A traffic user data analysis method and system

Abstract

The application discloses a kind of flow user data analysis method and system.According to the time point of user stop using flow, only the time point of using flow and corresponding used flow are retained, to obtain flow curve chart and flow discrete graph.The flow curve chart represents fitting multiple continuous flow data as curve.According to the abscissa corresponding to the fitted curve, the three-dimensional flow curve chart is cut to obtain multiple time discrete graphs.According to the multiple time discrete graphs, the time convolution network is input, and the flow discrete graph is convolved by three-dimensional convolution kernel, the change rule after each flow use can be found.And by fusing time flow three-dimensional graph, flow curve feature map and flow discrete feature map, the rule of user daily flow use can be found, so as to achieve the technical effect of more accurately analyzing the flow used by user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, in particular to a traffic user data analysis method and system. BACKGROUND

[0002] At present, with the development of the Internet, various industries have also developed Internet business, and currently, the user traffic data analysis can further analyze the user using traffic. The basic implementation method is to analyze and count the statistical indicators of interest to the user in the traffic analysis process. The time point at which the user stops using traffic, the time point at which the user uses traffic, and the characteristics corresponding to the change state of the traffic when the user uses traffic have an impact on the traffic analysis of the user. And the use of traffic has certain rules, so the traffic data analysis without considering the above characteristics is not accurate enough. SUMMARY

[0003] The purpose of the present application is to provide a traffic user data analysis method and system to solve the above problems existing in the prior art.

[0004] In a first aspect, the embodiments of the present application provide a traffic user data analysis method, comprising:

[0005] obtaining a user traffic use time point and corresponding user traffic; the user traffic use time point represents the time point of the traffic used by the user; the user traffic represents the traffic used by the user at the user traffic use time point;

[0006] Based on the user traffic use time point and corresponding user traffic, a plurality of curves are constructed to obtain a traffic curve graph and a traffic discrete graph; the traffic curve graph represents the smooth change image of the traffic at a plurality of user traffic use time points; the traffic discrete graph represents the image of marking the user traffic at a plurality of user traffic use time points;

[0007] Based on the traffic curve graph, the characteristics of the plurality of curves are projected to obtain a traffic curve feature graph;

[0008] Based on the traffic discrete graph, the traffic curve feature graph, the user traffic use time point and the corresponding user traffic, three-dimensional convolution is performed to obtain a traffic fusion feature graph;

[0009] The traffic fusion feature graph is input into a user discrimination network to discriminate the state of the user and obtain a user discrimination category; the user discrimination category represents the classification of the user according to the adaptive traffic.

[0010] Optionally, based on the user traffic use time point and corresponding user traffic, a plurality of curves are constructed to obtain a traffic curve graph and a traffic discrete graph, comprising:

[0011] Taking the user traffic usage time point as the abscissa and the user traffic usage as the ordinate, a two-dimensional traffic graph is obtained;

[0012] Based on the two-dimensional traffic graph, the position of the user's unused traffic is detected, and a traffic non-connection time point is obtained;

[0013] Based on the traffic non-connection time point, the two-dimensional traffic graph is segmented to obtain a plurality of segmented traffic graphs;

[0014] The plurality of segmented traffic graphs are arranged in order from early to late in time to obtain a traffic dispersion graph;

[0015] The points in the segmented traffic graph are fitted to obtain a fitting curve;

[0016] The fitting curve is drawn as an image to obtain a two-dimensional traffic curve graph; a plurality of two-dimensional traffic curve graphs are obtained corresponding to the plurality of segmented traffic graphs;

[0017] The plurality of two-dimensional traffic curve graphs are arranged in order from early to late in time to obtain a traffic curve graph.

