A population migration flow generation method and system based on deep learning and a terminal

Abstract

The application discloses a kind of population migration flow generation method, system and terminal based on deep learning, the method includes: the mobile phone signaling data of target resident is acquired, and classification is carried out according to meteorological data, while tracking the mobile phone signaling data of target resident in destination area;According to mobile phone signaling data and the deep learning model of multiple source data construction and training obtained population migration model is collected;Current mobile phone signaling data is input into population migration model, and the population migration flow in the target area is output.The application integrates multiple source data and deep learning technology, constructs the population migration model for predicting population migration flow, can accurately capture the dynamic change of resident during disaster, more fine description is carried out to the migration flow of different community or area in city, improves data precision and timeliness, realizes the accurate prediction of population migration flow of city resident during disaster.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information analysis, in particular to a population migration flow generation method and system based on deep learning, a terminal and a computer readable storage medium. BACKGROUND

[0002] Population migration flow is influenced by various factors, including changes in statistical caliber, acceleration of urbanization process, economic development imbalance, etc. These factors jointly act on the migration trend of population from rural areas to cities, and also accelerate the process of urbanization and change the social and economic structure.

[0003] However, the existing technology mainly relies on historical data and expert experience to analyze population migration flow during disasters, and this method has problems such as low data accuracy, lack of real-time performance and insufficient spatial fineness, and at the same time, traditional methods cannot fully utilize modern data sources (such as mobile signaling data, remote sensing image data, etc.) to make refined disaster population migration flow prediction.

[0004] Therefore, the prior art still needs to be improved and developed. SUMMARY

[0005] The main purpose of the present application is to provide a population migration flow generation method and system based on deep learning, a terminal and a computer readable storage medium, which aims to solve the problem of lack of data accuracy and real-time performance in the analysis of population migration flow in the prior art, and insufficient spatial fineness and single data source, thereby leading to inaccurate population migration flow analysis.

[0006] To achieve the above purpose, the present application provides a population migration flow generation method based on deep learning, which comprises the following steps:

[0007] Obtain multiple initial mobile signaling data of each resident in a target area, filter out multiple target residents according to all the initial mobile signaling data, and preprocess all the initial mobile signaling data of all the target residents to obtain multiple mobile signaling data of each target resident;

[0008] Obtain meteorological data of the target area, divide all the mobile signaling data into multiple first mobile signaling data and multiple second mobile signaling data according to the meteorological data, and obtain multiple third mobile signaling data of all the target residents;

[0009] Construct a first matrix according to multiple first mobile signaling data and multiple third mobile signaling data, construct a second matrix according to multiple second mobile signaling data and multiple third mobile signaling data, and obtain multiple source data, construct a deep learning model according to the multiple source data, the first matrix and the second matrix;

[0010] perform feature extraction on the plurality of first mobile phone signaling data and the plurality of second mobile phone signaling data to obtain a plurality of feature vectors and a plurality of geographical distances, and input all the feature vectors and all the geographical distances into the deep learning model for training to obtain a population migration model;

[0011] obtain a plurality of current mobile phone signaling data of a plurality of residents in the target area, and input the plurality of current mobile phone signaling data into the population migration model to output a population migration flow in the target area.

[0012] Optionally, the population migration flow generation method based on deep learning, wherein the plurality of initial mobile phone signaling data of each resident in the target area is obtained, a plurality of target residents is selected according to all the initial mobile phone signaling data, and all the initial mobile phone signaling data of all the target residents is preprocessed to obtain a plurality of mobile phone signaling data of each target resident, and the preprocessing specifically includes:

[0013] all initial mobile phone signaling data of all residents in the target area within a preset time is obtained, and if there is a plurality of initial mobile phone signaling data of a resident, the resident is taken as a target resident, wherein the initial mobile phone signaling data complete indicates that the initial mobile phone signaling data of the target resident is complete every day within the preset time in the target area;

[0014] the format of all the initial mobile phone signaling data of all the target residents is uniformly processed, and the data of each activity track anomaly is deleted to obtain a plurality of mobile phone signaling data of each target resident within the preset time.

[0015] Optionally, the population migration flow generation method based on deep learning, wherein the meteorological data of the target area is obtained, all the mobile phone signaling data is divided into a plurality of first mobile phone signaling data and a plurality of second mobile phone signaling data according to the meteorological data, and a plurality of third mobile phone signaling data of all the target residents is obtained, and the preprocessing specifically includes:

[0016] the meteorological data of the target area within the preset time is obtained, and the preset time is divided into a first preset time and a second preset time according to the meteorological data;

[0017] all the mobile phone signaling data within the first preset time is marked as a plurality of first mobile phone signaling data, and all the mobile phone signaling data within the second preset time is marked as a plurality of second mobile phone signaling data;

[0018] a plurality of third mobile phone signaling data of all the target residents is obtained, wherein the third mobile phone signaling data represents the mobile phone signaling data of the target resident after migrating to a destination area after the preset time.

