Intelligent positioning system for sudden infectious disease based on spatiotemporal distribution anomaly detection

Abstract

The application relates to the field of public health information technology, and particularly discloses an intelligent positioning system for sudden infectious diseases based on spatiotemporal distribution anomaly detection, which comprises a data acquisition module, a spatiotemporal feature modeling module, a spatiotemporal distribution anomaly detection module, an intelligent infectious disease positioning module, a transmission trend prediction module and a visualization module. The scheme introduces a multi-dimensional Gaussian triple modeling scheme, introduces expectation, entropy and hyper-entropy parameters to finely describe the center position, change range and instability of the spatiotemporal distribution features, combines multi-criteria similarity calculation and a dynamic cluster merging mechanism to realize accurate capture of weak anomalies, introduces a graph structure modeling method to construct an explanatory graph structure and a layer-by-layer attribution trajectory, introduces a trajectory volatility metric in the hierarchical clustering process to improve the robust detection capability of abnormal points, and realizes high-precision anomaly recognition and infection source positioning of sudden infectious diseases, and improves the intelligent level and timeliness of epidemic response.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of public health information technology, in particular to an intelligent positioning system for sudden infectious diseases based on spatio-temporal distribution anomaly detection. BACKGROUND

[0002] With the intensification of population mobility and the advancement of urbanization process, the frequency and speed of outbreak of sudden infectious diseases have significantly increased, which seriously threatens public health safety. In the prior art, some methods attempt to use clustering or anomaly detection technology for epidemic monitoring, but most of them have the following shortcomings:

[0003] (1) The existing clustering method ignores the dynamic uncertainty of spatio-temporal features, lacks modeling of spatio-temporal distribution feature uncertainty and evolution trend, and only clusters based on static feature similarity, which makes it difficult to effectively identify the weak but key spatio-temporal anomaly signals hidden in the development process of the epidemic;

[0004] (2) The existing clustering method has the problems of insufficient modeling of similarity structure and being easily affected by noise, lacks in-depth modeling of similarity structure and topological relationship between nodes, and is easily disturbed by boundary ambiguity, isolated points and structural noise in the clustering process, resulting in unstable anomaly detection results and low accuracy of infection source positioning and transmission prediction. SUMMARY

[0005] In view of the above situation, in order to overcome the defects of the prior art, the present application provides an intelligent positioning system for sudden infectious diseases based on spatio-temporal distribution anomaly detection. In view of the problem that the existing clustering method ignores the dynamic uncertainty of spatio-temporal features, the present application proposes a multi-dimensional Gaussian triple modeling scheme, introduces expectation, entropy and hyper-entropy parameters to finely describe the center position, variation range and instability of spatio-temporal distribution features, combines multi-criteria similarity calculation and dynamic cluster merging mechanism, and realizes accurate capture of weak anomalies; In view of the problem that the existing clustering method has insufficient modeling of similarity structure and is easily affected by noise, the present application introduces a graph structure modeling method to construct an explanatory graph structure and a layer-by-layer attribution trajectory, and introduces a trajectory volatility metric in the hierarchical clustering process to improve the robust detection capability of abnormal points and avoid boundary ambiguity or isolated point interference.

[0006] The technical scheme adopted by the present application is as follows: The intelligent positioning system for sudden infectious diseases based on spatio-temporal distribution anomaly detection provided by the present application comprises a data acquisition module, a spatio-temporal feature modeling module, a spatio-temporal distribution anomaly detection module, an infectious disease intelligent positioning module, a transmission trend prediction module and a visualization module, and specifically comprises the following contents:

[0007] The data acquisition module collects infectious disease related data, including case reporting data, personnel activity trajectory data, environmental meteorological data and regional population density data;

[0008] The spatiotemporal feature modeling module uses a spatiotemporal graph convolutional network to perform spatiotemporal encoding and modeling of infectious disease-related data, and extracts the spatiotemporal distribution features of infectious diseases;

[0009] The spatiotemporal distribution anomaly detection module is designed with a hierarchical clustering method that integrates multi-criteria similarity and dynamic cluster merging mechanism to detect anomalies in spatiotemporal distribution features and obtain anomaly detection results.

