Hospital infection propagation path tracing method based on multi-source data fusion

Abstract

The invention provides a hospital infection propagation path tracing method based on multi-source data fusion, and belongs to the technical field of infection propagation paths in hospitals, and the method comprises the steps: building a data set through collecting hospital multi-source infection monitoring data, constructing a multi-layer network propagation model comprising physical contact, spatial proximity and time co-occurrence, and carrying out the tracing of the multi-source infection propagation path. Community detection is carried out by using a random block model, infection propagation path identification is converted into graph coloring problem solving, an individual behavior heterogeneity model is established, and SEIR model propagation parameters are corrected by fusing individual characteristic parameters such as the immune state of a patient and the protection level of medical staff; a propagation situation awareness model based on a hybrid expert mechanism is adopted to optimize propagation parameter estimation, and a propagation parameter weight coefficient is obtained through a game optimization mechanism and Nash equilibrium solution for iterative optimization. And finally, outputting a high-confidence infection propagation path sequence and a propagation source positioning result by adopting a propagation path backtracking algorithm in combination with spatial-temporal correlation analysis.

Description

Technical Field

[0001] The present invention belongs to the technical field of hospital infection transmission paths, and in particular relates to a hospital infection transmission path tracing method based on multi-source data fusion. Background Art

[0002] Tracing the transmission paths of hospital-acquired infections is a key technical approach for infection control in healthcare facilities. Traditional methods rely primarily on epidemiological surveys and contact tracing, manually collecting patient contact histories, healthcare worker activity records, and environmental monitoring data to construct simple transmission relationship maps. Some advanced medical institutions have begun adopting personnel trajectory tracking technology based on RFID positioning systems, integrating them with electronic medical record systems to build a basic infection surveillance network. Traditional SEIR epidemiological models are used to analyze transmission dynamics and statistically identify possible infection sources and transmission paths. However, these traditional technical methods have significant drawbacks: First, manual survey-based methods are inefficient and prone to missing key transmission links, and the accuracy of tracing relies heavily on the experience of investigators. Second, simple location tracking technologies only record physical location information and lack comprehensive consideration of complex contact patterns and environmental factors, resulting in inaccurate transmission path reconstruction. Finally, traditional SEIR models assume a homogeneous population, ignoring the influence of individual behavioral differences and multi-layered transmission mechanisms, resulting in significant parameter estimation bias. Due to insufficient tracing accuracy, traditional technical methods are unable to accurately reconstruct the true transmission path of hospital infections. Especially when faced with complex transmission networks with small-world and scale-free characteristics, it is difficult to identify the true high-risk transmission paths and key transmission nodes, resulting in infection control measures that are not very targeted. Summary of the Invention

[0003] In view of this, the present invention provides a hospital infection transmission path tracing method based on multi-source data fusion, which can solve the technical problem in the existing technology that the hospital infection transmission path tracing accuracy is insufficient, resulting in the inability to effectively identify high-risk transmission paths.

[0004] The present invention is implemented as follows: The present invention provides a hospital infection transmission path tracing method based on multi-source data fusion, including: collecting hospital multi-source infection monitoring data, establishing a multi-dimensional data set including patient flow trajectory, medical staff contact records, pathogen gene sequence information and environmental monitoring data; constructing a multi-layer network transmission model, hierarchically modeling the physical contact network, spatial proximity network and temporal co-occurrence network, and identifying the small-world characteristics and scale-free characteristics in the network; using a random block model to perform community detection on the hierarchical network structure, converting the infection transmission path identification problem into a graph coloring problem for solution, optimizing the high-risk transmission path identification accuracy by minimizing the number of colorings, and calculating the transmission probability matrix between nodes; establishing individual behavior heterogeneity A novel SEIR model was developed, integrating the patient's immune status, medical staff's protection level and environmental disinfection frequency as individual characteristic parameters to modify the traditional SEIR model propagation parameters. A propagation situation awareness model was used to optimize the propagation parameter estimation, and a hybrid expert mechanism was used to dynamically select different parameter estimation experts. The number of activated experts was controlled by the capacity factor, and the infection rate, recovery rate and incubation period were adaptively adjusted based on the modified propagation parameters. A game optimization mechanism was established to conduct strategic games between the simulation results of different time periods and the actual observation results of the corresponding time periods, and the Nash equilibrium solution was used to obtain the propagation parameter weight coefficients. A propagation path backtracking algorithm was used, combined with spatiotemporal correlation analysis and network propagation simulation results, to output the most likely infection propagation path sequence and the propagation source location results.

[0005] Among them, the multidimensional dataset is specifically a four-dimensional data structure including time dimension, space dimension, individual dimension and environmental dimension. The time dimension records the specific time and duration of the infection event, the space dimension records the location coordinates and movement trajectory of the infected and susceptible persons, the individual dimension records the identity, health status and behavioral characteristics of the personnel, and the environmental dimension records the ward temperature, humidity, ventilation conditions and disinfection frequency.

[0006] Among them, the multi-layer network transmission model specifically abstracts the hospital infection transmission process into multiple interrelated network layers. The physical contact network layer describes the direct physical contact relationship between people, the spatial proximity network layer describes the proximity relationship between people with a spatial distance of less than 2m, and the temporal co-occurrence network layer describes the co-occurrence relationship between people appearing in the same spatial area within the same time period.

[0007] Among them, the small-world characteristic specifically refers to the fact that the shortest path length between most nodes in the network is small, and the network has a high clustering coefficient, which is manifested as a network topology feature with dense local connections and sparse global connections; the scale-free characteristic specifically refers to the fact that the node degree distribution in the network follows a power-law distribution, and there are a few high-degree nodes connecting a large number of other nodes.

[0008] Among them, the random block model is specifically a network community discovery algorithm, which assumes that nodes in the network belong to different potential communities, and the probability of connecting nodes within the same community is higher than the probability of connecting nodes between different communities; the graph coloring problem is specifically to assign a color to each vertex in the graph so that adjacent vertices have different colors and the total number of colors used at the same time is minimized.

[0009] Among them, the transmission probability matrix is ​​specifically a two-dimensional matrix that describes the probability of infection transmission between any two nodes in the network, and the matrix element value represents the probability of infection spreading from the source node to the target node; the community detection result is specifically structured data including community division identification, intra-community connection density, inter-community connection strength and a list of high-risk transmission paths.

[0010] Among them, the individual behavior heterogeneity model specifically takes into account the differentiated behavioral characteristics exhibited by different individuals during the transmission process, including the impact of susceptibility differences, contact pattern differences, and protective behavior differences on transmission dynamics; the individual characteristic parameters specifically include a parameter set including patient age, gender, underlying disease status, immunity level, years of work experience of medical staff, and use of protective equipment.

[0011] Among them, the transmission situation awareness model is specifically an intelligent transmission parameter estimation model based on a hybrid expert mechanism, which includes a multi-level hybrid expert architecture of an expert network layer, a gating network layer and a fusion output layer. The expert network layer includes infection rate estimation experts, recovery rate estimation experts, incubation period estimation experts and environmental factor estimation experts. The gating network layer dynamically calculates the weight of each expert through a soft routing mechanism.

[0012] Among them, the capacity factor is specifically a key parameter that controls the number of experts activated simultaneously in the propagation situation awareness model, and its value range is a positive integer between 1 and the total number of experts; it also includes designing a dynamic capacity adjustment function to adjust the capacity factor of the propagation situation awareness model, and calculating the complexity evaluation value based on the network complexity index, prediction accuracy index and computing resource occupancy rate.

[0013] Among them, when the complexity evaluation value is in the low complexity range, a linear increasing adjustment function is used to increase the capacity factor; when the complexity evaluation value is in the medium complexity range, a nonlinear smooth adjustment function is used to fine-tune the capacity factor; when the complexity evaluation value is in the high complexity range, an exponential decay adjustment function is used to reduce the capacity factor to control the computational overhead and avoid overfitting.

