An electromagnetic environment monitoring system and method based on deep learning

Abstract

The present invention relates to the technical field of electromagnetic environment monitoring, and discloses an electromagnetic environment monitoring system and method based on deep learning. A data acquisition module is used to divide a target area into M sub-areas, and within a first preset time period T days, electromagnetic wave data of the M sub-areas are collected t times at fixed intervals every day; a data processing module is used to preprocess the collected electromagnetic wave data and construct M feature matrices; a graph network module is used to construct a graph network for the electromagnetic wave data of the M sub-areas of the target area, and the graph network structure includes: nodes, initial features of nodes and edges between nodes; through the electromagnetic environment monitoring system based on deep learning, when there is no interference signal in the target area most of the time, abnormal signals are detected preferentially, thereby significantly reducing the consumption of computing resources, solving the problems of resource waste and low monitoring efficiency in the prior art, and enhancing the system's ability to cope with potential electromagnetic threats.

Description

Technical Field

[0001] The present invention relates to the field of electromagnetic environment monitoring, and more specifically, to an electromagnetic environment monitoring system and method based on deep learning. Background Art

[0002] Electromagnetic waves are a form of energy propagation generated by the interaction of electric and magnetic fields, and are widely present in nature and human activities. From natural lightning and cosmic radiation to human-invented radio broadcasting, television, mobile communications and radar, electromagnetic waves play a vital role in information transmission and environmental detection. With the rapid development of wireless communication technology, the electromagnetic environment has become increasingly complex and changeable, which also brings potential interference and safety risks.

[0003] However, in actual applications, the existing technology requires a large amount of computing resources to continuously monitor and identify electromagnetic wave data in the target area. For the cost-prioritized civilian monitoring field, the high computing resource requirements make the system difficult to implement. At the same time, most of the time there is no interference signal in the target area, and most of the invested computing resources do not bring benefits. The investment of computing resources is not proportional to the actual benefits, which leads to waste of resources and low efficiency. Summary of the invention

[0004] The present invention provides an electromagnetic environment monitoring system and method based on deep learning to solve the technical problems raised in the background technology.

[0005] The present invention provides an electromagnetic environment monitoring system and method based on deep learning, comprising:

[0006] Data collection module: It is used to divide the target area into M sub-areas, and collect electromagnetic wave data from the M sub-areas t times at a fixed interval every day within a first preset time period T days;

[0007] Electromagnetic wave data includes: frequency, amplitude, phase, polarization, signal-to-noise ratio, location coordinates and timestamp;

[0008] Data processing module: It is used to pre-process the collected electromagnetic wave data and construct M feature matrices. The j-th column and i-th row of the k-th feature matrix represent the electromagnetic wave data collected at the i-th time on the j-th day of the k-th sub-area, where 0≤k≤M, 0≤j≤T, and 0≤i≤t;

[0009] Graph network module: It is used to construct a graph network for the electromagnetic wave data of M sub-areas of the target area. The graph network structure includes: nodes, initial features of nodes and edges between nodes;

[0010] Anomaly detection module: It is used to input the M graph structure data of the first preset time period T days into the anomaly detection model respectively, and the output of the anomaly detection model represents the abnormal value of the electromagnetic wave data of the M sub-regions of the second preset time period G days in the i-th time period of the j-th day;

[0011] Identification module: It is used to identify abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G days.

[0012] Furthermore, T, M, t, and G are all custom parameters, wherein the first preset time period T days and the second preset time period G days have the same number of days.

[0013] Further, the pre-processing steps include:

[0014] Step S201: applying a mean filter method to remove noise from electromagnetic wave data;

[0015] Step S202: using a bandpass filter to filter out irrelevant frequency band data according to the target frequency band and bandwidth;

[0016] Step S203: Filling in missing values ​​in the electromagnetic wave data by taking the average value of the electromagnetic wave data collected in the previous time period and the electromagnetic wave data collected in the next time period;

[0017] Step S204: normalizing the electromagnetic wave data using the Z-score method.

