Heterogeneity feature-based epidemic propagation prediction method and system

Abstract

The invention discloses an epidemic disease propagation prediction method and system based on heterogeneity characteristics, and relates to the technical field of epidemic disease propagation prediction.The method comprises the steps that a time-space-human heterogeneous network is constructed, and the heterogeneity characteristics of epidemic disease propagation in different times, spaces and crowds are described based on the time-space-human heterogeneous network; constructing epidemic propagation event sequence characterization and rehabilitation event sequence characterization based on heterogeneity characteristics; respectively inputting the propagation event sequence representation and the rehabilitation event sequence representation into a pre-trained model for prediction to respectively obtain propagation force and rehabilitation force; the occurrence time, the place and the propagation number of the next epidemic propagation event are accurately predicted based on the propagation force, and the occurrence time, the place and the rehabilitation number of the next epidemic rehabilitation event are accurately predicted based on the rehabilitation force. The heterogeneity characteristic in the epidemic propagation process is really and reasonably considered, and the occurrence time, the occurrence place and the propagation number of the next event can be predicted more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of epidemic spread prediction, and in particular to an epidemic spread prediction method and system based on heterogeneous characteristics. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Studying the epidemic spread process and predicting the future spread trend of the epidemic are of far-reaching significance for solving these problems. Due to the effectiveness of epidemic spread prediction models in predicting epidemic spread and providing early warning, they have attracted the attention of many researchers. WOKermack and AGMcKendrick pioneered the introduction of the susceptible-infected-removed model, which divides the population into three categories and uses the mean field equation to calculate the number of infected people. Subsequently, complex network models, agent-based models, geographic information system models, and models based on random processes have been proposed to consider the influence of various factors in epidemic spread. However, these models often rely on mathematical assumptions and cannot accurately reflect the real laws of epidemic spread. To solve this problem, epidemic spread prediction models based on deep learning have been widely studied to improve the accuracy of epidemic spread prediction by learning the propagation characteristics of real data. However, the introduction of deep learning methods reduces the interpretability of the model, thereby weakening the credibility of the prediction results. Models based on point processes can simulate random transmission and recovery events in epidemic spread, making predictions more interpretable and credible. Therefore, point processes are combined with deep learning methods to predict epidemic spread.

[0004] However, existing point-process-based models mainly focus on the temporal dimension of epidemic spread events, lacking consideration of spatial heterogeneity and population heterogeneity. How to improve prediction accuracy by capturing the heterogeneous characteristics of epidemic spread in different time, space, and populations remains a challenging task. Summary of the invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides an epidemic spread prediction method and system based on heterogeneous characteristics, which truly and reasonably considers the heterogeneous characteristics in the epidemic spread process.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention provides a method for predicting the spread of an epidemic based on heterogeneous characteristics, comprising:

[0008] Constructing a "time-space-people" heterogeneous network, based on which the heterogeneous characteristics of the spread of the epidemic in different times, spaces and populations are described;

[0009] Based on the heterogeneous characteristics, a sequence representation of epidemic transmission events and a sequence representation of recovery events are constructed; the transmission event sequence representation and the recovery event sequence representation are respectively input into a pre-trained model for prediction to obtain transmission power and recovery power respectively;

[0010] Based on the transmission force, the time, location and number of transmission of the next epidemic transmission event are accurately predicted; based on the recovery force, the time, location and number of recovery of the next epidemic recovery event are accurately predicted.

[0011] A further technical solution is that the "time-space-people" heterogeneous network is represented by a graph, in which each node represents a heterogeneous propagation space with heterogeneous characteristics, including time characteristics, space characteristics and population characteristics.

[0012] A further technical solution is to construct a representation of the epidemic transmission / recovery event sequence as follows: construct an epidemic transmission / recovery event sequence, use a graph convolutional network to learn the node representation in the "time-space-human" heterogeneous network, embed the learned node representation into the transmission / recovery event sequence, and obtain the representation of the epidemic transmission / recovery event sequence.

[0013] A further technical solution is to improve the traditional deep Hawkes method and convert the propagation force λ of the heterogeneous propagation space v at time t I (t,v) is expressed as:

[0014]

[0015] h(t)=o j ⊙(2σ(2C(t))-1 )t>t j

[0016] f sp (x) = log(1 + exp(x))

[0017] Among them, h(t) is the hidden state of the event sequence, w T 、b v is a trainable parameter, o j is the output gate, σ is the sigmoid activation function, and C(t) represents the storage unit at time t.