[0014] Among them, the game optimization mechanism is specifically a parameter optimization strategy based on game theory. It takes the simulated prediction results of different time periods as game participants and the actual observation results as the game environment. Through strategy game and Nash equilibrium solution, the optimal weight distribution scheme of the prediction results of each time period is obtained; the Nash equilibrium is specifically a stable state in game theory where each participant chooses the optimal strategy given the strategies of other participants.

[0015] Among them, the transmission parameter weight coefficient is specifically a numerical weight used to quantify the importance of the transmission parameters in different time periods, which is calculated through the game optimization mechanism and used to guide the iterative training and parameter update process of the transmission situation awareness model; the final transmission parameter is specifically the final determined value of the infection rate, recovery rate, incubation period and environmental transmission coefficient obtained after processing by the game optimization mechanism.

[0016] Among them, the spatiotemporal correlation analysis specifically studies the correlation patterns of data in the time and space dimensions, and identifies the regular characteristics of the spatiotemporal distribution of infection events by calculating the spatiotemporal autocorrelation coefficient and cross-correlation function; the propagation path backtracking algorithm is specifically a path reconstruction algorithm based on dynamic programming and probabilistic reasoning, which reconstructs the most likely infection propagation path by reversely tracing the spatiotemporal trajectory of infection propagation.

[0017] Among them, the infection propagation path sequence is specifically a sequence of infection propagation nodes arranged in chronological order, and each node contains the infected person identification, infection time, infection location and propagation probability information; the transmission source positioning result is specifically the infection propagation starting node determined by the propagation path backtracing algorithm, which contains the source node identification, source location coordinates, source time and confidence information.

[0018] Among them, the network complexity index is specifically a comprehensive index used to quantify the complexity of the hierarchical network structure, which includes a weighted combination of network topology characteristic parameters such as the number of network nodes, edge connection density, clustering coefficient and average path length; the prediction accuracy index is specifically a quantitative index used to evaluate the accuracy of parameter estimation of the communication situation awareness model, which includes mean square error, mean absolute error and correlation coefficient.

[0019] Among them, the linear increasing adjustment function is specifically a linear function with a positive slope, which is used to gradually increase the capacity factor value in a low-complexity environment; the nonlinear smooth adjustment function is specifically a smooth nonlinear function based on the Sigmoid function or the hyperbolic tangent function; the exponential decay adjustment function is specifically an exponential function with a natural constant as the base, which is used to quickly reduce the capacity factor value in a high-complexity environment.

[0020] The present invention significantly improves the accuracy of tracing infection transmission paths by constructing a multi-layer network transmission model and a transmission situation awareness mechanism. This method collects patient flow trajectories, medical staff contact records, pathogen gene sequence information and environmental monitoring data, establishes a four-dimensional data structure, and constructs a hierarchical network architecture that includes physical contact, spatial proximity and temporal co-occurrence, comprehensively capturing multi-dimensional information in the transmission process. The present invention significantly improves the tracing accuracy through precise modeling: it uses a random block model and a graph coloring algorithm to automatically identify the network community structure, decomposes the complex transmission network into manageable community modules, and the high connection density characteristics within each community make the high-risk transmission path clearer and more identifiable; the individual behavior heterogeneity model integrates individual characteristic parameters such as the patient's immune status and the protection level of medical staff, corrects the homogeneity assumption of the traditional model, and makes the transmission parameter estimation more accurate; the transmission situation awareness model based on the hybrid expert mechanism works through multiple specialized sub-networks to dynamically select the optimal parameter estimation strategy, which significantly reduces the parameter estimation error. The present invention achieves effective identification of high-risk transmission paths through improved precision: multi-layer network modeling accurately depicts the complex topological structure of the hospital infection network, the game optimization mechanism obtains the optimal parameter weights through Nash equilibrium solution, and the transmission path backtracking algorithm combines spatiotemporal correlation analysis to output a transmission path sequence with higher confidence, solving the core technical problem of insufficient transmission path tracing accuracy leading to the failure of high-risk path identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flow chart of the method of the present invention.

[0022] Figure 2 Schematic diagram of the multi-layer network propagation model.

[0023] Figure 3 Schematic diagram of the multi-layer network communication model after adopting the graph coloring problem.

[0024] Figure 4 Schematic diagram of the communication situation awareness model.

[0025] Figure 5 Schematic diagram of the propagation path backtracking algorithm. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] like Figure 1 FIG. 1 is a flowchart of a method for tracing the transmission path of hospital infection based on multi-source data fusion provided by the present invention. The method includes the following steps:

[0028] S01. Collect hospital multi-source infection monitoring data to establish a multi-dimensional dataset containing patient flow trajectories, medical staff contact records, pathogen gene sequence information, and environmental monitoring data to obtain original transmission data;

[0029] S02. Constructing a multi-layer network propagation model, hierarchically modeling the physical contact network, spatial proximity network, and temporal co-occurrence network in the original propagation data, identifying small-world characteristics and scale-free features in the network, and generating a hierarchical network structure;

[0030] S03. Perform community detection on the hierarchical network structure using a random block model, transform the infection transmission path identification problem into a graph coloring problem for solution, optimize the accuracy of high-risk transmission path identification by minimizing the number of colorings, calculate the transmission probability matrix between nodes, and output the community detection results;

[0031] S04. Establish an individual behavior heterogeneity model, integrate the patient's immune status, medical staff's protection level, and environmental disinfection frequency in the original transmission data as individual characteristic parameters, and modify the traditional SEIR model transmission parameters to obtain modified transmission parameters;

[0032] S05. Optimize the estimation of propagation parameters using a propagation situation awareness model, dynamically select different parameter estimation experts using a hybrid expert mechanism, control the number of activated experts using a capacity factor, adaptively adjust the infection rate, recovery rate, and incubation period based on the modified propagation parameters, and output the optimized propagation parameters.

[0033] S06. Establish a game optimization mechanism, conduct strategic games by comparing simulation results of different time periods with actual observation results of corresponding time periods, and use Nash equilibrium to solve the propagation parameter weight coefficients, which are used to iteratively optimize the propagation situation awareness model to obtain the final propagation parameters;

[0034] S07. Using a propagation path backtracking algorithm, combined with the spatiotemporal correlation analysis of the original propagation data and the network propagation simulation results of the final propagation parameters, output the most likely infection propagation path sequence and the propagation source positioning result.

[0035] Among them, the multidimensional dataset is specifically a four-dimensional data structure that includes time dimension, space dimension, individual dimension and environmental dimension. The time dimension records the specific time and duration of the infection event, the space dimension records the location coordinates and movement trajectory of the infected and susceptible people, the individual dimension records the identity, health status and behavioral characteristics of the personnel, and the environmental dimension records environmental factors such as ward temperature, humidity, ventilation conditions and disinfection frequency.

[0036] like Figure 2As shown in the figure, the multi-layer network transmission model specifically abstracts the hospital infection transmission process into multiple interrelated network layers. Each layer of the network represents a different type of transmission relationship, and describes the complex transmission dynamics process through the inter-layer coupling mechanism. The physical contact network layer describes the direct physical contact relationship between people, the spatial proximity network layer describes the proximity relationship between people with a spatial distance of less than 2m, and the temporal co-occurrence network layer describes the co-occurrence relationship of people appearing in the same spatial area within the same time period. Figure 2 This is a schematic diagram of the multi-layer network propagation model, showing the three-layer architecture of the physical contact network layer, the spatial proximity network layer, and the temporal co-occurrence network layer, as well as the coupling relationship and network topology characteristics between them.

[0037] Among them, the small-world characteristic specifically refers to the fact that the shortest path length between most nodes in the network is small, and the network has a high clustering coefficient, which is manifested as a network topology feature with dense local connections and sparse global connections.

[0038] Among them, the scale-free feature specifically refers to the network degree distribution characteristic that the node degree distribution in the network follows a power-law distribution, where there are a few high-degree nodes connected to a large number of other nodes, while most nodes have only a small number of connections.

[0039] Among them, the hierarchical network structure is specifically a three-layer network architecture consisting of a physical contact network layer, a spatial proximity network layer, and a temporal co-occurrence network layer. Each layer of the network contains the same set of nodes but has a different edge connection pattern. The coupling strength parameter between layers describes the degree of mutual influence between different propagation mechanisms.