[0010] The infectious disease intelligent location module uses a graph neural network model to construct an infection source tracing map based on the anomaly detection results, and locates the potential infection source location;

[0011] The transmission trend prediction module constructs an infectious disease transmission localization model based on a graph diffusion mechanism based on possible sources of infection to obtain future risk transmission areas;

[0012] The visualization module graphically displays abnormal detection results, the location of potential infection sources, and areas of future risk of transmission.

[0013] Furthermore, the spatiotemporal distribution anomaly detection module employs a hierarchical clustering method that integrates multi-criteria similarity with a dynamic cluster merging mechanism to detect anomalies in spatiotemporal distribution features. Specifically, it includes the following steps:

[0014] Step S1: Input acquisition, receiving spatiotemporal distribution features, represented as follows:

[0015] ;

[0016] in, Represents a set of spatiotemporal distribution characteristics. Indicates the first A spatiotemporal distribution feature vector, Indicates an index;

[0017] Step S2: Extend the modeling by designing a Gaussian distribution mapping model that maps each spatiotemporal distribution feature vector to its corresponding multidimensional Gaussian triplet parameters, as shown below:

[0018] ;

[0019] in, For the first The expected value of a spatiotemporal distribution eigenvector represents the center of the spatiotemporal distribution. and For the first The entropy and hyperentropy of a spatiotemporal distribution eigenvector represent the range of uncertainty and the instability of the spatiotemporal distribution's changes. Indicates the first A spatiotemporal distribution feature vector;

[0020] Step S3: Calculate the similarity matrix, design a density similarity function based on the multi-dimensional Gaussian triplet parameters, and calculate the comprehensive similarity score between any two feature points. The density similarity function used is as follows:

[0021] ;

[0022] ;

[0023] wherein, represents the comprehensive similarity of the first and the second spatiotemporal distribution feature vector, is a smoothing factor to avoid division by zero constants, is a regulating function based on hyperentropy;

[0024] Step S4: Abnormal clustering, based on the correlation structure between spatiotemporal features, based on the similarity matrix between multi-dimensional Gaussian triplets, design a hierarchical clustering method that integrates multi-criteria similarity and dynamic cluster merging mechanism, perform anomaly detection on the similarity matrix, and obtain the abnormal clustering result;

[0025] Step S5: Fusion feature construction, combine the abnormal clustering result with the spatiotemporal distribution feature set to construct a fusion feature vector;

[0026] Step S6: Output the result, input the fusion feature vector into the classification model based on gradient boosting tree, perform anomaly discrimination, and output the anomaly detection result corresponding to each spatiotemporal distribution feature vector.

[0027] Further, step S4, a hierarchical clustering method that integrates multi-criteria similarity and dynamic cluster merging mechanism is designed to perform anomaly detection on the similarity matrix, which specifically includes the following steps:

[0028] Step S41: Multi-strategy similarity matrix expansion, introduce a multi-strategy similarity fusion mechanism based on the similarity matrix, and the fused similarity matrix is represented as:

[0029] ;

[0030] wherein, is the fused similarity matrix, , and represent adjustable parameters that satisfy , satisfy , represent the scale adjustment parameters of the Gaussian kernel function;

[0031] Step S42: initial cluster construction, based on the fused similarity matrix, define each spatio-temporal distribution feature vector corresponding to an initial cluster;

[0032] Step S43: construction of an explanatory similarity view, constructing an explanatory graph structure , the node set is , and the edge set is ;

[0033] Step S44: multi-level similarity fusion, introducing a clustering hierarchy definition, denoting the i-th level as , the level i represents the cluster division state formed after the i-th round of similarity merging, and a hierarchical clustering strategy based on adaptive adjustment of the multi-round similarity threshold is designed to perform layer-by-layer merging of the explanatory graph structure to obtain the final clustering structure;

[0034] Step S45: anomaly boundary detection, based on the final clustering structure, an anomaly detection method based on inter-layer cluster structure change analysis is designed to identify abnormal points;

[0035] Step S46: abnormal point output, the abnormal points identified in step S45 are arranged into an abnormal point set as potential spatio-temporal distribution abnormal points.

