Post-disaster population flow prediction method and device

Abstract

The invention provides a post-disaster population flow prediction method and device. The post-disaster population flow prediction method comprises the steps of obtaining population flow data of a target area according to predetermined space-time fine granularity; obtaining an input population flow dynamic graph according to the population flow data of the target area; wherein the input population flow dynamic graph at least comprises an initial abnormal population flow graph of the target area in the disaster peak period and an average normal population flow graph of the target area in the non-disaster period; inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; wherein the population flow prediction model is obtained by training population flow data of a sample area based on a coupled dynamic graph Sheng differential equation. According to the method, the population flow prediction model is constructed by adopting the coupled dynamic graph normal differential equation, and post-disaster population flow prediction with high accuracy is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and in particular to a method and device for predicting post-disaster population mobility. Background Art

[0002] With the acceleration of urbanization, extreme events (disasters) that frequently occur around the world have had an impact on the lives and safety of urban residents that cannot be ignored. Therefore, there is an urgent need to gain a deeper understanding of urban resilience, that is, the ability of a city to quickly restore its functions after being hit by an extreme event. Among them, population mobility within the city plays a vital role, which is mainly reflected in two aspects: on the one hand, urban population mobility reflects the behavior of urban residents in obtaining necessary resources (such as food and money) from residential areas to other functional areas, and is a key indicator for measuring residents' satisfaction with obtaining necessities of life; on the other hand, urban population mobility is intricately linked to the normal operation of urban infrastructure such as transportation facilities and office buildings, thus reflecting the recovery status of urban infrastructure. Generally speaking, cities with poor resilience usually take more time to recover to normal urban population mobility, which has a greater impact on the lives of urban residents and brings greater economic losses. Therefore, accurately predicting the recovery process of urban population mobility after a disaster can help discover high-risk urban communities, design better emergency response strategies, and ultimately build a smarter and more resilient city.

[0003] However, predicting the post-disaster recovery process of urban population mobility is also a difficult task, facing the following challenges: First, the post-disaster recovery of urban population mobility has a complex pattern, which is intricately intertwined with different disaster situations and their pre-disaster population mobility characteristics. However, its underlying mechanism is still unclear, and it is difficult to accurately model using only observable data. In addition, the population mobility change pattern under normal conditions is not very helpful for the population mobility change pattern under abnormal conditions. Second, the population mobility recovery process in different urban areas is not independent of each other. Inter-regional population mobility will also have a significant impact on its recovery process. For urban areas with large net population inflows, due to the large number of active populations, the recovery speed will be faster, which is conducive to the reconstruction and recovery process. The highly dynamic interaction process of human flow between regions further increases the difficulty of effectively modeling and predicting the dynamics of human flow recovery.

[0004] Existing modeling methods for the post-disaster recovery process of urban population mobility are mainly based on mathematical modeling methods, which make overly simplified assumptions about urban mobility and construct mathematical models with a very limited number of parameters. Although these models provide valuable knowledge about the recovery of urban population mobility after disasters and have good explanatory power, the strong simplifying assumptions and limited model parameters limit their ability to fully capture the diverse and complex recovery patterns of post-disaster urban transportation. In addition, all these models fail to effectively simulate the impact of inter-regional population mobility on their recovery.

[0005] At the same time, the rapid development of data-driven deep learning technologies represented by graph neural networks (GNNs) and Neural Ordinary Differential Equations (Neural ODEs) has further enhanced our ability to model the dynamics of nodes and edges that co-evolve in graph structures, and a number of algorithm inventions for learning graph data with time series characteristics have emerged, that is, using dynamic graphs to model multivariate time series and their interactions. It is feasible to use mainstream dynamic graph prediction methods in the post-disaster urban population flow recovery process, but there are large errors in the prediction results. The reason is that the idea of ​​mainstream dynamic graph prediction methods is to use the graph of a period of time in the past to predict the graph of a period of time in the future, relying on the graph signal evolution pattern learned from the historical graph sequence. However, this pattern only reflects the data fluctuations under normal conditions, and is of little help to the abnormal population flow recovery process after a disaster.

[0006] In summary, the existing technology has the problem of low accuracy. Summary of the invention

[0007] The present invention provides a method and device for predicting population mobility after a disaster, so as to solve the defect of low accuracy in the prior art and realize prediction of population mobility after a disaster with high accuracy.

[0008] The present invention provides a method for predicting population mobility after a disaster, comprising the following steps: Obtain population mobility data in the target area according to predetermined spatiotemporal granularity; An input population flow dynamic map is obtained according to the population flow data of the target area; wherein the input population flow dynamic map at least includes an initial abnormal population flow map during the disaster peak period of the target area and an average normal population flow map during the non-disaster period of the target area; Inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; The population flow prediction model is based on coupled dynamic graph neural ordinary differential equations and is trained using population flow data in the sample area.