[0018] Optionally, based on the traffic curve graph, the characteristics of a plurality of curves are projected to obtain a traffic curve characteristic graph, including:

[0019] The side of the traffic curve graph is segmented multiple times with length as the segmentation point to obtain a plurality of time dispersion graphs; the width of the time dispersion graph is equal to the width of the traffic curve graph; the length of the time dispersion graph is equal to the number of channels of the traffic curve graph;

[0020] The time dispersion graph is convolved to extract features to obtain a time curve feature; a plurality of time curve features are obtained corresponding to the plurality of time dispersion graphs;

[0021] The plurality of time curve features are input into a time convolution network in order from early to late in time point to obtain a traffic curve feature graph.

[0022] Optionally, based on the traffic dispersion graph, the traffic curve feature graph, the user traffic usage time point, and the corresponding user traffic, a three-dimensional convolution is performed to obtain a traffic fusion feature graph, including:

[0023] The user traffic usage time point is segmented at 24-hour intervals to obtain a plurality of user interval time periods and corresponding user interval traffic;

[0024] The user interval time period is taken as the abscissa and the corresponding user interval traffic is taken as the ordinate to draw an image to obtain a user interval two-dimensional graph; a plurality of user interval two-dimensional graphs are obtained corresponding to the plurality of user interval time periods;

[0025] superimpose the plurality of user interval two-dimensional graphs from early to late according to time points to obtain a time flow three-dimensional graph;

[0026] input the time flow three-dimensional graph into a three-dimensional convolution network, extract three-dimensional features of each part to obtain a time flow three-dimensional feature map;

[0027] based on the flow discrete graph, sequentially convolve according to time points from early to late using a three-dimensional convolution kernel to obtain a flow discrete feature map; the flow discrete feature map is a two-dimensional graph;

[0028] wherein the size of the three-dimensional convolution kernel in the three-dimensional convolution network and the number of convolutions make the size of the time flow three-dimensional feature map the same as the size of the flow curve feature map; the size of the three-dimensional convolution kernel corresponding to the flow discrete graph and the number of convolutions make the size of the time flow three-dimensional feature map the same as the size of the flow discrete feature map;

[0029] superimpose the time flow three-dimensional graph, the flow curve feature map and the flow discrete feature map to obtain a flow fusion feature map;

[0030] wherein the flow discrete feature map and the time flow three-dimensional graph are superimposed on the right side of the flow curve feature map, adjacent to the end of the curve.

[0031] Optionally, based on the flow discrete graph, sequentially convolving according to time points from early to late using a three-dimensional convolution kernel to obtain a flow discrete feature map, comprising:

[0032] obtain a three-dimensional convolution kernel; the width of the three-dimensional convolution kernel is equal to the width of the flow discrete graph;

[0033] the three-dimensional convolution kernel moves in the length and channel direction of the flow discrete graph with a step size of 1; the three-dimensional convolution kernel moves sequentially from early to late according to time points.

[0034] Optionally, based on the two-dimensional flow graph, detecting the position of the unused flow of the user to obtain a flow unconnected time point, comprising:

[0035] obtain a two-dimensional convolution kernel; the width of the two-dimensional convolution kernel is equal to the width of the two-dimensional flow graph;

[0036] on the two-dimensional flow graph, the two-dimensional convolution kernel convolves from left to right with a step size of 1 to extract features to obtain a first convolution feature vector;

[0037] input the first convolution feature vector into a first neural network to detect whether there is flow at the time point to obtain a flow unconnected time point.

[0038] Optionally, the two-dimensional flow graph is segmented based on the flow disconnection time point to obtain a plurality of segmented flow graphs, including:

[0039] The two-dimensional flow graph is sequentially searched from left to right, and the flow disconnection time point is detected to obtain a first local segmented flow graph and a first to-be-detected segmented flow graph; the time point corresponding to the first local segmented flow graph is earlier than the time point corresponding to the first to-be-detected segmented flow graph.

[0040] The first to-be-detected segmented flow graph is sequentially searched from left to right, and if a time point other than the flow disconnection time point is detected, the first to-be-detected segmented flow graph is segmented to obtain a second to-be-detected segmented flow graph; the time point corresponding to the second to-be-detected segmented flow graph is the time point at which the user connects to the flow.