[0040] Among them, the random block model is specifically a network community discovery algorithm, which is used to identify modular structures in the network by assuming that nodes in the network belong to different potential communities and the probability of connecting nodes within the same community is higher than the probability of connecting nodes between different communities.

[0041] like Figure 3 As shown, the graph coloring problem is a classic combinatorial optimization problem in graph theory. The goal is to assign a color to each vertex in the graph so that adjacent vertices have different colors and the total number of colors used is minimized. It is used to identify different transmission communities and high-risk path groupings in infection transmission networks. Figure 3 A schematic diagram of the multi-layer network propagation model after adopting the graph coloring problem, showing the random block model community detection process, using different colors to identify communities with different risk levels, and the calculation method of the propagation probability matrix.

[0042] Among them, the propagation probability matrix is ​​specifically a two-dimensional matrix that describes the probability of infection propagation between any two nodes in the network. The matrix element value represents the probability of infection propagation from the source node to the target node, which is used to quantify the propagation risk and predict the propagation path.

[0043] Among them, the community detection results specifically include structured data including community division identification, intra-community connection density, inter-community connection strength and a list of high-risk transmission paths, which are used to guide subsequent transmission parameter optimization and path tracing analysis.

[0044] Among them, the individual behavioral heterogeneity model specifically considers the differentiated behavioral characteristics exhibited by different individuals during the transmission process, including the impact of factors such as differences in susceptibility, contact patterns, and protective behavior on transmission dynamics.

[0045] Among them, individual characteristic parameters specifically include a parameter set of quantitative indicators such as patient age, gender, underlying disease status, immunity level, years of work experience of medical staff, use of protective equipment, type of environmental disinfectant and disinfection frequency.

[0046] Among them, the corrected transmission parameters are the corrected values ​​of key transmission dynamics parameters such as infection rate, recovery rate, incubation period and environmental transmission coefficient obtained by integrating individual characteristic parameters on the basis of the traditional SEIR model.

[0047] Among them, the communication situation awareness model is specifically an intelligent communication parameter estimation model based on a hybrid expert mechanism. By dynamically selecting and combining multiple parameter estimation sub-models, it realizes the adaptive identification and optimization adjustment of key parameters in complex communication environments.

[0048] like Figure 4 As shown in the figure, the specific structure of the transmission situation awareness model is a multi-level hybrid expert architecture, which includes an expert network layer, a gating network layer and a fusion output layer. The expert network layer includes multiple specialized sub-networks such as infection rate estimation experts, recovery rate estimation experts, incubation period estimation experts and environmental factor estimation experts. The gating network layer dynamically calculates the weight of each expert through a soft routing mechanism. The fusion output layer fuses the expert outputs according to the weight to generate the final transmission parameter estimation result. The capacity factor parameter controls the upper limit of the number of experts activated at the same time to balance computing resources and estimation accuracy.

[0049] The steps for establishing the training data set of the transmission situation awareness model specifically include historical infection event data collection, transmission parameter annotation, multidimensional feature extraction and data preprocessing. First, infection transmission case data from different periods are collected from the hospital information system, including original records such as infected person information, contact trajectory, environmental conditions and transmission time. Then, epidemiological experts annotate the transmission parameters of each case to determine the true infection rate, recovery rate and incubation period and other key parameter values. Then, multidimensional feature extraction is performed on the original data, including network topology features, time series features, spatial distribution features and individual attribute features. Finally, through preprocessing steps such as data cleaning, normalization and sample balancing, a high-quality data set suitable for hybrid expert model training is constructed.

[0050] The steps of training the propagation situation awareness model specifically include expert network pre-training, gated network training, end-to-end joint optimization and parameter tuning. In the expert network pre-training stage, each expert sub-network is trained separately using a sub-dataset so that each expert can achieve good initial performance in the parameter estimation task. In the gated network training stage, the mapping relationship between input features and expert selection is learned to train the weight distribution strategy of the gating mechanism. In the end-to-end joint optimization stage, the expert network and the gated network are integrated for joint training, and all network parameters are synchronously optimized through the back-propagation algorithm. In the parameter tuning stage, the optimal hyperparameter configuration is determined through methods such as cross-validation and grid search, including learning rate, regularization coefficient, capacity factor and expert network structure parameters.

[0051] Among them, the capacity factor is a key parameter that controls the number of experts activated simultaneously in the communication situation awareness model. Its value range is a positive integer between 1 and the total number of experts, and is used to balance the model calculation complexity and parameter estimation accuracy.

[0052] Among them, the optimized transmission parameters are specifically the optimized estimated values ​​of infection rate, recovery rate, incubation period and environmental transmission coefficient obtained after processing by the transmission situation awareness model. Compared with the corrected transmission parameters, they have higher estimation accuracy and stronger environmental adaptability.

[0053] Among them, the game optimization mechanism is specifically a parameter optimization strategy based on game theory. It takes the simulated prediction results of different time periods as game participants and the actual observation results as the game environment. Through strategic game and Nash equilibrium solution, the optimal weight distribution scheme of the prediction results of each time period is obtained.

[0054] Among them, Nash equilibrium is specifically a core concept in game theory, which represents the stable state in which each participant in a multi-party game chooses the optimal strategy given the strategies of other participants. It is used to find the optimal balance point of the prediction weights in each time period in the optimization of propagation parameters.

[0055] Among them, the propagation parameter weight coefficient is specifically used to quantify the numerical weight of the importance of the propagation parameters in different time periods. It is calculated through the game optimization mechanism and used to guide the iterative training and parameter update process of the propagation situation awareness model.

[0056] Among them, the final transmission parameters are the final determined values ​​of the infection rate, recovery rate, incubation period and environmental transmission coefficient obtained after processing by the game optimization mechanism, which are used for subsequent transmission path tracing and source location analysis.

[0057] Among them, spatiotemporal correlation analysis specifically studies the correlation patterns of data in the time and space dimensions, and identifies the regular characteristics of infection events in the spatiotemporal distribution by calculating the spatiotemporal autocorrelation coefficient and cross-correlation function.

[0058] like Figure 5 As shown in the figure, the propagation path backtracking algorithm is a path reconstruction algorithm based on dynamic programming and probabilistic reasoning. It reconstructs the most likely infection propagation path by reversely tracing the spatiotemporal trajectory of infection propagation and combining the propagation probability and spatiotemporal constraints. Figure 5 This is a schematic diagram of the propagation path backtracing algorithm, showing the complete process from spatiotemporal correlation analysis to Viterbi algorithm solution, and then to uncertainty quantification, and finally outputting the infection propagation path sequence and source location results.

[0059] Among them, the infection propagation path sequence is specifically a sequence of infection propagation nodes arranged in chronological order. Each node contains information such as the infected person identification, infection time, infection location and propagation probability, which is used to describe the complete propagation process of the infection in the network.

[0060] Among them, the transmission source positioning result is specifically the infection transmission starting node determined by the transmission path backtracking algorithm, which includes information such as the source node identification, source location coordinates, source time and confidence level.

[0061] A dynamic capacity adjustment function is designed, which is used to adjust the capacity factor of the propagation situation awareness model. The dynamic capacity adjustment function is calculated based on multiple data such as network complexity index, prediction accuracy index and computing resource occupancy rate to obtain a complexity evaluation value. When the complexity evaluation value belongs to different ranges, different weight adjustment functions are used to adjust the capacity factor of the propagation situation awareness model. The specific calculation method is: when the complexity evaluation value is in a low complexity range, a linear increasing adjustment function is used to increase the capacity factor to improve the model expression ability; when the complexity evaluation value is in a medium complexity range, a nonlinear smooth adjustment function is used to fine-tune the capacity factor to balance accuracy and efficiency; when the complexity evaluation value is in a high complexity range, an exponential decay adjustment function is used to reduce the capacity factor to control computing overhead and avoid overfitting.

[0062] Among them, the network complexity index is a comprehensive indicator used to quantify the complexity of the hierarchical network structure, which includes a weighted combination of network topology characteristic parameters such as the number of network nodes, edge connection density, clustering coefficient and average path length.