[0018] A further technical solution is that the pre-trained model is a long short-term memory network model, and the hidden state of the event sequence is obtained through the long short-term memory network model. The model takes the propagation event sequence representation and the rehabilitation event sequence representation as input respectively, and passes the data to three fully connected layers by using the sigmoid activation function to calculate the values ​​of the input gate, the forgetting gate and the output gate.

[0019] In a further technical solution, the long short-term memory network model uses the negative log-likelihood of the propagation event sequence and the recovery event sequence as a loss function.

[0020] A further technical solution is to accurately predict the time, location and number of transmission of the next epidemic transmission event based on the transmission force:

[0021] In the heterogeneous propagation space, the time when the next epidemic propagation event occurs is:

[0022]

[0023] At this point, the heterogeneous propagation space where the propagation event is most likely to occur is:

[0024]

[0025] The expected total number of propagation events occurring within the time window [t,T) is:

[0026]

[0027] in, Is the node v in is the probability of a propagation event occurring when , and V represents the set of all heterogeneous propagation spaces.

[0028] In a second aspect, the present invention provides an epidemic spread prediction system based on heterogeneous characteristics, comprising:

[0029] A heterogeneous feature acquisition module is configured to: construct a "time-space-human" heterogeneous network, and describe the heterogeneous features of the spread of the epidemic in different times, spaces and populations based on the "time-space-human" heterogeneous network;

[0030] A transmission / recovery prediction module is configured to: construct an epidemic transmission event sequence representation and a recovery event sequence representation based on the heterogeneous characteristics; input the transmission event sequence representation and the recovery event sequence representation into a pre-trained model for prediction, and obtain transmission and recovery respectively;

[0031] An epidemic transmission process prediction module is configured to: accurately predict the time, location and number of transmission of the next epidemic transmission event based on the transmission force, and accurately predict the time, location and number of recovery of the next epidemic recovery event based on the recovery force.

[0032] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for predicting epidemic spread based on heterogeneous characteristics as described in the first aspect.

[0033] In a fourth aspect, the present invention provides a computer 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, the steps in the method for predicting epidemic spread based on heterogeneous characteristics as described in the first aspect are implemented.

[0034] One or more of the above technical solutions have the following beneficial effects:

[0035] The present invention first constructs a "time-space-human" heterogeneous network to describe the heterogeneous characteristics of epidemic spread in different times, spaces and populations. The network can describe the "time-space-human" heterogeneous characteristics of epidemic spread by defining the transmission / recovery capabilities of different times, spaces and populations during the epidemic spread process; then, on the basis of the "time-space-human" heterogeneous network, a transmission / recovery estimation method based on the "time-space-human" heterogeneous characteristics is proposed. This method comprehensively considers the impact of the heterogeneity of time, space and population characteristics on the spread of epidemics, and the estimated transmission / recovery capabilities are closer to the real and complex transmission environment; finally, based on the estimated "time-space-human" transmission / recovery capabilities, a method for predicting epidemic spread events based on the "time-space-human" transmission / recovery capabilities is proposed, which can make more fine-grained predictions on the transmission / recovery capabilities, and can more accurately predict the time and place of the next event and the number of transmissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0037] Figure 1 is a schematic diagram of a heterogeneous propagation space in an embodiment of the present invention;

[0038] Figure 2 4 is a flow chart of an epidemic spread prediction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0042] Epidemic spread prediction is a hot research field that helps make intervention decisions in advance to avoid the outbreak of the spread scale. However, since it is difficult to capture the heterogeneous characteristics of epidemic spread in different time, space and populations, how to improve the prediction accuracy remains a challenging task.

[0043] Embodiment 1

[0044] like Figure 2 As shown, this embodiment discloses a method for predicting epidemic spread based on heterogeneity characteristics, which includes the following steps:

[0045] S1: Construct a “time-space-people” heterogeneous network, and describe the heterogeneous characteristics of the spread of epidemics in different time, space and populations based on the “time-space-people” heterogeneous network;

[0046] In this embodiment, a heterogeneous transmission space is defined: a heterogeneous transmission space refers to geographical or population areas with different levels of epidemic transmission defined by the health department based on various factors such as the distance from the infection source. Let v k ∈V represents the heterogeneous propagation space, where 0≤k≤N, N represents the total number of heterogeneous propagation spaces, and V represents the set of all heterogeneous propagation spaces.