[0009] According to a method for predicting post-disaster population mobility provided by the present invention, based on coupled dynamic graph neural ordinary differential equations, the population mobility prediction model is obtained by training using population mobility data in a sample area, specifically comprising: Obtain population mobility data in the sample area; According to the population flow data of the sample area, an initial abnormal population flow map of the sample area during the disaster peak period, an average normal population flow map of the sample area during the non-disaster period, and a real abnormal population flow map of the sample area during the recovery stage are obtained; Construct a basic prediction model based on coupled dynamic graph neural ordinary differential equations; The initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are used as input, and the real abnormal population flow map during the recovery stage of the sample area is used as a label to train the basic prediction model until the model converges to obtain a population flow prediction model.

[0010] According to a method for predicting post-disaster population mobility provided by the present invention, the basic prediction model includes a decoding module and an encoding module; Accordingly, the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are used as inputs, and the real abnormal population flow map during the recovery stage of the sample area is used as a label to train the basic prediction model until the model converges, thereby obtaining a population flow prediction model, which specifically includes: According to the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area, based on the coding module, the inter-area population flow and the intra-area population flow are coded to obtain a coding result; Decoding the encoding result based on the decoding module to obtain population flow prediction values ​​at multiple moments in the post-disaster recovery process; According to the population flow prediction value and the actual abnormal population flow map in the sample area recovery stage, back propagation is performed using a preset loss function to update the model parameters until the model converges, thereby obtaining a population flow prediction model.

[0011] According to a method for predicting post-disaster population mobility provided by the present invention, the encoding module comprises: ,

[0012] ,

[0013] ,

[0014] in, The dimension is The attenuation coefficient vector, Based on The function of express The encoded population mobility code within the abnormal region has the dimension of , Indicates Area of ​​the day Population mobility within Then After and The population mobility coding in the normal area after the same coding method, Represents the attribute value of each point in the normal population flow graph, represents the transformation matrix, represents the activation function, Used to indicate whether an edge exists. The dimension is The vector coefficients of Representation and Region There are areas where there is inter-regional population mobility. Softmax represents the Softmax activation function. represents an MLP, express The coded abnormal inter-regional population mobility code, Indicates Area of ​​the day To area population mobility, represents a multi-layer perceptron, represents the convolution operator of GCN, express The degree matrix of .

[0015] According to a method for predicting post-disaster population mobility provided by the present invention, the decoding module comprises: ,

[0016] ,

[0017] ,

[0018] in, and Respectively represent Population mobility within a region and the region To area Population mobility during the recovery period The predicted value for the day, express The coded population mobility code in the abnormal area, Represents the decoding function of the MLP represented by the superposition of two linear layers and two RELU activation functions.

[0019] According to a method for predicting post-disaster population mobility provided by the present invention, when the target area and the sample area are the same, the real abnormal population mobility map of the sample area in the recovery stage is the real abnormal population mobility map of the sample area in the early stage of the recovery stage; In the case where the target area and the sample area are different, the real abnormal population flow map of the sample area in the recovery stage is the real abnormal population flow map of the sample area in the full recovery stage.

[0020] The present invention also provides a post-disaster population flow prediction device, comprising the following modules: Data unit, used to obtain population flow data of the target area according to the predetermined spatiotemporal granularity; A conversion unit, used to obtain an input population flow dynamic map according to the population flow data of the target area; wherein the input population flow dynamic map at least includes an initial abnormal population flow map during the disaster peak period of the target area and an average normal population flow map during the non-disaster period of the target area; A prediction unit, used for inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; Among them, the population flow prediction model is based on a coupled dynamic graph neural ordinary differential equation and is trained using population flow data in a sample area.

[0021] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned methods for predicting post-disaster population flows is implemented.

[0022] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for predicting post-disaster population mobility.

[0023] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for predicting post-disaster population mobility.

[0024] The method and device for predicting post-disaster population mobility provided by the present invention obtains population mobility data of a target area according to a predetermined spatiotemporal granularity; obtains an input population mobility dynamic graph according to the population mobility data of the target area; wherein the input population mobility dynamic graph at least includes an initial abnormal population mobility graph during the peak period of the disaster in the target area and an average normal population mobility graph during the non-disaster period in the target area; the input population mobility dynamic graph is input into a pre-trained population mobility prediction model to obtain a population mobility prediction result; wherein the population mobility prediction model is based on a coupled dynamic graph neural ordinary differential equation and is trained using population mobility data of a sample area. The present invention uses a coupled dynamic graph neural ordinary differential equation to construct a population mobility prediction model, which can effectively model the diverse and highly volatile urban mobility in different regions, and jointly consider the urban mobility before and after the disaster, solving the problem that the prior art relies on overly simplified assumptions, thereby failing to effectively capture the impact of population mobility between regions, and it is difficult to use the normal flow change pattern to predict the abnormal flow recovery process after the disaster, resulting in low accuracy, and realizing a high-accuracy post-disaster population mobility prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 This is one of the flow charts of the post-disaster population flow prediction method provided by the present invention.

