Method and system for predicting mobile traffic of affected people based on mixed deep learning

Abstract

The invention provides a disaster-affected crowd movement traffic prediction method and system based on hybrid deep learning, and the method comprises the steps: obtaining the original movement traffic data of a target disaster-affected crowd, carrying out the data preprocessing of the original movement traffic data, obtaining the preprocessed movement traffic data, and carrying out the prediction of the movement traffic of the target disaster-affected crowd. Performing heterogeneity factor calculation on the pre-processed mobile traffic data to obtain mobile heterogeneity data, then performing spatial feature extraction on the pre-processed mobile traffic data based on a graph convolutional neural network algorithm to obtain mobile spatial feature data, and then performing spatial feature extraction on the pre-processed mobile traffic data based on a long-short-term memory network algorithm to obtain mobile spatial feature data; the method comprises the following steps: performing time feature extraction on spatial feature movement data and movement heterogeneity data to obtain movement time feature data, and finally performing crowd movement flow prediction on the movement spatial feature data and the movement time feature data according to a Seq2Seq sequence crowd prediction model to obtain a movement flow prediction result. And the accuracy and the reliability of the movement prediction of the crowd in the disaster are improved.

Description

Technical Field

[0001] The present application relates to the field of crowd movement prediction, and in particular to a method and system for predicting the movement flow of disaster-affected people based on hybrid deep learning. Background Art

[0002] In densely populated modern cities, the mobility patterns of people are of great significance for urban planning, traffic management and emergency response. Especially in recent years, under the frequent extreme weather events such as heavy rain, heavy snow and typhoons, in-depth understanding of the complex mobility patterns of urban people during extreme weather events and accurate prediction of the mobility patterns of people can significantly improve the efficiency of emergency response and reduce casualties and property losses.

[0003] In the existing technology, the traditional crowd movement prediction has the following main shortcomings: (1) Insufficient ability to handle nonlinear and complex flow patterns: Traditional model prediction methods perform poorly when dealing with complex nonlinear flow patterns. For example, although ARIMA and VAR models can capture certain time series characteristics, their prediction performance significantly decreases when faced with complex changes caused by disasters. These methods often assume that the changes in data are linear and cannot effectively handle the nonlinear characteristics of crowd flow during disasters. (2) Most deep learning models related to crowd movement prediction are directly transferred from the fields of image recognition or natural language processing: These models ignore the possible contribution of geographic information to prediction and fail to make full use of geographic information for more accurate predictions. (3) They are unable to adapt to crowd movement prediction in special disaster scenarios, especially for abnormal patterns under the influence of disasters such as urban floods. They cannot accurately predict crowd movement in sudden disaster scenarios and lack effective processing of abnormal movement patterns. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method and system for predicting the mobile flow of disaster-stricken people based on hybrid deep learning, which extracts spatiotemporal features through graph convolutional neural networks and long short-term memory networks, introduces heterogeneous factors for prediction, and forms a spatiotemporal heterogeneous graph convolution prediction module, which effectively captures the differences in crowd mobility patterns in different areas under special disaster scenarios, and improves the accuracy and reliability of crowd mobility prediction during disasters.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for predicting mobile traffic of disaster-affected people based on hybrid deep learning, comprising: Obtaining raw mobile traffic data of the target disaster-affected population; Preprocessing the original mobile traffic data to obtain preprocessed mobile traffic data of the target disaster-affected population; Based on the centrality measurement algorithm of undirected network nodes, the heterogeneity factor of the pre-processed mobile traffic data is calculated to obtain the mobile heterogeneity data of the target disaster-affected population, where the mobile heterogeneity data includes the degree centrality index data of the target disaster-affected population, the closeness centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population. Based on the graph convolutional neural network algorithm, the spatial features of the pre-processed mobile traffic data are extracted to obtain the mobile spatial feature data of the target disaster-affected population; Based on the long short-term memory network algorithm, the temporal features of spatial feature movement data and movement heterogeneity data are extracted to obtain the movement time feature data of the target disaster-affected population; According to the Seq2Seq sequence crowd prediction model, the crowd movement flow is predicted based on the mobile space feature data and mobile time feature data to obtain the mobile flow prediction results of the target disaster-stricken population.

[0006] Furthermore, in some embodiments, the original mobile traffic data includes the original time series data of the movement of the target disaster-affected population and the original geographical location data of the target disaster-affected population, and the original mobile traffic data is preprocessed to obtain the preprocessed mobile traffic data of the target disaster-affected population, including: Perform data cleaning on the original time series data to obtain stable time series data; Clean the original geographic location data to obtain stable geographic location data; Perform graph data abstraction on the geographical location stationary data and time series stationary data to obtain the spatiotemporal graph data of the target disaster-affected population; Calculate the adjacency matrix of the spatiotemporal graph data to obtain the spatiotemporal adjacency matrix of the target disaster-affected population; The data type of the spatiotemporal adjacency matrix of the target disaster-affected population is converted to obtain the preprocessed mobile flow data.

[0007] Furthermore, in some embodiments, the geographical location stationary data and the time series stationary data are subjected to graph data abstraction processing to obtain the spatiotemporal graph data of the target disaster-affected population, including: Based on the geographical location stable data, multiple disaster-affected area nodes of the target disaster-affected population are determined; According to the geographical proximity between the nodes of each disaster area, multiple disaster area edges of the target disaster-stricken population are determined; Determine the undirected network graph data of the target disaster-affected population based on each disaster-affected area node and each disaster-affected area edge; The time series stationary data is fused into the undirected network graph data in the time dimension to obtain the space-time graph data.

[0008] Furthermore, in some embodiments, based on the centrality measurement algorithm of undirected network nodes, the heterogeneity factor is calculated for the pre-processed mobile traffic data to obtain the mobile heterogeneity data of the target disaster-affected population, including: The degree centrality of undirected network nodes is calculated for the pre-processed mobile traffic data to obtain degree centrality index data; The closeness centrality of undirected network nodes is calculated for the pre-processed mobile traffic data to obtain the closeness centrality index data; The betweenness centrality of undirected network nodes is calculated for the pre-processed mobile traffic data to obtain the betweenness centrality index data; The heterogeneity vectors of degree centrality index data, closeness centrality index data and betweenness centrality index data are calculated to obtain moving heterogeneity data.

[0009] Further, in some embodiments, the Seq2Seq sequence population prediction model is obtained by the following steps: Initialize the original Seq2Seq sequence population model, which includes an LSTM model encoder and an LSTM model decoder; Obtaining original mobile traffic data of sample disaster-affected people, and performing data preprocessing on the original mobile traffic data to obtain sample mobile traffic data; Dividing the mobile traffic sample data into mobile traffic training data and mobile traffic verification data; According to the mobile traffic training data, the Seq2Seq sequence population original model is iteratively trained multiple times to obtain the Seq2Seq sequence population training model; According to the mobile traffic verification data, the Seq2Seq sequence population training model is verified and optimized to obtain the Seq2Seq sequence population prediction model.