[0036] Further, step S44 specifically includes the following steps:

[0037] Step S441: cluster similarity measurement construction, each node in the explanatory graph structure represents a spatio-temporal distribution feature vector, according to step S42, each node also constitutes a cluster, and the similarity between any two clusters is defined as:

[0038] ;

[0039] wherein represents the average similarity of clusters under the i-th level;

[0040] Step S442: graph structure layer merging, set an initial similarity threshold, when the similarity between any cluster pair satisfies , then and are merged into a new cluster , and the explanatory graph structure is updated to the next level , wherein the node set is the merged cluster;

[0041] ​​​Step S443: Threshold self-adaptive adjustment, introduce clustering variation index to update threshold adjustment factor, the threshold adjustment factor update formula is as follows:

[0042] ;

[0043] Wherein, Indicates The similarity threshold of the level, Indicates The similarity threshold of the level, Threshold adjustment factor, Clustering variation index;

[0044] Step S444: Termination decision, set the maximum number of iterations, when the maximum number of iterations is reached, output the final clustering structure , into step S45 to perform abnormal boundary detection.

[0045] Further, step S45, specifically includes the following steps:

[0046] Step S451: Belonging trajectory construction, for the unfolding step of step S44, record the label change of each spatiotemporal distribution feature vector belonging to the cluster in each level, define the level belonging trajectory as:

[0047] ;

[0048] Wherein, Indicates the Spatiotemporal distribution feature vector in the first layer to the Layer clustering label sequence, Indicates the Spatiotemporal distribution feature vector belongs to the cluster number of the Layer, Indicates the maximum level number of the clustering process;

[0049] Step S452: Trajectory fluctuation measure construction, define the trajectory fluctuation index, as follows:

[0050] ;

[0051] Wherein, Indicates the Spatiotemporal distribution feature vector trajectory fluctuation index, Indicating function, the current layer and the next layer cluster label is not consistent, then 1, otherwise 0;

[0052] Step S453: Abnormal point screening, set the fluctuation threshold, filter out the spatiotemporal distribution feature vector whose fluctuation is significantly higher than the overall average, recorded as:

[0053] ;

[0054] wherein, is an abnormal point, indicating a subset of spatiotemporal distribution features identified as an abnormal boundary point, represents an abnormality discrimination threshold of trajectory volatility.

[0055] The beneficial effects achieved by the above-mentioned scheme are as follows:

[0056] (1) In view of the problem that the existing clustering method ignores the dynamic uncertainty of spatiotemporal features, the present application proposes a multi-dimensional Gaussian triple modeling scheme, introduces expectation, entropy, and hyper-entropy parameters to finely describe the center position, variation range, and instability of spatiotemporal distribution features, combines multi-criteria similarity calculation and dynamic cluster merging mechanism, and realizes accurate capture of weak anomalies.

[0057] (2) In view of the problem that the existing clustering method is insufficient in modeling similarity structure and is easily affected by noise, the present application introduces a graph structure modeling method, constructs an explanatory graph structure and a layer-by-layer attribution trajectory, introduces trajectory volatility measurement in the hierarchical clustering process, improves the robust detection capability of abnormal points, and avoids the interference of boundary ambiguity or isolated points. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 is a schematic diagram of the intelligent positioning system for sudden infectious diseases based on spatiotemporal distribution anomaly detection proposed by the present application;

[0059] Figure 2 is a schematic diagram of the method flow for anomaly detection by the spatiotemporal distribution anomaly detection module proposed by the present application.

[0060] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0062] Embodiment one, refer to Figure 1 The intelligent positioning system for sudden infectious diseases based on spatiotemporal distribution anomaly detection provided by the present application comprises a data acquisition module, a spatiotemporal feature modeling module, a spatiotemporal distribution anomaly detection module, an infectious disease intelligent positioning module, a transmission trend prediction module, and a visualization module, and specifically comprises the following contents:

[0063] The data collection module collects infectious disease related data, including case reporting data, personnel activity trajectory data, environmental meteorological data and regional population density data;

[0064] The spatio-temporal feature modeling module uses a spatio-temporal graph convolution network to perform spatio-temporal coding and modeling on the infectious disease related data, and extracts spatio-temporal distribution features of the infectious disease;

[0065] The spatio-temporal distribution anomaly detection module designs a hierarchical clustering method that fuses multi-criteria similarity and dynamic cluster merging mechanism to perform anomaly detection on the spatio-temporal distribution features, and obtains an anomaly detection result;

[0066] The infectious disease intelligent positioning module uses a graph neural network model to construct an infection source tracing graph according to the anomaly detection result, and locates a potential infection source position;

[0067] The transmission trend prediction module constructs an infectious disease transmission positioning model based on a graph diffusion mechanism based on the possible infection source, and obtains a future risk transmission area;

[0068] The visualization module displays the anomaly detection result, the potential infection source position and the future risk transmission area through graphics.