[0018] Furthermore, T elements are generated at the upper and lower boundaries of the feature matrix, t elements are generated at the left and right boundaries of the feature matrix, and one element is generated at each of the four corners of the feature matrix to construct a (T+2)×(t+2) update matrix X;

[0019] Update the electromagnetic wave data of the qth element in the 1st column and the T+2th column in the matrix X to ;

[0020] Update the electromagnetic wave data of the pth element in the 1st and t+2nd rows of the matrix X to ;

[0021] Update the matrix X , , and The electromagnetic wave data of the elements at the four corners are ;

[0022] Among them, 2≤q≤t+1, 2≤p≤T+1, , and represent the first weight parameter, the second weight parameter and the third weight parameter respectively, , and They represent the first bias term, the second bias term and the third bias term respectively, and ReLU represents the ReLU activation function;

[0023] The nodes in the graph network structure establish a mapping relationship with the elements of the update matrix, and the initial characteristics of the nodes are represented by the electromagnetic wave data of the elements that establish a mapping relationship with them;

[0024] The edge between nodes is constructed in the following ways: constructing edges between nodes with mapping relationships established by adjacent elements;

[0025] Edges are constructed between nodes that establish mapping relationships between diagonally adjacent elements.

[0026] Furthermore, the anomaly detection module includes M anomaly detection models, which respectively input the graph network structure of the M sub-regions and the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days. The anomaly detection formula of the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days is as follows:

[0027] ;

[0028] ;

[0029] ;

[0030] in, represents the abnormal value of the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days, represents the electromagnetic wave data collected at the i-th time of the j-th day of the k-th sub-region in the second preset time period G days, represents the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the first preset time period T days, represents the electromagnetic wave data collected at the i+nth time of the j+mth day of the kth sub-area in the first preset time period T days, Represents the similarity value of adjacent nodes in the network structure of the electromagnetic wave data comparison graph collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days, It represents the similarity value of adjacent nodes in the network structure of the electromagnetic wave data comparison graph collected at the i-th moment of the j-th day in the k-th sub-region in the first preset time period T days, Respectively represent the fourth weight parameter, represents the sub-region abnormal threshold, represents the step activation function, Represents a multi-layer perceptron

[0031] Furthermore, the global anomaly detection step of the electromagnetic wave data collected at the i-th moment of the j-th day in the second preset time period G includes:

[0032] Step S301: Create a three-dimensional diagram of M sub-areas of the target area, and input the abnormal values ​​of the M sub-areas at the i-th moment of the j-th day in the second preset time period G days;

[0033] Step S302: If there are m sub-regions with an outlier value of 1 In the three-dimensional diagram, a straight line structure is formed, and the scores of the m sub-regions are recorded as , Expressed as ,in represents the first reward coefficient;

[0034] Step S303: If n sub-regions have an abnormal value of 1 If the Euclidean distance in the three-dimensional diagram is less than the distance threshold β, a clustering structure is formed, and the scores of the n sub-regions are recorded as , Expressed as ,in represents the first reward coefficient;

[0035] The Euclidean distance calculation formula is: ,in and Represents sub-areas and subregions The three-dimensional coordinates of

[0036] Step S304: If there are z sub-regions with an outlier value of 1 If no clustering structure or straight line structure is formed in the three-dimensional diagram, the score of the z sub-regions is recorded as , Expressed as ,in represents the penalty coefficient;

[0037] Step S305: Calculate the global outlier value, which is expressed as , where the sub-regions that belong to both the clustered structure and the linear structure are scored repeatedly. If the global outlier value is greater than the global outlier threshold, it is considered a global outlier. If the global outlier value is less than the global outlier threshold, it is considered a global normal.

[0038] Furthermore, the node feature updating step of the graph network structure includes:

[0039] Step S401: input the electromagnetic wave data collected from the kth sub-region at the i-th moment on the j-th day of the second time period G into the graph network structure to perform sub-region anomaly detection;

[0040] Step S402: updating the graph network structure according to the sub-region abnormal value judgment output by the abnormality detection module 104;

[0041] Step S404: if the sub-region abnormal value of the k-th sub-region at the i-th moment of the j-th day in the second preset time period G days is 0, then the electromagnetic wave data of the k-th sub-region at the i-th moment of the j-th day in the second preset time period G days and the electromagnetic wave data of the k-th sub-region at the i-th moment of the j-th day in the second preset time period T days are averaged to update the feature of the node in the j-th column and the i-th row in the k-th graph network structure;

[0042] Step S404: if the sub-region abnormal value of the k-th sub-region at the i-th moment on the j-th day in the second preset time period G days is 1, the feature of the node in the j-th column and the i-th row in the k-th graph network structure is not updated.