[0010] Furthermore, in some embodiments, the mobile traffic sample data includes time series sample data of the movement of the sample disaster-affected population and geographic location sample data of the sample disaster-affected population, and the mobile traffic sample data is divided into mobile traffic training data and mobile traffic verification data, including: According to the time series sample data, determine the disaster-affected time point of the sample disaster-affected population; Based on the time point of the disaster, the mobile traffic sample data is divided into the mobile traffic sample data of the sample disaster-affected population before the disaster, the mobile traffic sample data of the sample disaster-affected population during the disaster and after the disaster; Based on the cross-validation method, the mobile traffic sample data of a part of the sample disaster-affected population before the disaster are determined as the mobile traffic training data, and the mobile traffic sample data of another part of the sample disaster-affected population before the disaster and the mobile traffic sample data of the sample disaster-affected population during and after the disaster are determined as the mobile traffic verification data.

[0011] Furthermore, in some embodiments, the Seq2Seq sequence population original model is iteratively trained multiple times according to the mobile traffic training data to obtain a Seq2Seq sequence population training model, including: In each iterative training process of the Seq2Seq sequence population original model, the loss value between the predicted output value and the true output value of the Seq2Seq sequence population original model of the current iterative training is calculated according to the mean square error loss function to obtain the loss value of the population prediction model of each iteration; In each iterative training process of the original Seq2Seq sequence crowd model, the gradient of the loss value of the crowd prediction model of the current iterative training is calculated according to the back propagation algorithm to obtain the corresponding crowd prediction model gradient value; During each iteration training process of the Seq2Seq sequence population original model, the model parameters of the Seq2Seq sequence population original model of the current iteration training are updated according to the population prediction model gradient value of the current iteration training; In each iterative training process of the original model of the Seq2Seq sequence population, if the model parameters of the current iterative training meet the preset training requirements, the iterative training of the original model of the Seq2Seq sequence population is terminated to obtain the Seq2Seq sequence population training model.

[0012] To achieve the above purpose, a second aspect of the embodiment of the present application proposes a disaster-affected population mobile flow prediction system, comprising: A data acquisition module, which is used to acquire the original mobile traffic data of the target disaster-affected population; A data preprocessing module, the data preprocessing module is in communication with the data acquisition module, and the data preprocessing module is used to preprocess the original mobile traffic data to obtain preprocessed mobile traffic data of the target disaster-affected population; A heterogeneity module, the heterogeneity module is in communication connection with the data preprocessing module, and the heterogeneity module is used to calculate the heterogeneity factor of the preprocessed mobile traffic data based on the centrality measurement algorithm of the undirected network nodes, so as to obtain the mobile heterogeneity data of the target disaster-affected population, wherein the mobile heterogeneity data includes the degree centrality index data of the target disaster-affected population, the proximity centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population; The spatiotemporal prediction module is connected to the data preprocessing module in communication, and the spatiotemporal prediction module is connected to the heterogeneity module in communication, wherein: The spatiotemporal prediction module is used to extract spatial features of pre-processed mobile traffic data based on the graph convolutional neural network algorithm to obtain the mobile spatial feature data of the target disaster-affected population; The spatiotemporal prediction module is also used to extract temporal features from spatial feature mobility data and mobility heterogeneity data based on the long short-term memory network algorithm to obtain the mobility temporal feature data of the target disaster-affected population; The spatiotemporal prediction module is also used to predict the crowd movement flow based on the mobile space feature data and the mobile time feature data according to the Seq2Seq sequence crowd prediction model, and obtain the mobile flow prediction result of the target disaster-stricken population.

[0013] To achieve the above-mentioned purpose, the third aspect of the embodiments of the present application proposes an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the method for predicting the mobile flow of disaster-stricken people in the first aspect of the embodiments when executing the computer program.

[0014] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a storage medium, which is a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method for predicting the mobile flow of disaster-stricken people in the first aspect of the embodiments.

[0015] The embodiments of the present application have the following beneficial effects: the present application obtains the original mobile traffic data of the target disaster-affected population, and then performs data preprocessing on the original mobile traffic data to obtain the preprocessed mobile traffic data of the target disaster-affected population; then, based on the centrality measurement algorithm of the undirected network nodes, the heterogeneity factor is calculated on the preprocessed mobile traffic data to obtain the mobile heterogeneity data of the target disaster-affected population, wherein the mobile heterogeneity data includes the degree centrality index data of the target disaster-affected population, the closeness centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population; and further, by introducing the heterogeneity factor, the contribution of different special disaster scenarios to the mobility patterns of the population can be fully improved, thereby improving the processing capability of abnormal mobility patterns.

[0016] Furthermore, based on the graph convolutional neural network algorithm, spatial features are extracted from the preprocessed mobile flow data to obtain the mobile spatial feature data of the target disaster-stricken population. Then, based on the long short-term memory network algorithm, temporal features are extracted from the spatial feature mobile data and the mobile heterogeneity data to obtain the mobile time feature data of the target disaster-stricken population. Then, spatiotemporal features can be extracted through graph convolutional neural networks and long short-term memory networks to capture more detailed spatiotemporal dependencies. Finally, according to the Seq2Seq sequence crowd prediction model, crowd mobile flow is predicted for the mobile spatial feature data and the mobile time feature data to obtain the mobile flow prediction results of the target disaster-stricken population. Then, the Seq2Seq sequence architecture model can be used to process complex nonlinear flow data, which significantly improves the accuracy and reliability of crowd movement prediction during disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of a method for predicting mobile traffic of disaster-affected people based on hybrid deep learning provided by some embodiments of the present application; Figure 2 It is a data flow diagram of a method for predicting mobile traffic of disaster-affected people based on hybrid deep learning provided by some embodiments of the present application; Figure 3 Other embodiments of the present application provide Figure 1 Flow chart of step S120 in FIG. Figure 4 Other embodiments of the present application provide Figure 3 Flow chart of step S330 in FIG. Figure 5 Other embodiments of the present application provide Figure 1 Flow chart of step S130 in FIG. Figure 6 is a flowchart of generating a Seq2Seq sequence population prediction model provided by other embodiments of the present application; Figure 7 Some embodiments of the present application provide Figure 6 Flowchart of step S630; Figure 8 Some embodiments of the present application provide Figure 6 Flowchart of step S640; Fig. 9 A schematic diagram of the structure of a disaster-affected population mobile flow prediction system provided in some embodiments of the present application; Fig.10 It is a schematic diagram of the hardware structure of an electronic device provided in some embodiments of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] In the description of the present application, it should be understood that descriptions involving orientation, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0020] It should also be noted that in the description of this application, "several" means more than one, "more" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used to distinguish the technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0022] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0023] First, some nouns involved in this application are analyzed: Graph Convolution Networks (GCN) is an important model in the field of deep learning, which is specially used to process graph structure data. Graph structure data consists of nodes and edges, and is widely used in many fields such as crowd mobility networks and transportation networks. Graph convolution neural networks extract and aggregate features on graph structure data by simulating convolution operations, thereby realizing the analysis and prediction of graph data.

[0024] A node is a basic element in a graph and can represent an entity (such as a person or an object), a concept, or anything that can be associated with other elements.

[0025] An edge is a line connecting two nodes in a graph, which can represent the relationship or interaction between nodes. An edge can be directed or undirected, and can also carry weights.

[0026] Degree is the number of edges directly connected to a node in the graph. For an undirected graph, the degree of a node is the total number of edges connected to it; for a directed graph, the degree of a node is usually divided into in-degree and out-degree, where in-degree is the number of edges pointing to the node, and out-degree is the number of edges starting from the node.

[0027] Neighborhood: The set of all nodes directly connected to a given node. In graph neural networks, the representation of a node is often updated by aggregating information from its neighborhood.

