TW202025169A - System for detecting abnormality of people flow using mobile communication sevice by machine learning and method thereof - Google Patents

Abstract

The invention discloses a system for detecting abnormality of mobile communication people flow by machine learning and a method thereof. The system comprises a data reading module, a machine learning module, an anomaly detection module and a map marking module. The data reading module reads historical people flow data associated with a plurality of mobile communication devices from a grid and its adjacent grids. The machine learning module with a machine learning program performs machine learning on the historical people flow data to generate an estimate value of the people flow data of next time based on result of the machine learning. The anomaly detection module compares a difference value between the estimated value and an actual value of the people flow data for the next time, so that when the difference value is outside a threshold value, it is determined that the people flow data of the next time is abnormal people flow data. The map marking module automatically marks or warns an abnormal people flow area corresponding to the abnormal people flow data on an electronic map.

Description

System and method for detecting abnormal flow of mobile communication people based on machine learning

The present invention relates to a technology for detecting anomalies in the flow of people, in particular to a system and method for detecting anomalies in the flow of mobile communications based on machine learning.

In the past, the population density of various events (such as disasters) was calculated based on the data registered by the user registration, but this data is quite different from the actual population data. Therefore, in order to grasp the actual flow of people, technologies such as video image recognition or infrared sensors are needed, but these two technologies still have their shortcomings.

In the video image recognition technology, data can be easily obtained, but the image recognition technology needs to be strengthened, and the actual number cannot be counted. It can only use the crowd density classification, and the scope is limited to a small area.

In addition, in the infrared sensor technology, the number of people can be recorded through the sensor, but it is limited to a specific enclosed space, and the range is limited to a small area.

Therefore, how to provide technologies different from video image recognition and infrared sensors, using new technologies to quickly and/or accurately detect or determine the flow of people data or abnormal flow of people data, has become one of the major research courses for those skilled in the art. question.

The present invention provides a system and method for detecting abnormalities in mobile communication traffic based on machine learning. Machine learning programs, anomaly detection modules, and map identification modules can work together to detect or determine the next time (in the future) People flow data or abnormal flow data.

The system of the present invention for detecting abnormal traffic flow in mobile communication based on machine learning includes: a data reading module that reads historical traffic data associated with a plurality of mobile communication devices from a grid and adjacent grids; and a machine learning The module is equipped with a machine learning program to perform machine learning on the historical traffic data related to the plural mobile communication devices read by the data reading module, so that the machine learning program can generate associations with the plural mobile communication devices according to the results of machine learning The estimated value of the flow of people data at the next time; an anomaly detection module that compares the difference between the estimated value of the flow of people data generated by the machine learning program and the actual value at the next time to determine the difference When it is outside the threshold value or its range, the abnormality detection module determines that the next-time pedestrian flow data is abnormal pedestrian flow data; and a location icon marking module is automatically calibrated on the electronic map or a warning and anomaly detection module The abnormal pedestrian flow area corresponding to the determined abnormal pedestrian flow data.

The method for detecting abnormal traffic flow in mobile communication based on machine learning of the present invention includes: reading historical traffic flow data associated with a plurality of mobile communication devices from a grid and adjacent grids by a data reading module; learning by a machine The machine learning program of the module performs machine learning on the historical traffic data read by the data reading module associated with multiple mobile communication devices for the machine learning program According to the results of machine learning, the estimated value of the people flow data related to the plurality of mobile communication devices at a time is generated; an anomaly detection module compares the estimated value of the people flow data generated by the machine learning program with the actual value of the next time The difference between the two values, when the difference value is outside the threshold value or its range, the abnormality detection module determines that the pedestrian flow data at the next time is the abnormal pedestrian flow data; and the module is marked on the electronic map by a place icon Automatically calibrate or warn the abnormal pedestrian flow area corresponding to the abnormal pedestrian flow data determined by the abnormal detection module.

In order to make the above-mentioned features and advantages of the present invention more comprehensible, embodiments are specifically described below in conjunction with the accompanying drawings. In the following description, the additional features and advantages of the present invention will be partially explained, and these features and advantages will be partly obvious from the description, or can be learned by practicing the present invention. The features and advantages of the present invention are realized and achieved by means of the elements and combinations specifically pointed out in the scope of the patent application. It should be understood that the foregoing general description and the following detailed description are both illustrative and explanatory, and are not intended to limit the claimed scope of the present invention.