[0043] Furthermore, the recognition module recognizes the globally abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G, including:

[0044] Step S501, converting the electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G from the time domain to the frequency domain by fast Fourier transform, generating a spectrum diagram, and determining the frequency range of the abnormal signal;

[0045] Step S502, generating a time-frequency diagram by short-time Fourier transform of the electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G, and identifying the frequency mutation and frequency drift characteristics of the signal;

[0046] Step S503, identifying the modulation mode of the abnormal signal according to the result of time-frequency analysis;

[0047] Step S504: determine the location of the signal source by using the arrival time difference method in combination with the spatial locations between the sub-areas.

[0048] Furthermore, according to the recognition result of the global anomaly output by the recognition module, if there is no anomaly, no measures are taken. If there is an anomaly, the measures are to immediately issue an early warning to the relevant departments and mobilize relevant personnel to investigate the source of the abnormal electromagnetic wave data.

[0049] Furthermore, the electromagnetic environment monitoring method based on deep learning of the present invention comprises the following steps:

[0050] Step S601: Divide the target area into M sub-areas, and collect electromagnetic wave data t times from the M sub-areas at a fixed interval every day within a first preset time period T days;

[0051] Step S602: pre-processing the collected electromagnetic wave data and constructing M feature matrices;

[0052] Step S603: constructing a graph network for the electromagnetic wave data of the M sub-regions of the target region, wherein the graph network structure includes: nodes, initial features of the nodes, and edges between the nodes;

[0053] Step S604: inputting the M graph structure data of the first preset time period T days into the anomaly detection model respectively, and the output of the anomaly detection model represents the abnormal value of the electromagnetic wave data of the M sub-regions of the second preset time period G days in the i-th time period of the j-th day;

[0054] Step S605: Identify the abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G.

[0055] The beneficial effects of the present invention are as follows: through the electromagnetic environment monitoring system based on deep learning, when there is no interference signal in the target area most of the time, the consumption of computing resources is significantly reduced by prioritizing the detection of abnormal signals and then conducting in-depth analysis; at the same time, the dynamic update of electromagnetic wave data is realized by utilizing the graph network structure, and the global anomaly detection and signal source positioning technology are combined to effectively improve the accuracy and response speed of anomaly detection, thereby solving the problems of resource waste and low monitoring efficiency in the prior art and enhancing the system's ability to deal with potential electromagnetic threats. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a schematic diagram of an electromagnetic environment monitoring system based on deep learning of the present invention;

[0057] Figure 2 is a flow chart of preprocessing of collected electromagnetic wave data according to the present invention;

[0058] Figure 3 is a flow chart of global abnormality discrimination of a target area of ​​the present invention;

[0059] Figure 4 It is a flowchart of updating the graph network structure of the present invention;

[0060] Figure 5 is the electromagnetic wave data recognition flow chart of the present invention;

[0061] Figure 6 It is a flow chart of an electromagnetic environment monitoring method based on deep learning of the present invention.

[0062] In the figure: 101, data acquisition module; 102, data processing module; 103, graph network module; 104, anomaly detection module; 105, recognition module. DETAILED DESCRIPTION

[0063] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0064] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0065] like Figure 1 to Figure 6 As shown, a deep learning-based electromagnetic environment monitoring system and method include:

[0066] Data collection module 101: used to divide the target area into M sub-areas, and collect electromagnetic wave data t times from the M sub-areas at a fixed interval every day within a first preset time period T days;

[0067] Electromagnetic wave data includes: frequency, amplitude, phase, polarization, signal-to-noise ratio, location coordinates and timestamp;

[0068] Data processing module 102: used to pre-process the collected electromagnetic wave data and construct M feature matrices, where the j-th column and the i-th row of the k-th feature matrix represent the electromagnetic wave data collected at the i-th time on the j-th day of the k-th sub-region, where 0≤k≤M, 0≤j≤T, and 0≤i≤t;

[0069] Graph network module 103: used to construct a graph network for the electromagnetic wave data of the M sub-regions of the target region, wherein the graph network structure includes: nodes, initial features of nodes, and edges between nodes;

[0070] Anomaly detection module 104: used to input the M graph structure data of the first preset time period T days into the anomaly detection model respectively, and the output of the anomaly detection model represents the abnormal value of the electromagnetic wave data of the M sub-regions of the second preset time period G days in the i-th time period of the j-th day;

[0071] The identification module 105 is used to identify the abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G.