[0019] Optionally, the population migration flow generation method based on deep learning, wherein the first matrix is constructed according to the plurality of first mobile signaling data and the plurality of third mobile signaling data, the second matrix is constructed according to the plurality of second mobile signaling data and the plurality of third mobile signaling data, the multi-source data is obtained, and the deep learning model is constructed according to the multi-source data, the first matrix and the second matrix, and specifically includes:

[0020] determining the first target resident quantity of the target residents in the target area within the first preset time according to all the first mobile signaling data;

[0021] determining the second target resident quantity of the target residents in the target area within the second preset time according to all the second mobile signaling data;

[0022] determining the third target resident quantity of the target residents in the target area after the preset time according to all the third mobile signaling data;

[0023] obtaining the multi-source data in the target area, constructing the first matrix according to the first target resident quantity and the third target resident quantity, and constructing the second matrix according to the second target resident quantity and the third target resident quantity;

[0024] constructing the deep learning model according to the multi-source data, the first matrix and the second matrix.

[0025] Optionally, the population migration flow generation method based on deep learning, wherein the feature vector includes migration frequency and average migration distance;

[0026] the feature extraction is performed on the plurality of first mobile signaling data and the plurality of second mobile signaling data to obtain a plurality of feature vectors and a plurality of geographic distances, all the feature vectors and all the geographic distances are input into the deep learning model for training to obtain a population migration model, and specifically includes:

[0027] the feature extraction is performed on all the first mobile signaling data to obtain first coordinate data of all the target residents in the target area within the first preset time;

[0028] the feature extraction is performed on all the second mobile signaling data to obtain second coordinate data of all the target residents in the target area within the second preset time;

[0029] the migration frequency, the average migration distance and the geographic distance of all the target residents are calculated according to all the first coordinate data and all the second coordinate data;

[0030] The migration frequency, the average migration distance, and the geographical distance are input into the deep learning model for training to obtain a population migration model.

[0031] Optionally, the deep learning-based population migration flow generation method, wherein acquiring multiple current mobile phone signaling data of multiple residents within the target area, inputting them into the population migration model, and outputting the population migration flow within the target area specifically includes:

[0032] Acquire multiple current mobile phone signaling data of multiple residents within the target area, and extract current coordinate information from each current mobile phone signaling data, wherein the current coordinate information represents the current location of the resident corresponding to the current mobile phone signaling data;

[0033] Each of the current coordinate information is input into the population migration model for prediction, and the population migration flow of the target area is output, wherein the population migration flow represents the predicted number of residents in the target area who migrate to the destination area.

[0034] Optionally, the deep learning-based population migration flow generation method, wherein inputting each of the current coordinate information into the population migration model for prediction and outputting the population migration flow of the target area specifically includes:

[0035] Each of the current coordinate information is input into the feedforward neural network of the population migration model. The feedforward neural network analyzes the current coordinate information through an activation function and outputs the prediction score of the target area.

[0036] The population migration model uses a preset function to convert the predicted score into the population migration flow and output it.

[0037] Furthermore, to achieve the above objectives, the present invention also provides a deep learning-based population migration flow generation system, wherein the deep learning-based population migration flow generation system includes:

[0038] The data preprocessing module is used to acquire multiple initial mobile phone signaling data of each resident in the target area, filter out multiple target residents based on all the initial mobile phone signaling data, and preprocess all the initial mobile phone signaling data of all the target residents to obtain multiple mobile phone signaling data of each target resident.

[0039] The data classification module is used to acquire meteorological data of the target area, divide all the mobile phone signaling data into multiple first mobile phone signaling data and multiple second mobile phone signaling data according to the meteorological data, and acquire multiple third mobile phone signaling data of all the target residents;

[0040] a model construction module, configured to construct a first matrix according to the plurality of first mobile phone signaling data and the plurality of third mobile phone signaling data, construct a second matrix according to the plurality of second mobile phone signaling data and the plurality of third mobile phone signaling data, and obtain multi-source data, and construct a deep learning model according to the multi-source data, the first matrix and the second matrix;

[0041] a model training module, configured to perform feature extraction on the plurality of first mobile phone signaling data and the plurality of second mobile phone signaling data to obtain a plurality of feature vectors and a plurality of geographic distances, input all the feature vectors and all the geographic distances to the deep learning model for training to obtain a population migration model;

[0042] a migration prediction module, configured to obtain a plurality of current mobile phone signaling data of a plurality of residents in the target area, input the plurality of current mobile phone signaling data to the population migration model, and output a population migration flow in the target area.

[0043] In addition, to achieve the above object, the present application further provides a terminal, wherein the terminal comprises a memory, a processor, and a deep learning-based population migration flow generation program stored in the memory and executable on the processor, and the deep learning-based population migration flow generation program, when executed by the processor, implements the steps of the deep learning-based population migration flow generation method as described above.

[0044] In addition, to achieve the above object, the present application further provides a computer readable storage medium, wherein the computer readable storage medium stores a deep learning-based population migration flow generation program, and the deep learning-based population migration flow generation program, when executed by a processor, implements the steps of the deep learning-based population migration flow generation method as described above.