[0047] Heterogeneous transmission space is mainly used to measure the different transmission levels caused by spatial heterogeneity during the spread of an epidemic. In real life, the scale of heterogeneous transmission space can be scaled to communities, cities, or even provinces based on the spatial relationship of the population.

[0048] like Figure 1As shown, the present invention divides individuals into three different heterogeneous propagation spaces (v1, v2, v3) according to the geographical location information of the population. When t = 0, it can obtain the initial epidemic propagation characteristics. Among them, the propagation source (individual 0) is in the heterogeneous propagation space v1, and obviously has a higher propagation level; heterogeneous propagation spaces v2 and v3 do not have epidemic individuals, so they have a lower propagation level.

[0049] Constructing a “time-space-human” heterogeneous network model: The “time-space-human” heterogeneous network can be represented by a graph G(V,E), which includes an undirected graph of a heterogeneous propagation space set V and an edge set E. Where V represents the set of all heterogeneous propagation spaces, e mn ∈E represents the heterogeneous propagation space v m With v n The edge between two regions v m and v n Edges are created when they are geographically adjacent, and E represents edge e mn A collection of .

[0050] Each node in the "time-space-people" heterogeneous network represents a heterogeneous communication space with "time-space-people" heterogeneity characteristics. Among them, the time feature is the node weight ω that changes with time, which represents the time difference from the last event to time t; the spatial feature is the heterogeneous communication space represented by the node itself; the crowd characteristics include physical characteristics and social characteristics. Among them, the physical characteristics are represented by the density of the crowd ρ, which corresponds to the density of the crowd in the communication space; the social characteristics are represented by the contact probability of the crowd φ, which corresponds to the probability of contact between people in the communication space.

[0051] The present invention assumes that people in different transmission spaces at a certain moment have three states: susceptible (S), infected (I) and recovered (R). The present invention describes the epidemic transmission process based on the heterogeneous characteristics of "time-space-people" by modeling heterogeneous "time-space-people" epidemic transmission events and transmission / recovery capabilities. The epidemic transmission process includes "time-space-people" transmission events and recovery events.

[0052] Further, “time-space-people” transmission / recovery events: transmission events and recovery events are two types of events in which an epidemic spreads in different time, space, and populations, denoted by e I (t,v) and e R (t,v).e I The occurrence of (t,v) means that at time t, a person in space v is infected with the epidemic, and the person's state changes from S to I. Similarly, the recovery event e R The occurrence of (t,v) indicates that at time t a person in space v recovers and his state changes from I to R.

[0053] The occurrence rates of epidemic transmission events and recovery events vary in different times, spaces and populations. The present invention uses "time-space-people" epidemic transmission and recovery forces to describe these heterogeneous characteristics.

[0054] Further, “time-space-people” transmission / recovery power: “time-space-people” transmission power λ I (t,v) represents the rate at which a transmission event occurs in a certain time, space, and population, and the recovery capacity λ R (t,v) usually represents the rate at which recovery events occur in a certain time, space, and population.

[0055] S2: constructing an epidemic transmission event sequence representation and a recovery event sequence representation based on the heterogeneous characteristics; inputting the transmission event sequence representation and the recovery event sequence representation into a pre-trained model for prediction, and obtaining transmission power and recovery power respectively;

[0056] In this embodiment, a method for estimating the transmission / recovery capacity based on the heterogeneous characteristics of "time-space-people" is proposed. First, a "time-space-people" epidemic transmission / recovery event sequence representation is constructed. Then, a method for estimating the transmission / recovery capacity of the epidemic based on the "time-space-people" deep Hawkes is proposed.

[0057] S201: Construct a “time-space-people” representation of the epidemic transmission / recovery event sequence.

[0058] Epidemic event sequences include the time of occurrence of the event, the propagation space to which it belongs, and the characteristics of the population, implying the "time-space-people" information of the epidemic spread. The present invention constructs the epidemic "time-space-people" propagation event sequence as follows:

[0059] Q I ={e I (t1,v(1)),…,e I (t i ,v(i)),...,e I (t L ,v(L))}

[0060] Among them, Q I is the sequence of propagation events, t i and v(i) are the occurrence time and propagation space of the ith event, respectively, e I is the transmission event, L is the number of events. Similarly, the rehabilitation event sequence Q can also be constructed R .