[0028] The adjacency matrix is ​​a two-dimensional array used to represent the adjacency relationship between vertices in a graph. Let G=(V,E) be a graph, where V is the vertex set and E is the edge set. Each element A[i][j] in the adjacency matrix represents the connection between vertex i and vertex j.

[0029] Long Short-Term Memory (LSTM) is a special type of Recurrent Neural Network (RNN) designed to solve the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequence data. LSTM controls the flow of information by introducing a gating mechanism, enabling it to remember long-term dependencies.

[0030] The Sequence-to-Sequence (Seq2Seq) architecture model is a deep learning model architecture widely used in natural language processing (NLP) tasks. It can convert an input sequence into an output sequence and is often used in machine translation, speech recognition and other application scenarios with large amounts of nonlinear streaming data.

[0031] In densely populated modern cities, the mobility patterns of people are of great significance for urban planning, traffic management and emergency response. Especially in recent years, under the frequent extreme weather events such as heavy rain, heavy snow and typhoons, in-depth understanding of the complex mobility patterns of urban people during extreme weather events and accurate prediction of the mobility patterns of people can significantly improve the efficiency of emergency response and reduce casualties and property losses.

[0032] In the prior art, traditional crowd movement prediction generally includes the following methods: 1. Historical average method (HA), which is a simple time series prediction method that predicts future flows by calculating the average of historical data. Although this method is simple to calculate, it ignores the spatiotemporal dependency in the data, especially in disaster situations, and the prediction accuracy is low; 2. Using the autoregressive integrated moving average model (ARIMA), the ARIMA model is a widely used time series prediction method that can capture the autocorrelation characteristics in the data. However, when dealing with nonlinear changes during disasters, the prediction results obtained by the ARIMA model are often inaccurate; 3. Using the support vector regression method model (SVR), the data is fitted by constructing a nonlinear function. When dealing with large-scale data and complex nonlinear relationships, the support vector regression method model shows good performance. However, its computational complexity is high and it is insufficient in spatial feature extraction. 4. Using the fully connected long short-term memory network model (FC-LSTM), the fully connected long short-term memory network model introduces memory units to process long time series data, which can effectively capture time dependencies. However, the model has limitations in processing spatial features and cannot fully utilize spatial information for prediction; 5. Using the graph convolutional neural network model (GCN), the convolution operation can effectively capture spatial features, which is particularly suitable for data with complex spatial structures. However, the GCN model still has room for improvement when dealing with time dependencies and responding to sudden disaster changes.

[0033] In summary, traditional crowd movement prediction has the following main shortcomings: (1) Insufficient ability to handle nonlinear and complex flow patterns: Traditional model prediction methods perform poorly when dealing with complex nonlinear flow patterns. For example, although ARIMA and VAR models can capture certain time series characteristics, their prediction performance significantly decreases when faced with complex changes caused by disasters. These methods often assume that data changes are linear and cannot effectively handle the nonlinear characteristics of crowd flow during disasters. (2) Most deep learning models related to crowd movement prediction are directly transferred from the fields of image recognition or natural language processing: These models ignore the possible contribution of geographic information to prediction and fail to make full use of geographic information for more accurate predictions. (3) They are unable to adapt to crowd movement prediction in special disaster scenarios, especially for abnormal patterns under the influence of disasters such as urban floods. They cannot accurately predict crowd movement in sudden disaster scenarios and lack effective processing of abnormal movement patterns.

[0034] Based on this, an embodiment of the present application provides a method and medium for predicting mobile traffic of disaster-stricken people based on hybrid deep learning. The method obtains the original mobile traffic data of the target disaster-stricken people, and then preprocesses the original mobile traffic data to obtain the preprocessed mobile traffic data of the target disaster-stricken people. Then, based on the centrality measurement algorithm of the undirected network nodes, the heterogeneity factor is calculated for the preprocessed mobile traffic data to obtain the mobile heterogeneity data of the target disaster-stricken people, wherein the mobile heterogeneity data includes the degree centrality index data of the target disaster-stricken people, the closeness centrality index data of the target disaster-stricken people, and the betweenness centrality index data of the target disaster-stricken people. Furthermore, by introducing the heterogeneity factor, the contribution of different special disaster scenarios to the mobility patterns of the people can be fully improved, thereby improving the processing capability of abnormal mobility patterns.

[0035] Furthermore, based on the graph convolutional neural network algorithm, spatial features are extracted from the preprocessed mobile flow data to obtain the mobile spatial feature data of the target disaster-stricken population. Then, based on the long short-term memory network algorithm, temporal features are extracted from the spatial feature mobile data and the mobile heterogeneity data to obtain the mobile time feature data of the target disaster-stricken population. Then, spatiotemporal features can be extracted through graph convolutional neural networks and long short-term memory networks to capture more detailed spatiotemporal dependencies. Finally, according to the Seq2Seq sequence crowd prediction model, crowd mobile flow is predicted for the mobile spatial feature data and the mobile time feature data to obtain the mobile flow prediction results of the target disaster-stricken population. Then, the Seq2Seq sequence architecture model can be used to process complex nonlinear flow data, which significantly improves the accuracy and reliability of crowd movement prediction during disasters.

[0036] In the first aspect, a method for predicting mobile flow of disaster-stricken people based on hybrid deep learning in an embodiment of the present application is first described.

[0037] Reference Figure 1 and Figure 2 As shown, Figure 1 is a flowchart of a method for predicting mobile traffic of disaster-affected people based on hybrid deep learning provided by some embodiments of the present application, Figure 2 It is a data flow diagram of a method for predicting mobile traffic of disaster-stricken people based on hybrid deep learning provided in some embodiments of the present application. The method for predicting mobile traffic of disaster-stricken people based on hybrid deep learning may include but is not limited to steps S110 to S160.

[0038] Step S110: Acquire the original mobile traffic data of the target disaster-affected population.

[0039] It should be noted that the original mobile traffic data includes the original time series data of the movement of the target disaster-stricken people and the original geographical location data of the target disaster-stricken people. The time series original data includes the specific timestamp of each individual data point, the number of people passing through a certain place or area within a period of time, and the movement speed and direction of each individual data point at a certain point in time; the geographical location original data includes the longitude and latitude of the people, and the geographical environment around the people (such as: cities, mountains, plains, seaside, etc.).

[0040] Step S120: pre-processing the original mobile traffic data to obtain pre-processed mobile traffic data of the target disaster-affected population.

[0041] Specifically, the original mobile traffic data is cleaned, and then an undirected network graph is constructed based on the cleaned data to obtain the corresponding graph data. The adjacency matrix of the graph data is then calculated, and the data type of the corresponding adjacency matrix is ​​converted into tensor data that can be processed by deep learning to obtain the preprocessed mobile traffic data of the target disaster-stricken population.

[0042] Step S130: Based on the centrality measurement algorithm of the undirected network nodes, the heterogeneity factor of the pre-processed mobile flow data is calculated to obtain the mobile heterogeneity data of the target disaster-affected population.

[0043] Among them, the mobile heterogeneity data include the degree centrality index data of the target disaster-affected population, the closeness centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population.