[0063] Among them, the prediction accuracy index is specifically a quantitative indicator used to evaluate the accuracy of parameter estimation of the communication situation awareness model, which includes the comprehensive evaluation results of statistical indicators such as mean square error, mean absolute error and correlation coefficient.

[0064] Among them, the computing resource utilization rate is specifically the proportion of computing resources such as processor time, memory space and storage capacity occupied during the operation of the communication situation awareness model.

[0065] Among them, the complexity evaluation value is a comprehensive evaluation value obtained by weighted fusion of network complexity indicators, prediction accuracy indicators and computing resource occupancy rate, which is used to guide the selection of dynamic adjustment strategy of capacity factor.

[0066] Among them, the linear increasing adjustment function is specifically a linear function with a positive slope, which is used to gradually increase the capacity factor value in a low-complexity environment and improve the parameter estimation and expression capabilities of the communication situation awareness model.

[0067] Among them, the nonlinear smooth adjustment function is specifically a smooth nonlinear function based on the Sigmoid function or the hyperbolic tangent function, which is used to fine-tune the capacity factor value in a medium-complexity environment to achieve the optimal balance between model accuracy and computational efficiency.

[0068] Among them, the exponential decay adjustment function is specifically an exponential function with a natural constant as the base, which is used to quickly reduce the capacity factor value in a high-complexity environment, control the computational complexity of the propagation situation awareness model and prevent the occurrence of model overfitting.

[0069] The specific implementation of the above steps is described in detail below.

[0070] The specific implementation of step S01 involves comprehensively collecting and integrating various types of infection monitoring data within the hospital using a multi-dimensional data acquisition system. First, radio frequency identification technology and a Bluetooth beacon system are used to collect real-time location information of patients and medical staff, recording their movement trajectories and duration within the hospital. The acquisition frequency is set to every 5 seconds, and the positioning accuracy is controlled within 1 meter. Next, contact tracing sensors and smart wearable devices are used to record contact events between individuals, including contact time, contact distance, and contact duration. The contact distance threshold is set to 2 meters, and the contact duration threshold is set to 15 minutes. Gene sequencing technology and polymerase chain reaction methods are then used to obtain genetic sequence information of the pathogen, and pathogen type and mutation characteristics are identified by comparison with gene libraries. Simultaneously, an environmental monitoring sensor network is deployed to collect environmental parameters such as temperature, humidity, air quality, and disinfectant concentration within the ward. The sensor sampling interval is set to 10 minutes. Finally, these four types of data are correlated and integrated according to unified timestamps and spatial coordinates to construct a four-dimensional data structure consisting of time, space, individual, and environmental dimensions, forming a complete original transmission dataset.

[0071] The specific implementation of step S02 involves constructing a multi-layered infection transmission network model based on complex network theory. First, a physical contact network layer is constructed based on the collected contact record data, with individuals serving as network nodes and direct physical contact relationships serving as network edges. Edge weights are determined by contact frequency and contact intensity. Next, a spatial proximity network layer is constructed based on the location trajectory data. When two individuals appear within a spatial distance of less than 2 meters, a proximity edge is established. The edge weight is determined by the inverse of the spatial distance and the duration of their shared stay. Next, a temporal co-occurrence network layer is constructed using temporal co-occurrence information. Co-occurrence edges are established when multiple individuals appear in the same spatial region within the same time window. The time window is set to 30 minutes, and the spatial region is partitioned at a granularity of 10 square meters. Complex network analysis methods are then used to calculate topological metrics such as degree distribution, clustering coefficient, and average path length for each network layer to identify the network's small-world and scale-free characteristics. Small-world characteristics are determined by a clustering coefficient greater than 0.3 and an average path length less than logN, where N is the total number of network nodes. Scale-free features are identified by fitting the power-law exponent of the degree distribution, and the network is determined to be scale-free when the power-law exponent is between 2 and 3. Finally, the interaction strength between different network layers is described by the inter-layer coupling matrix, generating a multi-layer network propagation model with a hierarchical structure.

[0072] The specific implementation of step S03 involves using a stochastic block model algorithm to identify community structures and optimize transmission paths in a multi-layer network. First, the network is partitioned into communities using the maximum likelihood estimation method of the stochastic block model. An iterative optimization algorithm is then used to find the community assignment scheme that maximizes the network likelihood function. The stochastic block model assumes that the probability of connecting nodes within a community follows a Beta distribution, and that the probability of connecting nodes between different communities is significantly lower than the probability of connecting within a community. The threshold for the probability of connecting within a community is set to 0.6, and the threshold for the probability of connecting between communities is set to 0.1. The community detection problem is then transformed into a graph coloring problem for solution. A greedy coloring algorithm is used to assign a different color to each community, ensuring that adjacent communities have different colors while minimizing the total number of colors used. Based on the community partition results, a transmission probability matrix is ​​then calculated between nodes. The Markov Chain Monte Carlo method is then used to estimate the probability of infection spreading from any source node to a target node. The transmission probability calculation takes into account factors such as network distance, community affiliation, and connection strength, with probability values ​​ranging from 0 to 1. Finally, a structured community detection result is output, including community identifiers, intra-community connection density, inter-community connection strength, and high-risk transmission paths, providing a network topology foundation for subsequent analysis.

[0073] The specific implementation of step S04 is to construct a transmission dynamics correction model that integrates individual difference characteristics. First, collect individual characteristic information such as the patient's age, gender, underlying disease status, and immune level to establish an individual susceptibility assessment system. The age factor is set with different susceptibility coefficients according to the three intervals of 0-18 years old, 19-65 years old, and over 65 years old, which are 1.2, 1.0, and 1.5 respectively. The immune level is quantified by serum antibody titer measurement. The antibody titer is less than 10 2 The individual susceptibility coefficient was set at 1.8. Then, occupational characteristic parameters such as years of work experience, use of protective equipment, and vaccination status of medical staff were collected to construct a protection effect evaluation model. The protection coefficient was set to 0.2 when the protective equipment was fully worn, 0.6 when partially worn, and 1.0 when not worn. Then, environmental factor parameters such as the frequency of environmental disinfection, type and concentration of disinfectants were recorded to establish an environmental transmission risk assessment system. The environmental transmission coefficient was 0.3 when the disinfection frequency was more than 4 times a day, 0.6 when it was 2-4 times a day, and 1.0 when it was less than 2 times a day. Finally, the individual characteristic parameters were incorporated into the traditional susceptible-exposed-infected-recovered model, and the key parameters of the model such as infection rate, recovery rate and incubation period were corrected. The corrected infection rate is equal to the basic infection rate multiplied by the individual susceptibility coefficient and the environmental transmission coefficient. The recovery rate is adjusted according to the individual immune status and treatment conditions. The incubation period is determined based on the characteristics of the pathogen and individual physical differences.

[0074] The specific implementation of step S05 involves establishing a transmission situation awareness model based on a hybrid expert mechanism for parameter optimization and estimation. This model utilizes a multi-level hybrid expert architecture, consisting of three main components: an expert network layer, a gating network layer, and a fusion output layer. The expert network layer is designed with four functional sub-networks: infection rate estimation experts, recovery rate estimation experts, incubation period estimation experts, and environmental factor estimation experts. Each expert network utilizes a three-layer fully connected neural network structure, with the number of hidden layer neurons set to 128, 64, and 32, respectively. The gating network layer uses a soft routing mechanism to dynamically calculate the weight coefficients of each expert, ensuring that the sum of the weights is 1 through the Softmax function. The gating network utilizes a two-layer fully connected structure with 64 hidden layer neurons. The fusion output layer weights the outputs of each expert according to the gating weights to generate the final transmission parameter estimates. The capacity factor parameter controls the number of experts activated simultaneously, ranging from 1 to 4. A capacity factor of 2 indicates that the two most relevant expert networks are activated simultaneously. The model was trained using a gradient descent optimization algorithm with a learning rate of 0.001, a batch size of 32, and 1000 training rounds. By dynamically adjusting the activation state and weight distribution of each expert, the model achieved adaptive identification and accurate estimation of key parameters in complex propagation environments.