[0061] In order to obtain the representation of the "time-space-people" epidemic transmission event sequence, we must first learn the node representation in the "time-space-people" heterogeneous network, and then embed the node representation into the event sequence to finally obtain the representation of the event sequence.

[0062] Since the spread of an epidemic in a propagation space is often affected by the surrounding propagation spaces, the present invention uses a graph convolutional network to learn the representation of nodes in the "time-space-human" heterogeneous network. Let ρ i ,ω i , are the density value, time weight value and contact probability value of node i respectively, then the embedding representation of the initialized node is It can be expressed as:

[0063]

[0064] in, The initial representation vector of the node, F is the vector dimension; but An iterative method is used to calculate the embedding vector of the node at time t, where the l-th iteration can be expressed as:

[0065]

[0066] Among them, I is the identity matrix, A is the adjacency matrix of the "time-space-person" heterogeneous network, D is the degree matrix, and Y (0) To initialize the representation matrix, W l is a trainable network parameter, σ is a nonlinear activation function, A matrix constructed for properties such as node density, time weight, and contact probability. Here, in order to better maintain the node properties during the iterative update process, the node density, time weight, and contact probability properties are concatenated with the node representation. In the graph convolutional network, a self-loop is added to each node to consider the impact of node information on itself. Each row vector of Y represents the representation of the influence of node i on its surrounding nodes at time t.

[0067] Embed the learned node representation into the event sequence Q I or Q R By embedding it into the corresponding vector, we can get the representation of the "time-space-person" epidemic transmission event sequence. In other words, by embedding it into the vector corresponding to the transmission event sequence, we can get the transmission event sequence representation; by embedding it into the vector corresponding to the recovery event sequence, we can get the recovery event sequence representation.

[0068] S202: Epidemic transmission / recovery capacity estimation method based on "time-space-people" deep Hawkes.

[0069] A method for estimating the epidemic transmission / recovery capacity based on the "time-space-human" deep Hawkes is proposed. This method can use the "time-space-human" epidemic transmission event sequence as input to accurately predict the "time-space-human" epidemic transmission / recovery capacity.

[0070] The traditional deep Hawkes' description of propagation force can be expressed by the following formula:

[0071] λ I (t) = f sp (w T h(t))where f sp (x) = log(1 + exp(x)), h(t) is the hidden state of the event sequence, w T is a trainable parameter.

[0072] In this method, the traditional deep Hawkes method is improved by converting the propagation force λ of the heterogeneous propagation space v at time t into I (t,v) is expressed as:

[0073]

[0074] h(t)=o j ⊙(2σ(2C(t))-1 t>t j

[0075] Among them, h(t) is the hidden state of the event sequence, which can be obtained through the continuous time long short-term memory network (LSTM, Long Short-Term Memory); w T 、b v is a trainable parameter, o j is the output gate, σ is the sigmoid activation function, C(t) represents the storage unit at time t, and each storage unit decays exponentially to a steady-state value. The long short-term memory network model LSTM takes the previously acquired "time-space-person" epidemic transmission event sequence as input, and then passes the data to three fully connected layers using the sigmoid activation function σ to calculate the values ​​of the input gate, forget gate, and output gate. It maintains long-term dependencies by storing information across time steps.

[0076] The propagation force Λ in the entire space at time t I (t) can be expressed as:

[0077] Λ I (t)=∑ v∈V λ I (t,v).

[0078] Similarly, the rehabilitation event history vector and event occurrence time can be used to predict the rehabilitation force λ of the heterogeneous communication space v at time t R (t,v).