[0044] Specifically, based on the centrality measurement algorithm of undirected network nodes, the heterogeneity factor of the preprocessed mobile flow data is calculated to obtain the mobile heterogeneity data of the target disaster-stricken population, and then the spatial heterogeneity characteristics of population movement are accurately extracted, providing multi-dimensional feature data support for subsequent crowd flow prediction, thereby improving the accuracy of subsequent crowd flow prediction.

[0045] It should be noted that mobility heterogeneity data is essentially a data expression of geographic proximity, which reflects the spatial proximity between subdivided areas within the flow space. This data expression aims to capture the spatial heterogeneity characteristics of crowd flow, especially the changes in crowd mobility patterns in different geographical areas. Among them, the reason why geographic proximity can capture heterogeneity is that spatially close areas usually have similar geographical characteristics and environmental conditions. In disaster scenarios such as urban floods, geographic proximity is manifested as the impact of geographical elements (including terrain, road layout, and building distribution) on crowd mobility patterns. Therefore, mapping this proximity to mobility heterogeneity data helps to more accurately capture the changes in geographical characteristics between different regions, so that subsequent model predictions can more accurately reflect the differences in crowd mobility patterns in different regions under disaster scenarios.

[0046] At the same time, the step of calculating mobile heterogeneity data is specially designed for disaster scenarios such as urban floods. It can significantly improve the prediction accuracy of subsequent crowd flow, and then guide relevant emergency departments to better respond to abnormal crowd movement patterns in sudden emergencies and improve the efficiency of emergency management and decision-making.

[0047] Moreover, the step of calculating mobile heterogeneous data can significantly improve the generalization ability of model predictions and enhance the adaptability and robustness of the model to different disaster scenarios. Therefore, subsequent model predictions are not only applicable to flood disasters, but also to other disaster scenarios such as earthquakes and typhoons. Step S140: Based on the graph convolutional neural network algorithm, the spatial features of the pre-processed mobile traffic data are extracted to obtain the mobile spatial feature data of the target disaster-affected population.

[0048] Specifically, based on the graph convolutional neural network algorithm, the graph convolutional neural network model is initialized to preprocess the mobile traffic data. Input into the graph convolutional neural network model in the form of feature matrix to pre-process mobile traffic data Perform spatial feature extraction to obtain the mobile spatial feature data of the target disaster-affected population.

[0049] It should be noted that the graph convolutional neural network model includes multiple layers of graph convolutional neural networks. Each layer of graph convolutional neural network captures the adjacency matrix corresponding to each time segment in the pre-processed mobile traffic data to calculate the spatial characteristics of each audience area node, thereby accurately capturing the spatial dependency characteristics of the pre-processed mobile traffic data corresponding to different time segments.

[0050] For example, for a city graph containing N audience area nodes in a time segment , whose adjacency matrix is , the input feature matrix is ​​the preprocessed mobile traffic data The graph convolutional neural network model is calculated using the following propagation rules:

[0051] in, is the adjacency matrix with self-connection added, is the identity matrix, , is a layer-specific trainable weight matrix, is an activation function, such as , It is The activation matrix of the layer, .

[0052] Step S150: Based on the long short-term memory network algorithm, time feature extraction is performed on the spatial feature movement data and the movement heterogeneity data to obtain the movement time feature data of the target disaster-affected population.

[0053] Specifically, based on the long short-term memory network algorithm, the long short-term memory network model is initialized, and the spatial feature mobility data and mobility heterogeneity data are input into the graph convolutional neural network model in the form of feature matrix to accurately pre-process the mobile traffic data. Extract time features and obtain the movement time feature data of the target disaster-affected population.

[0054] It should be noted that the calculation formula of the long short-term memory network model is as follows:

[0055] in, is the LSTM network unit state, is the hidden state, It's the Gate of Forgetfulness. is the input gate, is the output gate, is the sigmoid function, and is the trainable weight matrix, is the bias term.

[0056] Step S160: According to the Seq2Seq sequence crowd prediction model, the crowd movement flow is predicted for the movement space feature data and the movement time feature data to obtain the movement flow prediction result of the target disaster-affected population.

[0057] Specifically, a Seq2Seq sequence crowd prediction model is obtained, and then the mobile space feature data and the mobile time feature data are input into the Seq2Seq sequence crowd prediction model, so that the Seq2Seq sequence crowd prediction model can predict the future crowd mobility flow, and obtain the mobile flow prediction result of the target disaster-stricken population. Then, the Seq2Seq sequence crowd prediction model is used for analysis and prediction, which can improve the processing capability of data sparsity and nonlinear flow data, and improve the analysis and processing capability of complex and precise mobile space feature data and mobile time feature data, thereby realizing accurate prediction of crowd movement during special disasters.

[0058] Among them, the mobile traffic prediction results include the future movement direction of the current disaster-stricken people, the future movement speed of the current disaster-stricken people, the current number of people affected by the disaster, the future mobile traffic trend of the current disaster-stricken people, etc., which can enable relevant personnel to accurately perform corresponding emergency management and decision-making based on the above data, which is conducive to protecting the safety of people during disasters.

[0059] Reference Figure 3 As shown, Figure 3 Other embodiments of the present application provide Figure 1 In the flowchart of step S120 in FIG. 1 , the method for predicting mobile traffic flow of disaster-affected people may include but is not limited to steps S310 to S350.

[0060] Step S310: Clean the original time series data to obtain stable time series data.

[0061] Specifically, firstly, the missing values ​​of the original time series data are processed, and then the outliers are processed to obtain the first cleaned data of the time series; then, the first cleaned data of the time series are detrended by the difference or linear regression method to obtain the second cleaned data of the time series, so that the short-term fluctuations and periodic changes of the second cleaned data of the time series are more obvious; finally, the second cleaned data of the time series are smoothed by the moving average window to obtain the stable data of the time series, thereby providing high-quality data support for subsequent complex data analysis and improving the accuracy and reliability of the time series related data.

[0062] Step S320: Clean the original geographic location data to obtain stable geographic location data.

[0063] Specifically, the missing values ​​of the original geographic location data are first processed, and then the outliers are processed to obtain the first cleaned data of the geographic location; then, the first cleaned data of the geographic location are regionally aggregated through grid unit division to obtain stable geographic location data, thereby providing high-quality data support for subsequent complex data analysis and improving the accuracy and reliability of geographic location related data.

[0064] Step S330: Perform graph data abstraction processing on the geographical location stationary data and the time series stationary data to obtain the spatiotemporal graph data of the target disaster-affected population.

[0065] Specifically, assuming that the geographical location of the target population is city A, that is, the geographical location stationary data is the relevant geographical location data of the target population in city A, then the entire area of ​​city A in the geographical location stationary data is discretized into a disaster area group, which includes N disaster areas of city A. The data structure of the disaster area group is composed of the disaster area node set And for any two nodes of city A and nodes ,node and nodes The edges between nodes are defined as and nodes The geographical proximity between them.

[0066] Next, a spatiotemporal graph is constructed based on the affected area nodes, affected area edges, and corresponding time series stationary data (such as timestamps, etc.) of the affected area group. , and obtain the corresponding spatiotemporal graph data of the target disaster-affected population, where: is the set of nodes in the disaster area, is the edge set of the disaster-affected area, It is the time information corresponding to the time segment.

[0067] Step S340: Calculate the adjacency matrix of the spatiotemporal graph data to obtain the spatiotemporal adjacency matrix of the target disaster-affected population.