[0072] In one embodiment of the present invention, T, M, t, and G are all custom parameters, wherein the first preset time period T days and the second preset time period G days have the same number of days. Preferably, the first preset time period T days is set to all five working days in the first week, t is set to 1 hour, the second preset time period G is set to all five working days in the second week, and M is set according to the number of sub-areas actually divided into the target area.

[0073] In one embodiment of the present invention, the pre-processing step, such as Figure 2 ,include:

[0074] Step S201: applying a mean filter method to remove noise from electromagnetic wave data;

[0075] Step S202: using a bandpass filter to filter out irrelevant frequency band data according to the target frequency band and bandwidth;

[0076] Step S203: Filling in missing values ​​in the electromagnetic wave data by taking the average value of the electromagnetic wave data collected in the previous time period and the electromagnetic wave data collected in the next time period;

[0077] Step S204: normalizing the electromagnetic wave data using the Z-score method.

[0078] In one embodiment of the present invention, the load power data may also be normalized by a minimum-maximum normalization method.

[0079] In one embodiment of the present invention, T elements are generated at the upper and lower boundaries of the feature matrix, t elements are generated at the left and right boundaries of the feature matrix, and one element is generated at each of the four corners of the feature matrix to construct a (T+2)×(t+2) update matrix X;

[0080] Update the electromagnetic wave data of the qth element in the 1st column and the T+2th column in the matrix X to ;

[0081] Update the electromagnetic wave data of the pth element in the 1st and t+2nd rows of the matrix X to ;

[0082] Update the matrix X , , and The electromagnetic wave data of the elements at the four corners are ;

[0083] Among them, 2≤q≤t+1, 2≤p≤T+1, , and represent the first weight parameter, the second weight parameter and the third weight parameter respectively, , and They represent the first bias term, the second bias term and the third bias term respectively, and ReLU represents the ReLU activation function;

[0084] The nodes in the graph network structure establish a mapping relationship with the elements of the update matrix, and the initial characteristics of the nodes are represented by the electromagnetic wave data of the elements that establish a mapping relationship with them;

[0085] The edge between nodes is constructed in the following ways: constructing edges between nodes with mapping relationships established by adjacent elements;

[0086] Edges are constructed between nodes that establish mapping relationships between diagonally adjacent elements.

[0087] It should be noted that, for example, the update matrix X is partially expressed as: ;in and , , and There are edges between them. and , , and There are edges between them.

[0088] It should be noted that, for example and Respectively represent the convolution kernels of size 1×3 and 3×1, represents a convolution kernel of size 3×3, , and The step length is 1.

[0089] In one embodiment of the present invention, the anomaly detection module 104 includes M anomaly detection models, which respectively input the graph network structure of the M sub-regions and the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days. The anomaly detection formula of the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days is as follows:

[0090] ;

[0091] ;

[0092] ;

[0093] in, represents the abnormal value of the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days, represents the electromagnetic wave data collected at the i-th time of the j-th day of the k-th sub-region in the second preset time period G days, represents the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the first preset time period T days, represents the electromagnetic wave data collected at the i+nth time of the j+mth day of the kth sub-area in the first preset time period T days, Represents the similarity value of adjacent nodes in the network structure of the electromagnetic wave data comparison graph collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days, It represents the similarity value of adjacent nodes in the network structure of the electromagnetic wave data comparison graph collected at the i-th moment of the j-th day in the k-th sub-region in the first preset time period T days, Respectively represent the fourth weight parameter, represents the sub-region abnormal threshold, represents the step activation function, Represents a multi-layer perceptron

[0094] In one embodiment of the present invention, the global anomaly detection step of the electromagnetic wave data collected at the i-th moment of the j-th day in the second preset time period G days is as follows: Figure 3 ,include:

[0095] Step S301: Create a three-dimensional diagram of M sub-areas of the target area, and input the abnormal values ​​of the M sub-areas at the i-th moment of the j-th day in the second preset time period G days;

