A user identification system based on SIM card

Abstract

The present invention discloses a user identification system based on a SIM card. A connection structure of a tree is constructed based on similarity relationships to obtain a similarity tree. The first convolution kernel and the second convolution kernel are then used to detect the relationship between adjacent time points in a day, and the similarity tree is pruned to obtain connection time points that conform to the user's traffic usage pattern. The convolution obtains the relationship between the usage traffic and the change in position of adjacent days at the same time, and prunes again, repeating the convolution kernel pruning operation. This allows a time point on the second day to be found that best conforms to the user's traffic usage pattern on the first day. Thus, the user's usage pattern is found, and the points of the pattern are constructed as a straight line. If the traffic usage does not conform to the user's traffic usage pattern, the parameters of the straight line are input into a neural discriminant network to detect user anomalies. This can significantly improve the security and efficiency of user identity authentication and provide a strong security guarantee for mobile communication networks.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a user identification system based on a SIM card. Background Art

[0002] With the rapid development of mobile communication technology, user authentication has become a critical component of network security. The SIM card (Subscriber Identity Module) is a smart card used to store user identity information and is widely used in mobile communication devices. While traditional SIM card-based authentication systems are widely used, they still have certain security vulnerabilities, such as susceptibility to SIM card cloning. SIM cards are susceptible to theft, and conventional SIM card authentication systems cannot provide early warning of such theft. Therefore, the market urgently needs a more secure and efficient user identification system. Summary of the Invention

[0003] The purpose of the present invention is to provide a user identification system based on a SIM card to solve the above problems existing in the prior art.

[0004] An embodiment of the present invention provides a SIM card-based user identification system, including:

[0005] Acquisition module: acquires a traffic matrix; the traffic matrix represents the traffic used by the user at multiple locations at multiple time points;

[0006] Similar area detection module: Based on the traffic matrix, a three-dimensional matrix is ​​constructed to detect areas with similar traffic usage status, and multiple similar time points are obtained; the similar time points represent the time points at which users have similar daily behaviors;

[0007] Line construction module: constructs a straight line based on multiple similar time points to obtain a three-dimensional flow line;

[0008] Identification module: based on the three-dimensional flow line, determines whether the user is abnormal;

[0009] Dynamic token generation module: If the user is abnormal, a dynamic identity token is generated in real time based on multiple similar time points to verify the user.

[0010] Optionally, the method of constructing a three-dimensional matrix based on the traffic matrix, detecting areas with similar traffic usage status, and obtaining multiple similar time points includes:

[0011] Cutting the traffic matrix according to fixed time lengths to obtain multiple traffic cutting matrices;

[0012] Obtain a first traffic cutting matrix and a second traffic cutting matrix; the first traffic cutting matrix and the second traffic cutting matrix are two traffic cutting matrices at adjacent time points among a plurality of traffic cutting matrices;

[0013] Based on the first traffic cutting matrix and the second traffic cutting matrix, detecting time points in which traffic is similar in the two traffic cutting matrices, and obtaining a plurality of similarity trees;

[0014] Through the adjacent convolutional network, based on multiple similarity trees, the first traffic cutting matrix and the second traffic cutting matrix, the similarity states of traffic changes at adjacent time points are judged to obtain multiple similar time points.

[0015] Optionally, the method of detecting time points at which traffic in the two traffic cutting matrices is similar based on the first traffic cutting matrix and the second traffic cutting matrix to obtain multiple similarity trees includes:

[0016] Extracting the position and flow corresponding to each time point in the first flow cutting matrix to obtain a first position flow set; extracting the position and flow corresponding to each time point in the second flow cutting matrix to obtain a second position flow set;

[0017] Calculating cluster similarity based on the locations and traffic, clustering the first location traffic set and the second location traffic set to obtain multiple cluster sets;

[0018] Using the value of the first position flow set in a cluster set as a first detection value;

[0019] The first detection value is used as the root node of the tree, and the value of the second position flow set in the cluster set is used as the child node of the root node to obtain a similar tree; multiple values ​​in the first position flow set correspond to multiple similar trees.