[0041] The second to-be-detected segmented flow graph is sequentially searched from left to right, and the flow disconnection time point is detected to obtain a second local segmented flow graph and a third to-be-detected segmented flow graph.

[0042] A plurality of local segmented flow graphs are obtained by multiple searches; the plurality of local segmented flow graphs include the first local segmented flow graph and the second local segmented flow graph.

[0043] The local segmented flow graph is supplemented with 0 at both ends to obtain a segmented flow graph.

[0044] Optionally, one segmented flow graph corresponds to one time point.

[0045] Optionally, the user discrimination network is a convolutional neural network.

[0046] In a second aspect, an embodiment of the present application provides a flow user data analysis system, including:

[0047] An acquisition module obtains a user flow use time point and corresponding user flow; the user flow use time point represents a time point at which the user uses the flow; and the user flow represents the flow used by the user at the user flow use time point.

[0048] A curve module constructs a plurality of curves based on the user flow use time point and the corresponding user flow to obtain a flow curve graph and a flow discrete graph; the flow curve graph represents a smooth change image of the flow at a plurality of user flow use time points; and the flow discrete graph represents an image in which the user flow is labeled at a plurality of user flow use time points.

[0049] A curve feature module projects features of the plurality of curves based on the flow curve graph to obtain a flow curve feature graph.

[0050] The fusion module: based on the traffic discrete graph, the traffic curve feature graph, the user traffic use time point and the corresponding user traffic use, three-dimensional convolution is carried out to obtain a traffic fusion feature graph;

[0051] The category module: the traffic fusion feature graph is input into a user discrimination network to discriminate the state of the user to obtain a user discrimination category; the user discrimination category represents that the user is classified according to adaptive traffic.

[0052] Compared with the prior art, the embodiment of the application has the following beneficial effects:

[0053] The embodiment of the application also provides a traffic user data analysis method and system, the method comprising: obtaining a user traffic use time point and corresponding user traffic use; the user traffic use time point represents a time point of traffic use of a user; the user traffic use represents traffic used by the user at the user traffic use time point; based on the user traffic use time point and the corresponding user traffic use, a plurality of curves are constructed to obtain a traffic curve graph and a traffic discrete graph; the traffic curve graph represents a smooth change image of traffic at a plurality of user traffic use time points; the traffic discrete graph represents an image of marking the user traffic use at a plurality of user traffic use time points; based on the traffic curve graph, the features of the plurality of curves are projected to obtain a traffic curve feature graph; based on the traffic discrete graph, the traffic curve feature graph, the user traffic use time point and the corresponding user traffic use, three-dimensional convolution is carried out to obtain a traffic fusion feature graph; the traffic fusion feature graph is input into a user discrimination network to discriminate the state of the user to obtain a user discrimination category; the user discrimination category represents that the user is classified according to adaptive traffic.

[0054] In the application, the time point at which the user stops using traffic is segmented, only the time point at which traffic is used and the corresponding used traffic are retained to obtain a traffic curve graph and a traffic discrete graph. The traffic curve graph represents that a plurality of continuously used traffic data are fitted into a curve. According to the abscissa corresponding to the fitted curve, the three-dimensional traffic curve graph is cut to obtain a plurality of time discrete graphs. According to the plurality of time discrete graphs, a time convolution network is input, and the traffic discrete graph is convolved through a three-dimensional convolution kernel to find the time length of a time point after reusing traffic from the starting time point of the reused traffic, and then find a time point of the same time length as the starting time point of the distance traffic from the last time of using traffic, so as to find the change of the traffic corresponding to the two time points, so as to find the change rule after each time of using traffic. And by fusing the time traffic three-dimensional graph, the traffic curve feature graph and the traffic discrete feature graph, the daily use traffic rule of the user can be found, so that the technical effect of more accurately analyzing the used traffic of the user is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is a flow user data analysis method flowchart provided by an embodiment of the application. DETAILED DESCRIPTION

[0056] The application will be described in detail below with reference to the drawings.