[0069] Embodiment two, see Figure 2 , the embodiment is based on the above embodiment, the spatio-temporal distribution anomaly detection module, a hierarchical clustering method that fuses multi-criteria similarity and dynamic cluster merging mechanism is designed to perform anomaly detection on the spatio-temporal distribution features, comprising the following steps:

[0070] Step S1: input acquisition, receiving spatio-temporal distribution features, represented as follows:

[0071] ;

[0072] Among them, represents a set of spatio-temporal distribution features, represents the th spatio-temporal distribution feature vector, represents the index;

[0073] Step S2: extension modeling, a Gaussian distribution mapping model is designed to map each spatio-temporal distribution feature vector to the corresponding multi-dimensional Gaussian triple parameters, represented as follows:

[0074] ;

[0075] Among them, is the th spatio-temporal distribution feature vector Expectation value, representing the spatio-temporal distribution center, and are the entropy and hyperentropy of the spatiotemporal distribution feature vector, representing the uncertainty range of the spatiotemporal distribution and the instability degree of the spatiotemporal distribution change, representing the first spatiotemporal distribution feature vector;

[0076] Step S3: calculating a similarity matrix, designing a density similarity function based on the multi-dimensional Gaussian triplet parameters, and calculating the comprehensive similarity score between any two feature points, wherein the density similarity function used is as follows:

[0077]

[0078]

[0079] wherein, representing the comprehensive similarity between the first spatiotemporal distribution feature vector and the second spatiotemporal distribution feature vector, is a smoothing factor to avoid a constant of division by zero, is a regulating function based on the hyperentropy;

[0080] Step S4: abnormal clustering, based on the correlation structure between the spatiotemporal features, based on the similarity matrix between the multi-dimensional Gaussian triplets, designing a hierarchical clustering method that fuses the multi-criteria similarity and the dynamic cluster merging mechanism, performing abnormal detection on the similarity matrix, and obtaining the abnormal clustering result;

[0081] Step S5: fusion feature construction, combining the abnormal clustering result with the spatiotemporal distribution feature set to construct a fusion feature vector;

[0082] Step S6: outputting the result, inputting the fusion feature vector into a classification model based on gradient boosting tree, performing abnormal discrimination, and outputting the abnormal detection result corresponding to each spatiotemporal distribution feature vector.

[0083] In this embodiment, the spatiotemporal distribution data of cases in 5 regions is collected, and after modeling by the ST-GCN, 5 spatiotemporal distribution feature vectors are extracted, which are defined as: , including the following vectors:

[0084] (0.85, 0.40, 0.70);

[0085] (0.80, 0.42, 0.68);

[0086] (0.10, 0.85, 0.15);

[0087] (0.12, 0.82, 0.20);

[0088] ​​(0.60, 0.38, 0.65);

[0089] The Gaussian mapping model is constructed through the sample set, and the triplet parameters of each vector are obtained:

[0090] Triplet parameter table

[0091]

[0092] ;

[0093] Take ;

[0094] The similarity merging threshold is set to decrease from 0.95 to 0.70, and the point pairs with similarity higher than the current threshold are merged in each round. The system generates the following hierarchical attribution trajectory:

[0095] Hierarchical attribution trajectory table

[0096]

[0097] represents the cluster identifier of the i-th point in the j-th layer; The trajectory volatility of each point is constructed, and the volatility threshold is set;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] The system sets the volatility threshold to 1, and obtains the abnormal boundary point set;

[0104] The original is spliced with its trajectory volatility index Δi to form a fusion feature vector.

[0105] In the third embodiment, based on the above-mentioned embodiments, a hierarchical clustering method of fusing multi-criteria similarity and dynamic cluster merging mechanism is designed in step S4. The similarity matrix is subjected to abnormal detection, and specifically includes the following steps:

[0106] Step S41: Multi-strategy similarity matrix expansion, a multi-strategy similarity fusion mechanism is introduced based on the similarity matrix, and the fused similarity matrix is represented as:

[0107] ; ​

[0108] wherein, is the fused similarity matrix, , and denote adjustable parameters, satisfying , satisfying , denotes a scale adjustment parameter of the Gaussian kernel function;

[0109] Step S42: initial cluster construction, based on the fused similarity matrix, defining each spatiotemporal distribution feature vector corresponding to an initial cluster;