1‧‧‧A system for detecting anomalies in mobile communications based on machine learning

10‧‧‧Data reading module

20‧‧‧Data Processing Module

30‧‧‧Machine Learning Module

31‧‧‧Machine Learning Model

32‧‧‧First density layer

33‧‧‧Second density layer

34‧‧‧LSTM (Long Short Term Memory) layer

40‧‧‧Anomaly Detection Module

41‧‧‧Analysis of the flow of people

50‧‧‧Ground icon fixed module

51‧‧‧electronic map

52‧‧‧Anomalous area map

53‧‧‧Region name

60‧‧‧Display Module

A‧‧‧Base station

B‧‧‧Mobile communication device

C1‧‧‧Grid

C2‧‧‧Adjacent grid

N‧‧‧Grid

R‧‧‧Distance

T‧‧‧Time Range

Steps S1 to S5‧‧‧

Figure 1 is a schematic diagram of the architecture of the system for detecting abnormalities in mobile communication based on machine learning of the present invention; Figures 2A and 2B are grid diagrams of different representations of the present invention; Figure 3A is an implementation of the present invention In the example, a machine learning model constructed by, for example, a deep neural network (DNN); Figure 3B is a machine learning model constructed by, for example, a recurrent neural network (RNN) in an embodiment of the present invention; Figure 4 is an analysis diagram of pedestrian flow data according to an embodiment of the present invention; Figure 5 is a schematic diagram of marking abnormal pedestrian flow areas, abnormal area maps and area names on an electronic map in an embodiment of the present invention; and Figure 6 shows The flow chart of the method for detecting abnormalities in mobile communication based on machine learning of the present invention.

The following describes the implementation of the present invention with specific specific embodiments. Those familiar with this technology can easily understand the other advantages and effects of the present invention from the contents disclosed in this specification, and can also be implemented by other different specific embodiments. Or apply.

Figure 1 is a schematic diagram of the architecture of the system 1 for detecting abnormalities in mobile communication based on machine learning of the present invention. As shown in the figure, the system 1 for detecting abnormal flow of people in mobile communication based on machine learning may include a data reading module 10, a data processing module 20, a machine learning module 30, which are connected to each other and transmit data or messages. There are at least six modules including an anomaly detection module 40, a ground icon positioning module 50, and a display module 60.

The above six modules can be constructed or composed using hardware, firmware, software, or a combination thereof. For example, the data reading module 10 may be a hardware data reader or a software data reading program, the data processing module 20 may be a hardware data processor or a software data processing program, and the machine learning module 30 It can be a software machine learning program. The anomaly detection module 40 can be a hardware anomaly detector or a software anomaly detection program. The map icon setting module 50 can be a software map program. The display module 60 can be Hardware display, software display program or their combination.

The data reading module 10 in Figure 1 can read historical traffic data associated with a plurality of mobile communication devices B from a grid C1 and an adjacent grid C2 (see Figures 2A and 2B). Please refer to page The historical traffic data shown in Figures 3A and 3B, or the historical traffic data from time T00 to T23 shown in Figure 4. The aforementioned mobile communication device B can be, for example, a smart phone, a smart watch, a tablet or a notebook computer, etc., and the time (such as T00 to T23...) can be in unlimited units, such as every 1 minute, every 10 minutes, every 1 hour Or wait every 1 day.

In terms of grid data filling, the data processing module 20 in Figure 1 can perform data processing on the historical pedestrian flow data of the grid C1 and the adjacent grid C2 read by the data reading module 10; if the grid C1 and The historical traffic data at all times in the adjacent grid C2 is missing, the first value (such as 0) is used to fill in the historical traffic data at all times; and if the historical traffic in the grid C1 and the adjacent grid C2 is only part of the time If the data is missing, the second value (for example, the average value) is used to fill in the historical flow data for part of the time. Moreover, in the time range T, the data processing module 20 can obtain the data of the previous few hours, and the minimum value (for example, 1) means that only the data of the current time is taken, and the data of the previous time is not taken. In the spatial range R, the data processing module 20 can take the data of the adjacent grid C2 of the grid C1, and the minimum value (for example, 0) means that only the data of the grid C1 is taken instead of the data of the adjacent grid C2 .

The machine learning module 30 in Fig. 1 may have a machine learning program to perform machine learning on the historical traffic data associated with the plural mobile communication device B read by the data reading module 10, so that the machine learning program can perform machine learning according to the machine learning. The result is an estimated value of the traffic data associated with the plural mobile communication device B at a time. Please refer to the next time shown in Figure 3A and Figure 3B The flow of people between time, or the flow of people at the next time T24 shown in Figure 4. The next time may represent the next time point, the next time period, or the next time slot, such as the next 10 minutes or the next 1 hour.

