the multiple random feature vectors can also be calculated in other ways. For example, in another example of the present application, the process of calculating the distance function between the training vibration feature vector and the multiple random feature vectors includes: calculating the L2 distance function between the training vibration feature vector and the multiple random feature vectors. Among them, the L2 distance function, also known as the least square error (LSE), is the sum of the squares of the difference between the target value and the estimated value, also called the Euclidean distance. It should be understood that compared with the L1 distance function, the L2 distance function has a stable solution.

[0085] In the other example mentioned above, by calculating the L2 distance between each position of the partial depth feature map and the global feature map, the feature difference between each position of the partial depth feature map and the global depth feature map can be reflected in the spatial distance dimension, thereby excluding the part with excessively large difference in the spatial distance dimension when constructing the vibration vector.

[0086] In step S160, multiple constructed vibration vectors are obtained by back propagation of gradients by minimizing the distance function. That is, multiple updated vibration data, i.e., constructed vibration vectors, are obtained by back propagation of gradients by minimizing the distance between the training vibration feature vector and the multiple random feature vectors. It should be understood that the purpose of obtaining multiple constructed vibration vectors by back propagation of gradients by minimizing the distance function is to strengthen the correlation between the enhanced samples and the original samples.

[0087] In step S170, the deep neural network is trained using the constructed vibration vector and the training vibration vector as enhanced training data sets. It should be understood that training the deep neural network using the constructed vibration vector and the training vibration vector can improve the deep neural network to extract high-dimensional features for promoting data classification.

[0088] Specifically, in an embodiment of the present application, the process of training the deep neural network with the constructed vibration vector and the training vibration vector includes: first, passing the constructed vibration vector and the training vibration vector through the deep neural network to obtain an enhanced vibration feature vector, that is, extracting high-dimensional features of the constructed vibration vector and the training vibration vector through the deep neural network; then, passing the enhanced feature vector through a classifier (such as a Softmax classification function) to obtain a classification loss function; and then, updating the classifier based on the classification loss function.

[0089] Figure 4 The figure shows a flowchart of a neural network training method for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application, in which the constructed vibration vector and the training vibration vector are used as enhanced training data sets to train the deep neural network. Figure 4 As shown, the deep neural network is trained using the constructed vibration vector and the training vibration vector as enhanced training data sets, including: S210, passing the constructed vibration vector and the training vibration vector through the deep neural network to obtain an enhanced vibration feature vector; S220, passing the enhanced feature vector through a classifier to obtain a classification loss function; and, S230, updating the classifier based on the classification loss function.

[0090] According to another aspect of the present application, a method for controlling underground coal mine fire extinguishing based on a deep neural network is also provided.

[0091] Figure 5 The figure shows a flow chart of a method for controlling underground coal mine fire based on a deep neural network according to an embodiment of the present application. Figure 5 As shown, according to the underground coal mine fire extinguishing control method based on deep neural network according to the embodiment of the present application, it includes: S310, obtaining the vibration signal of the mine wall to be detected; S320, inputting the vibration signal into the deep neural network and Softmax function trained according to the neural network training method for underground coal mine fire extinguishing control based on vibration data as described above, the output of the Softmax function is a first probability representing the turning on of the automatic fire extinguishing device and a second probability representing the turning on of the automatic fire extinguishing device; and, S330, determining whether to turn on the automatic fire extinguishing device based on the first probability and the second probability.

[0092] In summary, the neural network training method for underground coal mine fire extinguishing control based on vibration data and the underground coal mine fire extinguishing control method based on deep neural network of the embodiment of the present application are explained, which detects underground gas explosions based on the corresponding vibration of the mine wall. In particular, in the training process of the neural network for underground coal mine fire extinguishing control, the obtained training feature vector is enhanced in the feature space and the enhanced sample data is obtained through back propagation of the deep neural network, and then the model is trained in a way of combining the enhanced sample data with the original training feature vector, so as to enhance the robustness of the model.