[0045] In the present application, a plurality of initial mobile phone signaling data of each resident in the target area is obtained, a plurality of target residents is screened out according to all the initial mobile phone signaling data, and all the initial mobile phone signaling data of all the target residents is preprocessed to obtain a plurality of mobile phone signaling data of each target resident; meteorological data of the target area is obtained, all the mobile phone signaling data is divided into a plurality of first mobile phone signaling data and a plurality of second mobile phone signaling data according to the meteorological data, and a plurality of third mobile phone signaling data of all the target residents is obtained; a first matrix is constructed according to a plurality of the first mobile phone signaling data and a plurality of the third mobile phone signaling data, a second matrix is constructed according to a plurality of second mobile phone signaling data and a plurality of the third mobile phone signaling data, and multi-source data is obtained, a deep learning model is constructed according to the multi-source data, the first matrix and the second matrix; a plurality of feature vectors and a plurality of geographic distances are obtained by performing feature extraction on a plurality of the first mobile phone signaling data and a plurality of the second mobile phone signaling data, and all the feature vectors and all the geographic distances are input into the deep learning model for training to obtain a population migration model; a plurality of current mobile phone signaling data of a plurality of residents in the target area is obtained and input into the population migration model, and the population migration flow in the target area is output. The present application integrates multi-source data and deep learning technology to construct a population migration model for predicting population migration flow, which can accurately capture the dynamic changes of residents during disasters, more accurately depict the migration flow of different communities or regions in the city, improve the data accuracy and timeliness, and realize accurate prediction of the population migration flow of urban residents during disasters. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is a flowchart of a preferred embodiment of the population migration flow generation method based on deep learning of the present application;

[0047] Figure 2 is a structure diagram of a preferred embodiment of the population migration flow generation system based on deep learning of the present application;

[0048] Figure 3 is a running environment schematic diagram of a preferred embodiment of the terminal of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below with reference to the drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0050] The population migration flow generation method based on deep learning according to the preferred embodiment of the present application, as shown in Figure 1 the population migration flow generation method based on deep learning includes the following steps:

[0051] Step S10, obtaining a plurality of initial mobile phone signaling data of each resident in the target area, screening a plurality of target residents according to all the initial mobile phone signaling data, and preprocessing all the initial mobile phone signaling data of all the target residents to obtain a plurality of mobile phone signaling data of each target resident.

[0052] Among them, through the obtained initial mobile phone signaling data of each resident in the target area, the residence change characteristics of each resident within a certain time (for example, during the disaster) are tracked, and the frequency of acquisition can be selected by the user as needed (for example, acquisition every 15 minutes), to identify the community change of the residence of the resident at a high frequency, capture the space-time migration characteristics of the resident, improve the real-time performance of information acquisition, provide reliable samples for subsequent model training, and thus improve the accuracy of migration flow prediction.

[0053] Specifically, all initial mobile phone signaling data of all residents in the target area within a preset time is obtained, and if there are multiple initial mobile phone signaling data of a resident, the resident is taken as a target resident, wherein the initial mobile phone signaling data complete indicates that the initial mobile phone signaling data of the target resident within a preset time in the target area is complete every day; the format of all initial mobile phone signaling data of all target residents is uniformly processed, and data with each activity track exception is deleted, to obtain a plurality of mobile phone signaling data of each target resident within a preset time.

[0054] Among them, through relevant data materials and ArcGIS (ArcGIS Platform, geographic information system) spatial analysis, the boundaries of each urban community plot in the target area are identified; then the initial mobile phone signaling data of the residents in the target area within a preset time is obtained, including micro-population space-time activity data with geographic coordinate information, and the data is preprocessed to unify the format, delete device information without label ID and activity track abnormal data; according to the geographic information coordinates of the expressway and the main road, the related coordinate data is deleted, and finally the mobile phone signaling data is obtained; wherein, in order to protect the privacy of residents and the uniformity of data, the collection time interval of coordinate information is set to every 15 minutes, and only the mobile phone user samples that are tracked to the space-time coordinates completely every day are retained.

[0055] Further, if the mobile signaling data of a certain resident for several days is incomplete within a preset time, the mobile signaling data of the resident will not be considered and will be deleted, thereby improving the integrity of the data and avoiding the impact of incomplete data on subsequent model training. According to the mobile signaling data of each user, the most frequently visited spatial grid of each mobile user per day is identified, representing the main activity area, and each coordinate data can be represented by a vector <p, tp>, where p represents the spatial grid coordinates of the resident at time tp. By using high-frequency mobile signaling data and real-time remote sensing image data, the dynamic changes of residents during disasters can be accurately captured, improving data accuracy and timeliness.

[0056] Step S20, obtaining meteorological data of the target area, dividing all the mobile signaling data into a plurality of first mobile signaling data and a plurality of second mobile signaling data according to the meteorological data, and obtaining a plurality of third mobile signaling data of all the target residents.

[0057] Specifically, the meteorological data of the target area within the preset time is obtained, and the preset time is divided into a first preset time and a second preset time according to the meteorological data; all mobile signaling data within the first preset time is marked as a plurality of first mobile signaling data, and all mobile signaling data within the second preset time is marked as a plurality of second mobile signaling data; a plurality of third mobile signaling data of all the target residents is obtained, wherein the third mobile signaling data represents the mobile signaling data of the target residents after migrating to the destination area after the preset time.