[0079] During LSTM training, the goal is to maximize the probability of an event occurring in a sequence. Therefore, the negative log-likelihood of the transmission and recovery event sequences is used as the loss function. To achieve this, the probability of an event occurring is first derived based on the previously calculated “time-space-person” transmission / recovery forces. Based on λ I (t,v), the probability f of the next propagation event occurring in space v at time t I (t,v) can be expressed as:

[0080]

[0081] Therefore, no matter in which propagation space it occurs, the probability F of the next propagation event occurring at time t is I (t) is:

[0082]

[0083] Furthermore, given the next event time t i In the case of , the probability of a propagation event occurring in the propagation space v is:

[0084]

[0085] Then, the propagation event sequence Q I ={e I (t1,v(1)),…,e I (t i ,v(i)),…e I (t L ,v(L))} is:

[0086]

[0087] So, the negative log-likelihood is:

[0088]

[0089] Assume there are N propagation event sequences The goal is to find the network parameters that minimize the loss function:

[0090]

[0091] During training, there are two different types of event sequences, namely propagation event sequences and rehabilitation event sequences. The minimization of the loss function can be expanded as follows:

[0092]

[0093] It should be noted that M is the number of rehabilitation event sequences, l R (·) is the negative log-likelihood of the sequence of recovery events and can be derived in the same way as for transmission events.

[0094] S3: Based on the transmission force, accurately predict the time, location and number of transmission of the next epidemic transmission event; based on the recovery force, accurately predict the time, location and number of recovery of the next epidemic recovery event.

[0095] In this embodiment, according to the estimated "time-space-people" transmission power / recovery power, a method for predicting epidemic transmission events based on "time-space-people" transmission power / recovery power is proposed, which can more accurately predict the time and place of the next event and the number of transmissions. Transmission events and recovery events are predicted separately, for example, the number of transmission events is predicted using transmission power, and the number of recovery events is predicted using recovery power, and then the predicted number of transmission events-predicted number of recovery events=current number of newly added transmissions.

[0096] The time of occurrence of the next propagation event in each propagation space v It can be calculated using the following formula:

[0097]

[0098] In addition, no matter in which heterogeneous propagation space the next propagation event occurs, for:

[0099]

[0100] At this point, the heterogeneous propagation space where the propagation event is most likely to occur for:

[0101]

[0102] in, Is the node v in The probability of a propagation event occurring at time t, V represents the set of all heterogeneous propagation spaces. The expected number of propagation events occurring in region v within the time window [t, T) is:

[0103]

[0104] The expected total number of propagation events occurring within the time window [t,T) is:

[0105]

[0106] In addition, the heterogeneous propagation space v in the time window [t L ,T) is:

[0107]

[0108] where I0(v) is the number of propagations in a given sequence of events, where t≤t L The cumulative number of propagations in all heterogeneous propagation spaces [t0, T) is:

[0109]

[0110] Similarly, the time of the next recovery event in each heterogeneous propagation space can be estimated That is, the number of individuals E expected to recover in different heterogeneous propagation spaces within the time window [t,T) R (t,T,v), and the total number of expected recovered individuals In addition, the cumulative number R(T,v) in the recovery state in [t0, T) in the heterogeneous propagation space v can be calculated similarly:

[0111]

[0112] Where R0(v) is the number of recoveries in a given event sequence, E R (t, T, v) is the expected number of rehabilitation events occurring in region v within the time window [t, T), λ R (t, v) is the “time-space-person” recovery capacity of the heterogeneous communication space v at time t.

[0113] The cumulative number of recoveries R(T) in all heterogeneous propagation spaces in time [t0, T) is:

[0114]

[0115] Among them, Λ R (t)=∑ v∈V λ R (t, v) is the propagation force of the entire space at time t.

[0116] Finally, the expected number of transmissions at time T can be expressed as:

[0117] I now (T) = I(T) - R(T).

[0118] In summary, the time, space, and number of transmission of future events can be calculated, thereby achieving accurate prediction of epidemic spread based on the heterogeneous characteristics of "time-space-people".

[0119] Embodiment 2

[0120] This embodiment discloses an epidemic spread prediction system based on heterogeneity characteristics, including:

[0121] A heterogeneous feature acquisition module is configured to: construct a "time-space-human" heterogeneous network, and describe the heterogeneous features of the spread of the epidemic in different times, spaces and populations based on the "time-space-human" heterogeneous network;

[0122] A transmission / recovery prediction module is configured to: construct an epidemic transmission event sequence representation and a recovery event sequence representation based on the heterogeneous characteristics; input the transmission event sequence representation and the recovery event sequence representation into a pre-trained model for prediction, and obtain transmission and recovery respectively;

[0123] An epidemic transmission process prediction module is configured to: accurately predict the time, location and number of transmission of the next epidemic transmission event based on the transmission force, and accurately predict the time, location and number of recovery of the next epidemic recovery event based on the recovery force.