[0027] Figure 2 This is the second flow chart of the post-disaster population flow prediction method provided by the present invention.

[0028] Figure 3 This is the third flow chart of the post-disaster population flow prediction method provided by the present invention.

[0029] Figure 4 This is the fourth flow chart of the post-disaster population flow prediction method provided by the present invention.

[0030] Figure 5 It is a structural schematic diagram of the post-disaster population flow prediction device provided by the present invention.

[0031] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Combine the following Figure 1-Figure 4 The present invention describes the method for predicting population mobility after a disaster. Figure 1 FIG. 1 is one of the flow charts of the method for predicting population mobility after a disaster provided by the present invention, such as Figure 1 As shown, the method comprises the following steps: Step 110: Obtain population flow data of the target area according to a predetermined spatiotemporal granularity.

[0034] It should be noted that the goal of the present invention is to predict multiple population flow maps in the post-disaster recovery process given at least one abnormal population flow map on the day of the disaster and at least one real normal population flow map. On this basis, the target area refers to the city or region to be predicted, and the population flow data of the target area includes the abnormal population flow data of the city or region to be predicted and the real normal population flow data. These population flow data do not need to be normalized and are only in units of people.

[0035] Spatiotemporal granularity includes time granularity and space granularity. The present invention does not limit the time granularity and space granularity. The time granularity can be a day or other time intervals; the space granularity can be a street, a county, or even a province. After determining the space granularity and time granularity, the population flow data should also be aggregated to the corresponding spatiotemporal granularity.

[0036] In some embodiments, the population flow data of the target area includes initial abnormal population flow data during the disaster peak period and average normal population flow data during the non-disaster period.

[0037] It should be noted that in the present invention, the focus is on the changing process of daily population flow, so the population flow map here is Representative Population flow diagram for the day, where each point represents an area, the features on it represent the population flow within the area, and the lines between the points, i.e. the edges, represent the population flow between areas. The definitions of the related concepts of population flow attribute values ​​are as follows: Definition 1 (Intra-regional population mobility): Area of ​​the day Population mobility within It represents the total number of people who move only in this area during the day.

[0038] Definition 2 (Inter-regional population mobility): Area of ​​the day To area Population mobility available Indicates that it represents the number of people from the region during the day. Come to the area The total number of people.

[0039] It is observed that the population flow in a city will decrease to the lowest value and enter an abnormal state on the day of the disaster, and will slowly return to normal in the next few days. Therefore, different from the traditional dynamic graph prediction method that predicts a future graph sequence based on a past graph sequence, the present invention focuses on starting from the abnormal population flow graph when the disaster strikes, and predicting the population flow graph during the recovery period based on the normal population flow graph. The prediction problem of the normal population flow graph and the dynamic recovery of population flow after the disaster is defined as follows: Definition 3 (Normal Population Flow Graph): Given a sequence of normal population flow graphs during a non-disaster period , the normal population flow diagram is defined as Specifically, the attribute value of each point, that is, the population flow in the normal area, is defined as , and the attribute value of each edge, i.e., the normal inter-regional population flow, is defined as .

[0040] Definition 4 (Prediction of post-disaster population mobility recovery dynamics): Given the abnormal population mobility graph on the first day of the disaster And the normal population flow map , predict the next Population mobility during the post-disaster recovery period .

[0041] Step 120: Obtain an input population flow dynamic map based on the population flow data of the target area; wherein the input population flow dynamic map at least includes an initial abnormal population flow map during the disaster peak period of the target area and an average normal population flow map during the non-disaster period of the target area.

[0042] In step 120, the population flow data of the target area is converted into an input population flow dynamic map. Based on the description of step 110, the population flow data of the target area includes the initial abnormal population flow data during the disaster peak period and the average normal population flow data during the non-disaster period. Accordingly, the input population flow dynamic map includes the initial abnormal population flow map during the disaster peak period of the target area and the average normal population flow map during the non-disaster period of the target area.

[0043] Step 130: Input the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; wherein the population flow prediction model is based on a coupled dynamic graph neural ordinary differential equation and is trained using population flow data in a sample area.

[0044] In step 130, the obtained input population flow dynamic graph is input into the population flow prediction model, and the population flow prediction result of the target area is output. The population flow prediction model mentioned in the present invention is a coupled dynamic graph ODE network (CDGON), such as Figure 2 As shown, the present invention uses a physics-inspired machine learning paradigm to construct a neural ODE function guided by a spatiotemporal decay model (ST DecayModel), which can effectively model the diverse and highly volatile urban mobility in different regions, and jointly consider the urban mobility before and after the disaster. At the same time, another set of coupled neural ordinary differential equation functions is introduced to describe the dynamics of inter-regional population mobility as the edges in the formed dynamic graph to characterize the complex post-disaster urban mobility. These two sets of neural ODE functions jointly model the co-evolution process of nodes and edges in the dynamic graph of intra-regional and inter-regional population mobility. In terms of the specific calculation process, the network model first encodes the normal and abnormal regional population mobility and the abnormal inter-regional population mobility, and then takes the normal regional population mobility as a guide, considers the co-evolution of the edges of the points, solves the neural ordinary differential equation inspired by the ST Decay Model to obtain the population mobility coding sequence at multiple moments in the post-disaster recovery process, and finally decodes it to achieve the prediction of the post-disaster population mobility recovery process.