[0020] Optionally, the method of determining similar states of traffic changes at adjacent time points based on the first traffic cutting matrix and the second traffic cutting matrix through the adjacent convolutional network to obtain multiple similar time points includes:

[0021] Obtaining a first eigenvector and a second eigenvector based on the first flow cutting matrix and the second flow cutting matrix;

[0022] Superimposing the first flow cutting matrix and the second flow cutting matrix to obtain a first three-dimensional matrix;

[0023] Through the third convolution kernel, the change between two adjacent days is extracted from the time point from early to late on the first three-dimensional matrix with a step size of 1 to obtain the third eigenvector;

[0024] The length of the third convolution kernel is 2; the width of the third convolution kernel is 2; the third convolution kernel performs convolution in the direction of the surface formed by the columns and pages of the first three-dimensional matrix;

[0025] Calculate the similarity of the values ​​in the third eigenvector, mark the time points corresponding to similar values; set the time points corresponding to the similar values ​​to the same label;

[0026] Pruning multiple similar trees, retaining root nodes and child nodes with the same label, to obtain a first similar tree;

[0027] Based on the first feature vector and the second feature vector, detecting time points with similar usage traffic, and obtaining a plurality of second similarity trees;

[0028] In the first similarity tree, the same root node and child nodes as those in the second similarity tree are retained and pruned;

[0029] Pruning is performed multiple times until the root node of the similar tree contains only one child node, and the time points corresponding to the root node and the child node are marked as similar time points.

[0030] Optionally, performing user identity detection based on the traffic matrix to obtain an identity detection value includes:

[0031] According to the time points from early to late, the first convolution kernel performs convolution on the first flow cutting matrix with a step size of 1, extracts the change relationship between flow and position between two adjacent time points, and obtains the first eigenvector;

[0032] The width of the first convolution kernel is equal to the number of rows of the first flow cutting matrix; the length of the first convolution kernel is equal to 2;

[0033] According to the time points from early to late, the second convolution kernel performs convolution on the second flow cutting matrix with a step size of 1, extracts the change relationship between flow and position between two adjacent time points, and obtains the second eigenvector;

[0034] The width of the second convolution kernel is equal to the number of rows of the second flow cutting matrix; the length of the second convolution kernel is equal to 2.

[0035] Optionally, the adjacent convolutional network includes a first convolution kernel, a second convolution kernel, a third convolution kernel, an average structure, and a pruning structure;

[0036] The input of the first convolution kernel is a first flow cutting matrix; the input of the second convolution kernel is a second flow cutting matrix;

[0037] The outputs of the first convolution kernel and the second convolution kernel are inputs of the average structure;

[0038] The input of the third convolution kernel is the first three-dimensional matrix;

[0039] The output of the averaging structure and the output of the third convolution kernel are inputs of the pruning structure;

[0040] The pruning structure is self-connected;

[0041] The output of the pruning structure is similar time points.

[0042] Optionally, the parameters of the first convolution kernel and the second convolution kernel are the same;

[0043] The first convolution kernel and the second convolution kernel are trained, and the two convolution kernels with different parameters are averaged using an average structure to obtain the trained first convolution kernel and the second convolution kernel.

[0044] Optionally, constructing a straight line based on multiple similar time points to obtain a three-dimensional flow line includes:

[0045] The multiple similar time points are fitted into a straight line using the least square method to obtain a three-dimensional flow straight line.

[0046] Optionally, judging whether the user is abnormal based on the three-dimensional flow line includes:

[0047] The coefficients of the three-dimensional flow line are input into a trained neural discriminant network to determine whether the user is abnormal.

[0048] Optionally, if the user is abnormal, verifying the user based on multiple similar time points includes:

[0049] Randomly extract a similar time point from multiple similar time points as the detection time point;

[0050] Sending the detection time point to the user;

[0051] Receive a user-answered location; the user-answered location represents the user's location at the detection time point input by the user;

[0052] If the position corresponding to the detection time point is equal to the user's answer position, it is considered that the verification is successful.