At the same time, the machine learning program of the machine learning module 30 can further establish the flow characteristics of the complex grid C1 to estimate the flow data or the number of people flow under the complex area at a time based on the flow characteristics of the complex grid C1.

The anomaly detection module 40 in Figure 1 can compare the difference between the estimated value and the actual value (observed value) of the flow data generated by the machine learning program at the next time, so that the difference value is at the threshold (see Fig. 4) or outside the range, the abnormality detection module 40 determines that the pedestrian flow data at the next time is abnormal pedestrian flow data.

The location icon marking module 50 in Figure 1 can automatically calibrate or warn on the electronic map 51 (see Figure 5) the abnormal pedestrian flow area corresponding to the abnormal pedestrian flow data determined by the abnormal detection module 40 (see Figure 5) Dark or red grid) and normal pedestrian flow areas (see the light or light blue grid shown in Figure 5).

The display module 60 in Fig. 1 can automatically display the corresponding abnormal area map 52 or the area name 53 according to the abnormal pedestrian flow area marked on the electronic map 51 by the map marking module 50 (see Fig. 5).

Figures 2A and 2B are grid diagrams N of different representations of the present invention. As shown in Figures 2A, 2B and 1, the data reading module 10 can read historical traffic data associated with a plurality of mobile communication devices B from a grid C1 and an adjacent grid C2. Cell C1 can correspond to at least two mobile communication devices B and at least one base station A, and at least two mobile communication devices B can communicate with at least one base station A. At the same time, a grid C1 has For example, the area of "(2R+1)x(2R+1)", R represents the spatial range (i.e. distance, such as 0.5 km), 2R+1 represents the length or width of the grid C1, and T represents the time range (e.g. Figure 4 shows the time from T00 to T23 of historical traffic data).

Figure 3A is a machine learning model 31 constructed by, for example, a deep neural network (DNN) in an embodiment of the present invention, and Figure 3B is a machine learning model 31 constructed by, for example, a recurrent neural network (Recurrent Neural Network) in an embodiment of the present invention. Neural Network; RNN) constructed machine learning model 31.

In the present invention, the machine learning module 30 can use various neural networks, such as deep neural network (DNN), recurrent neural network (RNN), or convolutional neural network (Convolutional neural network; CNN) to construct a machine learning The model 31 is used to generate a machine learning program according to the machine learning model 31. The aforementioned deep neural network (DNN) can be a Deep Fully-connected Neural Network, and a recurrent neural network (RNN) can be a recurrent neural network with Long Short-Term Memory (LSTM) Network, Convolutional Neural Network (CNN) can be a convolutional neural network with map image input (map image input) or time sequence data input (time sequence data input). However, the present invention is not limited to this.

For example, in Figure 3A, the machine learning module 30 can use a deep neural network (DNN) to construct a machine learning model 31 to generate historical flow data through the first density layer 32 and the second density layer 33 sequentially. People flow data for the next time. In Figure 3B, the machine learning module 30 can use a recurrent neural network (RNN) with long short-term memory (LSTM) to construct a machine learning model 31 to generate historical human flow data through a complex LSTM layer 34 People flow data for the next time.

In the above figures 3A and 3B, the historical flow data can include the flow data within the range of the formula "(2R+1)x(2R+1)x T" shown in Figure 3A to Figure 3B. The next time The flow data of may include the flow data of the time under the formula "(2R+1)x(2R+1)x(T+1)" shown in Figure 3A to Figure 3B. R represents the spatial range (i.e. distance), 2R+1 represents the length or width of the grid C1, T represents the time range (such as the time T00 to T23 of the historical traffic data shown in Figure 4), the historical traffic data indicates the known or already The calculated flow data, and the flow data at the next time (such as time T24 in Figure 4) represents the flow data that is unknown, to be calculated or to be estimated.

Figure 4 is a data analysis diagram 41 of the flow of people according to an embodiment of the present invention. As shown in Figure 4, the anomaly detection module 40 in Figure 1 can analyze historical traffic data (such as traffic data from time T00 to T23) to create an "estimated value" based on the analysis result of the historical traffic data , "Normal (i.e. normal flow of people data)", "Abnormal (i.e. abnormal flow of people data)" and "Threshold" data analysis chart 41. In addition, the anomaly detection module 40 can also set a threshold value based on the analysis result of the historical traffic data, and the threshold value can be changed with the time of the historical traffic data or the number of people (the number of people).

At the same time, the anomaly detection module 40 in Figure 1 can compare the difference between the estimated value and the actual value (observed value) of the pedestrian flow data generated by the machine learning program at the next time (such as time T24) to When the difference value is outside the threshold value or its range, the abnormality detection module 40 determines that the pedestrian flow data at the next time is "abnormal (that is, abnormal pedestrian flow data)".