[0096] Step S302: If there are m sub-regions with an outlier value of 1 In the three-dimensional diagram, a straight line structure is formed, and the scores of the m sub-regions are recorded as , Expressed as ,in represents the first reward coefficient;

[0097] Step S303: If n sub-regions have an abnormal value of 1 If the Euclidean distance in the three-dimensional diagram is less than the distance threshold β, a clustering structure is formed, and the scores of the n sub-regions are recorded as , Expressed as ,in represents the first reward coefficient;

[0098] The Euclidean distance calculation formula is: ,in and Represents sub-areas and subregions The three-dimensional coordinates of

[0099] Step S304: If there are z sub-regions with an outlier value of 1 If no clustering structure or straight line structure is formed in the three-dimensional diagram, the score of the z sub-regions is recorded as , Expressed as ,in represents the penalty coefficient;

[0100] Step S305: Calculate the global outlier value, which is expressed as , where the sub-regions that belong to both the clustered structure and the linear structure are scored repeatedly. If the global outlier value is greater than the global outlier threshold, it is considered a global outlier. If the global outlier value is less than the global outlier threshold, it is considered a global normal.

[0101] It should be noted that Matlab software tools are used to establish the three-dimensional diagram of the M sub-regions of the target region and to determine the aggregation structure and the straight line structure in the three-dimensional diagram.

[0102] In one embodiment of the present invention, the node feature updating step of the graph network structure is as follows: Figure 4 ,include:

[0103] Step S401: input the electromagnetic wave data collected from the kth sub-region at the i-th moment on the j-th day of the second time period G into the graph network structure to perform sub-region anomaly detection;

[0104] Step S402: updating the graph network structure according to the sub-region abnormal value judgment output by the abnormality detection module 104;

[0105] Step S404: if the sub-region abnormal value of the k-th sub-region at the i-th moment of the j-th day in the second preset time period G days is 0, then the electromagnetic wave data of the k-th sub-region at the i-th moment of the j-th day in the second preset time period G days and the electromagnetic wave data of the k-th sub-region at the i-th moment of the j-th day in the second preset time period T days are averaged to update the feature of the node in the j-th column and the i-th row in the k-th graph network structure;

[0106] Step S404: if the sub-region abnormal value of the k-th sub-region at the i-th moment on the j-th day in the second preset time period G days is 1, the feature of the node in the j-th column and the i-th row in the k-th graph network structure is not updated.

[0107] In one embodiment of the present invention, the identification module 105 identifies the globally abnormal electromagnetic wave data at the i-th moment of the j-th day in the second preset time period G, such as Figure 5 ,include:

[0108] Step S501, converting the electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G from the time domain to the frequency domain by fast Fourier transform, generating a spectrum diagram, and determining the frequency range of the abnormal signal;

[0109] Step S502, generating a time-frequency diagram by short-time Fourier transform of the electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G, and identifying the frequency mutation and frequency drift characteristics of the signal;

[0110] Step S503, identifying the modulation mode of the abnormal signal according to the result of time-frequency analysis;

[0111] Step S504: determine the location of the signal source by using the arrival time difference method in combination with the spatial locations between the sub-areas.

[0112] It should be noted that the existing solutions for electromagnetic wave data recognition are not unique, and they are all conventional technical means, which will not be elaborated here.

[0113] In one embodiment of the present invention, according to the recognition result of the global anomaly output by the recognition module 105, if there is no anomaly, no measures are taken. If there is an anomaly, the measures are to immediately issue an early warning to the relevant departments and mobilize relevant personnel to investigate the source of the abnormal electromagnetic wave data.

[0114] In one embodiment of the present invention, the electromagnetic environment monitoring method based on deep learning is as follows: Figure 6 , including the following steps:

[0115] Step S601: Divide the target area into M sub-areas, and collect electromagnetic wave data t times from the M sub-areas at a fixed interval every day within a first preset time period T days;

[0116] Step S602: pre-processing the collected electromagnetic wave data and constructing M feature matrices;

[0117] Step S603: constructing a graph network for the electromagnetic wave data of the M sub-regions of the target region, wherein the graph network structure includes: nodes, initial features of the nodes, and edges between the nodes;

[0118] Step S604: inputting the M graph structure data of the first preset time period T days into the anomaly detection model respectively, and the output of the anomaly detection model represents the abnormal value of the electromagnetic wave data of the M sub-regions of the second preset time period G days in the i-th time period of the j-th day;

[0119] Step S605: Identify the abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G.