[0053] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0054] An embodiment of the present invention also provides a user identification system based on a SIM card, the system comprising: an acquisition module: acquiring a traffic matrix; the traffic matrix represents the traffic used by the user at multiple time points and multiple locations; a similar area detection module: constructing a three-dimensional matrix based on the traffic matrix, detecting areas with similar traffic usage status, and obtaining multiple similar time points; the similar time points represent time points at which users have similar daily behaviors; a straight line construction module: constructing a straight line based on multiple similar time points to obtain a three-dimensional traffic line; a judgment module: judging whether the user is abnormal based on the three-dimensional traffic line; a dynamic token generation module: if the user is abnormal, generating a dynamic identity token in real time based on multiple similar time points to verify the user.

[0055] The present invention first detects time points on two adjacent days where traffic usage and location are similar, then constructs a tree connection structure based on the similarity relationship to obtain a similarity tree. A first convolution kernel and a second convolution kernel are then used to detect the relationship between adjacent time points within a day, pruning the similarity tree to obtain connected time points that conform to the user's traffic usage patterns. The first and second traffic cut matrices corresponding to two adjacent days are used to construct a three-dimensional matrix, namely the first three-dimensional matrix. This matrix is ​​then convolved to obtain the relationship between traffic usage and location for adjacent days at the same time. Pruning is then performed again, and the convolution kernel pruning operation is repeated. This results in finding a time point on the second day that best matches the user's traffic usage patterns on the first day. This results in the user's usage patterns being identified, and the points of the pattern being constructed as a straight line. If the traffic usage conforms to the user's traffic usage patterns, the parameters of the line are input into a neural discriminant network to detect normal user behavior. If the traffic usage does not conform to the user's traffic usage patterns, the parameters of the line are input into a neural discriminant network to detect user anomalies. Therefore, user detection is performed based on whether the user is abnormal, and the positions corresponding to adjacent time points of the random user, which are known to the user and difficult to crawl, are used as identity tokens to verify the user. This can determine whether the user using the SIM card is the original user based on the user's traffic usage patterns, significantly improving the security and efficiency of user identity authentication, and providing a strong technical effect of security protection for mobile communication networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flowchart of a method corresponding to a SIM card-based user identification system provided by an embodiment of the present invention.

[0057] Figure 2 This is a structural diagram of an adjacent convolutional network in a SIM card-based user identification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The present invention will be described in detail below with reference to the accompanying drawings.

[0059] Example 1

[0060] like Figure 1 As shown, an embodiment of the present invention provides a user identification system based on a SIM card. The system 1 includes an acquisition module, a similar area detection module, a straight line construction module, a discrimination module and a dynamic token generation module.

[0061] The acquisition module is used for S101: acquiring a traffic matrix; the traffic matrix represents the traffic used by the user at multiple locations at multiple time points.

[0062] The rows of the traffic matrix represent multiple time points; the columns of the traffic matrix represent multiple locations; the values ​​in the traffic matrix represent traffic; and a row of the traffic matrix contains only one value, representing the traffic spent at one location at one time point.

[0063] The location represents an area, and the area is divided by the location of the base station receiving the traffic.

[0064] In this embodiment, the flow rate used at one location at one point in time refers to the flow rate used at one location within one minute.

[0065] The similar area detection module is used in S102: based on the traffic matrix, a three-dimensional matrix is ​​constructed to detect areas with similar traffic usage status, and multiple similar time points are obtained; the similar time points represent time points at which users have similar behaviors every day.

[0066] The straight line construction module is used in S103 to construct a straight line based on multiple similar time points to obtain a three-dimensional flow straight line;

[0067] The determination module is used for S104: determining whether the user is abnormal based on the three-dimensional flow line;

[0068] The dynamic token generation module is used in S105: if the user is abnormal, a dynamic identity token is generated in real time based on multiple similar time points to verify the user.

[0069] Optionally, the method of constructing a three-dimensional matrix based on the traffic matrix, detecting areas with similar traffic usage status, and obtaining multiple similar time points includes:

[0070] The traffic matrix is ​​cut according to a fixed time length to obtain a plurality of traffic cut matrices.

[0071] In this embodiment, the fixed time length is 1 day.

[0072] The traffic cutting matrix is ​​a two-dimensional image; the number of rows of multiple traffic cutting matrices is the same.