[0057] Embodiment 1

[0058] As shown in the drawings, an embodiment of the application provides a flow user data analysis method, which comprises the following steps. Figure 1

[0059] S101: obtaining a user flow usage time point and corresponding user usage flow; the user flow usage time point represents a time point of flow used by a user; the user usage flow represents flow used by the user at the user flow usage time point.

[0060] S102: constructing a plurality of curves based on the user flow usage time point and the corresponding user usage flow to obtain a flow curve graph and a flow discrete graph; the flow curve graph represents a smooth change image of flow at a plurality of user flow usage time points; the flow discrete graph represents an image of marking the user usage flow at a plurality of user flow usage time points.

[0061] The flow curve graph and the flow discrete graph are three-dimensional graphs.

[0062] S103: projecting features of the plurality of curves based on the flow curve graph to obtain a flow curve feature graph.

[0063] The flow analysis feature graph is a three-dimensional graph.

[0064] S104: performing three-dimensional convolution based on the flow discrete graph and the flow curve feature graph to obtain a flow fusion feature graph.

[0065] The flow fusion feature graph is a three-dimensional graph.

[0066] S105: inputting the flow fusion feature graph into a user discrimination network to discriminate a state of the user to obtain a user discrimination category; the user discrimination category represents classification of the user according to adaptive flow.

[0067] Optionally, the step of constructing a plurality of curves based on the user flow usage time point and the corresponding user usage flow to obtain a flow curve graph and a flow discrete graph comprises the following steps.

[0068] Taking the user flow usage time point as the abscissa and the user usage flow as the ordinate to obtain a two-dimensional flow graph.

[0069] ​The two-dimensional flow graph marks a point corresponding to a user flow usage time point and user flow usage.

[0070] Based on the two-dimensional flow graph, the position of the user flow not in use is detected to obtain a flow disconnection time point.

[0071] The position of the user flow not in use is a user flow usage time point in which the user flow is 0.

[0072] Based on the flow disconnection time point, the two-dimensional flow graph is segmented to obtain a plurality of segmented flow graphs.

[0073] The plurality of segmented flow graphs are arranged in order from early to late in time to obtain a flow dispersion graph.

[0074] The points in the segmented flow graph are fitted to obtain a fitting curve.

[0075] In this embodiment, a polynomial fitting algorithm is used for fitting.

[0076] The fitting curve is drawn as an image to obtain a two-dimensional flow curve graph; and a plurality of two-dimensional flow curve graphs are obtained corresponding to the plurality of segmented flow graphs.

[0077] The fitting curve is drawn using a plot function.

[0078] The fitting curve is drawn according to the corresponding abscissa and ordinate in the two-dimensional flow graph.

[0079] The plurality of two-dimensional flow curve graphs are arranged in order from early to late in time to obtain a flow curve graph.

[0080] Optionally, based on the flow curve graph, a plurality of curve features are projected to obtain a flow curve feature graph, including:

[0081] The side of the flow curve graph is segmented multiple times using the length as a segmentation point to obtain a plurality of time dispersion graphs; the width of the time dispersion graph is equal to the width of the flow curve graph; and the length of the time dispersion graph is equal to the number of channels of the flow curve graph.

[0082] The time dispersion graph is a two-dimensional graph; and the width of the time dispersion graph is equal to the width of the flow curve graph.

[0083] The flow curve graph is obtained by superimposing a plurality of two-dimensional flow curve graphs, so that when viewed from the front, it is a curve, and when the side of the flow curve graph is segmented using the length of the flow curve graph as a segmentation point, an image with a plurality of discrete points when viewed from the side is obtained, which is a time dispersion graph.