[0068] Specifically, we first obtain each time segment in the spatiotemporal graph data and construct the corresponding static spatial adjacency matrix in each time segment. Then, in the time dimension, we combine the corresponding static spatial adjacency matrices in all time segments into a one-dimensional array in chronological order to obtain the spatiotemporal adjacency matrix of the target disaster-affected population.

[0069] The static spatial adjacency matrix is ​​a two-dimensional array. The rows and columns of the static spatial adjacency matrix represent the nodes of each disaster-stricken area. The elements in the matrix ( , ) represents a node and nodes The connection status in the same time segment, and this element can be used to indicate whether there is a connection by 0 and 1. , ) has a value of 1, which means the node and nodes connected to each other within the same time segment.

[0070] Step S350: converting the data type of the spatiotemporal adjacency matrix of the target disaster-affected population to obtain pre-processed mobile flow data.

[0071] Specifically, the data type of the spatiotemporal adjacency matrix of the target disaster-affected population is converted into tensor data that can be processed by deep learning to obtain preprocessed mobile traffic data.

[0072] It should be noted that for any node , pre-processing mobile traffic data corresponding to the movement of people Indicates time interval Internal Node The goal of crowd movement prediction is to The historical population movement observations within a time interval are: , estimated time interval The amount of people moving . In form, in the past Preprocessed mobile traffic data within a time interval , predicting future crowd movement .

[0073] Reference Figure 4 As shown, Figure 4 Other embodiments of the present application provide Figure 3 In the flowchart of step S330 in FIG. 1 , the method for predicting mobile traffic flow of disaster-affected people may include but is not limited to steps S410 to S440.

[0074] Step S410: Determine multiple disaster-affected area nodes of the target disaster-affected population based on the geographical location stability data.

[0075] In some possible embodiments, in city A, multiple disaster-affected area nodes of the target disaster-affected population in city A are determined based on the geographical environment of the geographical location stable data. For example, based on the geographical environment with a large altitude difference in city A, low-lying area nodes in city A are identified and the low-lying area nodes are determined as disaster-affected area nodes. According to the geographical environment of the urban transportation lines in City A, the subway station entrance nodes in City A are identified, and the subway station entrance nodes are determined as the disaster area nodes. ; According to the geographical environment of the mountains in City A, identify the landslide-prone nodes in City A and determine the landslide-prone nodes as the disaster-affected area nodes According to the geographical environment of the water flow trend in city A, the water flow convergence nodes in city A are identified and the water flow convergence nodes are determined as the disaster area nodes. .

[0076] Step S420: Determine multiple disaster-affected area edges of the target disaster-affected population based on the geographic adjacency between the nodes in the disaster-affected areas.

[0077] In some possible embodiments, in city A, by using the distance threshold method, the geographical adjacency between the nodes of each disaster area is compared to determine multiple disaster area edges of the target disaster population. For example, the distance threshold for adjacency judgment is defined as 0.005 degrees longitude and latitude. Nodes in the affected area If the distance is less than 0.005 degrees, the node in the affected area is determined Nodes in the affected area Mutually adjacent, that is, determine the nodes in the disaster area Nodes in the affected area The connecting line between them is the disaster area edge ab; when the disaster area node Nodes in the affected area If the distance is less than 0.005 degrees, the node in the affected area is determined Nodes in the affected area Mutually adjacent, that is, determine the nodes in the disaster area Nodes in the affected area The connecting line between them is the disaster area edge bc; when the disaster area node Nodes in the affected area If the distance is less than 0.005 degrees, the node in the affected area is determined Nodes in the affected area Mutually adjacent, that is, determine the nodes in the disaster area Nodes in the affected area The connecting line between them is the disaster area edge cd; when the disaster area node Nodes in the affected area If the distance is greater than 0.005 degrees, the node is determined With Node Not adjacent to each other.

[0078] Step S430: Determine the undirected network graph data of the target disaster-affected population according to each disaster-affected area node and each disaster-affected area edge.

[0079] In some possible embodiments, according to the nodes in the disaster area , Nodes in the affected area , Nodes in the affected area , Nodes in the affected area As well as the disaster-affected area edge ab, the disaster-affected area edge bc, and the disaster-affected area edge cd, the undirected network graph data of the target disaster-affected population is determined as [disaster-affected area node - Nodes in the affected areas , nodes in the affected area - Nodes in the affected areas , nodes in the affected area - Nodes in the affected areas ].

[0080] Step S440: Fuse the time series stationary data into the undirected network graph data in the time dimension to obtain the spatiotemporal graph data.

[0081] Specifically, the time series stationary data and the undirected network graph data are aligned in time dimension, and then based on each time segment, the relevant sequence features of the time series stationary data are correlated with the nodes of each disaster-stricken area. At the same time, the relevant sequence features of the time series stationary data are correlated with the edges of each disaster-stricken area to obtain the spatiotemporal graph data. , for example: Nodes in the disaster-affected areas The characteristic data of the number of people is 100 and the nodes of the disaster-affected areas Associate, time segment Nodes in the disaster-affected areas The characteristic data of the number of people is 150 and the nodes of the disaster-affected area Associate; associate time segments Nodes in the disaster-affected areas The direction of crowd movement is -> The characteristic data of is associated with the disaster area edge ab; the time segment Nodes in the disaster-affected areas The direction of crowd movement is -> The feature data is associated with the disaster-affected area edge ab.

[0082] It should be noted that each time segment are time segments divided into equal intervals in the time domain, where each time segment The interval may be 1 hour, 2 hours, or 30 minutes, which is not specifically limited in this application.

[0083] Reference Figure 5 As shown, Figure 5 Other embodiments of the present application provide Figure 1In the flowchart of step S130 in FIG. 1 , the method for predicting mobile flow of disaster-affected people may include but is not limited to steps S510 to S540.

[0084] Step S510: Calculate the degree centrality of undirected network nodes on the pre-processed mobile traffic data to obtain degree centrality index data.

[0085] Specifically, the degree centrality of the undirected network nodes is calculated for the pre-processed mobile traffic data to obtain the degree centrality index data, where the degree in the degree centrality index data is the degree centrality calculation formula of the nodes in each disaster-stricken area, as shown below:

[0086] in, For the The degree centrality index of the nodes in the disaster-affected areas, is the total number of nodes in the disaster-affected area, For the The degree of the nodes in the disaster area.

[0087] Step S520: Calculate the closeness centrality of undirected network nodes on the pre-processed mobile traffic data to obtain closeness centrality index data.

[0088] Specifically, the proximity centrality of the undirected network nodes is calculated for the pre-processed mobile traffic data to obtain proximity centrality index data, wherein the proximity centrality calculation formula of each disaster-affected area node in the proximity centrality index data is as follows:

[0089] in, For the The closeness centrality index of the nodes in the disaster-affected areas, is the total number of nodes in the disaster area, From the nodes in the affected area To the disaster area node The number of shortest paths, Nodes passing through the disaster area The number of shortest paths.

[0090] Step S530: Calculate the betweenness centrality of undirected network nodes on the pre-processed mobile traffic data to obtain betweenness centrality index data.

[0091] Specifically, the betweenness centrality of the undirected network nodes is calculated for the pre-processed mobile traffic data to obtain the betweenness centrality index data, wherein the betweenness centrality calculation formula of each disaster-stricken area node in the betweenness centrality index data is as follows:

[0092] in, For the The betweenness centrality index of the nodes in the disaster-affected areas, Indicates the affected area node The proximity to other disaster-affected area nodes in the undirected network. The smaller the value, the closer the disaster-affected area nodes are. The closer to other disaster-affected area nodes, is the total number of nodes in the disaster-affected area.