[0120] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. An electromagnetic environment monitoring system based on deep learning, characterized in that: include: Data collection module: It is used to divide the target area into M sub-areas, and collect electromagnetic wave data from the M sub-areas t times at a fixed interval every day within a first preset time period T days; Electromagnetic wave data includes: frequency, amplitude, phase, polarization, signal-to-noise ratio, location coordinates and timestamp; Data processing module: It is used to pre-process the collected electromagnetic wave data and construct M feature matrices. The j-th column and i-th row of the k-th feature matrix represent the electromagnetic wave data collected at the i-th time on the j-th day of the k-th sub-area, where 0≤k≤M, 0≤j≤T, and 0≤i≤t; Graph network module: It is used to construct a graph network for the electromagnetic wave data of the M sub-areas of the target area. The graph network structure includes: nodes, initial features of nodes and edges between nodes. The nodes in the graph network structure establish a mapping relationship with the elements of the update matrix. The initial features of the nodes are represented by the electromagnetic wave data of the elements with which the mapping relationship is established. The edges between the nodes are constructed in the following ways: edges are constructed between nodes with mapping relationships established by adjacent elements, and edges are constructed between nodes with mapping relationships established by diagonally adjacent elements. Anomaly detection module: it is used to input the M graph structure data of the first preset time period T days into the anomaly detection model respectively, and the output of the anomaly detection model represents the abnormal value of the electromagnetic wave data of the M sub-regions in the i-th time period of the j-th day in the second preset time period G days, wherein the anomaly detection module includes M anomaly detection models, and the M anomaly detection models respectively input the graph network structure of the M sub-regions and the electromagnetic wave data collected at the i-th time of the j-th day in the k-th sub-region in the second preset time period G days; Identification module: used to identify abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G; Generate T elements at the upper and lower boundaries of the feature matrix, generate t elements at the left and right boundaries of the feature matrix, and generate one element at each of the four corners of the feature matrix to construct a (T+2)×(t+2) update matrix X; Update the electromagnetic wave data of the qth element in the 1st column and the T+2th column in the matrix X to ； Update the electromagnetic wave data of the pth element in the 1st and t+2nd rows of the matrix X to ； Update the matrix X , , and The electromagnetic wave data of the elements at the four corners are ; Among them, 2≤q≤t+1, 2≤p≤T+1, , and represent the first weight parameter, the second weight parameter and the third weight parameter respectively, , and They represent the first bias term, the second bias term and the third bias term respectively, and ReLU represents the ReLU activation function; The anomaly detection formula for the electromagnetic wave data collected at the i-th moment of the j-th day in the k-th sub-area in the second preset time period G is as follows: ； ； ； in, represents the abnormal value of the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days, represents the electromagnetic wave data collected at the i-th time of the j-th day of the k-th sub-region in the second preset time period G days, represents the electromagnetic wave data collected at the i-th moment of the j-th day of the k-th sub-region in the first preset time period T days, represents the electromagnetic wave data collected at the i+nth time of the j+mth day of the kth sub-area in the first preset time period T days, Represents the similarity value of adjacent nodes in the network structure of the electromagnetic wave data comparison graph collected at the i-th moment of the j-th day of the k-th sub-region in the second preset time period G days, It represents the similarity value of adjacent nodes in the network structure of the electromagnetic wave data comparison graph collected at the i-th moment of the j-th day in the k-th sub-region in the first preset time period T days, Respectively represent the fourth weight parameter, represents the sub-region abnormal threshold, represents the step activation function, represents a multi-layer perceptron; The global anomaly detection step of the electromagnetic wave data collected at the i-th moment of the j-th day in the second preset time period G includes: Step S301: Create a three-dimensional diagram of M sub-areas of the target area, and input the abnormal values ​​of the M sub-areas at the i-th moment of the j-th day in the second preset time period G days; Step S302: If there are m sub-regions with an outlier value of 1 In the three-dimensional diagram, a straight line structure is formed, and the scores of the m sub-regions are recorded as , Expressed as ,in represents the first reward coefficient; Step S303: If n sub-regions have an abnormal value of 1 If the Euclidean distance in the three-dimensional diagram is less than the distance threshold β, a clustering structure is formed, and the scores of the n sub-regions are recorded as , Expressed as ,in represents the first reward coefficient; The Euclidean distance calculation formula is: ,in and Represents sub-areas and subregions The three-dimensional coordinates of Step S304: If there are z sub-regions with an outlier value of 1 If no clustering structure or straight line structure is formed in the three-dimensional diagram, the score of the z sub-regions is recorded as , Expressed as ,in represents the penalty coefficient; Step S305: Calculate the global outlier value, which is expressed as , where the sub-regions that belong to both clustered structures and linear structures are scored repeatedly, and the global outlier value is considered to be greater than the global outlier threshold if it is considered to be globally outlier, and the global outlier value is considered to be less than the global outlier threshold if it is considered to be globally normal; The node feature updating steps of the graph network structure include: Step S401: input the electromagnetic wave data collected from the kth sub-region at the i-th moment on the j-th day of the second time period G into the graph network structure to perform sub-region anomaly detection; Step S402: updating the graph network structure according to the sub-region outlier value judgment output by the outlier detection module; Step S404: if the sub-region abnormal value of the k-th sub-region at the i-th moment of the j-th day in the second preset time period G days is 0, then the electromagnetic wave data of the k-th sub-region at the i-th moment of the j-th day in the second preset time period G days and the electromagnetic wave data of the k-th sub-region at the i-th moment of the j-th day in the second preset time period T days are averaged to update the feature of the node in the j-th column and the i-th row in the k-th graph network structure; Step S404: if the sub-region abnormal value of the k-th sub-region at the i-th moment on the j-th day in the second preset time period G days is 1, the feature of the node in the j-th column and the i-th row in the k-th graph network structure is not updated; The identification module identifies the globally abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G, including: Step S501, converting the electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G from the time domain to the frequency domain by fast Fourier transform, generating a spectrum diagram, and determining the frequency range of the abnormal signal; Step S502, generating a time-frequency diagram by short-time Fourier transform of the electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G, and identifying the frequency mutation and frequency drift characteristics of the signal; Step S503, identifying the modulation mode of the abnormal signal according to the result of time-frequency analysis; Step S504: determine the location of the signal source by using the arrival time difference method in combination with the spatial locations between the sub-areas.