Figure 5 shows the marking of people on the electronic map 51 in one embodiment of the present invention Schematic diagram of abnormal flow area (see dark or red grid), abnormal area map 52 and area name 53.

As shown in the figure, the location icon marking module 50 in Figure 1 can automatically calibrate the abnormal pedestrian flow area corresponding to the abnormal pedestrian flow data determined by the abnormal detection module 40 on the electronic map 51 in Figure 5 (see dark or red Grid) and normal pedestrian flow areas (see light-colored or light-blue grids), and each grid may have, for example, 0.25 square kilometers (km ² ). At the same time, the display module 60 in FIG. 1 can automatically display the corresponding abnormal area map 52 or the area name 53 based on the abnormal pedestrian flow area marked on the electronic map 51 in the fifth diagram by the map-marking module 50.

Figure 6 is a schematic flow chart of the method for detecting abnormal traffic in mobile communication based on machine learning of the present invention. Please refer to Figures 1 to 5 together. At the same time, the main technical content of the method of the present invention for detecting abnormal traffic in mobile communication based on machine learning is as follows, and the remaining technical content is the same as the detailed description of the above-mentioned Figures 1 to 5, and will not be repeated here.

As shown in step S1 in FIG. 6, a data reading module 10 reads historical traffic data associated with a plurality of mobile communication devices B from a grid C1 and an adjacent grid C2.

In this step, a data processing module 20 can also perform data processing on the historical people flow data read by the data reading module 10, where, if the historical people flow data in the grid C1 and the adjacent grid C2 at all times If there is any missing, the first value (such as 0) is used to fill in the historical traffic data at all times. If the historical traffic data of only part of the time in the grid C1 and the adjacent grid C2 is missing, the second value (such as average Value) to fill in historical traffic data for part of the time.

As shown in step S2 in Figure 6, the machine learning program of a machine learning module 30 performs machine learning on the historical traffic data read by the data reading module 10 and associated with the plural mobile communication device B for the machine The learning program generates an estimated value of people flow data related to the plurality of mobile communication devices B at a time according to the results of the machine learning.

As shown in step S3 in Fig. 6, an anomaly detection module 40 compares the difference between the estimated value of the next-time pedestrian flow data generated by the machine learning program and the actual value (observed value) to When the difference value is outside the threshold value or its range, the abnormality detection module 40 determines that the pedestrian flow data at the next time is abnormal pedestrian flow data.

In this step, the anomaly detection module 40 can also analyze the historical people flow data to set a threshold value based on the analysis result of the historical people flow data, and the threshold value can be based on the time or the number of people (the number of people flow) of the historical people flow data. Change and change.

As shown in step S4 in FIG. 6, the location icon marking module 50 automatically marks or warns the abnormal pedestrian flow area corresponding to the abnormal pedestrian flow data determined by the abnormality detection module 40 on the electronic map 51.

As shown in step S5 in FIG. 6, a display module 60 automatically displays the corresponding abnormal area map 52 or area name 53 based on the abnormal pedestrian flow area marked on the electronic map 51 by the local icon marking module 50.

In summary, the system and method for detecting abnormal traffic in mobile communication based on machine learning of the present invention can have the following characteristics, advantages or technical effects:

1. The present invention can cooperate with machine learning programs, anomaly detection modules, and map identification modules to quickly and/or accurately detect or determine (Future) Crowd flow data or abnormal flow data at the next time.

2. The present invention makes use of the popularity of mobile communication devices (almost one device per person), and can grasp the distribution status of the flow of people associated with multiple mobile communication devices on a large scale.

3. The present invention uses the huge amount of data from the mobile communication device to extract the flow data, and the machine learning module (machine learning program) automatically determines whether the time or area of the flow data is abnormal, so as to automatically mark or warn the abnormal area of the flow, and improve the Response efficiency of abnormal data on the flow of people in various events (such as emergency disaster events).

4. The present invention can be applied to any event that requires the detection of abnormal data on the flow of people, such as emergency disasters such as earthquakes, wind disasters, and floods, or general events such as gatherings, parades, elections, and concerts.

The above embodiments are only illustrative of the principles, features and effects of the present invention, and are not intended to limit the scope of implementation of the present invention. Anyone who is familiar with the art can comment on the above without departing from the spirit and scope of the present invention. Modifications and changes to the implementation form. Any equivalent changes and modifications made by using the contents disclosed in the present invention should still be covered by the scope of the patent application. Therefore, the protection scope of the present invention should be as listed in the scope of patent application.