[0058] Among them, after preprocessing all the mobile signaling data, according to the obtained meteorological data, the preset time is divided into three categories, for example, through the meteorological data, the disaster period is divided into the pre-disaster period, the disaster period and the post-disaster period, and according to the three types of time, all mobile signaling data can be divided into three types, which are the first mobile signaling data in the pre-disaster period, the second mobile signaling data in the disaster period and the third mobile signaling data in the post-disaster period.

[0059] Step S30, constructing a first matrix according to a plurality of the first mobile signaling data and a plurality of the third mobile signaling data, constructing a second matrix according to a plurality of second mobile signaling data and a plurality of the third mobile signaling data, and obtaining multi-source data, constructing a deep learning model according to the multi-source data, the first matrix and the second matrix.

[0060] Specifically, the first target resident quantity of the target residents in the target region within the first preset time is determined according to all the first mobile phone signaling data; the second target resident quantity of the target residents in the target region within the second preset time is determined according to all the second mobile phone signaling data; the third target resident quantity of the target residents in the target region after the preset time is determined according to all the third mobile phone signaling data; the multi-source data in the target region is acquired, the first matrix is constructed according to the first target resident quantity and the third target resident quantity, and the second matrix is constructed according to the second target resident quantity and the third target resident quantity; and the deep learning model is constructed according to the multi-source data, the first matrix and the second matrix.

[0061] The target region represents the region at the time of disaster occurrence, and the destination region represents the destination of the residents in the target region after the disaster occurrence. The first mobile phone signaling data and the third mobile phone signaling data can be used to determine the resident quantity during the pre-disaster period and the post-disaster period. The first matrix is constructed according to the two kinds of data, which represents the resident quantity in the target region that moves from one city to another city during the pre-disaster period. Similarly, the second matrix represents the resident quantity in the target region that moves from one city to another city during the disaster period. Further, in order to compare the linear correlation of the two period matrices, the Pearson correlation coefficient can be used to quantify the linear relationship, thereby providing better samples for subsequent model training.

[0062] In step S40, the first mobile phone signaling data and the second mobile phone signaling data are subjected to feature extraction to obtain a plurality of feature vectors and a plurality of geographic distances, and all the feature vectors and all the geographic distances are input into the deep learning model for training to obtain a population migration model.

[0063] The feature vectors include migration frequency and average migration distance. The migration frequency represents how many residents move out of the target region per day, and the average migration distance represents the distance moved by the residents within a day. According to the mobile phone signaling data of the users, in addition to the feature vectors and the geographic distances, community space information and social and economic data in the target region can also be acquired for model training, thereby more finely describing the migration flow of different communities or regions within the city and enhancing the spatial resolution of the model.

[0064] Specifically, feature extraction is performed on all the first mobile phone signaling data to obtain first coordinate data of all the target residents in the target region within the first preset time; feature extraction is performed on all the second mobile phone signaling data to obtain second coordinate data of all the target residents in the target region within the second preset time; the migration frequency, the average migration distance, and the geographic distance of all the target residents are calculated according to all the first coordinate data and all the second coordinate data; and the migration frequency, the average migration distance, and the geographic distance are input into the deep learning model for training to obtain a population migration model.

[0065] In the method, a deep learning model is constructed, first coordinate data of residents in a target region during a pre-disaster period is extracted from first mobile phone signaling data, second coordinate data of residents in the target region during a disaster period is extracted from second mobile phone signaling data, migration frequency, average migration distance, and geographic distance (i.e., geographic distance of each resident from a starting position in the target region to a destination region) in the target region are calculated, and the migration frequency, the average migration distance, and the geographic distance are input into the constructed deep learning model for training to obtain a population migration model for generating a population migration flow. In the training process, cross entropy can be used as a loss function, and multiple cycles (for example, 20 cycles) of training can be performed. The optimizer selects an RMSProp (Root Mean Square Propagation) algorithm, the momentum of the training process can be set to 0.9, the learning rate is 5x10-6, and the batch size is 64 starting positions. Further, a negative sampling method can be used to randomly select up to 512 target positions for each starting position, thereby reducing the training time.

[0066] Further, in the training process, the mobile phone signaling data, community spatial information, and socio-economic data of residents are integrated to establish a multi-dimensional and multi-level disaster population migration flow prediction model, so that the prediction result is more accurate and reliable.

[0067] Step S50, a plurality of current mobile phone signaling data of a plurality of residents in the target region are obtained and input into the population migration model, and a population migration flow in the target region is output.

[0068] Specifically, a plurality of current mobile phone signaling data of a plurality of residents in the target region are obtained, and current coordinate information in each of the current mobile phone signaling data is extracted, wherein the current coordinate information represents a current position of a resident corresponding to the current mobile phone signaling data; each of the current coordinate information is input into the population migration model for prediction, and a population migration flow of the target region is output, wherein the population migration flow represents a predicted population quantity of residents in the target region migrating to the destination region.