[0084] The time moment discrete graph represents a plurality of curves from a starting point corresponding to a flow rate, or a plurality of time points corresponding to a flow rate at the same time length from the starting point.

[0085] The time moment discrete graph is convolved to extract features to obtain time moment curve features.

[0086] In this embodiment, a convolutional neural network (CNN) is used for convolution.

[0087] The plurality of time moment curve features are sequentially input into a time convolution network according to time points from early to late to obtain a flow rate curve feature map.

[0088] In this embodiment, a time convolution network (TCN) is used to extract the relationship of the flow rate at a plurality of time points.

[0089] Optionally, based on the flow rate discrete graph, the flow rate curve feature map, the user flow rate use time point, and the corresponding user flow rate, a three-dimensional convolution is performed to obtain a flow rate fusion feature map, including:

[0090] The user flow rate use time point is segmented at a time interval of 24 hours to obtain a plurality of user interval time periods and corresponding user interval flow rates.

[0091] A user interval two-dimensional graph is obtained by taking the user interval time period as the abscissa and the corresponding user interval flow rate as the ordinate.

[0092] In this embodiment, a Scatter function can be used to plot discrete points.

[0093] The plurality of user interval two-dimensional graphs are superimposed according to time points from early to late to obtain a time flow rate three-dimensional graph.

[0094] The time flow rate three-dimensional graph is input into a three-dimensional convolution network to extract three-dimensional features of each part to obtain a time flow rate three-dimensional feature map.

[0095] Based on the flow rate discrete graph, a three-dimensional convolution kernel is sequentially convolved according to time points from early to late to obtain a flow rate discrete feature map; the flow rate discrete feature map is a two-dimensional graph.

[0096] In the three-dimensional convolution network, the size of the three-dimensional convolution kernel and the number of convolutions are the same as the size of the time flow rate three-dimensional feature map and the size of the flow rate curve feature map; the size of the three-dimensional convolution kernel corresponding to the flow rate discrete graph and the number of convolutions are the same as the size of the time flow rate three-dimensional feature map and the size of the flow rate discrete feature map.

[0097] Wherein, in the three-dimensional convolutional network, the size of the three-dimensional convolution kernel is 2*2*2, the number of convolutions is 3, and the size of the three-dimensional convolution kernel corresponding to the flow discrete graph is 4*n*4, the number of convolutions is 1, the length and width of the flow discrete feature graph and the time flow three-dimensional feature graph are equal, and the size is the same, wherein n represents the width of the flow discrete feature graph.

[0098] Wherein, the size of the flow discrete graph, the time flow three-dimensional graph and the flow curve graph is the same by adding 0.

[0099] The time flow three-dimensional graph, the flow curve feature graph and the flow discrete feature graph are superimposed to obtain a flow fusion feature graph.

[0100] Wherein, the flow discrete feature graph and the time flow three-dimensional graph are superimposed on the right side of the flow curve feature graph, adjacent to the end of the curve.

[0101] Optionally, based on the flow discrete graph, a three-dimensional convolution kernel is used to convolve in time order from early to late to obtain a flow discrete feature graph, comprising:

[0102] A three-dimensional convolution kernel is obtained; the width of the three-dimensional convolution kernel is equal to the width of the flow discrete graph.

[0103] The three-dimensional convolution kernel moves in the length and channel direction of the flow discrete graph with a step size of 1; the three-dimensional convolution kernel moves in time order from early to late.

[0104] Optionally, based on the two-dimensional flow graph, the position of the user's unused flow is detected to obtain a flow unconnected time point, comprising:

[0105] A two-dimensional convolution kernel is obtained; the width of the two-dimensional convolution kernel is equal to the width of the two-dimensional flow graph.

[0106] Wherein, in the embodiment, the length of the two-dimensional convolution kernel is 2.

[0107] On the two-dimensional flow graph, the two-dimensional convolution kernel is convolved from left to right with a step size of 1 to extract features to obtain a first convolution feature vector.