2. The electromagnetic environment monitoring system based on deep learning according to claim 1, characterized in that: T, M, t, and G are all custom parameters, where the number of days in the first preset time period, T days, and the number of days in the second preset time period, G days, are the same.

3. The electromagnetic environment monitoring system based on deep learning according to claim 1, characterized in that: The preprocessing steps include: Step S201: applying a mean filter method to remove noise from electromagnetic wave data; Step S202: using a bandpass filter to filter out irrelevant frequency band data according to the target frequency band and bandwidth; Step S203: Filling in missing values ​​in the electromagnetic wave data by taking the average value of the electromagnetic wave data collected in the previous time period and the electromagnetic wave data collected in the next time period; Step S204: normalizing the electromagnetic wave data using the Z-score method.

4. The electromagnetic environment monitoring system based on deep learning according to claim 3 is characterized in that: According to the recognition results of global anomalies output by the recognition module, no measures will be taken if there is no anomaly. If there is an anomaly, the measures taken are to immediately issue an early warning to the relevant departments and mobilize relevant personnel to investigate the source of the abnormal electromagnetic wave data.

5. An electromagnetic environment monitoring method based on deep learning, which is applicable to an electromagnetic environment monitoring system based on deep learning as claimed in any one of claims 1 to 4, characterized in that: The following steps are involved: Step S601: Divide the target area into M sub-areas, and collect electromagnetic wave data t times from the M sub-areas at a fixed interval every day within a first preset time period T days; Step S602: pre-processing the collected electromagnetic wave data and constructing M feature matrices; Step S603: constructing a graph network for the electromagnetic wave data of the M sub-regions of the target region, wherein the graph network structure includes: nodes, initial features of the nodes, and edges between the nodes; Step S604: inputting the M graph structure data of the first preset time period T days into the anomaly detection model respectively, and the output of the anomaly detection model represents the abnormal value of the electromagnetic wave data of the M sub-regions of the second preset time period G days in the i-th time period of the j-th day; Step S605: Identify the abnormal electromagnetic wave data at the i-th moment on the j-th day in the second preset time period G.

Images