[0045] It should be noted that the purpose of the spatiotemporal decay model (STDecay Model) is to fit the recovery curve of population mobility after the peak of large-scale natural disasters such as typhoons and cold waves. Its formula can be expressed as follows: ,

[0046] in, Indicators of population mobility that represent normality, is a function that decays with time, called the time decay term, and Indicates disaster The day is The severity of the region, It indicates the area The weighted sum of the disaster severity in the surrounding areas. Since the severity of disasters tends to decrease during the recovery period, the spatiotemporal attenuation model shows that as time goes by, the abnormal population flow during the disaster will eventually return to normal. How to capture the attenuation factor during this period The process of change is the key.

[0047] The population flow prediction model is further described below. In some embodiments, based on the coupled dynamic graph neural ordinary differential equation, the population flow prediction model is trained using the population flow data of the sample area, specifically including: Step 210: Obtaining population flow data of the sample area; Step 220: obtaining an initial abnormal population flow map of the sample area during the disaster peak period, an average normal population flow map of the sample area during the non-disaster period, and a real abnormal population flow map of the sample area during the recovery stage according to the population flow data of the sample area; Step 230: constructing a basic prediction model based on coupled dynamic graph neural ordinary differential equations; Step 240: The initial abnormal population flow map of the sample area during the disaster peak period and the average normal population flow map of the sample area during the non-disaster period are used as input, and the real abnormal population flow map of the sample area during the recovery stage is used as a label to train the basic prediction model until the model converges to obtain a population flow prediction model.

[0048] Specifically, in step 210, the population flow data of the sample area is obtained. It should be noted that the population flow data of the sample area used for reference is obtained. The population flow data of the sample area includes the initial abnormal population flow data of the sample area during the disaster peak period, the average normal population flow data of the sample area during the non-disaster period, and the real abnormal population flow data of the sample area during the recovery stage.

[0049] It is understandable that the method provided by the present invention can use the existing population flow data of multiple cities in a certain area in the early recovery stage to train the model, and extrapolate the population flow status in the middle and late recovery stages of the same group of cities (i.e., the target area and the sample area are the same), such as Figure 3 As shown in ; it is also possible to use the existing population flow data of multiple cities in a certain area during the entire recovery phase to train the model, and use the trained model to carry out post-disaster population flow recovery prediction in another group of cities in another area (that is, the target area and the sample area are different). This is equivalent to using the experience of other cities in responding to disasters to predict the recovery process of the target city, such as Figure 4 shown.

[0050] On this basis, in some embodiments, when the target area and the sample area are the same, the real abnormal population flow map of the sample area in the recovery stage is the real abnormal population flow map of the sample area in the early stage of the recovery stage; In the case where the target area and the sample area are different, the real abnormal population flow map of the sample area in the recovery stage is the real abnormal population flow map of the sample area in the full recovery stage.

[0051] Specifically, the population flow data of multiple cities in a certain area at the early stage of recovery are used to extrapolate the population flow conditions in the same group of cities in the middle and late stages of recovery. That is, when the target area and the sample area are the same, the real and abnormal population flow map of the sample area in the recovery stage can be simply the population flow map obtained from the real and abnormal population flow data at the early stage of recovery.

[0052] With the help of the experience of population mobility recovery patterns in other cities or regions after the same type of disasters, we can help predict the population mobility of a certain city or region during the entire recovery period after a certain type of disaster. That is, when the target area and the sample area are different, the three parts of data that need to be collected are similar to but not exactly the same as those when the target area and the sample area are the same. The difference is that not only the real abnormal population mobility data in the early stage of the recovery stage is needed, but the real abnormal population mobility data in the full recovery stage is needed to obtain a population mobility map.

[0053] In step 220, the collected data is divided into a third part: the initial abnormal population flow data of the sample area during the disaster peak period, the average normal population flow data during the non-disaster period, and the real abnormal population flow data during the recovery stage, which are converted into a population flow map to obtain the initial abnormal population flow map of the sample area during the disaster peak period, the average normal population flow map of the sample area during the non-disaster period, and the real abnormal population flow map of the sample area during the recovery stage.

[0054] In step 230, a basic prediction model is constructed based on coupled dynamic graph neural ordinary differential equations. In this step, based on the ST Decay Model, the present invention proposes a coupled dynamic graph neural ordinary differential equation network CDGON, which can start from the initial abnormal population flow after the disaster and combine the normal population flow to realize the prediction of the recovery dynamics of the population flow after the disaster. The present invention will explain the basic prediction model, i.e., the coupled dynamic graph neural ordinary differential equation network CDGON, in subsequent embodiments.