[0069] According to the current mobile phone signaling data of the residents in the target area, an input feature vector (current coordinate information) is constructed, including the feature vector of the starting position, the feature vector of the destination position and the geographical distance between the two, which is input into the population migration model to output the population migration flow of the target area. The generated migration flow data can provide accurate data support for urban emergency management and post-disaster reconstruction, help to develop more effective disaster response strategies and resource allocation schemes, and improve the disaster response capability and resilience of the city.

[0070] Further, each of the current coordinate information is input into the feedforward neural network of the population migration model, and the feedforward neural network analyzes the current coordinate information through an activation function to output a prediction score of the target area; and the population migration model converts the prediction score into the population migration flow through a preset function and outputs the population migration flow.

[0071] The model uses input features to calculate the migration scores of each destination in a specific area, and outputs a migration score vector, which can be used for subsequent migration flow analysis or probability conversion. Specifically, the model calculation process includes three steps: 1. Construct an input feature vector; 2. The Kolmogorov-Arnold network in the model realizes complex nonlinear mapping through univariate function decomposition and combination; 3. The migration scores output by the model are further normalized or ranked as needed to analyze the trend of the migration flow.

[0072] The feedforward neural network is constructed using Kolmogorov-Arnold networks, which are a type of neural network model designed based on the Kolmogorov-Arnold representation theorem, with 15 hidden layers, of which the bottom 6 layers are 256-dimensional and the remaining 9 layers are 128-dimensional. The width parameter of the Kolmogorov-Arnold network can be set to [input_dim, 128, 64, n_destinations], the grid parameter can be set to 5 to control the complexity of the edge function, and the k parameter can be set to 3 to control the number of connections per node.

[0073] After the feedforward neural network outputs the prediction score of the target area, an activation function (such as softmax, a normalized exponential function) is used to convert the prediction score into a probability value, thereby generating a population migration flow. The activation function is a learnable edge function, and each edge uses an independent activation function. The network finally outputs a migration score vector s = [s1, s2, …, sn] for each destination, which reflects the attractiveness or migration trend of each destination.

[0074] The present application can accurately capture the dynamic changes of residents during disasters, more finely depict the migration flow of different communities or regions in the city, improve the data accuracy and timeliness, and realize the accurate prediction of the migration flow of urban residents during disasters.

[0075] Further, as shown in the figure, Figure 2 Based on the above deep learning-based population migration flow generation method, the present application also correspondingly provides a deep learning-based population migration flow generation system, wherein the deep learning-based population migration flow generation system comprises:

[0076] The data preprocessing module 51 is configured to obtain a plurality of initial mobile phone signaling data of each resident in the target area, filter a plurality of target residents according to all the initial mobile phone signaling data, and preprocess all the initial mobile phone signaling data of all the target residents to obtain a plurality of mobile phone signaling data of each target resident.

[0077] The data classification module 52 is configured to obtain meteorological data of the target area, classify all the mobile phone signaling data into a plurality of first mobile phone signaling data and a plurality of second mobile phone signaling data according to the meteorological data, and obtain a plurality of third mobile phone signaling data of all the target residents.

[0078] The model construction module 53 is configured to construct a first matrix according to a plurality of the first mobile phone signaling data and a plurality of the third mobile phone signaling data, construct a second matrix according to a plurality of the second mobile phone signaling data and a plurality of the third mobile phone signaling data, obtain a plurality of source data, and construct a deep learning model according to the plurality of source data, the first matrix and the second matrix.

[0079] The model training module 54 is configured to extract features from a plurality of the first mobile phone signaling data and a plurality of the second mobile phone signaling data to obtain a plurality of feature vectors and a plurality of geographic distances, input all the feature vectors and all the geographic distances into the deep learning model for training to obtain a population migration model.

[0080] The migration prediction module 55 is configured to obtain a plurality of current mobile phone signaling data of a plurality of residents in the target area, input the plurality of current mobile phone signaling data into the population migration model, and output the population migration flow in the target area.

[0081] Further, as shown in the figure, Figure 3 Based on the above deep learning-based population migration flow generation method and system, the present application also correspondingly provides a terminal, which comprises a processor 10, a memory 20 and a display 30. Figure 3Only some components of the terminal are shown, but it should be understood that all the shown components are not required, and more or less components can be implemented instead.

[0082] The memory 20 can be an internal storage unit of the terminal in some embodiments, such as a hard disk or a memory of the terminal. The memory 20 can also be an external storage device of the terminal in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the terminal. Further, the memory 20 can include both the internal storage unit and the external storage device of the terminal. The memory 20 is used to store application software and various data installed on the terminal, such as program codes of the terminal, etc. The memory 20 can also be used to temporarily store data that has been output or will be output. In an embodiment, the memory 20 stores a deep learning-based population migration flow generation program 40, which can be executed by the processor 10 to implement the deep learning-based population migration flow generation method in the present application.

[0083] The processor 10 can be a Central Processing Unit (CPU), a microprocessor or other data processing chip in some embodiments, which is used to run program codes or process data stored in the memory 20, such as to execute the deep learning-based population migration flow generation method, etc.