[0073] Obtain a first traffic cutting matrix and a second traffic cutting matrix; the first traffic cutting matrix and the second traffic cutting matrix are two traffic cutting matrices at adjacent time points among a plurality of traffic cutting matrices;

[0074] Based on the first traffic cutting matrix and the second traffic cutting matrix, detecting time points in which traffic is similar in the two traffic cutting matrices, and obtaining a plurality of similarity trees;

[0075] Through the adjacent convolutional network, based on multiple similarity trees, the first traffic cutting matrix and the second traffic cutting matrix, the similarity states of traffic changes at adjacent time points are judged to obtain multiple similar time points.

[0076] Optionally, the method of detecting time points at which traffic in the two traffic cutting matrices is similar based on the first traffic cutting matrix and the second traffic cutting matrix to obtain multiple similarity trees includes:

[0077] Extracting the position and flow corresponding to each time point in the first flow cutting matrix to obtain a first position flow set; extracting the position and flow corresponding to each time point in the second flow cutting matrix to obtain a second position flow set;

[0078] The locations and traffic flows are used together to calculate cluster similarity, and the first location traffic set and the second location traffic set are clustered to obtain a plurality of cluster sets.

[0079] In this embodiment, k-means is used for clustering. When calculating the similarity of clusters, the sum of the Euclidean distance calculated by location and the Euclidean distance calculated by traffic is used to determine the similarity, thereby performing clustering to obtain multiple cluster sets.

[0080] The value of the first position traffic set in a cluster set is used as a first detection value.

[0081] The first detection value represents a value extracted from a cluster set and belonging to a first location traffic set;

[0082] The first detection value is used as the root node of the tree, and the value of the second position flow set in the cluster set is used as the child node of the root node to obtain a similar tree; multiple values ​​in the first position flow set correspond to multiple similar trees.

[0083] Since a cluster set includes multiple values ​​in the first location traffic set and multiple values ​​in the second location traffic set, the construction of the similarity tree can reflect the similarity relationship between the values ​​in the first location traffic set and the values ​​in the second location traffic set.

[0084] The number of similar trees is equal to the number of elements in the first position flow set.

[0085] Optionally, the method of determining similar states of traffic changes at adjacent time points based on the first traffic cutting matrix and the second traffic cutting matrix through the adjacent convolutional network to obtain multiple similar time points includes:

[0086] A first eigenvector and a second eigenvector are obtained based on the first flow cutting matrix and the second flow cutting matrix.

[0087] The first feature vector represents a change state of the flow rate used between adjacent time points of a day.

[0088] The first flow cutting matrix and the second flow cutting matrix are superimposed to obtain a first three-dimensional matrix.

[0089] Through the third convolution kernel, the changes between two adjacent days are extracted from the time point from early to late on the first three-dimensional matrix with a step size of 1 to obtain the third eigenvector.

[0090] The length of the third convolution kernel is 2; the width of the third convolution kernel is 2; and the third convolution kernel performs convolution in the direction of a surface formed by columns and pages of the first three-dimensional matrix.

[0091] Here, because the number of pages of the superimposed first three-dimensional matrix is ​​equal to 2, the length of the third convolution kernel is set to 2.

[0092] Calculate the similarity of the values ​​in the third eigenvector, mark the time points corresponding to similar values; set the time points corresponding to the similar values ​​to the same label;

[0093] Values ​​whose absolute value of the subtraction is less than 1 are considered similar. For example, if the third eigenvector is [4, 2, 8, 2, 7], 2-2 = 0, and 0 is less than 1, the first "2" corresponds to subscript 1, and the second "2" corresponds to subscript 3. The time points corresponding to subscripts 1 and 0 are marked as 0.

[0094] Pruning is performed on multiple similar trees, and root nodes and child nodes with the same label are retained to obtain a first similar tree.

[0095] In the similarity tree, pruning is performed to retain the root nodes and child nodes with the same label, which means that the time points with the same flow rate changes are retained. In the similarity tree, the flow rates used in adjacent days are retained.

[0096] Based on the first feature vector and the second feature vector, time points with similar usage traffic are detected to obtain a plurality of second similarity trees.

[0097] The method for obtaining the second similarity tree is the same as the method for obtaining the similarity tree, which is modified from determining that the traffic in the two traffic cutting matrices is the same to determining that the traffic changes in the related feature vectors are the same.