[0093] Exemplary Systems

[0094] Figure 6 A block diagram of a neural network training system for underground coal mine fire extinguishing control based on vibration data according to an embodiment of the present application is illustrated.

[0095] like Figure 6 As shown, according to an embodiment of the present application, a neural network training system 600 for underground coal mine fire extinguishing control based on vibration data includes: a training vibration signal acquisition unit 610, used to acquire a training vibration signal, wherein the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment; a training vibration signal conversion unit 620, used to convert the training vibration signal obtained by the training vibration signal acquisition unit 610 into a training vibration vector; a training vibration feature vector generation unit 630, used to encode the training vibration vector obtained by the training vibration signal conversion unit 620 using a deep neural network as an encoder to obtain a training vibration feature vector; and a random feature vector generation unit 640, used to generate a training vibration feature vector based on the training vibration feature vector obtained by the training vibration feature vector generation unit 630. A Gaussian distribution is constructed and a plurality of random feature vectors are randomly extracted from the Gaussian distribution; a distance function calculation unit 650 is used to calculate the distance function between the training vibration feature vector obtained by the training vibration feature vector generation unit 630 and the plurality of random feature vectors obtained by the random feature vector generation unit 640; a construction vibration vector generation unit 660 is used to obtain a plurality of construction vibration vectors by gradient back propagation by minimizing the distance function obtained by the distance function calculation unit 650; and a deep neural network training unit 670 is used to train the deep neural network using the construction vibration vector obtained by the construction vibration vector generation unit 660 and the training vibration vector obtained by the training vibration signal conversion unit 620 as an enhanced training data set.

[0096] In one example, in the training system 600, the training vibration signal conversion unit 620 is further used to: convert the training vibration signal into a training vibration vector by vibration vectorization.

[0097] In one example, in the training system 600, the distance function calculation unit 650 is further used to: calculate the L1 distance function between the training vibration feature vector and the multiple random feature vectors.

[0098] In one example, in the training system 600, the distance function calculation unit 650 is further used to: calculate the L2 distance function between the training vibration feature vector and the multiple random feature vectors.

[0099] In one example, in the above training system 600, as Figure 7 As shown, the deep neural network training unit 670 includes: an enhanced vibration feature vector generating subunit 671, used to pass the constructed vibration vector and the training vibration vector through a deep neural network to obtain an enhanced vibration feature vector; a classification loss function generating subunit 672, used to pass the enhanced feature vector obtained by the enhanced vibration feature vector generating subunit 671 through a classifier to obtain a classification loss function; and a classifier updating subunit 673, used to update the classifier based on the classification loss function obtained by the classification loss function generating subunit 672.

[0100] In one example, in the above-mentioned training system 600, the classification loss function generating subunit 672 is further used to: pass the enhanced feature vector through a fully connected layer to obtain a classification feature vector; and input the classification feature vector into a Softmax classification function to obtain the classification loss function.

[0101] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the training system 600 have been described in detail above. Figures 1 to 4 The training method of the neural network for underground coal mine fire extinguishing control based on vibration data has been described in detail, and therefore, its repeated description will be omitted.

[0102] As described above, the training system 600 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for underground coal mine fire extinguishing control. In one example, the training system 600 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the training system 600 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the training system 600 can also be one of the many hardware modules of the terminal device.

[0103] Alternatively, in another example, the training system 600 and the terminal device may also be separate devices, and the training system 600 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0104] According to another aspect of the present application, an underground coal mine fire extinguishing control system based on a deep neural network is also provided.

[0105] Figure 8 The figure shows a block diagram of a coal mine fire extinguishing control system based on a deep neural network according to an embodiment of the present application. Figure 8 As shown, according to an underground coal mine fire extinguishing control system 800 based on a deep neural network according to an embodiment of the present application, it includes: a vibration signal acquisition unit 810, used to acquire the vibration signal of the mine wall to be detected; a classification unit 820, used to input the vibration signal obtained by the vibration signal acquisition unit 810 into a deep neural network and a Softmax function trained according to the training method of the neural network for underground coal mine fire extinguishing control based on vibration data as described above, the output of the Softmax function is a first probability representing turning on the automatic fire extinguishing device and a second probability representing not turning on the automatic fire extinguishing device; and a control result generation unit 830, used to determine whether to turn on the automatic fire extinguishing device based on the first probability and the second probability obtained by the classification unit 820.