[0075] The specific implementation of step S06 involves iteratively optimizing the propagation parameter weights using a game-theoretic optimization mechanism. First, the propagation parameter estimation results for different time periods are used as game participants, with each time period's prediction corresponding to a strategy choice. Then, using actual observed infection spread data as an objective criterion for the game environment, the game payoff is measured by calculating the deviation between the predicted and actual results for each time period. This deviation metric uses a weighted combination of root mean square error (RMS) and mean absolute percentage error (MAPE), with weights set to 0.6 and 0.4. Next, a multi-party game model is constructed, with each participant's strategy space and payoff function defined. A Nash equilibrium algorithm is then used to find the equilibrium point that maximizes the payoff for all participants. The Nash equilibrium solution is solved using an iterative optimal response algorithm, with the iterative termination condition being a strategy change of less than 0.01 over 10 consecutive iterations. Based on the Nash equilibrium solution, weight coefficients for the propagation parameters for each time period are calculated. The weight coefficients reflect the reliability and importance of the prediction results for each time period. Finally, the calculated weight coefficients are used to perform feedback optimization on the propagation situation awareness model. A weighted loss function guides the update of model parameters, achieving continuous improvement in model performance. The final propagation parameters obtained after game optimization have higher prediction accuracy and stronger generalization ability.

[0076] The specific implementation of step S07 utilizes a propagation path backtracking algorithm based on dynamic programming to locate the source and reconstruct the path. First, a spatiotemporal correlation analysis is performed on the original propagation data. The autocorrelation coefficients and cross-correlation functions between different locations and time points are calculated to identify the clustering patterns and diffusion patterns of infection events in their spatiotemporal distribution. The spatiotemporal autocorrelation analysis utilizes the Moran index and Geary coefficient. A Moran index greater than 0.3 indicates significant spatial clustering. A state transition probability matrix is ​​then constructed based on the final propagation parameters to describe the propagation and transition patterns of infection states within the network. The Viterbi algorithm is then used to determine the most likely propagation path sequence. Dynamic programming is then used to identify the propagation trajectory with the highest probability among all possible paths. The state space of the Viterbi algorithm is composed of all nodes in the network, the observation sequence is composed of infection events arranged in chronological order, and the state transition probabilities are determined by the propagation probability matrix. The Markov Chain Monte Carlo method is then used to quantify the uncertainty of the path reconstruction results, calculating the confidence interval and credibility score for each candidate path. Finally, a sequence of infection transmission paths arranged in chronological order is output. Each path node contains detailed information such as the infected person's identification, infection time, infection location, and transmission probability. At the same time, the location coordinates, starting time, and confidence level of the transmission source are determined, providing a scientific basis for infection control decisions.

[0077] The detailed structure of the transmission situation awareness model adopts a hierarchical hybrid expert architecture. The bottom layer consists of four functional expert networks, responsible for parameter estimation of infection rate, recovery rate, incubation period, and environmental factors, respectively. The infection rate estimation expert uses a recurrent neural network structure to process time-series transmission data. The network has three layers, each containing 64 long-short-term memory units, and the activation function is the hyperbolic tangent function. The recovery rate estimation expert uses a convolutional neural network structure to extract spatial transmission features. The network consists of two convolutional layers and two fully connected layers with a convolution kernel size of 3×3 and 32 and 64 feature maps, respectively. The incubation period estimation expert uses a residual network structure to process complex individual differences. The network is 18 layers deep, with each residual block consisting of two convolutional layers and one skip connection. The environmental factor estimation expert uses an attention mechanism network structure to fuse multidimensional environmental features. It consists of eight attention heads and a four-layer transformer encoder. The middle layer is a gated network with a multilayer perceptron architecture. The input layer receives the fused feature vector, the hidden layer contains 128 neurons, and the output layer generates weight coefficients for the four experts. The top layer is the fusion output layer, which sums the outputs of each expert according to their weights to generate the final propagation parameter estimation results.

[0078] The detailed steps of establishing the training dataset include four stages: data collection, preprocessing, feature engineering, and quality control. In the data collection stage, infection transmission case data from the hospital information management system for the past five years were extracted, including original records such as patient basic information, diagnosis records, treatment process, contact history, and environmental monitoring data. The data volume reached 10 5 records of infection events. In the preprocessing phase, the raw data was cleaned and standardized, records with more than 30% missing values ​​were removed, numerical features were normalized to zero mean, and categorical features were converted to one-hot encoding. In the feature engineering phase, multidimensional feature vectors such as network topology features, time series statistical features, spatial distribution features, and individual attribute features were extracted. The total feature dimension was 512, and the principal component analysis method was used to reduce the dimension to 128. In the quality control phase, the true transmission parameter value of each case was determined through manual annotation by epidemiological experts, and a quality assessment system for parameter annotation was established. The annotation consistency coefficient was required to reach above 0.85. Finally, a complete dataset consisting of training set, validation set, and test set was constructed, and the sample ratio was divided into 7:2:1 to ensure the representativeness and generalization ability of the dataset.

[0079] It should be noted that the key technical ideas of the present invention include: a multi-layer network propagation model, community detection and graph coloring optimization based on a random block model, and a propagation situation awareness model of a hybrid expert mechanism.

[0080] Multi-layer network transmission model: The multi-layer network transmission model constructed by the present invention decomposes the hospital infection transmission process into three interrelated network levels: physical contact, spatial proximity, and temporal co-occurrence. Each layer of the network captures different types of transmission relationships and describes the synergy of different transmission modes through the inter-layer coupling mechanism. Compared with the traditional single network modeling method, the multi-layer network model can simultaneously characterize the complex interaction process of multiple transmission mechanisms such as direct contact transmission, air transmission, and environmental media transmission. Traditional methods often simplify complex transmission processes into a single contact network or spatial network, ignoring the mutual influence and superposition effect between different transmission modes. Through hierarchical modeling and inter-layer coupling, the multi-layer network model can accurately reflect the true complexity of infection transmission in the hospital environment, significantly improving the accuracy of transmission dynamics modeling and the accuracy of transmission path tracing.

[0081] Community detection and graph coloring optimization based on random block model: The present invention adopts random block model to perform network community detection. By assuming that the probability of node connection within the same community is higher than the probability of node connection between different communities, it automatically identifies the modular structure in the propagation network and converts the infection propagation path identification into a graph coloring optimization problem for solution. Traditional propagation path identification methods mainly rely on statistical analysis and manual judgment, lack of systematic network structure analysis capabilities, and it is difficult to discover community structures and high-risk propagation clusters hidden in complex networks. The random block model can automatically discover potential communities in the network through probabilistic modeling, identify dense propagation areas and propagation bridge nodes, and the graph coloring algorithm can accurately divide different propagation communities and identify key propagation paths by minimizing the optimization goal of the number of colorings. Compared with traditional manual analysis, this method is more objective and systematic, and can discover complex propagation patterns that are difficult to detect through manual analysis.

[0082] Propagation situation awareness model based on hybrid expert mechanism: The propagation situation awareness model established by the present invention adopts a hybrid expert mechanism. The multiple specialized sub-networks of the expert network layer respectively handle the estimation tasks of different propagation parameters such as infection rate, recovery rate, incubation period and environmental factors. The gated network layer dynamically calculates the weight of each expert through the soft routing mechanism, and the fusion output layer fuses the expert outputs by weight to generate the final propagation parameter estimation results. Traditional parameter estimation methods usually use a single statistical model or empirical formula, which cannot adapt to the complex and changeable propagation conditions in the hospital environment. The core advantage of the hybrid expert mechanism is that it can automatically select the most suitable parameter estimation strategy according to different propagation scenarios. Each expert sub-network is used to handle different types of propagation parameter estimation problems, and the gating mechanism ensures that the most relevant expert combination is activated in different situations. Compared with the traditional static parameter setting, this intelligent parameter estimation method has stronger adaptability and accuracy, and can dynamically respond to the impact of changes in the hospital environment on the propagation parameters.