[0108] The first convolution feature vector is input into a first neural network to detect whether there is flow at the time point to obtain a flow unconnected time point.

[0109] Optionally, based on the flow unconnected time point, the two-dimensional flow graph is segmented to obtain a plurality of segmented flow graphs, comprising:

[0110] In the two-dimensional flow diagram, from left to right, the flow unconnected time point is detected, cutting is performed, and a first local segmented flow diagram and a first to-be-detected segmented flow diagram are obtained; the time point corresponding to the first local segmented flow diagram is earlier than the time point corresponding to the first to-be-detected segmented flow diagram.

[0111] In the first to-be-detected segmented flow diagram, from left to right, if a time point that is not a flow unconnected time point is detected, cutting is performed, and a second to-be-detected segmented flow diagram is obtained; the starting time point of the second to-be-detected segmented flow diagram is a time point at which the user connects to the flow;

[0112] In the second to-be-detected segmented flow diagram, from left to right, the flow unconnected time point is detected, cutting is performed, and a second local segmented flow diagram and a third to-be-detected segmented flow diagram are obtained.

[0113] The multiple segmented flow diagrams are obtained through multiple searches.

[0114] The region of the time point in which no discrete point exists is cut and deleted, and thus multiple segmented flow diagrams that are retained after cutting are obtained.

[0115] Optionally, one segmented flow diagram corresponds to one time point.

[0116] Optionally, the user discrimination network is a convolutional neural network.

[0117] The user discrimination network is a convolutional neural network (CNN), and is a user discrimination network that is trained according to a labeled user discrimination category.

[0118] Embodiment 2

[0119] Based on the flow user data analysis method described above, the embodiment of the application further provides a flow user data analysis system, which comprises an acquisition module, a curve module, a curve feature module, a fusion module, and a category module.

[0120] The acquisition module is configured to obtain user flow use time points and corresponding user used flows; the user flow use time points represent time points at which the user uses the flow; and the user used flows represent the flows used by the user at the user flow use time points.

[0121] The curve module is configured to construct multiple curves based on the user flow use time points and the corresponding user used flows, and obtain a flow curve diagram and a flow discrete diagram; the flow curve diagram represents an image of smooth changes of the flow at the multiple user flow use time points; and the flow discrete diagram represents an image in which the user used flows are labeled at the multiple user flow use time points.

[0122] The curve feature module is configured to project features of a plurality of curves based on the flow curve diagram to obtain a flow curve feature diagram.

[0123] The fusion module is configured to perform three-dimensional convolution based on the flow dispersion diagram, the flow curve feature diagram, the user flow usage time point, and the corresponding user usage flow to obtain a flow fusion feature diagram.

[0124] The category module is configured to input the flow fusion feature diagram into a user discrimination network to discriminate a state of the user to obtain a user discrimination category; the user discrimination category represents that the user is classified according to adaptive flow.