[0055] In step 240, the initial abnormal population flow map of the sample area during the disaster peak period and the average normal population flow map of the sample area during the non-disaster period constructed in step 220 are input into the model as input, and the real abnormal population flow map of the sample area during the recovery stage is used as the label supervision model parameter update. After multiple iterations of updating the parameters to make the model converge, the population flow prediction model can be obtained.

[0056] Furthermore, in some embodiments, during the training process, since the prediction goal of the present invention is to achieve accurate prediction of the post-disaster population mobility recovery process, in this process, considering that the value of inter-regional population mobility is smaller than the value of intra-regional population mobility, in order to take into account the intra-regional and inter-regional population mobility and achieve accurate prediction of both, the loss function of the embodiment of the present invention uses a hyperparameter To amplify the weight of edge prediction, so as to ensure that the model does not abandon the prediction of population flow between regions, as follows

[0057] ,

[0058] in, Represents the number of days required for post-disaster recovery. In each training, by inputting the initial abnormal population flow within and between regions and the normal population flow within a region, the predicted population flow value for the next few days (i.e., the recovery period) can be obtained, and the loss is calculated between the actual value, and finally the gradient is calculated, back propagation is carried out, and the model parameters are updated.

[0059] During use, for extended predictions in the same city, that is, when the target area and the sample area are the same, the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are continued to be used to input the model, and the number of integral steps of the differential equation is extended to generate the extrapolated population flow forecast value in the middle and late stages of the recovery phase. For cross-city migration predictions, it is necessary to obtain the initial abnormal population flow map during the disaster peak period of the target area and the average normal population flow map during the non-disaster period of the target area to input the model. The model will generate population flow data for the corresponding length of the recovery phase according to the set parameters.

[0060] The basic prediction model is described below. In some embodiments, the basic prediction model includes a decoding module and an encoding module; Accordingly, the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are used as inputs, and the real abnormal population flow map during the recovery stage of the sample area is used as a label to train the basic prediction model until the model converges, thereby obtaining a population flow prediction model, which specifically includes: According to the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area, based on the coding module, the inter-area population flow and the intra-area population flow are coded to obtain a coding result; Decoding the encoding result based on the decoding module to obtain population flow prediction values ​​at multiple moments in the post-disaster recovery process; According to the population flow prediction value and the actual abnormal population flow map in the sample area recovery stage, back propagation is performed using a preset loss function to update the model parameters until the model converges, thereby obtaining a population flow prediction model.

[0061] Specifically, in the specific construction process, the same encoder needs to be used for normal and abnormal population flow graphs so that they can be aligned in high-dimensional space. After obtaining the high-dimensional embedding of the initial abnormal population flow and the normal population flow, the population flow at multiple moments in the future is obtained by solving the Neural Ordinary Differential Equation inspired by the ST Decay Model, and finally the final prediction result is obtained through the same decoder.

[0062] The encoding and decoding process of relevant values ​​in the population flow map is as follows Figure 2 As shown in the figure, the neural ordinary differential equation designed by ST DecayModel, namely the encoding module, is elaborated in detail. The equation mainly consists of two parts: the physics-informed dynamic module (PIDM) and the node-edge interaction module (NEIM).

[0063] The population flow map defined previously has both edge features and point features, representing the population flow between regions and within regions respectively. However, due to the large volatility and sparseness of population flow between regions, for example, before a disaster, some people Refuge Go to land, and after the disaster there will be Go to The number of people in the area has increased significantly, while under normal circumstances there may be almost no one Go to However, the population flow value within a region is more stable and dense than the population flow value between regions and has a more obvious recovery process. Therefore, the present invention only applies PIDM to the population flow within a region.

[0064] Some prior art studies have used negative logarithmic exponential functions To represent the time decay term in the ST Decay Model , and the present invention believes that the spatial attenuation term actually decays with time. Therefore, it is assumed here that the entire spatiotemporal attenuation factor , and then derive the formula as follows: ,

[0065] in, Represents the attenuation coefficient. Based on this, the present invention designs the PIDM as follows: ,

[0066] in, express The encoded population flow code in the abnormal region is set to have a dimension of , similarly Then After and The population mobility within the normal region is coded after the same encoder.

[0067] Since the changing patterns and speeds of each component of population mobility coding may be different, the dimension is also The attenuation coefficient vector Used to replace the original At the same time, in order to avoid The square term in The numerical explosion caused by this is used in advance based on Function The original population flow encoding vector Convert to a scalar. Finally Represents a transformation matrix, and adds a Activation function.