[0093] Step S540: Calculate the heterogeneity vector of the degree centrality index data, the closeness centrality index data, and the betweenness centrality index data to obtain moving heterogeneity data.

[0094] Specifically, the heterogeneity vector is calculated for the degree centrality index data, the closeness centrality index data, and the betweenness centrality index data to obtain the mobile heterogeneity data, wherein the calculation formula of the mobile heterogeneity data is as follows:

[0095] in, Represents different centrality indicators (degree centrality indicator data, closeness centrality indicator data, betweenness centrality indicator data), " " indicates cardinality, referring to the set The number of elements in the set. Only degree centrality index data, closeness centrality index data, and betweenness centrality index data are included in the set The cardinality of is 3; Represented as a set of heterogeneous vectors, i.e., moving heterogeneous data is a set of heterogeneous vectors.

[0096] Reference Figure 6 As shown, Figure 6 It is a flowchart of generating a Seq2Seq sequence crowd prediction model provided by other embodiments of the present application. The method for predicting mobile flow of disaster-affected people may include but is not limited to steps S610 to S650.

[0097] Step S610: Initialize the original Seq2Seq sequence population model.

[0098] Among them, the original model of the Seq2Seq sequence population includes an LSTM model encoder and an LSTM model decoder. The LSTM model encoder is used to read the input sequence of the nonlinear stream and convert it into an ordered context encoding vector. The LSTM model decoder is used to read the LSTM model encoder output by the LSTM model encoder and generate the corresponding prediction sequence.

[0099] It should be noted that in initializing the original Seq2Seq sequence population model, the calculation formula of the LSTM model encoder is configured as follows:

[0100] in, Represented as an LSTM model encoder, Represented as the context encoding vector output by the encoder of the t-th LSTM model, is the input sequence of the t-th nonlinear flow.

[0101] It should be noted that in initializing the original Seq2Seq sequence population model, the calculation formula of the LSTM model decoder is configured as follows:

[0102] in, Represented as an LSTM model decoder, Represented as the predicted sequence output by the decoder of the t-th LSTM model, is the tth context encoding vector.

[0103] Step S620: obtaining the original mobile traffic data of the sample disaster-affected population, and performing data preprocessing on the original mobile traffic data to obtain sample mobile traffic data.

[0104] The mobile traffic sample data include time series sample data of the movement of the sample disaster-affected population and geographic location sample data of the sample disaster-affected population.

[0105] Specifically, the raw data of mobile traffic is first cleaned to remove missing values, outliers and duplicate values ​​of the raw data of mobile traffic, and then data conversion is performed on the cleaned data (including: unified timestamp conversion, etc.), and finally the converted data is normalized or standardized to obtain mobile traffic sample data.

[0106] Step S630: Divide the mobile traffic sample data into mobile traffic training data and mobile traffic verification data.

[0107] The mobile traffic sample data include time series sample data of the movement of the sample disaster-affected population and geographic location sample data of the sample disaster-affected population.

[0108] It should be noted that, through the cross-validation method, the mobile traffic sample data is divided into mobile traffic training data and mobile traffic verification data, which can improve the generalization ability and stability of the model.

[0109] Step S640: According to the mobile traffic training data, the Seq2Seq sequence population original model is iteratively trained multiple times to obtain a Seq2Seq sequence population training model.

[0110] Specifically, the mobile traffic training data is input into the eq2Seq sequence population original model to perform multiple iterative training on the Seq2Seq sequence population original model to obtain the Seq2Seq sequence population training model, wherein in each iteration, the Seq2Seq sequence population original model is trained using a subset of the mobile traffic training data through small batch gradient descent, thereby avoiding overfitting of the Seq2Seq sequence population original model.

[0111] Step S650: According to the mobile traffic verification data, the Seq2Seq sequence population training model is verified and optimized to obtain a Seq2Seq sequence population prediction model.

[0112] Specifically, the mobile traffic training data is input into the Seq2Seq sequence population original model for prediction in the same way as the training data, the predicted output sequence is compared with the actual output sequence in the mobile traffic verification data, the performance index is calculated, and then the model is adjusted and optimized multiple times according to the corresponding performance index. Moreover, it is necessary to re-verify on the mobile traffic verification data each time to evaluate the improvement effect of the model until the performance of the model on the mobile traffic verification data is satisfactory, and the Seq2Seq sequence population prediction model is obtained, thereby ensuring the stability and reliability of the Seq2Seq sequence population prediction model in the actual environment.

[0113] Reference Figure 7 As shown, Figure 7 Some embodiments of the present application provide Figure 6 In the flowchart of step S630, the method for predicting mobile traffic of disaster-affected people may include but is not limited to steps S710 to S730.

[0114] Step S710: Determine the disaster-affected time point of the sample disaster-affected population based on the time series sample data.

[0115] Specifically, the disaster-affected time point of the sample disaster-affected population is determined based on the time series sample data, and the disaster-affected time point is the time point when the disaster occurs.

[0116] Step S720: Based on the disaster time point, the mobile traffic sample data is divided into mobile traffic sample data of the sample disaster-affected population before the disaster, and mobile traffic sample data of the sample disaster-affected population during and after the disaster.

[0117] Specifically, based on the disaster time point, the mobile traffic sample data is divided into mobile traffic sample data of the sample disaster-affected population before the disaster, and mobile traffic sample data of the sample disaster-affected population during and after the disaster.

[0118] Step S730: Based on the cross-validation method, the mobile traffic sample data of a part of the sample disaster-stricken population before the disaster is determined as the mobile traffic training data, and the mobile traffic sample data of another part of the sample disaster-stricken population before the disaster and the mobile traffic sample data of the sample disaster-stricken population during and after the disaster are determined as the mobile traffic verification data.

[0119] In some embodiments, based on the cross-validation method, the mobile traffic sample data of 80% of the sample disaster-affected people before the disaster are determined as mobile traffic training data, and the mobile traffic sample data of the remaining 20% ​​of the sample disaster-affected people before the disaster and the mobile traffic sample data of all the sample disaster-affected people during and after the disaster are determined as mobile traffic verification data.

[0120] Reference Figure 8 As shown, Figure 8 Some embodiments of the present application provide Figure 6 In the flowchart of step S640, the method for predicting mobile traffic of disaster-affected people may include but is not limited to steps S810 to S840.

[0121] Step S810: During each iterative training of the Seq2Seq sequence population original model, the loss value between the predicted output value and the true output value of the Seq2Seq sequence population original model of the current iterative training is calculated according to the mean square error loss function to obtain the loss value of the population prediction model of each iteration.

[0122] Specifically, in each model iteration training, a batch of mobile traffic training data is input into the Seq2Seq sequence population original model, and the Seq2Seq sequence population original model performs forward propagation to generate a predicted output sequence. Then, the mean square error loss function is used to calculate the loss between the predicted output value and the true output value sequence. For each model output sequence, the square of the difference between the predicted output value and the true output value is calculated, and then these square differences are averaged to obtain the mean square error loss value for each iteration.