[0083] The synergistic effect of the above three technical ideas has produced a significant amplification effect of technical advantages. The multi-layer network model provides a more accurate and comprehensive network structure foundation for community detection and parameter estimation, enabling the random block model to perform community division on a more accurate network representation, while providing richer feature information for the hybrid expert mechanism. The community detection results, in turn, provide guidance for the optimization of the inter-layer coupling parameters of the multi-layer network, helping the hybrid expert mechanism to better understand the local propagation characteristics of the network. The precise propagation parameters estimated by the hybrid expert mechanism further optimize the propagation dynamics description of the multi-layer network model and improve the accuracy of community detection. This positive feedback loop mechanism between the three enables the entire system to be continuously optimized and improved, forming an adaptive propagation path tracing system. Compared with the limitations of the traditional methods in which the components are independent of each other and lack collaborative optimization, the collaborative mechanism of the present invention achieves a significant improvement in overall performance and fundamentally solves the technical problem of insufficient accuracy in propagation path tracing.

[0084] Specifically, the principle of the present invention is: the fundamental reason why the technical solution of the present invention can solve the core technical problems is that it has established a high-precision multi-dimensional propagation modeling system and an intelligent parameter optimization mechanism. The root cause of the insufficient traceability accuracy of traditional methods lies in the limitations of single-dimensional modeling and static parameter estimation. The four-dimensional data structure constructed by the present invention through multi-source data fusion includes time, space, individual and environmental dimensions, providing a comprehensive data foundation for accurate modeling. The multi-layer network propagation model decomposes the complex hospital infection propagation process into three interrelated network levels of physical contact, spatial proximity and temporal co-occurrence. Each layer of the network captures different types of propagation relationships, and the inter-layer coupling mechanism describes the synergistic effect of different propagation modes. This hierarchical modeling method can accurately portray the complex propagation dynamics process in a real hospital environment.

[0085] The introduction of the stochastic block model is a key technological breakthrough in improving traceability accuracy. By assuming that the probability of connecting nodes within the same community is higher than that between different communities, it automatically identifies modular structures in the transmission network, transforming the identification of infection transmission paths into a graph coloring optimization problem. By minimizing the number of colorings, it accurately delineates transmission communities, making high-risk transmission paths more prominent given the high connection density within the community. The individual behavioral heterogeneity model incorporates individual characteristic parameters such as patient age, gender, underlying disease status, and immune status, and modifies key parameters in the traditional SEIR model, such as infection rate, recovery rate, and incubation period. This allows the model to reflect the differentiated transmission behaviors of different individuals, significantly improving the accuracy of transmission parameter estimation.

[0086] The hybrid expert mechanism employed in the propagation situation awareness model is the core technology for achieving high-precision parameter estimation. Multiple specialized subnetworks within the expert network layer handle the estimation of different propagation parameters. The gated network layer dynamically calculates the weights of each expert through a soft routing mechanism, ensuring the selection of the most appropriate parameter estimation strategy for different propagation scenarios. Dynamic adjustment of the capacity factor further balances model complexity and estimation accuracy. The game optimization mechanism uses Nash equilibrium to solve for optimal propagation parameter weights by comparing simulation results from different time periods with actual observations. This achieves continuous model optimization and precision improvement, ensuring the reliable identification of high-risk propagation paths.

[0087] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0088] The specific implementation of step S01 is the same as above and will not be described in detail here.

[0089] In step S02, the determination of the small-world characteristic is specifically expressed as follows:

[0090] C>0.3∧L <logN;

[0091] Where C is the network clustering coefficient; L is the average path length of the network; and N is the total number of network nodes.

[0092] The identification of scale-free features is achieved by fitting the degree distribution, which can be expressed as follows:

[0093] P(k)=Ak -γ ;

[0094] Where P(k) is the probability of occurrence of a node with degree k; A is the normalization constant; γ is the power law exponent, and when 2<γ<3, it is determined to be a scale-free network.

[0095] The calculation formula of interlayer coupling strength is:

[0096]

[0097] Where, is the coupling strength between node i and node j between the αth layer and the βth layer networks; is the connection indicator function between node i and node j in the αth layer network at time t; T is the total length of observation time.

[0098] In step S03, the likelihood function of the random block model is specifically expressed as:

[0099]

[0100] Where L(θ) is the network likelihood function; For the community i and community z j The connection probability between ij is the adjacency matrix element; z i is the community identifier of node i.

[0101] The propagation probability matrix calculation formula is:

[0102]

[0103] Where, P ij is the probability of propagation from node i to node j; w ij is the edge weight between nodes i and j; d ij is the network distance between nodes i and j; λ is the distance attenuation parameter, ranging from 1 to 5; N i is the set of neighbor nodes of node i.

[0104] In step S04, the calculation formula for the corrected infection rate is:

[0105] β corrected =β base ·S age ·S immune ·E env ;

[0106] Where, β corrected is the corrected infection rate; β base is the basic infection rate, obtained through historical data statistics; S age is the age susceptibility coefficient; S immune is the immune susceptibility coefficient; E env is the environmental propagation coefficient.

[0107] The age susceptibility coefficient is calculated as:

[0108]

[0109] Where age is the patient's age in years.

[0110] The immune susceptibility coefficient is determined based on the antibody titer:

[0111]

[0112] Where titer is the serum antibody titer, which is determined by enzyme-linked immunosorbent assay.

[0113] The protection factor calculation formula is:

[0114]

[0115] The environmental propagation coefficient is calculated as:

[0116]

[0117] The formula for calculating the corrected recovery rate is:

[0118] γ corrected =γ base ·R immune ·R treatment ;

[0119] Where, γ corrected is the corrected recovery rate; γ base is the basic recovery rate; R immune is the individual immune recovery coefficient; R treatment is the treatment condition correction factor.

[0120] In step S05, the gate network weight calculation formula is:

[0121]

[0122] Where g i is the gating weight of the i-th expert; W g is the gated network weight matrix, with a dimension of d×K; x is the input feature vector, with a dimension of d; b g is the bias vector with dimension K; K is the total number of experts, which is set to 4; d is the feature dimension, which is set to 128.

[0123] The expert selection strategy under capacity factor constraints is:

[0124] E active =TopK(g, C);

[0125] Where, E active is the set of activated experts; C is the capacity factor parameter, ranging from 1 to 4; TopK is the function of taking the top C experts with the largest weights.

[0126] The final parameter estimation results are calculated as:

[0127]

[0128] Where, is the estimated propagation parameter vector; f i (x) is the output of the i-th expert network, which contains the estimated values ​​of infection rate, recovery rate, incubation period and environmental factors.

[0129] In step S06, the game payoff function is defined as:

[0130] U t =-[α·RMSE t +(1-α)·MAPE t ];

[0131] Where U t is the game profit in period t; RMSE t is the root mean square error of the tth period; MAPE t is the mean absolute percentage error of the t-th period; α is the weight coefficient, which is set to 0.6; t is the period index.

[0132] The formula for calculating the root mean square error is:

[0133]

[0134] Where n is the number of observation samples; is the predicted value of the i-th sample in the t-th period; is the observation value of the i-th sample in the t-th period.

[0135] The mean absolute percentage error is calculated as:

[0136]

[0137] The Nash equilibrium condition is expressed as:

[0138]

[0139] Where w i is the weight coefficient of the i-th period; T is the total number of periods.

[0140] The normalization constraint of the weight coefficient is:

[0141]

[0142] In step S07, the spatiotemporal autocorrelation coefficient is calculated as follows:

[0143]

[0144] Where I is the Moran index; n is the number of spatial units; w ij is the spatial weight matrix element; x i is the observation value of the i-th unit; is the mean of the observations.

[0145] The state transition probability of the Viterbi algorithm is calculated as:

[0146] δ t (j) = max 1≤i≤N [δ t-1 (i) a ij ]·b j (o t );

[0147] Where, δt (j) is the maximum probability of state j at time t; a ij is the probability of transitioning from state i to state j; b j (o t ) generates observation o for state j t The probability of; N is the total number of states.

[0148] In the dynamic capacity adjustment function, the complexity evaluation value calculation formula is:

[0149] C eval =w1·C network +w2·C accuracy +w3·C resource ;

[0150] Where C eval is the complexity evaluation value; C network is the network complexity index; C accuracy is the prediction accuracy index; C resource is the resource utilization rate; w1, w2, and w3 are weight coefficients, satisfying w1+w2+w3=1. The default values ​​are w1=0.4, w2=0.3, and w3=0.3.