[0124] Embodiment 3

[0125] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of embodiment 1 when executing the program.

[0126] Embodiment 4

[0127] The purpose of this embodiment is to provide a computer-readable storage medium, a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the method of embodiment 1 are performed.

[0128] The steps involved in the apparatus of the above embodiments 3 and 4 correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0129] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0131] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. Epidemic spread prediction method based on heterogeneity characteristics, characterized in that: include: Constructing a "time-space-people" heterogeneous network, based on which the heterogeneous characteristics of the spread of the epidemic in different times, spaces and populations are described; Constructing an epidemic transmission event sequence representation and a recovery event sequence representation based on the heterogeneous characteristics; Inputting the transmission event sequence representation and the recovery event sequence representation into a pre-trained model for prediction, respectively, to obtain the transmission power and the recovery power; Based on the transmission force, the time, location and number of transmission of the next epidemic transmission event are accurately predicted; based on the recovery force, the time, location and number of recovery of the next epidemic recovery event are accurately predicted.

2. The epidemic spread prediction method based on heterogeneous characteristics according to claim 1, characterized in that: The "time-space-people" heterogeneous network is represented by a graph, in which each node represents a heterogeneous propagation space with heterogeneous characteristics, including time characteristics, space characteristics and population characteristics.

3. The method for predicting epidemic spread based on heterogeneous characteristics according to claim 1, characterized in that: Constructing the representation of the epidemic transmission / recovery event sequence is specifically as follows: constructing the epidemic transmission / recovery event sequence, using a graph convolutional network to learn the node representation in the "time-space-human" heterogeneous network, embedding the learned node representation into the transmission / recovery event sequence, and obtaining the representation of the epidemic transmission / recovery event sequence.

4. The method for predicting epidemic spread based on heterogeneous characteristics according to claim 1, characterized in that: The traditional deep Hawkes method is improved by converting the propagation force λ of the heterogeneous propagation space v at time t into I (t, v) is expressed as: h(t)=o j ⊙(2σ(2C(t))-1)t>t j f sp (x)=log(1+exp(x)) Among them, h(t) is the hidden state of the event sequence, w T 、b v is a trainable parameter, o j is the output gate, σ is the sigmoid activation function, and C(t) represents the storage unit at time t.

5. The method for predicting epidemic spread based on heterogeneous characteristics according to claim 1, characterized in that: The pre-trained model is a long short-term memory network model, through which the hidden state of the event sequence is obtained. The model takes the propagation event sequence representation and the rehabilitation event sequence representation as input respectively, and passes the data to three fully connected layers using the sigmoid activation function to calculate the values ​​of the input gate, the forgetting gate and the output gate.

6. The method for predicting epidemic spread based on heterogeneous characteristics according to claim 5, characterized in that: The LSTM network model uses the negative log-likelihood of the propagation event sequence and the rehabilitation event sequence as the loss function.

7. The method for predicting epidemic spread based on heterogeneous characteristics according to claim 1, characterized in that: Based on the transmission force, the time, location and number of transmission of the next epidemic transmission event are accurately predicted as follows: In the heterogeneous propagation space, the time when the next epidemic propagation event occurs is: At this point, the heterogeneous propagation space where propagation events are most likely to occur is: The expected total number of propagation events occurring within the time window [t,T) is: in, Is the node v in is the probability of a propagation event occurring when , and V represents the set of all heterogeneous propagation spaces.

8. Epidemic spread prediction system based on heterogeneity characteristics, characterized by: include: A heterogeneous feature acquisition module is configured to: construct a "time-space-human" heterogeneous network, and describe the heterogeneous features of the spread of the epidemic in different times, spaces and populations based on the "time-space-human" heterogeneous network; A transmission / recovery prediction module is configured to: construct an epidemic transmission event sequence representation and a recovery event sequence representation based on the heterogeneous characteristics; input the transmission event sequence representation and the recovery event sequence representation into a pre-trained model for prediction, and obtain transmission and recovery respectively; An epidemic transmission process prediction module is configured to: accurately predict the time, location and number of transmission of the next epidemic transmission event based on the transmission force, and accurately predict the time, location and number of recovery of the next epidemic recovery event based on the recovery force.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the epidemic spread prediction method based on heterogeneous characteristics as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the epidemic spread prediction method based on heterogeneous characteristics as described in any one of claims 1 to 7 are implemented.

Images