[0123] Step S820: During each iterative training of the Seq2Seq sequence crowd original model, the gradient of the crowd prediction model loss value of the current iterative training is calculated according to the back propagation algorithm to obtain the corresponding crowd prediction model gradient value.

[0124] Specifically, in each model iteration training, when the mean square error loss value of the current iteration is obtained, the chain rule is used to backward calculate the gradient of each parameter in the original Seq2Seq sequence crowd model layer by layer to obtain the gradient value of the crowd prediction model of the current iteration, where the gradient calculation process includes all layers of the LSTM model encoder and the LSTM model decoder, as well as the connection layer between the LSTM model encoder and the LSTM model decoder (such as the attention mechanism layer).

[0125] Step S830: During each iterative training process of the Seq2Seq sequence population original model, the model parameters of the Seq2Seq sequence population original model of the current iterative training are updated according to the population prediction model gradient value of the current iterative training.

[0126] Specifically, in each model iteration training, when the gradient value of the population prediction model of the current iteration is obtained, the exponential moving average method is used to calculate the first-order moment (i.e., the mean of the gradient values) and the second-order moment (i.e., the mean of the square of the gradient values) of each population prediction model gradient value, and then the first-order moment and the second-order moment of the gradient value of each population prediction model are bias-corrected. Finally, the corrected first-order moment and second-order moment are used to update the model parameters of the Seq2Seq sequence population original model. By using exponential moving average and bias correction, the computational efficiency and stability of the Seq2Seq sequence population original model can be improved.

[0127] Step S840: During each iterative training process of the Seq2Seq sequence population original model, if the model parameters of the current iterative training meet the preset training requirements, the iterative training of the Seq2Seq sequence population original model is terminated to obtain the Seq2Seq sequence population training model.

[0128] Specifically, during each iterative training process of the original model of the Seq2Seq sequence population, if the model parameters of the current iterative training meet the preset training requirements (such as: the loss value calculated by the loss function of the model tends to converge, the number of iterative training reaches the preset training threshold, etc.), the iterative training of the original model of the Seq2Seq sequence population is terminated to obtain the Seq2Seq sequence population training model.

[0129] In a second aspect, the present application also provides a disaster-affected population mobile traffic prediction system, referring to Fig. 9 , Fig. 9 A schematic diagram of a disaster-affected population mobile flow prediction system provided in some embodiments of the present application, wherein the disaster-affected population mobile flow prediction system 900 includes: Data acquisition module 910, the data acquisition module 910 is used to obtain the original mobile traffic data of the target disaster-affected population; A data preprocessing module 920, which is in communication with the data acquisition module 910, and is used to preprocess the original mobile traffic data to obtain preprocessed mobile traffic data of the target disaster-affected population; A heterogeneity module 930, the heterogeneity module 930 is in communication connection with the data preprocessing module 920, and the heterogeneity module 930 is used to calculate the heterogeneity factor of the preprocessed mobile traffic data based on the centrality measurement algorithm of the undirected network node, and obtain the mobile heterogeneity data of the target disaster-affected population, wherein the mobile heterogeneity data includes the degree centrality index data of the target disaster-affected population, the proximity centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population; The spatiotemporal prediction module 940 is connected in communication with the data preprocessing module 920, and the spatiotemporal prediction module 940 is connected in communication with the heterogeneity module 930, wherein: The spatiotemporal prediction module 940 is used to extract spatial features of the pre-processed mobile traffic data based on a graph convolutional neural network algorithm to obtain mobile spatial feature data of the target disaster-affected population; The spatiotemporal prediction module 940 is also used to extract time features from the spatial feature movement data and the movement heterogeneity data based on the long short-term memory network algorithm to obtain movement time feature data of the target disaster-affected population; The spatiotemporal prediction module 940 is also used to predict the movement flow of the crowd based on the movement space feature data and the movement time feature data according to the Seq2Seq sequence crowd prediction model, and obtain the movement flow prediction result of the target disaster-affected population.

[0130] The above-mentioned disaster-affected population mobile flow prediction system 900 and the above-mentioned disaster-affected population mobile flow prediction method are based on the same inventive concept. The above process describes the disaster-affected population mobile flow prediction method of the embodiment of the present application, which will not be repeated here. The disaster-affected population mobile flow prediction system 900 obtains the original mobile flow data of the target disaster-affected population, and then performs data preprocessing on the original mobile flow data to obtain the preprocessed mobile flow data of the target disaster-affected population. Then, based on the centrality measurement algorithm of the undirected network node, the heterogeneity factor is calculated for the preprocessed mobile flow data to obtain the mobile heterogeneity data of the target disaster-affected population, wherein the mobile heterogeneity data includes the degree centrality index data of the target disaster-affected population, the proximity centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population. Furthermore, by introducing the heterogeneity factor, the contribution of different special disaster scenarios to the population mobility pattern can be fully improved, and the processing capability of abnormal mobility patterns is improved. Furthermore, based on the graph convolutional neural network algorithm, spatial features are extracted from the preprocessed mobile flow data to obtain the mobile spatial feature data of the target disaster-stricken population. Then, based on the long short-term memory network algorithm, temporal features are extracted from the spatial feature mobile data and the mobile heterogeneity data to obtain the mobile time feature data of the target disaster-stricken population. Then, spatiotemporal features can be extracted through graph convolutional neural networks and long short-term memory networks to capture more detailed spatiotemporal dependencies. Finally, according to the Seq2Seq sequence population prediction model, the mobile spatial feature data and the mobile time feature data are used to predict the population mobile flow to obtain the mobile flow prediction results of the target disaster-stricken population. Then, the Seq2Seq sequence architecture can also process complex nonlinear flow data, which significantly improves the accuracy and reliability of population movement prediction during disasters.

[0131] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned disaster population mobile flow prediction method based on hybrid deep learning when executing the computer program. The electronic device can be any smart terminal including a mobile phone, a tablet computer, a car computer, etc.

[0132] See also Fig.10 , Fig.10 : is a schematic diagram of the hardware structure of an electronic device provided in some embodiments of the present application, and the electronic device includes: The processor 1001 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the mobile flow prediction method for disaster-affected people based on hybrid deep learning provided in the embodiment of the present application; The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other applications. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the mobile flow prediction method for disaster-affected people based on hybrid deep learning provided in the embodiments of this application; Input / output interface 1003, used to implement information input and output; Communication interface 1004, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.); A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 ); The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0133] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method for predicting the mobile flow of disaster-stricken people based on hybrid deep learning provided in an embodiment of the present application is implemented.

[0134] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0135] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0136] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0137] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0138] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0139] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be applied to sequential implementations other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0140] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0141] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0142] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0143] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0144] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes multiple instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store programs.

[0145] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for predicting mobile traffic of disaster-affected people based on hybrid deep learning, characterized in that: include: Obtaining raw mobile traffic data of the target disaster-affected population; Performing data preprocessing on the original mobile traffic data to obtain preprocessed mobile traffic data of the target disaster-affected population; Based on the centrality measurement algorithm of undirected network nodes, the pre-processed mobile traffic data is subjected to heterogeneity factor calculation to obtain the mobile heterogeneity data of the target disaster-affected population, wherein the mobile heterogeneity data includes the degree centrality index data of the target disaster-affected population, the closeness centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population; Based on a graph convolutional neural network algorithm, spatial feature extraction is performed on the pre-processed mobile traffic data to obtain mobile spatial feature data of the target disaster-affected population; Based on the long short-term memory network algorithm, the temporal features of the spatial feature movement data and the movement heterogeneity data are extracted to obtain the movement time feature data of the target disaster-affected population; According to the Seq2Seq sequence crowd prediction model, crowd movement flow prediction is performed on the movement space feature data and the movement time feature data to obtain the movement flow prediction result of the target disaster-affected population.