Claims

1. A traffic user data analysis method, characterized by, The method comprises the following steps: obtaining user traffic usage time points and corresponding user traffic usage; taking the user traffic usage time points as the abscissa and the user traffic usage as the ordinate to obtain a two-dimensional traffic graph; detecting the position of the user's unused traffic based on the two-dimensional traffic graph to obtain a traffic disconnection time point; segmenting the two-dimensional traffic graph based on the traffic disconnection time point to obtain a plurality of segmented traffic graphs; arranging the plurality of segmented traffic graphs in sequence according to the time from early to late to obtain a traffic discrete graph; fitting the points in the segmented traffic graph to obtain a fitting curve; drawing the fitting curve as an image to obtain a two-dimensional traffic curve graph; a plurality of segmented traffic graphs correspond to obtain a plurality of two-dimensional traffic curve graphs; arranging the plurality of two-dimensional traffic curve graphs in sequence according to the time from early to late to obtain a traffic curve graph; segmenting the side of the traffic curve graph multiple times with long as the segmentation point to obtain a plurality of time discrete graphs; the width of the time discrete graph is equal to the width of the traffic curve graph; the length of the time discrete graph is equal to the number of channels of the traffic curve graph; convolving the time discrete graph to extract features to obtain a time curve feature; a plurality of time discrete graphs correspond to obtain a plurality of time curve features; inputting the plurality of time curve features in sequence according to the time point from early to late into a time convolution network to obtain a traffic curve feature map; segmenting the user traffic usage time points at 24-hour intervals to obtain a plurality of user interval time periods and corresponding user interval traffic; taking the user interval time period as the abscissa and the corresponding user interval traffic as the ordinate to draw an image to obtain a user interval two-dimensional graph; a plurality of user interval time periods correspond to obtain a plurality of user interval two-dimensional graphs; superimposing the plurality of user interval two-dimensional graphs according to the time point from early to late to obtain a time traffic three-dimensional graph; inputting the time traffic three-dimensional graph into a three-dimensional convolution network to extract the three-dimensional features of each part to obtain a time traffic three-dimensional feature map; obtaining a three-dimensional convolution kernel; the width of the three-dimensional convolution kernel is equal to the width of the traffic discrete graph; the three-dimensional convolution kernel moves in the length and channel direction of the traffic discrete graph with a step size of 1 to obtain a traffic discrete feature map; the three-dimensional convolution kernel moves in sequence according to the time point from early to late; the traffic discrete feature map is a two-dimensional graph; wherein the size of the three-dimensional convolution kernel and the number of convolutions in the three-dimensional convolution network make the size of the time traffic three-dimensional feature map the same as the size of the traffic curve feature map; the size of the three-dimensional convolution kernel and the number of convolutions corresponding to the traffic discrete graph make the size of the time traffic three-dimensional feature map the same as the size of the traffic discrete feature map; superimposing the time traffic three-dimensional feature map, the traffic curve feature map and the traffic discrete feature map to obtain a traffic fusion feature map; wherein the traffic discrete feature map and the time traffic three-dimensional feature map are superimposed on the right side of the traffic curve feature map adjacent to the end of the curve; inputting the traffic fusion feature map into a user discrimination network to discriminate the state of the user to obtain a user discrimination category; the user discrimination category represents classifying the user according to the adaptive traffic.

2. The traffic subscriber data analysis method of claim 1, wherein, The position where the user does not use the traffic is detected based on the two-dimensional traffic graph, and a traffic disconnection time point is obtained, including: Obtaining a two-dimensional convolution kernel; the width of the two-dimensional convolution kernel is equal to the width of the two-dimensional traffic graph; The two-dimensional convolution kernel is convolved from left to right on the two-dimensional traffic graph with a step size of 1, features are extracted, and a first convolution feature vector is obtained; The first convolution feature vector is input into a first neural network to detect whether there is traffic at the time point, and a traffic disconnection time point is obtained.

3. The traffic subscriber data analysis method of claim 1, wherein, The two-dimensional traffic graph is segmented based on the traffic disconnection time point, and a plurality of segmented traffic graphs are obtained, including: The two-dimensional traffic graph is sequentially searched from left to right, and when a traffic disconnection time point is detected, the two-dimensional traffic graph is cut to obtain a first local segmented traffic graph and a first to-be-detected segmented traffic graph; the time point corresponding to the first local segmented traffic graph is earlier than the time point corresponding to the first to-be-detected segmented traffic graph; The first to-be-detected segmented traffic graph is sequentially searched from left to right, and if a time point other than the traffic disconnection time point is detected, the first to-be-detected segmented traffic graph is cut to obtain a second to-be-detected segmented traffic graph; the second to-be-detected segmented traffic graph has a starting time point that is the time point when the user connects to the traffic; The second to-be-detected segmented traffic graph is sequentially searched from left to right, and when a traffic disconnection time point is detected, the second to-be-detected segmented traffic graph is cut to obtain a second local segmented traffic graph and a third to-be-detected segmented traffic graph; A plurality of local segmented traffic graphs are obtained by multiple searches; the plurality of local segmented traffic graphs include the first local segmented traffic graph and the second local segmented traffic graph; The local segmented traffic graph is padded with zeros at both ends to obtain a segmented traffic graph.