[0151] The calculation formula of network complexity index is:

[0152]

[0153] Where |V| is the number of network nodes; |E| is the number of network edges; D is the network diameter; C cluster is the clustering coefficient; L avg is the average path length; N norm is the normalization factor.

[0154] The linear increasing adjustment function is:

[0155] C factor =C min +k·C eval ;

[0156] Where C factor is the adjusted capacity factor; C min is the minimum capacity factor, which is set to 1; k is the increasing slope parameter, which ranges from 0.5 to 2.0.

[0157] The nonlinear smoothing adjustment function uses the Sigmoid function:

[0158]

[0159] Where C maxis the maximum capacity factor, which is set to 4; β is the smoothing parameter, which ranges from 1 to 10; μ is the center parameter, which is set to 0.5.

[0160] The exponential decay adjustment function is:

[0161] C factor =C max ·exp(-γ·C eval );

[0162] Where γ is the attenuation coefficient, which ranges from 0.1 to 1.0.

[0163] To better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: In March, a multidrug-resistant Klebsiella pneumoniae infection occurred in the respiratory ward of a city's tertiary hospital. Researchers used a hospital infection transmission path tracing method based on multi-source data fusion to analyze and trace the source of this infection event.

[0164] The researchers first deployed a multi-dimensional data collection system to comprehensively collect hospital infection monitoring data. Radio frequency identification readers and Bluetooth beacon systems were installed in 32 wards and 3 nurse stations in the respiratory ward to track the real-time location of 136 patients and 47 medical staff. The positioning accuracy was controlled within 0.8m, and the data collection frequency was set to once every 5 seconds. Personnel contact events were recorded through smart wearable devices, with the contact distance threshold set to 2m and the contact time threshold set to 15 minutes. The researchers used gene sequencing technology to perform whole genome sequencing on 16 strains of Klebsiella pneumoniae isolated from patient sputum and environmental samples to obtain complete gene sequence information. At the same time, 47 environmental monitoring sensors were deployed in the ward to collect environmental parameters such as temperature, humidity, air quality and disinfectant concentration at intervals of 10 minutes. After 7 consecutive days of data collection, data containing 1.2×10 6 Track records, 8.7×10 3 contact events, 16 gene sequence samples, and 4.8×10 4 The four-dimensional data structure of environmental monitoring data points.

[0165] The researchers constructed a multi-layered infection transmission network model based on complex network theory. The physical contact network layer was constructed based on contact record data. The network consists of 183 nodes and 2.4×10 3 The edge weight is determined by the contact frequency and contact intensity. A spatial proximity network layer is constructed based on the location trajectory data. When the spatial distance between people is less than 2m, a proximity relationship edge is established. The network contains 5.1×10 3 The temporal co-occurrence network layer is constructed using temporal co-occurrence information. The time window is set to 30 minutes and the spatial region division granularity is 10m. 2, the network contains 3.7×10 3 edges. By calculating the topological indices of each layer of the network, it is found that the clustering coefficient of the physical contact network is 0.42 and the average path length is 3.8, satisfying the small-world property determination conditions C>0.3∧L<logN. The fitting result of the degree distribution shows that the power-law exponent γ = 2.3, meeting the scale-free feature recognition condition 2<γ<3. The calculation result of the inter-layer coupling strength shows that the coupling strength between the physical contact layer and the spatial proximity layer is 0.67, and the coupling strength with the time co-occurrence layer is 0.54.

[0166] The stochastic block model algorithm is used to identify the community structure of the multi-layer network. Community division is performed by the maximum likelihood estimation method. After iterative optimization, the network is divided into 8 propagation communities, with the intra-community connection probability being 0.73 and the inter-community connection probability being 0.08. The community detection problem is transformed into a graph coloring problem for solution. The greedy coloring algorithm is used to assign different color identifiers to each community, and finally 6 colors are used to complete the coloring. Based on the community division results, the propagation probability matrix between nodes is calculated, and the distance decay parameter λ is set to 2.5. Table 1 shows the propagation probabilities between some high-risk nodes.

[0167] Table 1 Propagation probability matrix between high-risk nodes

[0168]

[0169] A modified propagation dynamics model integrating individual difference features is constructed. Individual characteristic information such as patient age, gender, underlying disease status, and immune level is collected to establish an individual susceptibility assessment system. Among 136 patients, there are 23 patients aged 0 - 18, 87 patients aged 19 - 65, and 26 patients aged over 65, and the corresponding age susceptibility coefficients are 1.2, 1.0, and 1.5 respectively. Through serum antibody titer determination, it is found that the antibody titers of 89 patients are lower than 10 2 , and the susceptibility coefficient is set to 1.8. Among 47 medical staff, 32 who wear protective equipment completely have a protection coefficient of 0.2, 11 who wear protective equipment partially have a protection coefficient of 0.6, and 4 who do not wear protective equipment have a protection coefficient of 1.0. The investigation of environmental disinfection frequency shows that there are 18 wards with more than 4 disinfections per day, and the environmental transmission coefficient is 0.3; there are 12 wards with 2 - 4 disinfections per day, and the environmental transmission coefficient is 0.6; there are 2 wards with less than 2 disinfections per day, and the environmental transmission coefficient is 1.0. The basic infection rate is obtained by statistical analysis of historical data as 0.12. After correction by individual characteristic parameters, the corrected infection rate of high-risk patients reaches 0.32, and the corrected infection rate of low-risk patients is 0.08.

[0170] A transmission situation awareness model based on a hybrid expert mechanism was established for parameter optimization and estimation. The model utilizes four functionalized expert networks. The infection rate estimation expert utilizes a three-layer recurrent neural network, each containing 64 long-short-term memory units. The recovery rate estimation expert utilizes a convolutional neural network consisting of two convolutional layers and two fully connected layers, with a convolution kernel size of 3×3. The incubation period estimation expert utilizes an 18-layer residual network architecture. The environmental factor estimation expert utilizes a transformer encoder with eight attention heads. The gating network utilizes a two-layer fully connected architecture with 64 hidden layer neurons. The capacity factor parameter is set to 3, and the three most relevant expert networks are activated simultaneously. The model is trained using a gradient descent optimization algorithm with a learning rate of 0.001 and a batch size of 32. Convergence was achieved after 1000 training epochs. The optimized infection rate estimation accuracy improved to 0.94, the recovery rate estimation accuracy reached 0.91, and the incubation period estimation accuracy reached 0.89.

[0171] A game-theoretic optimization mechanism was used to iteratively optimize the weights of the propagation parameters. The estimated propagation parameters for seven time periods served as game participants, and the observed infection spread data served as the standard game environment. The weight coefficient α in the game payoff function was set to 0.6, and the root mean square error (RMS) and mean absolute percentage error (MAPE) between the predicted and actual results for each time period were calculated. The RMS error for the first time period was 0.083, and the MAPE was 12.4%; the RMS error for the seventh time period was 0.067, and the MAPE was 9.8%. Using a Nash equilibrium solver, after 15 iterations, convergence was achieved, resulting in the weight coefficients for the propagation parameters for each time period being 0.08, 0.12, 0.15, 0.18, 0.19, 0.16, and 0.12, respectively. The weight coefficients were used to feedback optimize the propagation situation awareness model, ultimately achieving a propagation parameter prediction accuracy of 0.96.

[0172] A propagation path backtracking algorithm based on dynamic programming was used to locate the source and reconstruct the path. Spatiotemporal correlation analysis revealed a Moran index of 0.41, indicating significant spatial clustering. A state transition probability matrix was constructed, and the Viterbi algorithm was used to determine the most likely propagation path sequence. The algorithm identified three primary propagation paths. Table 2 shows the node sequence for the highest-probability propagation path.