[0084] The display 30 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, etc. in some embodiments. The display 30 is used to display information of the terminal and to display visualized user interfaces. The components 10-30 of the terminal communicate with each other through a system bus.

[0085] In an embodiment, the following steps are implemented when the processor 10 executes the deep learning-based population migration flow generation program 40 in the memory 20:

[0086] A plurality of initial mobile phone signaling data of each resident in a target area are obtained, a plurality of target residents are filtered out according to all the initial mobile phone signaling data, and all the initial mobile phone signaling data of all the target residents are preprocessed to obtain a plurality of mobile phone signaling data of each target resident;

[0087] acquire meteorological data of the target area, divide all the mobile signaling data into a plurality of first mobile signaling data and a plurality of second mobile signaling data according to the meteorological data, and acquire a plurality of third mobile signaling data of all the target residents;

[0088] construct a first matrix according to the plurality of first mobile signaling data and the plurality of third mobile signaling data, construct a second matrix according to the plurality of second mobile signaling data and the plurality of third mobile signaling data, and acquire multi-source data, construct a deep learning model according to the multi-source data, the first matrix and the second matrix;

[0089] extract features from the plurality of first mobile signaling data and the plurality of second mobile signaling data to obtain a plurality of feature vectors and a plurality of geographic distances, and input all the feature vectors and all the geographic distances into the deep learning model for training to obtain a population migration model;

[0090] acquire a plurality of current mobile signaling data of a plurality of residents in the target area, and input the plurality of current mobile signaling data into the population migration model to output a population migration flow in the target area.

[0091] In the method, the plurality of initial mobile signaling data of each resident in the target area is acquired, a plurality of target residents is screened out according to all the initial mobile signaling data, and all the initial mobile signaling data of all the target residents is preprocessed to obtain a plurality of mobile signaling data of each target resident, and the preprocessing specifically includes:

[0092] all initial mobile signaling data of all residents in the target area within a preset time is acquired, and if a plurality of initial mobile signaling data of a resident is complete, the resident is taken as a target resident, wherein the initial mobile signaling data complete means that initial mobile signaling data of the target resident is complete every day within the preset time in the target area;

[0093] all the initial mobile signaling data of all the target residents is uniformly processed, and data of each activity track anomaly is deleted to obtain a plurality of mobile signaling data of each target resident within the preset time.

[0094] In the method, the meteorological data of the target area is acquired, all the mobile signaling data is divided into a plurality of first mobile signaling data and a plurality of second mobile signaling data according to the meteorological data, and a plurality of third mobile signaling data of all the target residents is acquired, and the method specifically includes:

[0095] acquire meteorological data of the target area within the preset time, and divide the preset time into a first preset time and a second preset time according to the meteorological data;

[0096] Mark all the mobile phone signaling data in the first preset time as a plurality of first mobile phone signaling data, and mark all the mobile phone signaling data in the second preset time as a plurality of second mobile phone signaling data;

[0097] Obtain a plurality of third mobile phone signaling data of all the target residents, wherein the third mobile phone signaling data represents the mobile phone signaling data of the target residents after migrating to the target area after the preset time.

[0098] According to the plurality of first mobile phone signaling data and the plurality of third mobile phone signaling data, a first matrix is constructed, according to the plurality of second mobile phone signaling data and the plurality of third mobile phone signaling data, a second matrix is constructed, and multi-source data is obtained, according to the multi-source data, the first matrix and the second matrix, a deep learning model is constructed, and specifically includes:

[0099] According to all the first mobile phone signaling data, a first target resident quantity of the target residents in the target area within the first preset time is determined;

[0100] According to all the second mobile phone signaling data, a second target resident quantity of the target residents in the target area within the second preset time is determined;

[0101] According to all the third mobile phone signaling data, a third target resident quantity of the target residents in the target area after the preset time is determined;

[0102] Obtain multi-source data in the target area, construct a first matrix according to the first target resident quantity and the third target resident quantity, and construct a second matrix according to the second target resident quantity and the third target resident quantity;

[0103] According to the multi-source data, the first matrix and the second matrix, a deep learning model is constructed.

[0104] The feature vector includes: migration frequency and average migration distance;

[0105] The plurality of first mobile phone signaling data and the plurality of second mobile phone signaling data are extracted to obtain a plurality of feature vectors and a plurality of geographic distances, and all the feature vectors and all the geographic distances are input into the deep learning model for training to obtain a population migration model, and specifically includes:

[0106] The first coordinate data of all the target residents in the target area within the first preset time is obtained by extracting the features of all the first mobile phone signaling data;

[0107] characteristic extraction is performed on all the second mobile phone signaling data to obtain second coordinate data of all the target residents in the target region within the second preset time;

[0108] According to all the first coordinate data and all the second coordinate data, the migration frequency, the average migration distance, and the geographic distance of all the target residents are calculated.

[0109] The migration frequency, the average migration distance, and the geographic distance are input into the deep learning model for training to obtain a population migration model.