[0068] On this basis, it is obviously not enough to only model the population flow within the region. Although the population flow between regions is more volatile and complex, it is obviously difficult to use traditional methods to model it. Data-driven methods represented by deep learning naturally have advantages in solving this type of modeling. Therefore, in addition to designing ordinary differential equations for population flow within a region, it is also necessary to design ordinary differential equations for population flow between regions. Prior to this, considering that the topological structure of the population flow graph in actual situations is dynamic, we can first calculate the possible edge values ​​at each moment based on its encoding. If the edge value is less than This means that this edge does not exist, as follows: ,

[0069] ,

[0070] in, Used to indicate whether an edge exists. Note that It does not represent a prediction To The population flow value of the land, the ordinary differential equation about the boundary value, that is, the design of the divine ordinary differential equation about the boundary in NEIM is as follows: ,

[0071] in, represents the vector concatenation operation, express Encoded abnormal inter-regional population mobility.

[0072] It can be found that through the self-evolution function of the edge And link point information aggregation function , the equation can learn the changing patterns from the factors that may affect the changes in population mobility between regions as comprehensively as possible.

[0073] Since population mobility between regions will have different degrees of impact on population mobility within a region, for example, a large number of people who left a city to escape will return to the city after a disaster. The number of people returning determines how quickly the population mobility within the city area recovers.

[0074] Therefore, for the ordinary differential equation of population mobility within a region, the interaction function of population mobility between regions is added on the basis of the original PIDM. In other words, the coding module can be expressed as follows: ,

[0075] in, and Similarly, the dimension is The vector coefficients of Representation and Region There is inter-regional population mobility between regions. The ordinary differential equation is similar to, Since then, it also consists of two parts: self-evolution and side information aggregation. In particular, here is given and The calculation method is as follows: ,

[0076] ,

[0077] in, Through A layer of GCN for the convolution operator and a layer of multi-layer perceptron And the absolute value maximum normalization method is used to obtain each The values ​​of the components are restricted to This allows the present invention to capture the unexpected attenuation of population mobility recovery. Represents an MLP (Multilayer Perceptron), express Activation function. express The degree matrix of .

[0078] Furthermore, by solving the ordinary differential equations of population flow coding within and between regions, the population flow coding of all regions and between regions in the post-disaster recovery process is obtained, and the coding results are obtained. The integration process is as follows:

[0079] The integration is carried out at equal intervals to obtain the population flow code of the population flow recovery process after the disaster, and the coding result is obtained: as well as The coding results include the coding of population mobility within abnormal regions and the coding of population mobility between abnormal regions.

[0080] Finally, through the decoding process of the following decoding module, the final population flow prediction value can be obtained. It is worth noting that It is also used in the edge decoding process to ensure that the final topology is consistent with the integration process, as follows, and That is, respectively Population mobility within a region and the region To area Population mobility during the recovery period The predicted value of the day. The decoding module includes: ,

[0081] ,

[0082] in, and Respectively represent Population mobility within a region and the region To area Population mobility during the recovery period The predicted value for the day, express The coded population mobility code in the abnormal area, Represents the decoding function of the MLP represented by the superposition of two linear layers and two RELU activation functions.

[0083] The method for predicting population mobility after a disaster provided by the present invention obtains population mobility data of a target area according to a predetermined spatiotemporal granularity; obtains an input population mobility dynamic graph according to the population mobility data of the target area; wherein the input population mobility dynamic graph at least includes an initial abnormal population mobility graph during the peak period of a disaster in the target area and an average normal population mobility graph during a non-disaster period in the target area; the input population mobility dynamic graph is input into a pre-trained population mobility prediction model to obtain a population mobility prediction result; wherein the population mobility prediction model is based on a coupled dynamic graph neural ordinary differential equation and is trained using population mobility data of a sample area. The present invention uses a coupled dynamic graph neural ordinary differential equation to construct a population mobility prediction model, which can effectively model the diverse and highly volatile urban mobility in different regions, and jointly consider the urban mobility before and after a disaster, solving the problem that the prior art relies on overly simplified assumptions, thereby failing to effectively capture the impact of population mobility between regions, and it is difficult to use the normal flow change pattern to predict the abnormal flow recovery process after a disaster, resulting in low accuracy, and realizing a high-accuracy post-disaster population mobility prediction.

[0084] The following is a description of the post-disaster population flow prediction device provided by the present invention. The post-disaster population flow prediction device described below and the post-disaster population flow prediction method described above can be referred to each other. Figure 5 As shown, the device comprises: A data unit 510 is used to obtain population flow data of a target area according to a predetermined spatiotemporal granularity; The conversion unit 520 is used to obtain an input population flow dynamic map according to the population flow data of the target area; wherein the input population flow dynamic map at least includes an initial abnormal population flow map during the disaster peak period of the target area and an average normal population flow map during the non-disaster period of the target area; A prediction unit 530, used for inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; The population flow prediction model is based on coupled dynamic graph neural ordinary differential equations and is trained using population flow data in the sample area.