1‧‧‧A system for detecting anomalies in mobile communications based on machine learning

10‧‧‧Data reading module

20‧‧‧Data Processing Module

30‧‧‧Machine Learning Module

40‧‧‧Anomaly Detection Module

50‧‧‧Ground icon fixed module

60‧‧‧Display Module

A‧‧‧Base station

B‧‧‧Mobile communication device

C1‧‧‧Grid

C2‧‧‧Adjacent grid

Claims (11)

A system for detecting abnormal human flow in mobile communication based on machine learning includes: a data reading module that reads historical human flow data associated with a plurality of mobile communication devices from a grid and adjacent grids; and a machine learning module The group is equipped with a machine learning program to perform machine learning on the historical traffic data associated with the plurality of mobile communication devices read by the data reading module, so that the machine learning program generates a correlation with the machine learning result based on the machine learning result. The estimated value of people flow data at one time under multiple mobile communication devices; an anomaly detection module that compares the difference between the estimated value and actual value of the next time people flow data generated by the machine learning program Value, when the difference value is outside the threshold value or its range, the abnormality detection module determines that the pedestrian flow data at the next time is abnormal pedestrian flow data; and a location-based identification module is automatically displayed on the electronic map Mark or warn the abnormal pedestrian flow area corresponding to the abnormal data of the pedestrian flow determined by the abnormal detection module. The system according to claim 1, wherein the grid corresponds to at least two mobile communication devices and at least one base station, and the at least two mobile communication devices communicate with the at least one base station. For example, the system described in item 1 of the scope of patent application further includes a data processing module for processing the historical traffic data of the grid and adjacent grids read by the data reading module, wherein , If the historical traffic data at all times in the grid and adjacent grids are missing, The first value is used to fill in the historical traffic data at all times, and if the historical traffic data for only part of the time in the grid and adjacent grids is missing, the second value is used to fill in the historical traffic data at that part of the time data. The system described in item 1 of the scope of patent application, wherein the machine learning module further establishes the flow characteristics of the complex grid to estimate the flow of people or the flow of people at a time under the complex area based on the flow characteristics of the complex grid Quantity. Such as the system described in item 1 of the scope of patent application, wherein the machine learning module uses deep neural network (DNN), recurrent neural network (RNN) or convolutional neural network (CNN) to construct a machine learning Model to generate the machine learning program according to the machine learning model. For example, the system described in item 1 of the scope of patent application, wherein the anomaly detection module further analyzes the historical traffic data to set the threshold value according to the analysis result of the historical traffic data, and the threshold value follows the The historical flow of people data changes according to the time or number of people. The system described in item 1 of the scope of patent application further includes a display module, which automatically displays the corresponding abnormal area map or area name based on the abnormal pedestrian flow area marked on the electronic map by the local icon positioning module. A method for detecting abnormal traffic flow in mobile communication based on machine learning includes: reading historical traffic data associated with a plurality of mobile communication devices from a grid and adjacent grids by a data reading module; and a machine learning model The group’s machine learning program reads the data The historical traffic data associated with the plurality of mobile communication devices read by the module is subjected to machine learning, so that the machine learning program generates a prediction of the traffic data related to the plurality of mobile communication devices at a time based on the results of the machine learning. Estimate; compare the difference between the estimated value and the actual value of the next time flow data generated by the machine learning program by an anomaly detection module, so that the difference is within the threshold or its range When outside, the abnormality detection module determines that the next-time pedestrian flow data is abnormal data; and is automatically calibrated or warned on the electronic map by a local icon marking module and the pedestrian flow determined by the abnormality detection module The abnormal pedestrian flow area corresponding to the abnormal data. For example, the method described in item 8 of the scope of patent application further includes data processing by a data processing module on the grid and the historical pedestrian flow data of adjacent grids read by the data reading module, wherein: If the historical traffic data at all times in the grid and adjacent grids are missing, the first value is used to fill in the historical traffic data at all times, and if the grid and adjacent grids are only part of the time If the historical traffic data is missing, the second value will be used to fill in the historical traffic data for that part of the time. For example, the method described in item 8 of the scope of patent application further includes analyzing the historical traffic data by the anomaly detection module to set the threshold value based on the analysis result of the historical traffic data, and the threshold value follows the The historical flow of people data changes according to the time or number of people. The method described in item 8 of the scope of patent application further includes that a display module automatically displays the corresponding abnormal area map or area name according to the abnormal pedestrian flow area marked on the electronic map by the local icon positioning module.

Images