[0173] Table 2 Node sequence of the highest probability propagation path

[0174] Propagation Order Node ID Time of infection site of infection Probability of transmission 1 N003 03-15 08:30 Ward 312 1.000 2 P001 03-15 14:20 Ward 312 0.378 3 P015 03-16 09:45 Ward 315 0.234 4 N012 03-16 16:10 Nurse Station B 0.203 5 P027 03-17 11:30 Ward 318 0.298 6 P043 03-17 19:55 Ward 318 0.267

[0175] The Markov Chain Monte Carlo method was used to quantify the uncertainty of the path reconstruction results. The confidence interval of the main transmission path was [0.823, 0.947], with a credibility score of 0.89. The source of transmission was ultimately determined to be medical staff member N003, with the source coordinates (127.3, 89.7), the start time was 03-15 08:30, and the confidence level was 0.92.

[0176] The dynamic capacity adjustment function adjusts the capacity factor based on network complexity, prediction accuracy, and computing resource utilization. The calculated network complexity index is 0.67, the prediction accuracy index is 0.94, the computing resource utilization is 73%, and the complexity assessment value is 0.72. Because the assessment value is in the medium complexity range, a Sigmoid nonlinear smoothing adjustment function is used with a smoothing parameter β set to 5 and a center parameter μ set to 0.5. The adjusted capacity factor is 2.8.

[0177] Traditional methods for tracing the transmission paths of hospital-acquired infections rely primarily on epidemiological surveys and manual analysis. These methods involve interviews with infected patients and healthcare workers to collect contact histories and activity trajectories. Experts then use their experience to determine possible transmission paths. This approach suffers from incomplete information collection, strong subjectivity, and poor timeliness. It typically takes 5-7 days to initially pinpoint the source of transmission, with an accuracy rate of approximately 65%. The present invention utilizes multi-source data fusion and intelligent algorithm optimization to achieve automated identification and precise location of transmission paths. In terms of data collection completeness, this method achieves continuous, real-time monitoring through a multi-dimensional sensor network, achieving 98% data coverage, compared to only 60% for traditional methods. In terms of analytical accuracy, this method achieves 92% accuracy in locating the source of transmission, a 15% improvement over traditional methods. Furthermore, this method can complete a complete transmission path analysis within two days, reducing analysis time by 60% compared to traditional methods. In terms of transmission parameter estimation accuracy, this method achieves a comprehensive estimation accuracy of 96%, compared to only 78% for traditional empirical estimation methods, an 18% improvement. These technological advances provide more scientific and accurate decision-making support for hospital infection prevention and control, and effectively improve the emergency response capabilities to infection outbreaks.

[0178] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A hospital infection transmission path tracing method based on multi-source data fusion, characterized by: include: Collect multi-source infection monitoring data from hospitals and establish a multi-dimensional data set including patient flow trajectories, medical staff contact records, pathogen gene sequence information and environmental monitoring data; construct a multi-layer network transmission model, hierarchically model the physical contact network, spatial proximity network and temporal co-occurrence network, and identify the small-world characteristics and scale-free characteristics in the network; use the random block model to perform community detection on the hierarchical network structure, transform the infection transmission path identification problem into a graph coloring problem for solution, optimize the accuracy of high-risk transmission path identification by minimizing the number of colorings, and calculate the transmission probability matrix between nodes; establish an individual behavior heterogeneity model, integrate the patient's immune status, medical staff's protection level and environmental disinfection frequency as individual characteristic parameters, and modify the traditional SEIR model transmission parameters; use the transmission situation awareness model to optimize the transmission parameter estimation, use the hybrid expert mechanism to dynamically select different parameter estimation experts, control the number of activated experts through the capacity factor, and adaptively adjust the infection rate, recovery rate and incubation period based on the modified transmission parameters; A game optimization mechanism is established to conduct strategic games through the simulation results of different time periods and the actual observation results of the corresponding time periods, and the Nash equilibrium solution is used to obtain the weight coefficients of the propagation parameters; a propagation path backtracking algorithm is used, combined with spatiotemporal correlation analysis and network propagation simulation results, to output the most likely infection propagation path sequence and the propagation source location results.

2. The hospital infection transmission path tracing method based on multi-source data fusion according to claim 1 is characterized in that: The multidimensional dataset is specifically a four-dimensional data structure including time dimension, space dimension, individual dimension and environmental dimension. The time dimension records the specific time and duration of the infection event, the space dimension records the location coordinates and movement trajectory of the infected and susceptible persons, the individual dimension records the identity, health status and behavioral characteristics of the personnel, and the environmental dimension records the ward temperature, humidity, ventilation conditions and disinfection frequency.

3. The hospital infection transmission path tracing method based on multi-source data fusion according to claim 2 is characterized in that: The multi-layer network transmission model specifically abstracts the hospital infection transmission process into multiple interrelated network layers. The physical contact network layer describes the direct physical contact relationship between people, the spatial proximity network layer describes the proximity relationship between people with a spatial distance of less than 2m, and the temporal co-occurrence network layer describes the co-occurrence relationship between people appearing in the same spatial area within the same time period.

4. The hospital infection transmission path tracing method based on multi-source data fusion according to claim 3 is characterized in that: The small-world characteristic specifically refers to the fact that the shortest path length between most nodes in the network is small, and the network has a high clustering coefficient, which is manifested as a network topology feature with dense local connections and sparse global connections; the scale-free characteristic specifically refers to the fact that the node degree distribution in the network follows a power-law distribution, with a small number of high-degree nodes connecting a large number of other nodes.

5. The hospital infection transmission path tracing method based on multi-source data fusion according to claim 4 is characterized in that: The random block model is specifically a network community discovery algorithm that assumes that nodes in the network belong to different potential communities, and the probability of connecting nodes within the same community is higher than the probability of connecting nodes between different communities; the graph coloring problem is specifically to assign a color to each vertex in the graph so that adjacent vertices have different colors and the total number of colors used simultaneously is minimized.

6. The hospital infection transmission path tracing method based on multi-source data fusion according to claim 5 is characterized in that: The transmission probability matrix is ​​specifically a two-dimensional matrix that describes the probability of infection transmission between any two nodes in the network, and the matrix element values ​​represent the probability of infection spreading from the source node to the target node; the community detection results are specifically structured data including community division identification, intra-community connection density, inter-community connection strength and a list of high-risk transmission paths.

7. The method for tracing the transmission path of hospital infection based on multi-source data fusion according to claim 6 is characterized in that: The individual behavior heterogeneity model specifically takes into account the differentiated behavioral characteristics exhibited by different individuals during the transmission process, including the impact of differences in susceptibility, contact patterns, and protective behavior on transmission dynamics; the individual characteristic parameters specifically include a set of parameters including patient age, gender, underlying disease status, immunity level, years of work experience of medical staff, and use of protective equipment.

8. The hospital infection transmission path tracing method based on multi-source data fusion according to claim 7 is characterized in that: The transmission situation awareness model is specifically an intelligent transmission parameter estimation model based on a hybrid expert mechanism, which includes a multi-level hybrid expert architecture of an expert network layer, a gating network layer and a fusion output layer. The expert network layer includes infection rate estimation experts, recovery rate estimation experts, incubation period estimation experts and environmental factor estimation experts. The gating network layer dynamically calculates the weight of each expert through a soft routing mechanism.

9. The method for tracing the transmission path of hospital infection based on multi-source data fusion according to claim 8 is characterized in that: The capacity factor is specifically a key parameter that controls the number of experts activated simultaneously in the propagation situation awareness model, and its value range is a positive integer between 1 and the total number of experts. It also includes designing a dynamic capacity adjustment function to adjust the capacity factor of the propagation situation awareness model, and calculating the complexity evaluation value based on the network complexity index, the prediction accuracy index and the computing resource occupancy rate.

10. The hospital infection transmission path tracing method based on multi-source data fusion according to claim 9 is characterized in that: When the complexity evaluation value is in the low complexity range, a linear increasing adjustment function is used to increase the capacity factor; when the complexity evaluation value is in the medium complexity range, a nonlinear smooth adjustment function is used to fine-tune the capacity factor; when the complexity evaluation value is in the high complexity range, an exponential decay adjustment function is used to reduce the capacity factor to control the computational overhead and avoid overfitting.

Images