[0085] According to a post-disaster population flow prediction device provided by the present invention, based on a coupled dynamic graph neural ordinary differential equation, the population flow data of a sample area is used for training to obtain the population flow prediction model, which specifically includes: Obtain population mobility data in the sample area; According to the population flow data of the sample area, an initial abnormal population flow map of the sample area during the disaster peak period, an average normal population flow map of the sample area during the non-disaster period, and a real abnormal population flow map of the sample area during the recovery stage are obtained; Construct a basic prediction model based on coupled dynamic graph neural ordinary differential equations; The initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are used as input, and the real abnormal population flow map during the recovery stage of the sample area is used as a label to train the basic prediction model until the model converges to obtain a population flow prediction model.

[0086] According to a post-disaster population flow prediction device provided by the present invention, the basic prediction model includes a decoding module and an encoding module; Accordingly, the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are used as inputs, and the real abnormal population flow map during the recovery stage of the sample area is used as a label to train the basic prediction model until the model converges, thereby obtaining a population flow prediction model, which specifically includes: According to the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area, based on the coding module, the inter-area population flow and the intra-area population flow are coded to obtain a coding result; Decoding the encoding result based on the decoding module to obtain population flow prediction values ​​at multiple moments in the post-disaster recovery process; According to the population flow prediction value and the actual abnormal population flow map in the sample area recovery stage, back propagation is performed using a preset loss function to update the model parameters until the model converges, thereby obtaining a population flow prediction model.

[0087] According to a post-disaster population flow prediction device provided by the present invention, the encoding module comprises: ,

[0088] ,

[0089] ,

[0090] in, The dimension is The attenuation coefficient vector, Based on The function of express The encoded population mobility code within the abnormal region has the dimension of , Indicates Area of ​​the day Population mobility within Then After and The population mobility coding in the normal area after the same coding method, Represents the attribute value of each point in the normal population flow diagram, represents the transformation matrix, represents the activation function, Used to indicate whether an edge exists. The dimension is The vector coefficients of Representation and Region There are areas where there is inter-regional population mobility. Softmax represents the Softmax activation function. represents an MLP, express The coded abnormal inter-regional population mobility code, Indicates Area of ​​the day To area population mobility, represents a multi-layer perceptron, represents the convolution operator of GCN, express The degree matrix of .

[0091] According to a post-disaster population flow prediction device provided by the present invention, the decoding module comprises: ,

[0092] ,

[0093] ,

[0094] in, and Respectively represent Population mobility within a region and the region To area Population mobility during the recovery period The predicted value for the day, express The coded population mobility code in the abnormal area, Represents the decoding function of the MLP represented by the superposition of two linear layers and two RELU activation functions.

[0095] According to a post-disaster population flow prediction device provided by the present invention, when the target area and the sample area are the same, the real abnormal population flow map of the sample area in the recovery stage is the real abnormal population flow map of the sample area in the early stage of the recovery stage; In the case where the target area and the sample area are different, the real abnormal population flow map of the sample area in the recovery stage is the real abnormal population flow map of the sample area in the full recovery stage.

[0096] The post-disaster population flow prediction device provided by the present invention obtains the population flow data of the target area according to the predetermined spatiotemporal granularity; obtains the input population flow dynamic graph according to the population flow data of the target area; wherein the input population flow dynamic graph at least includes the initial abnormal population flow graph of the target area during the disaster peak period and the average normal population flow graph of the target area during the non-disaster period; the input population flow dynamic graph is input into the pre-trained population flow prediction model to obtain the population flow prediction result; wherein the population flow prediction model is based on the coupled dynamic graph neural ordinary differential equation and is trained using the population flow data of the sample area. The present invention uses the coupled dynamic graph neural ordinary differential equation to construct a population flow prediction model, which can effectively model the diversified and highly volatile urban mobility in different regions, and jointly consider the urban mobility before and after the disaster, so as to solve the problem that the prior art relies on over-simplified assumptions, thereby failing to effectively capture the impact of population flow between regions, and it is difficult to use the normal flow change pattern to predict the abnormal flow recovery process after the disaster, resulting in low accuracy, and realizes the post-disaster population flow prediction with high accuracy.

[0097] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the post-disaster population flow prediction method, which includes: obtaining the population flow data of the target area according to the predetermined spatiotemporal granularity; obtaining the input population flow dynamic graph according to the population flow data of the target area; wherein the input population flow dynamic graph at least includes the initial abnormal population flow graph of the target area during the disaster peak period and the average normal population flow graph of the target area during the non-disaster period; inputting the input population flow dynamic graph into the pre-trained population flow prediction model to obtain the population flow prediction result; wherein the population flow prediction model is based on the coupled dynamic graph neural ordinary differential equation and is trained using the population flow data of the sample area.

[0098] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0099] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the post-disaster population flow prediction method provided by the above-mentioned methods, which includes: obtaining population flow data of the target area according to a predetermined spatiotemporal granularity; obtaining an input population flow dynamic graph based on the population flow data of the target area; wherein the input population flow dynamic graph at least includes an initial abnormal population flow graph during the disaster peak period of the target area and an average normal population flow graph during the non-disaster period of the target area; inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; wherein the population flow prediction model is based on a coupled dynamic graph neural ordinary differential equation and is trained using population flow data of a sample area.