[0110] Step S43: construction of an explanatory similarity view, constructing an explanatory graph structure wherein, the node set is , and the edge set is ;

[0111] Step S44: multi-level similarity fusion, introducing a clustering hierarchy definition, denoting the level representing the cluster division state formed after the round of similarity merging, and designing a hierarchical clustering strategy based on adaptive adjustment of the multi-round similarity threshold to perform layer-by-layer merging of the explanatory graph structure to obtain the final clustering structure;

[0112] Step S45: anomaly boundary detection, based on the final clustering structure, designing an anomaly detection method based on inter-layer cluster structure change analysis to identify abnormal points;

[0113] Step S46: abnormal point output, arranging the abnormal points identified in step S45 into an abnormal point set as potential spatiotemporal distribution abnormal points.

[0114] In this embodiment, the core code used is as follows:

[0115] import numpy as np

[0116] import networkx as nx

[0117] from sklearn.metrics.pairwise import cosine_similarity

[0118] from scipy.spatial.distance import cdist

[0119] # --------------------------

[0120] # 1. Initialize data (5 samples)

[0121] # --------------------------

[0122] F = np.array([

[0123] [0.80, 0.40, 0.70],

[0124] [0.82, 0.39, 0.72],

[0125] [0.12, 0.85, 0.10],

[0126] [0.10, 0.82, 0.13],

[0127] [0.60, 0.35, 0.68] ])

[0129] # Simulate Gaussian triplet parameters (Ex, En, He)

[0130] Ex = F.copy()

[0131] En = np.random.rand(5) * 0.1 + 0.05 # Random entropy values En_i ∈ [0.05,0.15]

[0132] He = np.random.rand(5) * 0.2 + 0.1 # Random hyper-entropy values He_i ∈ [0.1,0.3]

[0133] epsilon = 1e-5

[0134] lambda_ = 0.5

[0135] alpha, beta, gamma = 0.4, 0.3, 0.3

[0136] # --------------------------

[0137] # 2. Multi-criteria similarity fusion

[0138] # --------------------------

[0139] n = F.shape[0]

[0140] Sim = np.zeros((n, n))

[0141] Cos = cosine_similarity(F)

[0142] K = np.exp(-λ * cdist(F, F, metric='sqeuclidean'))

[0143] for i in range(n):

[0144] for j in range(n):

[0145] diff_sq = np.sum((Ex[i] - Ex[j]) ** 2)

[0146] density_sim = np.exp(-diff_sq / (En[i] + En[j] + epsilon))

[0147] phi = 1 / (1 + abs(He[i] - He[j]))

[0148] Sim[i, j] = density_sim * phi

[0149] S_fused = α * Sim + β * Cos + γ * K

[0150] # --------------------------

[0151] # 3. Construct the interpretive graph structure

[0152] # --------------------------

[0153] θ = 0.98 # Similarity threshold

[0154] G = nx.Graph()

[0155] for i in range(n):

[0156] G.add_node(i)

[0157] for j in range(i + 1, n):

[0158] if S_fused[i, j] > θ:

[0159] G.add_edge(i, j, weight=S_fused[i, j])

[0160] # --------------------------

[0161] # 4. Hierarchical Clustering Simulation

[0162] # --------------------------

[0163] def hierarchical_clustering(graph, max_level=3):

[0164] clusters_per_level = []

[0165] current_graph = graph.copy()

[0166] for l in range(max_level):

[0167] communities = list(nx.connected_components(current_graph))

[0168] label_map = {node: idx for idx, com in enumerate(communities)for node in com}

[0169] clusters_per_level.append(label_map)

[0170] # Dynamically adjust the threshold to reduce density and merge weak edges

[0171] new_graph = nx.Graph()

[0172] new_graph.add_nodes_from(current_graph.nodes)

[0173] for u, v, d in current_graph.edges(data=True):

[0174] if d["weight"] > θ - (l + 1) * 0.03:

[0175] new_graph.add_edge(u, v, weight=d["weight"])

[0176] current_graph = new_graph

[0177] return clusters_per_level

[0178] cluster_labels = hierarchical_clustering(G)

[0179] # --------------------------

[0180] # 5. Trajectory volatility detection (anomaly recognition)

[0181] # --------------------------

[0182] def calc_volatility(cluster_labels):

[0183] volatility = np.zeros(n)