[0106] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above detection system 800 have been described in detail above. Figure 5 The description of the underground coal mine fire extinguishing control method based on deep neural network has been introduced in detail, and therefore, its repeated description will be omitted.

[0107] As described above, the detection system 800 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for underground coal mine fire extinguishing control. In one example, the detection system 800 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the detection system 800 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the detection system 800 can also be one of the many hardware modules of the terminal device.

[0108] Alternatively, in another example, the detection system 800 and the terminal device may also be separate devices, and the detection system 800 may be connected to the terminal device via a wired and / or wireless network, and transmit the interactive information in accordance with an agreed data format.

[0109] Exemplary Electronic Devices

[0110] Below, reference Fig. 9 To describe an electronic device according to an embodiment of the present application.

[0111] Fig. 9 A block diagram of an electronic device according to an embodiment of the present application is illustrated.

[0112] like Fig. 9 As shown, the electronic device 10 includes one or more processors 11 and a memory 12 .

[0113] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0114] The memory 12 may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the training method of the neural network for underground coal mine fire extinguishing control based on vibration data of each embodiment of the present application described above, or the functions of the underground coal mine fire extinguishing control method based on deep neural network and / or other desired functions. Various contents such as distance function, classification loss function, etc. may also be stored in the computer-readable storage medium.

[0115] In one example, the electronic device 10 may further include: an input system 13 and an output system 14, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0116] The input system 13 may include, for example, a keyboard, a mouse, and the like.

[0117] The output system 14 can output various information to the outside, including underground coal mine fire extinguishing control results, etc. The output system 14 can include, for example, a display, a speaker, a printer, a communication network and its connected remote output device, etc.

[0118] Of course, to simplify, Fig. 9 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 10 may also include any other appropriate components.

[0119] Exemplary computer program products and computer-readable storage media

[0120] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the training method of a neural network for underground coal mine fire extinguishing control based on vibration data according to various embodiments of the present application described in the above "Exemplary Method" section of this specification, or the steps in the functions of the underground coal mine fire extinguishing control method based on a deep neural network.

[0121] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0122] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, causes the processor to execute the training method of a neural network for underground coal mine fire extinguishing control based on vibration data described in the above “Exemplary Method” section of this specification, or the steps in the underground coal mine fire extinguishing control method based on a deep neural network.

[0123] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, system or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0124] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0125] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0126] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0127] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0128] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A neural network training method for underground coal mine fire extinguishing control based on vibration data, characterized in that: include: Acquire a training vibration signal, wherein the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment; converting the training vibration signal into a training vibration vector; encoding the training vibration vector using a deep neural network as an encoder to obtain a training vibration feature vector; constructing a Gaussian distribution based on the training vibration feature vector and randomly extracting a plurality of random feature vectors from the Gaussian distribution; calculating a distance function between the training vibration feature vector and the plurality of random feature vectors; Obtaining a plurality of constructed vibration vectors by back propagation of gradients by minimizing the distance function; Training the deep neural network using the constructed vibration vector and the training vibration vector as enhanced training data sets; The method of training the deep neural network by using the constructed vibration vector and the training vibration vector as enhanced training data sets includes: Passing the constructed vibration vector and the training vibration vector through a deep neural network to obtain an enhanced vibration feature vector; Passing the enhanced vibration feature vector through a classifier to obtain a classification loss function; and Updating the classifier based on the classification loss function; Wherein, passing the enhanced vibration feature vector through a classifier to obtain a classification loss function comprises: Passing the enhanced vibration feature vector through a fully connected layer to obtain a classification feature vector; and Inputting the classification feature vector into a Softmax classification function to obtain the classification loss function; Wherein, calculating the distance function between the training vibration feature vector and the plurality of random feature vectors includes: An L1 distance function or an L2 distance function between the training vibration feature vector and the plurality of random feature vectors is calculated.