[0100] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the post-disaster population flow prediction method provided by the above-mentioned methods, the method comprising: obtaining population flow data of a target area according to a predetermined spatiotemporal granularity; obtaining an input population flow dynamic graph according to the population flow data of the target area; wherein the input population flow dynamic graph at least includes an initial abnormal population flow graph of the target area during the disaster peak period and an average normal population flow graph of the target area during the non-disaster period; inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; wherein the population flow prediction model is based on a coupled dynamic graph neural ordinary differential equation, and is trained using population flow data of a sample area.

[0101] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0102] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting population mobility after a disaster, characterized in that: include: Obtain population mobility data in the target area according to predetermined spatiotemporal granularity; An input population flow dynamic map is obtained according to the population flow data of the target area; wherein the input population flow dynamic map at least includes an initial abnormal population flow map during the disaster peak period of the target area and an average normal population flow map during the non-disaster period of the target area; Inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; The population flow prediction model is based on coupled dynamic graph neural ordinary differential equations and is trained using population flow data in the sample area.

2. The method for predicting post-disaster population mobility according to claim 1, characterized in that: Based on the coupled dynamic graph neural ordinary differential equation, the population flow prediction model is obtained by training using the population flow data of the sample area, which specifically includes: Obtain population mobility data in the sample area; According to the population flow data of the sample area, an initial abnormal population flow map of the sample area during the disaster peak period, an average normal population flow map of the sample area during the non-disaster period, and a real abnormal population flow map of the sample area during the recovery stage are obtained; Construct a basic prediction model based on coupled dynamic graph neural ordinary differential equations; The initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are used as input, and the real abnormal population flow map during the recovery stage of the sample area is used as a label to train the basic prediction model until the model converges to obtain a population flow prediction model.

3. The method for predicting post-disaster population mobility according to claim 1, characterized in that: The basic prediction model includes a decoding module and an encoding module; Accordingly, the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area are used as inputs, and the real abnormal population flow map during the recovery stage of the sample area is used as a label to train the basic prediction model until the model converges, thereby obtaining a population flow prediction model, which specifically includes: According to the initial abnormal population flow map during the disaster peak period of the sample area and the average normal population flow map during the non-disaster period of the sample area, based on the coding module, the inter-area population flow and the intra-area population flow are coded to obtain a coding result; Decoding the encoding result based on the decoding module to obtain population flow prediction values ​​at multiple moments in the post-disaster recovery process; According to the population flow prediction value and the actual abnormal population flow map in the sample area recovery stage, back propagation is performed using a preset loss function to update the model parameters until the model converges, thereby obtaining a population flow prediction model.

4. The method for predicting post-disaster population mobility according to claim 3, characterized in that: The encoding module comprises: ， ， ， in, The dimension is The attenuation coefficient vector, Based on The function of express The encoded population mobility code within the abnormal region has the dimension of , Indicates Area of ​​the day Population mobility within Then After and The population mobility coding in the normal area after the same coding method, Represents the attribute value of each point in the normal population flow graph, represents the transformation matrix, represents the activation function, Used to indicate whether an edge exists. The dimension is The vector coefficients of Representation and Region There are areas where there is inter-regional population mobility. Softmax represents the Softmax activation function. represents an MLP, express The coded abnormal inter-regional population mobility code, Indicates Area of ​​the day To area population mobility, represents a multi-layer perceptron, represents the convolution operator of GCN, express The degree matrix of .

5. The method for predicting post-disaster population mobility according to claim 3, characterized in that: The decoding module comprises: ， ， ， in, and Respectively represent Population mobility within a region and the region To area Population mobility during the recovery period The predicted value for the day, express The coded population mobility code in the abnormal area, Represents the decoding function of the MLP represented by the superposition of two linear layers and two RELU activation functions.

6. The method for predicting post-disaster population mobility according to claim 1, characterized in that: In the case where the target area and the sample area are the same, the real abnormal population flow map of the sample area in the recovery stage is the real abnormal population flow map of the sample area in the early stage of the recovery stage; In the case where the target area and the sample area are different, the real abnormal population flow map of the sample area in the recovery stage is the real abnormal population flow map of the sample area in the full recovery stage.

7. A post-disaster population flow prediction device, characterized in that: include: Data unit, used to obtain population flow data of the target area according to the predetermined spatiotemporal granularity; A conversion unit, used to obtain an input population flow dynamic map according to the population flow data of the target area; wherein the input population flow dynamic map at least includes an initial abnormal population flow map during the disaster peak period of the target area and an average normal population flow map during the non-disaster period of the target area; A prediction unit, used for inputting the input population flow dynamic graph into a pre-trained population flow prediction model to obtain a population flow prediction result; The population flow prediction model is based on coupled dynamic graph neural ordinary differential equations and is trained using population flow data in the sample area.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the post-disaster population flow prediction method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the post-disaster population flow prediction method as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the post-disaster population flow prediction method as claimed in any one of claims 1 to 6 is implemented.

Images