2. The method for predicting mobile traffic of disaster-affected people according to claim 1, characterized in that: The original mobile traffic data includes the original time series data of the movement of the target disaster-affected population and the original geographical location data of the target disaster-affected population. The data preprocessing of the original mobile traffic data to obtain the preprocessed mobile traffic data of the target disaster-affected population includes: Performing data cleaning on the original time series data to obtain stable time series data; Performing data cleaning on the original geographical location data to obtain stable geographical location data; Performing graph data abstraction processing on the geographical location steady data and the time series steady data to obtain spatiotemporal graph data of the target disaster-affected population; Performing adjacency matrix calculation on the spatiotemporal graph data to obtain the spatiotemporal adjacency matrix of the target disaster-affected population; The data type of the spatiotemporal adjacency matrix of the target disaster-affected population is converted to obtain the pre-processed mobile traffic data.

3. The method for predicting mobile traffic of disaster-affected people according to claim 2, characterized in that: The step of performing graph data abstraction processing on the geographical location steady data and the time series steady data to obtain the spatiotemporal graph data of the target disaster-affected population includes: Determining multiple disaster-affected area nodes of the target disaster-affected population according to the geographical location stability data; Determining multiple disaster-affected area edges of the target disaster-affected population according to the geographic adjacency between the nodes of each disaster-affected area; Determine the undirected network graph data of the target disaster-affected population according to each of the disaster-affected area nodes and each of the disaster-affected area edges; The time series stationary data is fused into the undirected network graph data in the time dimension to obtain the space-time graph data.

4. The method for predicting mobile traffic of disaster-affected people according to claim 1, characterized in that: The centrality measurement algorithm based on the undirected network nodes calculates the heterogeneity factor of the pre-processed mobile traffic data to obtain the mobile heterogeneity data of the target disaster-affected population, including: Performing degree centrality calculation of undirected network nodes on the pre-processed mobile traffic data to obtain the degree centrality index data; Performing closeness centrality calculation of undirected network nodes on the pre-processed mobile traffic data to obtain the closeness centrality index data; Performing betweenness centrality calculation of undirected network nodes on the pre-processed mobile traffic data to obtain the betweenness centrality index data; Heterogeneity vector calculation is performed on the degree centrality index data, the closeness centrality index data, and the betweenness centrality index data to obtain the moving heterogeneity data.

5. The method for predicting mobile traffic of disaster-affected people according to claim 1, characterized in that: The Seq2Seq sequence population prediction model is obtained by the following steps: Initialize a Seq2Seq sequence population original model, wherein the Seq2Seq sequence population original model includes an LSTM model encoder and an LSTM model decoder; Obtaining original mobile traffic data of sample disaster-affected people, and performing data preprocessing on the original mobile traffic data to obtain sample mobile traffic data; Dividing the mobile traffic sample data into mobile traffic training data and mobile traffic verification data; According to the mobile traffic training data, the Seq2Seq sequence population original model is iterated multiple times to obtain a Seq2Seq sequence population training model; According to the mobile traffic verification data, the Seq2Seq sequence population training model is verified and optimized to obtain a Seq2Seq sequence population prediction model.

6. The method for predicting mobile traffic of disaster-affected people according to claim 5, characterized in that: The mobile traffic sample data includes time series sample data of the movement of the sample disaster-affected population and geographical location sample data of the sample disaster-affected population, and the mobile traffic sample data is divided into mobile traffic training data and mobile traffic verification data, including: Determining the disaster-affected time point of the sample disaster-affected population according to the time series sample data; Based on the disaster time point, the mobile traffic sample data is divided into the mobile traffic sample data of the sample disaster-affected population before the disaster, and the mobile traffic sample data of the sample disaster-affected population during and after the disaster; Based on the cross-validation method, the mobile traffic sample data of a part of the sample disaster-stricken population before the disaster is determined as the mobile traffic training data, and the mobile traffic sample data of another part of the sample disaster-stricken population before the disaster and the mobile traffic sample data of the sample disaster-stricken population during and after the disaster are determined as mobile traffic verification data.

7. The method for predicting mobile traffic of disaster-affected people according to claim 5, characterized in that: The method of performing multiple iterations of training on the Seq2Seq sequence population original model according to the mobile traffic training data to obtain a Seq2Seq sequence population training model includes: In each iterative training process of the Seq2Seq sequence population original model, the loss value between the predicted output value and the true output value of the Seq2Seq sequence population original model of the current iterative training is calculated according to the mean square error loss function to obtain the population prediction model loss value of each iteration; During each iterative training process of the Seq2Seq sequence population original model, a gradient calculation is performed on the loss value of the population prediction model of the current iterative training according to the back propagation algorithm to obtain a corresponding population prediction model gradient value; During each iterative training process of the Seq2Seq sequence population original model, updating the model parameters of the Seq2Seq sequence population original model of the current iterative training according to the population prediction model gradient value of the current iterative training; During each iterative training process of the Seq2Seq sequence population original model, if the model parameters of the current iterative training meet the preset training requirements, the iterative training of the Seq2Seq sequence population original model is terminated to obtain the Seq2Seq sequence population training model.

8. A disaster-affected population mobility prediction system, characterized in that: include: A data acquisition module, the data acquisition module is used to acquire the original mobile traffic data of the target disaster-affected population; A data preprocessing module, the data preprocessing module is in communication with the data acquisition module, and the data preprocessing module is used to perform data preprocessing on the original mobile traffic data to obtain preprocessed mobile traffic data of the target disaster-affected population; A heterogeneity module, the heterogeneity module is in communication connection with the data preprocessing module, the heterogeneity module is used to calculate the heterogeneity factor of the preprocessed mobile traffic data based on the centrality measurement algorithm of the undirected network node, and obtain the mobile heterogeneity data of the target disaster-affected population, wherein the mobile heterogeneity data includes the degree centrality index data of the target disaster-affected population, the proximity centrality index data of the target disaster-affected population, and the betweenness centrality index data of the target disaster-affected population; A spatiotemporal prediction module, the spatiotemporal prediction module is communicatively connected to the data preprocessing module, and the spatiotemporal prediction module is communicatively connected to the heterogeneity module, wherein: The spatiotemporal prediction module is used to extract spatial features of the pre-processed mobile traffic data based on a graph convolutional neural network algorithm to obtain mobile spatial feature data of the target disaster-affected population; The spatiotemporal prediction module is also used to extract time features from the spatial feature movement data and the movement heterogeneity data based on a long short-term memory network algorithm to obtain movement time feature data of the target disaster-affected population; The spatiotemporal prediction module is also used to predict the crowd movement flow of the mobile space feature data and the mobile time feature data according to the Seq2Seq sequence crowd prediction model, so as to obtain the mobile flow prediction result of the target disaster-stricken population.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method for predicting mobile flow of disaster-affected people as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program executable by a processor, and when the program executable by the processor is executed by the processor, it is used to implement the method for predicting the mobile flow of disaster-affected people as described in any one of claims 1 to 7.

Images