[0184] for i in range(n):

[0185] labels = [cl[i] for cl in cluster_labels]

[0186] volatility[i] = sum(labels[l] != labels[l + 1] for l in range(len(labels) - 1))

[0187] return volatility

[0188] volatility = calc_volatility(cluster_labels)

[0189] θ_delta = np.mean(volatility) + np.std(volatility)

[0190] anomaly_indices = np.where(volatility >= θ_delta)[0]

[0191] print("Anomaly indices:", anomaly_indices.tolist())。

[0192] Embodiment four, based on the above embodiment, step S44, specifically includes the following steps:

[0193] Step S441: inter-cluster similarity measure construction, each node in the explanatory graph structure represents a spatiotemporal distribution feature vector, according to step S42, each node also constitutes a cluster, the similarity between any two clusters is defined as:

[0194] ;

[0195] Wherein, represents the average similarity of the clusters and under the level;

[0196] Step S442: graph structure level merging, set the initial similarity threshold, when any cluster pair satisfies , then and are merged into a new cluster , and the explanatory graph structure is updated to the next level , wherein the node set is the merged cluster;

[0197] Step S443: threshold adaptive adjustment, introduce a clustering variation index to update the threshold adjustment factor, the threshold adjustment factor update formula is as follows:

[0198] ;

[0199] Wherein, represents the similarity threshold under the level, represents the similarity threshold under the level, is the threshold adjustment factor, is the clustering variation index;

[0200] Step S444: termination judgment, set the maximum number of iterations, when the maximum number of iterations is reached, output the final clustering structure , pass into step S45 to perform abnormal boundary detection.

[0201] Embodiment five, based on the above embodiment, step S45, specifically includes the following steps:

[0202] Step S451: attribution trajectory construction, for the expansion step of step S44, record the label change of each spatiotemporal distribution feature vector in each level to which cluster it belongs, define the level attribution trajectory as:

[0203] ;

[0204] wherein, represents the first spatiotemporal distribution feature vector in the first layer to the layer cluster label sequence, represents the first spatiotemporal distribution feature vector in the first layer cluster number, represents the maximum hierarchical number of the clustering process;

[0205] Step S452: trajectory fluctuation metric construction, defining the trajectory fluctuation index, denoted as follows:

[0206] ;

[0207] wherein, represents the trajectory fluctuation index of the first spatiotemporal distribution feature vector, is an indicator function, which is 1 if the current layer and the next layer cluster label are inconsistent, otherwise 0;

[0208] Step S453: abnormal point screening, setting a fluctuation threshold, screening out the spatiotemporal distribution feature vector whose fluctuation is significantly higher than the overall average, denoted as:

[0209] ;

[0210] wherein, is an abnormal point, representing a subset of spatiotemporal distribution features identified as abnormal boundary points, represents an abnormality discrimination threshold of trajectory fluctuation.

[0211] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0212] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

[0213] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired thereby, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution are not creative, and should belong to the protection scope of the present application.