4. The traffic subscriber data analysis method of claim 1, wherein, One segmented traffic graph corresponds to one time point.

5. The traffic subscriber data analysis method of claim 1, wherein, The user discrimination network is a convolutional neural network.

6. A traffic subscriber data analysis system characterized by, Including: An acquisition module obtains user traffic use time points and corresponding user traffic; A curve module obtains a two-dimensional traffic graph by taking the user traffic use time points as the horizontal coordinates and the user traffic as the vertical coordinates; A position where the user does not use the traffic is detected based on the two-dimensional traffic graph, and a traffic disconnection time point is obtained; The two-dimensional traffic graph is segmented based on the traffic disconnection time point, and a plurality of segmented traffic graphs are obtained; A plurality of segmented traffic graphs are sequentially arranged in order from early to late in time to obtain a traffic scatter plot; Points in the segmented traffic graph are fitted to obtain a fitting curve; The fitting curve is drawn as an image to obtain a two-dimensional traffic curve graph; a plurality of two-dimensional traffic curve graphs are obtained corresponding to the plurality of segmented traffic graphs; A plurality of two-dimensional traffic curve graphs are sequentially arranged in order from early to late in time to obtain a traffic curve graph; A curve feature module divides the side of the traffic curve graph multiple times with a long segment as a division point to obtain a plurality of time scatter plots; the width of the time scatter plot is equal to the width of the traffic curve graph; the length of the time scatter plot is equal to the number of channels of the traffic curve graph; The time scatter plot is convolved to extract features to obtain a time curve feature; A plurality of time curve features are obtained corresponding to the plurality of time scatter plots; The plurality of time curve features are sequentially input into a time convolution network in order from early to late in time to obtain a traffic curve feature map; The fusion module: the user traffic usage time point is segmented in 24 hours as time interval, to obtain a plurality of user interval time period and corresponding user interval traffic; The user interval time period is taken as the abscissa, and the corresponding user interval traffic is taken as the ordinate. An image is drawn to obtain a user interval two-dimensional graph. A plurality of user interval time periods correspond to obtain a plurality of user interval two-dimensional graphs; The plurality of user interval two-dimensional graphs are superimposed according to the time point from early to late to obtain a time traffic three-dimensional graph; The time traffic three-dimensional graph is input into a three-dimensional convolution network to extract three-dimensional features of each part to obtain a time traffic three-dimensional feature map; A three-dimensional convolution kernel is obtained; The width of the three-dimensional convolution kernel is equal to the width of the traffic discrete graph; The three-dimensional convolution kernel is moved in the long and channel direction of the traffic discrete graph with a step of 1 to obtain a traffic discrete feature map. The three-dimensional convolution kernel is sequentially moved from early to late in the time point. The traffic discrete feature map is a two-dimensional graph; The size and the number of times of convolution of the three-dimensional convolution kernel in the three-dimensional convolution network make the size of the time traffic three-dimensional feature map same as the size of the traffic curve feature map. The size and the number of times of convolution of the three-dimensional convolution kernel corresponding to the traffic discrete graph make the size of the time traffic three-dimensional feature map same as the size of the traffic discrete feature map; The time traffic three-dimensional feature map, the traffic curve feature map and the traffic discrete feature map are superimposed to obtain a traffic fusion feature map; The traffic discrete feature map and the time traffic three-dimensional feature map are superimposed on the right side of the traffic curve feature map, adjacent to the end of the curve. The category module: the traffic fusion feature map is input into a user discrimination network to discriminate the state of the user to obtain a user discrimination category. The user discrimination category represents that the user is classified according to the adaptive traffic.

Images