[0098] Since the second similarity tree detects the relationship between two time points, one child node corresponds to two adjacent time points.

[0099] In the first similarity tree, the root node and child nodes that are the same as those in the second similarity tree are retained and pruned.

[0100] The subnodes of the first similarity tree and the second similarity tree that are connected to the same subnode are retained, and the subnodes of the first similarity tree that do not exist in the second similarity tree are deleted.

[0101] For example, if the root node value in the first similarity tree corresponds to the traffic volume at the fourth time point on the first day, and the child node values ​​correspond to the traffic volume at the third, fifth, and seventh time points on the second day. The root node in the second similarity tree corresponds to the traffic volume changes at the fourth and fifth time points on the first day, and the child node values ​​correspond to the traffic volume changes at the fifth and sixth time points on the second day. In this case, the child node corresponding to the fifth time point on the second day in the first similarity tree is retained, and the other nodes are deleted.

[0102] Pruning is performed multiple times until the root node of the similar tree contains only one child node, and the time points corresponding to the root node and the child node are marked as similar time points.

[0103] In this embodiment, the fourth time point on the first day and the fifth time point on the second day are marked as similar time points.

[0104] Optionally, performing user identity detection based on the traffic matrix to obtain an identity detection value includes:

[0105] According to the time points from early to late, the first convolution kernel performs convolution on the first flow cutting matrix with a step size of 1, extracts the change relationship between flow and position between two adjacent time points, and obtains the first eigenvector;

[0106] The width of the first convolution kernel is equal to the number of rows of the first flow cutting matrix; the length of the first convolution kernel is equal to 2;

[0107] According to the time points from early to late, the second convolution kernel performs convolution on the second flow cutting matrix with a step size of 1, extracts the change relationship between flow and position between two adjacent time points, and obtains the second eigenvector;

[0108] The width of the second convolution kernel is equal to the number of rows of the second flow cutting matrix; the length of the second convolution kernel is equal to 2.

[0109] The first convolution kernel and the second convolution kernel are two-dimensional convolution kernels.

[0110] Optionally, the adjacent convolutional network includes a first convolution kernel, a second convolution kernel, a third convolution kernel, an average structure and a pruning structure.

[0111] The structural diagram of the adjacent convolutional network is as follows Figure 2 shown.

[0112] The input of the first convolution kernel is a first flow cutting matrix; the input of the second convolution kernel is a second flow cutting matrix;

[0113] The outputs of the first convolution kernel and the second convolution kernel are inputs of the average structure;

[0114] The input of the third convolution kernel is the first three-dimensional matrix;

[0115] The output of the averaging structure and the output of the third convolution kernel are inputs of the pruning structure;

[0116] The pruning structure is self-connected;

[0117] The output of the pruning structure is similar time points.

[0118] Among them, the labeled similar time points are used to train adjacent convolutional networks.

[0119] Optionally, the parameters of the first convolution kernel and the second convolution kernel are the same;

[0120] The first convolution kernel and the second convolution kernel are trained, and the two convolution kernels with different parameters are averaged using an average structure to obtain the trained first convolution kernel and the second convolution kernel.

[0121] Among them, because the first convolution kernel and the second convolution kernel can distinguish the changes of the first flow cutting matrix and the second flow cutting matrix, an adjacent convolution network capable of similar discrimination is obtained, and a convolution kernel for between discrimination can also be obtained.

[0122] Optionally, constructing a straight line based on multiple similar time points to obtain a three-dimensional flow line includes:

[0123] The multiple similar time points are fitted into a straight line using the least square method to obtain a three-dimensional flow straight line.

[0124] Optionally, judging whether the user is abnormal based on the three-dimensional flow line includes:

[0125] The coefficients of the three-dimensional flow line are input into a trained neural discriminant network to determine whether the user is abnormal.

[0126] The neural discriminant network is a fully-connected neural network (FCNN), which is trained using labeled values ​​that mark whether the user is abnormal.