2. The neural network training method for underground coal mine fire extinguishing control based on vibration data according to claim 1, wherein: Converting the training vibration signal into a training vibration vector comprises: The training vibration signal is converted into a training vibration vector by vibration vectorization.

3. A method for controlling underground coal mine fire extinguishing based on deep neural network, characterized in that: include: Acquire a vibration signal of a mine wall to be detected; Inputting the vibration signal into a deep neural network and a Softmax function trained by the neural network training method for underground coal mine fire extinguishing control based on vibration data according to any one of claims 1 to 2, wherein the output of the Softmax function is a first probability representing turning on the automatic fire extinguishing device and a second probability representing not turning on the automatic fire extinguishing device; and Whether to activate an automatic fire extinguishing device is determined based on the first probability and the second probability.

4. A neural network training system for underground coal mine fire extinguishing control based on vibration data, characterized in that: include: A training vibration signal acquisition unit, used to acquire a training vibration signal, wherein the training vibration signal is obtained by simulating an underground coal mine explosion in an experimental environment; a training vibration signal conversion unit, configured to convert the training vibration signal obtained by the training vibration signal acquisition unit into a training vibration vector; a training vibration feature vector generating unit, configured to encode the training vibration vector obtained by the training vibration signal converting unit by using a deep neural network as an encoder to obtain a training vibration feature vector; a random feature vector generating unit, configured to construct a Gaussian distribution based on the training vibration feature vector obtained by the training vibration feature vector generating unit and randomly extract a plurality of random feature vectors from the Gaussian distribution; a distance function calculation unit, configured to calculate a distance function between the training vibration feature vector obtained by the training vibration feature vector generation unit and the plurality of random feature vectors obtained by the random feature vector generation unit; a construction vibration vector generating unit, configured to obtain a plurality of construction vibration vectors by back propagation of gradients by minimizing the distance function obtained by the distance function calculating unit; and A deep neural network training unit, configured to train the deep neural network using the construction vibration vector obtained by the construction vibration vector generating unit and the training vibration vector obtained by the training vibration signal converting unit as an enhanced training data set; The deep neural network training unit includes: an enhanced vibration feature vector generation subunit, used to pass the construction vibration vector and the training vibration vector through a deep neural network to obtain an enhanced vibration feature vector; a classification loss function generation subunit, used to pass the enhanced vibration feature vector obtained by the enhanced vibration feature vector generation subunit through a classifier to obtain a classification loss function; and a classifier updating subunit, used to update the classifier based on the classification loss function obtained by the classification loss function generation subunit; The classification loss function generating subunit is used to: pass the enhanced vibration feature vector through a fully connected layer to obtain a classification feature vector; and input the classification feature vector into a Softmax classification function to obtain the classification loss function; Wherein, calculating the distance function between the training vibration feature vector and the plurality of random feature vectors includes: An L1 distance function or an L2 distance function between the training vibration feature vector and the plurality of random feature vectors is calculated.

5. A coal mine fire extinguishing control system based on deep neural network, characterized in that: include: A vibration signal acquisition unit, used to acquire a vibration signal of a mine wall to be detected; a classification unit, configured to input the vibration signal obtained by the vibration signal acquisition unit into a deep neural network and a Softmax function trained by the neural network training method for underground coal mine fire extinguishing control based on vibration data according to any one of claims 1 to 2, wherein the output of the Softmax function is a first probability representing turning on the automatic fire extinguishing device and a second probability representing not turning on the automatic fire extinguishing device; and A control result generating unit is used to determine whether to start the automatic fire extinguishing device based on the first probability and the second probability obtained by the classification unit.

6. An electronic device comprising: processor; as well as A memory, in which computer program instructions are stored, and when the computer program instructions are executed by the processor, the processor executes the neural network training method for underground coal mine fire extinguishing control based on vibration data as described in any one of claims 1-2 or the underground coal mine fire extinguishing control method based on deep neural network as described in claim 3.

Images