[0110] The method comprises the following steps:

[0111] A plurality of current mobile phone signaling data of a plurality of residents in the target region are obtained, and current coordinate information in each current mobile phone signaling data is extracted, wherein the current coordinate information represents the current position of the resident corresponding to the current mobile phone signaling data.

[0112] Each current coordinate information is input into the population migration model for prediction, and a population migration flow of the target region is output, wherein the population migration flow represents the predicted population number of the residents in the target region migrating to the target region.

[0113] The method comprises the following steps:

[0114] Each current coordinate information is input into a feedforward neural network of the population migration model, the feedforward neural network analyzes the current coordinate information through an activation function, and a predicted score of the target region is output.

[0115] The population migration model converts the predicted score into the population migration flow through a preset function and outputs the population migration flow.

[0116] The application further provides a computer readable storage medium, wherein the computer readable storage medium stores a deep learning-based population migration flow generation program, and the deep learning-based population migration flow generation program, when executed by a processor, implements the steps of the deep learning-based population migration flow generation method.

[0117] In summary, the present application provides a population migration flow generation method based on deep learning and related equipment, the method comprising: obtaining multiple initial mobile phone signaling data of each resident in a target area, screening multiple target residents according to all the initial mobile phone signaling data, and preprocessing all the initial mobile phone signaling data of all the target residents to obtain multiple mobile phone signaling data of each target resident; obtaining meteorological data of the target area, dividing all the mobile phone signaling data into multiple first mobile phone signaling data and multiple second mobile phone signaling data according to the meteorological data, and obtaining multiple third mobile phone signaling data of all the target residents; constructing a first matrix according to multiple first mobile phone signaling data and multiple third mobile phone signaling data, constructing a second matrix according to multiple second mobile phone signaling data and multiple third mobile phone signaling data, and obtaining multiple source data, constructing a deep learning model according to the multiple source data, the first matrix and the second matrix; extracting features from multiple first mobile phone signaling data and multiple second mobile phone signaling data to obtain multiple feature vectors and multiple geographic distances, and inputting all the feature vectors and all the geographic distances into the deep learning model for training to obtain a population migration model; obtaining multiple current mobile phone signaling data of multiple residents in the target area and inputting the multiple current mobile phone signaling data into the population migration model to output a population migration flow in the target area. The present application integrates multiple source data and deep learning technology to construct a population migration model for predicting population migration flow, can accurately capture the dynamic changes of residents during disasters, more accurately describes the migration flow of different communities or regions in the city, improves data accuracy and timeliness, and realizes accurate prediction of population migration flow of urban residents during disasters.

[0118] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that processes, methods, articles, or terminals that comprise a list of elements do not only include those elements, but also other elements that are not expressly listed, or other elements inherent in such processes, methods, articles, or terminals. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or terminal that includes the element.

[0119] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable computer-readable storage medium, and the program can include the processes of the above-mentioned method embodiments when executed. The computer-readable storage medium can be a memory, a disk, an optical disk, etc.

[0120] It is to be understood that the application is not limited to the examples described above, which can be modified or adapted in several ways by those skilled in the art without departing from the scope of the present application, as defined by the appended claims.