[0127] Optionally, if the user is abnormal, verifying the user based on multiple similar time points includes:

[0128] Randomly extract a similar time point from multiple similar time points as the detection time point;

[0129] Sending the detection time point to the user;

[0130] Receive a user-answered location; the user-answered location represents the user's location at the detection time point input by the user;

[0131] If the position corresponding to the detection time point is equal to the user's answer position, it is considered that the verification is successful.

[0132] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general-purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing this type of system. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the above description of specific languages ​​is for the purpose of disclosing the best mode of the present invention.

[0133] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0134] The various component embodiments of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to an embodiment of the present invention. The present invention can also be implemented as a device or device program (e.g., a computer program and a computer program product) for executing a part or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

Claims

1. A user identification system based on a SIM card, characterized in that: include; Acquisition module: acquires a traffic matrix; the traffic matrix represents the traffic used by the user at multiple locations at multiple time points; Similar area detection module: Based on the traffic matrix, a three-dimensional matrix is ​​constructed to detect areas with similar traffic usage status and obtain multiple similar time points; The similar time points represent the time points at which the user's behavior is similar every day; Line construction module: constructs a straight line based on multiple similar time points to obtain a three-dimensional flow line; Identification module: based on the three-dimensional flow line, determines whether the user is abnormal; Dynamic token generation module: If the user is abnormal, a dynamic identity token is generated in real time based on multiple similar time points to verify the user; Based on the traffic matrix, a three-dimensional matrix is ​​constructed to detect areas with similar traffic usage status, and multiple similar time points are obtained, including: Cutting the traffic matrix according to fixed time lengths to obtain multiple traffic cutting matrices; Obtain a first traffic cutting matrix and a second traffic cutting matrix; the first traffic cutting matrix and the second traffic cutting matrix are two traffic cutting matrices at adjacent time points among a plurality of traffic cutting matrices; Based on the first traffic cutting matrix and the second traffic cutting matrix, detecting time points in which traffic is similar in the two traffic cutting matrices, and obtaining a plurality of similarity trees; Through the adjacent convolutional network, based on multiple similarity trees, the first flow cutting matrix and the second flow cutting matrix, the similarity of flow changes at adjacent time points is determined to obtain multiple similar time points; The method detects time points in which similar traffic is used in the two traffic cutting matrices based on the first traffic cutting matrix and the second traffic cutting matrix, and obtains multiple similarity trees, including: Extracting the position and flow corresponding to each time point in the first flow cutting matrix to obtain a first position flow set; extracting the position and flow corresponding to each time point in the second flow cutting matrix to obtain a second position flow set; Calculating cluster similarity based on the locations and traffic, clustering the first location traffic set and the second location traffic set to obtain multiple cluster sets; Using the value of the first position flow set in a cluster set as a first detection value; The first detection value is used as the root node of the tree, and the value of the second position flow set in the cluster set is used as the child node of the root node to obtain a similar tree; multiple values ​​in the first position flow set correspond to multiple similar trees.

2. A SIM card-based user identification system according to claim 1, characterized in that: The method uses the adjacent convolutional network to determine the similarity of traffic changes at adjacent time points based on the first traffic cutting matrix and the second traffic cutting matrix, and obtains multiple similar time points, including: Obtaining a first eigenvector and a second eigenvector based on the first flow cutting matrix and the second flow cutting matrix; Superimposing the first flow cutting matrix and the second flow cutting matrix to obtain a first three-dimensional matrix; Through the third convolution kernel, the change between two adjacent days is extracted from the time point from early to late on the first three-dimensional matrix with a step size of 1 to obtain the third eigenvector; The length of the third convolution kernel is 2; the width of the third convolution kernel is 2; the third convolution kernel performs convolution in the direction of the surface formed by the columns and pages of the first three-dimensional matrix; Calculate the similarity of the values ​​in the third eigenvector, mark the time points corresponding to similar values; set the time points corresponding to the similar values ​​to the same label; Pruning multiple similar trees, retaining root nodes and child nodes with the same label, to obtain a first similar tree; Based on the first feature vector and the second feature vector, detecting time points with similar usage traffic, and obtaining a plurality of second similarity trees; In the first similarity tree, the same root node and child nodes as those in the second similarity tree are retained and pruned; Pruning is performed multiple times until the root node of the similar tree contains only one child node, and the time points corresponding to the root node and the child node are marked as similar time points.