Claims

1. An intelligent positioning system for sudden infectious diseases based on spatiotemporal distribution anomaly detection, characterized in that: The system comprises a data collection module, a space-time feature modeling module, a space-time distribution anomaly detection module, an intelligent infectious disease positioning module, a transmission trend prediction module, and a visualization module, and specifically comprises the following contents: The data collection module collects infectious disease related data; The space-time feature modeling module uses a space-time graph convolution network to perform space-time coding and modeling on the infectious disease related data, and extracts space-time distribution features of the infectious disease; The space-time distribution anomaly detection module designs a hierarchical clustering method fusing multi-criteria similarity and a dynamic cluster merging mechanism to perform anomaly detection on the space-time distribution features, and obtains an anomaly detection result; The intelligent infectious disease positioning module uses a graph neural network model to construct an infection source tracing graph according to the anomaly detection result, and locates a potential infection source position; The transmission trend prediction module constructs an infectious disease transmission positioning model based on a graph diffusion mechanism based on the possible infection source, and obtains a future risk transmission area; The visualization module displays the anomaly detection result, the potential infection source position, and the future risk transmission area through graphics. The space-time distribution anomaly detection module designs a hierarchical clustering method fusing multi-criteria similarity and a dynamic cluster merging mechanism to perform anomaly detection on the space-time distribution features, and specifically comprises the following steps: Step S1: input acquisition, receiving space-time distribution features, represented as follows: ； wherein, represents a set of spatio-temporal distribution features, represents a first spatio-temporal distribution feature vector, represents an index; Step S2: extension modeling, a Gaussian distribution mapping model is designed to map each space-time distribution feature vector into corresponding multi-dimensional Gaussian triplet parameters, represented as follows: ； wherein, is the expectation value of the spatiotemporal distribution feature vector of the thspatiotemporal distribution, representing the spatiotemporal distribution center, and is the entropy and hyperentropy of the spatiotemporal distribution feature vector of the thspatiotemporal distribution, representing the uncertainty range of the spatiotemporal distribution and the instability degree of the spatiotemporal distribution change, represents the spatiotemporal distribution feature vector of the thspatiotemporal distribution; Step S3: calculate the similarity matrix, based on the multi-dimensional Gaussian triplet parameters, a density similarity function is designed to calculate the comprehensive similarity score between any two feature points, and the density similarity function is represented as follows: ； ； wherein, denotes the combined similarity of the spatio-temporal feature vectors of the first and the second, is a smoothing factor to avoid division by zero, is a regulating function based on hyperentropy. Step S4: anomaly clustering, based on the correlation structure between the space-time features, based on the similarity matrix between the multi-dimensional Gaussian triplets, a hierarchical clustering method fusing multi-criteria similarity and a dynamic cluster merging mechanism is designed to perform anomaly detection on the similarity matrix, and an anomaly clustering result is obtained; Step S5: fusion feature construction, combining the anomaly clustering result with the space-time distribution feature set to construct a fusion feature vector; Step S6: output the result, input the fusion feature vector into a classification model based on gradient boosting tree for anomaly discrimination, and output the anomaly detection result corresponding to each space-time distribution feature vector. 2.The intelligent outbreak infectious disease positioning system based on spatio-temporal distribution anomaly detection of claim 1, wherein: Step S4, a hierarchical clustering method fusing multi-criteria similarity and a dynamic cluster merging mechanism is designed to perform anomaly detection on the similarity matrix, specifically comprising the following steps: Step S41: multi-strategy similarity matrix expansion, a multi-strategy similarity fusion mechanism is introduced based on the similarity matrix, and the fused similarity matrix is represented as: ； wherein, is the similarity matrix after fusion, , and denotes an adjustable parameter, satisfying , satisfies , denotes a scale adjustment parameter of the Gaussian kernel function; Step S42: initial clustering cluster construction, based on the fused similarity matrix, each space-time distribution feature vector is defined as an initial clustering cluster; Step S43: constructing an interpretative similarity view, constructing an interpretative graph structure wherein the set of nodes is , and the set of edges is ; Step S44: multi-level similarity fusion, introduce the definition of clustering hierarchy, record the hierarchical representation performs the first The cluster division state formed after the round similarity merging, a hierarchical clustering strategy based on multi-round similarity threshold adaptive adjustment is designed, which performs layer-by-layer merging on the explanatory graph structure to obtain the final clustering structure; Step S45: anomaly boundary detection, based on the final clustering structure, an anomaly detection method based on inter-layer cluster structure change analysis is designed to identify abnormal points; Step S46: output the abnormal points, the abnormal points identified in step S45 are arranged into an abnormal point set as potential space-time distribution abnormal points. 3.The intelligent outbreak infectious disease positioning system based on spatio-temporal distribution anomaly detection of claim 2, wherein: Step S44 specifically comprises the following steps: Step S441: inter-cluster similarity measurement construction, each node in the explanatory graph structure represents a spatiotemporal distribution feature vector, according to step S42, each node also constitutes a cluster, and the similarity between any two clusters is defined as: ； wherein represents hierarchical sub-clusters and clusters average similarity; Step S442: graph structure level merging, set initial similarity threshold, when any cluster pair satisfies , then merge and into new cluster , and update explanatory graph structure to next level , where node set is the merged cluster; Step S443: threshold adaptive adjustment, a clustering variation index is introduced to update the threshold adjustment factor, and the threshold adjustment factor updating formula is as follows: ； wherein, represents a similarity threshold at a hierarchy level, represents a similarity threshold at a hierarchy level, is a threshold adjustment factor, is a cluster variation index; Step S444: termination determination, set the maximum number of iterations, when the maximum number of iterations is reached, output the final clustering structure , the incoming step S45 performs exception boundary detection.

Images