Claims

1. A deep learning-based population migration flow generation method, characterized in that, The population migration flow generation method based on deep learning comprises: Obtaining multiple initial mobile signaling data of each resident in a target area, screening multiple target residents according to all the initial mobile signaling data, and preprocessing all the initial mobile signaling data of all the target residents to obtain multiple mobile signaling data of each target resident; Obtaining meteorological data of the target area, dividing all the mobile signaling data into multiple first mobile signaling data and multiple second mobile signaling data according to the meteorological data, and obtaining multiple third mobile signaling data of all the target residents; Constructing a first matrix according to multiple first mobile signaling data and multiple third mobile signaling data, constructing a second matrix according to multiple second mobile signaling data and multiple third mobile signaling data, and obtaining multiple source data, constructing a deep learning model according to the multiple source data, the first matrix and the second matrix; The method comprises the following steps: Determining the first target resident number of the target residents in the target area within a first preset time according to all the first mobile signaling data; Determining the second target resident number of the target residents in the target area within a second preset time according to all the second mobile signaling data; Determining the third target resident number of the target residents in the target area after a preset time according to all the third mobile signaling data; Obtaining multiple source data in the target area, constructing a first matrix according to the first target resident number and the third target resident number, and constructing a second matrix according to the second target resident number and the third target resident number; Constructing a deep learning model according to the multiple source data, the first matrix and the second matrix; Performing feature extraction on multiple first mobile signaling data and multiple second mobile signaling data to obtain multiple feature vectors and multiple geographic distances, inputting all the feature vectors and all the geographic distances into the deep learning model for training to obtain a population migration model; Obtaining multiple current mobile signaling data of multiple residents in the target area and inputting the data into the population migration model to output the population migration flow in the target area. 2.The deep learning-based population migration flow generation method of claim 1, wherein, The method comprises the following steps: Obtaining multiple initial mobile signaling data of each resident in a target area, screening multiple target residents according to all the initial mobile signaling data, and preprocessing all the initial mobile signaling data of all the target residents to obtain multiple mobile signaling data of each target resident; Obtaining all initial mobile signaling data of all residents in a target area within a preset time, if there are multiple initial mobile signaling data of a resident, the resident is taken as a target resident, wherein the initial mobile signaling data completeness indicates that the initial mobile signaling data of the target resident in the target area within the preset time is complete every day; Uniformly processing the format of all initial mobile signaling data of all target residents, and deleting data of each activity track exception, to obtain multiple mobile signaling data of each target resident within the preset time. 3.The deep learning-based population migration flow generation method of claim 2, wherein, The weather data of the target area is obtained, all mobile signaling data is divided into multiple first mobile signaling data and multiple second mobile signaling data according to the weather data, and multiple third mobile signaling data of all target residents are obtained, specifically including: Obtaining the weather data of the target area within the preset time, and dividing the preset time into a first preset time and a second preset time according to the weather data; Marking all mobile signaling data in the first preset time as multiple first mobile signaling data, and marking all mobile signaling data in the second preset time as multiple second mobile signaling data; Obtaining multiple third mobile signaling data of all target residents, wherein the third mobile signaling data represents the mobile signaling data of the target resident after migrating to the target area after the preset time. 4.The deep learning-based population migration flow generation method of claim 3, wherein, The feature vector includes: migration frequency and average migration distance; The feature extraction is performed on multiple first mobile signaling data and multiple second mobile signaling data to obtain multiple feature vectors and multiple geographic distances, and all feature vectors and all geographic distances are input into the deep learning model for training to obtain a population migration model, specifically including: Feature extraction is performed on all first mobile signaling data to obtain first coordinate data of all target residents in the target area within the first preset time; Feature extraction is performed on all second mobile signaling data to obtain second coordinate data of all target residents in the target area within the second preset time; According to all first coordinate data and all second coordinate data, the migration frequency, the average migration distance and the geographic distance of all target residents are calculated; The migration frequency, the average migration distance and the geographic distance are input into the deep learning model for training to obtain a population migration model. 5.The deep learning-based population migration flow generation method of claim 1, wherein, Obtaining multiple current mobile signaling data of multiple residents in the target area, and inputting into the population migration model to output the population migration flow in the target area, specifically including: Obtaining multiple current mobile signaling data of multiple residents in the target area, and extracting current coordinate information in each current mobile signaling data, wherein the current coordinate information represents the current position of the corresponding resident of the current mobile signaling data; inputting each of the current coordinate information into the population migration model for prediction, and outputting a population migration flow of the target area, wherein the population migration flow represents a predicted population number of residents in the target area migrating to the target area. 6.The deep learning-based population migration flow generation method of claim 5, wherein, The inputting each of the current coordinate information into the population migration model for prediction, and outputting a population migration flow of the target area, specifically comprises: inputting each of the current coordinate information into a feedforward neural network of the population migration model, the feedforward neural network analyzing the current coordinate information through an activation function, and outputting a predicted score of the target area; The population migration model converts the predicted score into the population migration flow through a preset function and outputs.

7. A deep learning based population migration flow generation system, characterized by, The deep learning-based population migration flow generation system applied to the deep learning-based population migration flow generation method of any one of claims 1-6, the deep learning-based population migration flow generation system comprising: a data preprocessing module configured to obtain a plurality of initial mobile phone signaling data of each resident in a target area, filter a plurality of target residents according to all the initial mobile phone signaling data, and preprocess all the initial mobile phone signaling data of all the target residents to obtain a plurality of mobile phone signaling data of each target resident; a data classification module configured to obtain meteorological data of the target area, classify all the mobile phone signaling data into a plurality of first mobile phone signaling data and a plurality of second mobile phone signaling data according to the meteorological data, and obtain a plurality of third mobile phone signaling data of all the target residents; a model construction module configured to construct a first matrix according to a plurality of the first mobile phone signaling data and a plurality of the third mobile phone signaling data, construct a second matrix according to a plurality of second mobile phone signaling data and a plurality of the third mobile phone signaling data, and obtain multi-source data, and construct a deep learning model according to the multi-source data, the first matrix, and the second matrix; a model training module configured to extract features from a plurality of the first mobile phone signaling data and a plurality of the second mobile phone signaling data to obtain a plurality of feature vectors and a plurality of geographic distances, input all the feature vectors and all the geographic distances into the deep learning model for training to obtain a population migration model; a migration prediction module configured to obtain a plurality of current mobile phone signaling data of a plurality of residents in the target area, and input the plurality of current mobile phone signaling data into the population migration model to output a population migration flow in the target area.

8. A terminal, characterized by comprising: The terminal comprises a memory, a processor, and a deep learning-based population migration flow generation program stored on the memory and executable on the processor, and the deep learning-based population migration flow generation program, when executed by the processor, implements the steps of the deep learning-based population migration flow generation method of any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a deep learning-based population migration flow generation program, and the deep learning-based population migration flow generation program, when executed by a processor, implements the steps of the deep learning-based population migration flow generation method of any one of claims 1-6.

Images