3. A SIM card-based user identification system according to claim 2, characterized in that: The performing user identity detection based on the traffic matrix to obtain an identity detection value includes: According to the time points from early to late, the first convolution kernel performs convolution on the first flow cutting matrix with a step size of 1, extracts the change relationship between flow and position between two adjacent time points, and obtains the first eigenvector; The width of the first convolution kernel is equal to the number of rows of the first flow cutting matrix; the length of the first convolution kernel is equal to 2; According to the time points from early to late, the second convolution kernel performs convolution on the second flow cutting matrix with a step size of 1, extracts the change relationship between flow and position between two adjacent time points, and obtains the second eigenvector; The width of the second convolution kernel is equal to the number of rows of the second flow cutting matrix; the length of the second convolution kernel is equal to 2.

4. A SIM card-based user identification system according to claim 3, characterized in that: The adjacent convolutional network includes a first convolution kernel, a second convolution kernel, a third convolution kernel, an average structure and a pruning structure; The input of the first convolution kernel is a first flow cutting matrix; The input of the second convolution kernel is a second flow cutting matrix; The outputs of the first convolution kernel and the second convolution kernel are inputs of the average structure; The input of the third convolution kernel is the first three-dimensional matrix; The output of the averaging structure and the output of the third convolution kernel are inputs of the pruning structure; The pruning structure is self-connected; The output of the pruning structure is similar time points.

5. A SIM card-based user identification system according to claim 4, characterized in that: The parameters of the first convolution kernel and the second convolution kernel are the same; The first convolution kernel and the second convolution kernel are trained, and the two convolution kernels with different parameters are averaged using an average structure to obtain the trained first convolution kernel and the second convolution kernel.

6. A SIM card-based user identification system according to claim 1, characterized in that: The method of constructing a straight line based on multiple similar time points to obtain a three-dimensional flow straight line includes: The multiple similar time points are fitted into a straight line using the least square method to obtain a three-dimensional flow straight line.

7. A SIM card-based user identification system according to claim 1, characterized in that: The determining whether the user is abnormal based on the three-dimensional flow line includes: The coefficients of the three-dimensional flow line are input into a trained neural discriminant network to determine whether the user is abnormal.

8. The SIM card-based user identification system according to claim 1, characterized in that: If the user is abnormal, verify the user based on multiple similar time points, including: Randomly extract a similar time point from multiple similar time points as the detection time point; Sending the detection time point to the user; Receive a user-answered location; the user-answered location represents the user's location at the detection time point input by the user; If the position corresponding to the detection time point is equal to the user's answer position, the verification is considered successful; Based on the traffic matrix, a three-dimensional matrix is ​​constructed to detect areas with similar traffic usage status, and multiple similar time points are obtained, including: Cutting the traffic matrix according to fixed time lengths to obtain multiple traffic cutting matrices; Obtain a first traffic cutting matrix and a second traffic cutting matrix; the first traffic cutting matrix and the second traffic cutting matrix are two traffic cutting matrices at adjacent time points among a plurality of traffic cutting matrices; Based on the first traffic cutting matrix and the second traffic cutting matrix, detecting time points in which traffic is similar in the two traffic cutting matrices, and obtaining a plurality of similarity trees; Through the adjacent convolutional network, based on multiple similarity trees, the first flow cutting matrix and the second flow cutting matrix, the similarity of flow changes at adjacent time points is determined to obtain multiple similar time points; The method detects time points in which similar traffic is used in the two traffic cutting matrices based on the first traffic cutting matrix and the second traffic cutting matrix, and obtains multiple similarity trees, including: Extracting the position and flow corresponding to each time point in the first flow cutting matrix to obtain a first position flow set; extracting the position and flow corresponding to each time point in the second flow cutting matrix to obtain a second position flow set; Calculating cluster similarity based on the locations and traffic, clustering the first location traffic set and the second location traffic set to obtain multiple cluster sets; Using the value of the first position flow set in a cluster set as a first detection value; The first detection value is used as the root node of the tree, and the value of the second position flow set in the cluster set is used as the child node of the root node to obtain a similar tree; multiple values ​​in the first position flow set correspond to multiple similar trees.

Images