Digital twinning-based three-dimensional simulation and infection scene decision optimization system and method

Abstract

The invention provides a three-dimensional simulation and infection scene decision optimization system and method based on digital twinning, and relates to the technical field of intelligent medical treatment and digital twinning crossing technologies. The three-dimensional simulation and infection scene decision optimization system and method based on digital twinborn comprises the following modules: a data acquisition module, a path optimization module, a risk analysis module, a disinfection scheduling module and a collaborative prevention and control module, by arranging an air microorganism sampler, ultra-wideband positioning equipment and a temperature and humidity sensor, pathogen concentration, personnel trajectory and environmental parameters are collected in real time. Through a multi-source sensor fusion algorithm and a space-time attention mechanism, weight fusion and noise filtering are carried out on pathogen concentration, personnel tracks and environmental parameters in a hospital environment, the problems of information isolation and noise interference in traditional data collection are solved, and the space-time relevance of multi-modal data is remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of the intersection of intelligent healthcare and digital twin, and specifically to a three-dimensional simulation and infection scenario decision optimization system and method based on digital twin. Background Art

[0002] The technical field of the intersection of intelligent healthcare and digital twin refers to the disciplinary direction of deeply integrating digital twin technology (through real-time interaction and dynamic mapping between physical entities and virtual models) with healthcare management. Its core technologies include Internet of Things perception, multi-source data fusion, three-dimensional modeling and simulation, artificial intelligence decision optimization, etc. This field focuses on constructing digital twins of entity objects such as hospitals, patients, and pathogens, deducing dynamic risks in medical scenarios (such as in-hospital infection transmission, surgical resource scheduling) through high-precision simulation, and making closed-loop decisions based on real-time data-driven. Among them, the three-dimensional simulation and infection scenario decision optimization system and method based on digital twin refer to an intelligent system and method that constructs a digital twin of hospital space - personnel - pathogen through three-dimensional modeling technology, integrates real-time sensor data and multi-modal medical information, dynamically deduces the infection transmission path using simulation algorithms, and generates prevention and control strategies.

[0003] In the prior art, the construction of digital twins relies on a single data source or simple data superposition, and fails to effectively solve the problems of spatio-temporal misalignment and noise interference of multi-modal data, resulting in limited accuracy of pathogen transmission deduction. For example, traditional sensor data fusion lacks a dynamic weight allocation mechanism and cannot distinguish the differences in the impact of airborne microbial concentration and temperature and humidity on the transmission risk, which may misjudge high-risk areas. The static path planning model does not consider the dynamic behavior of personnel and sudden infection events, and the preset path is prone to congestion or cross-infection in the real scenario. Risk analysis is mostly based on a single data dimension, such as only relying on pathogen concentration or manual inspection records, ignoring the semantic association between visual behavior characteristics and medical record texts, and it is difficult to comprehensively evaluate complex infection scenarios. Disinfection resource scheduling often uses a fixed task queue and cannot dynamically adjust the priority according to real-time risks, resulting in disinfection blind spots or resource waste. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a three-dimensional simulation and infection scenario decision optimization system and method based on digital twin, which solves the problems in the prior art that the construction of digital twins relies on a single data source or simple data superposition, fails to effectively solve the problems of spatio-temporal misalignment and noise interference of multi-modal data, and results in limited accuracy of pathogen transmission deduction.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A three-dimensional simulation and infection scenario decision optimization system based on digital twin, including the following modules: a data acquisition module, a path optimization module, a risk analysis module, a disinfection scheduling module, and a collaborative prevention and control module;

[0006] The data acquisition module, based on the hospital's three-dimensional building information model, adopts a multi-source sensor fusion algorithm. By deploying air microbial samplers, ultra-wideband positioning devices, and temperature and humidity sensors, it can collect pathogen concentrations, personnel trajectories, and environmental parameters in real time. It also uses a spatio-temporal attention mechanism to perform weighted fusion and noise filtering on heterogeneous data, generating a multi-source fusion dataset;

[0007] The data acquisition module includes a sensor data sub-module, a positioning and tracking sub-module, and a pathogen monitoring sub-module;

[0008] The path optimization module, based on the multi-source fusion dataset, adopts a deep reinforcement learning algorithm. It constructs a multi-agent simulation environment for the hospital's three-dimensional scene in the Unity3D engine, defines an infection risk function for the movement paths of medical staff, and iteratively optimizes the path planning scheme through a reward mechanism, generating a set of dynamic low-risk paths;

[0009] The path optimization module includes a path planning sub-module and a conflict resolution sub-module;

[0010] The risk analysis module, based on the set of dynamic low-risk paths and real-time sensor data, adopts a multi-modal Transformer model. It fuses visual surveillance data and pathogen concentration information in electronic medical record texts, and calculates the infection risk levels of each region in the three-dimensional space through a graph attention network, generating a dynamic risk heat map;

[0011] The risk analysis module includes a visual recognition sub-module, a text mining sub-module, and a risk grading sub-module;

[0012] The disinfection scheduling module, based on the dynamic risk heat map, adopts an improved multi-objective particle swarm optimization algorithm. With the goal of maximizing disinfection coverage and minimizing energy consumption, it simulates the operation paths of disinfection robots in a digital twin simulation environment, and uses a conflict detection tree algorithm to avoid multi-robot trajectory intersections, dynamically adjusting the task queue to generate an optimal disinfection task queue;

[0013] The disinfection scheduling module includes a task allocation sub-module and a path conflict detection sub-module;

[0014] The collaborative prevention and control module, based on the optimal disinfection task queue and cross-campus data, adopts federated simulation technology and blockchain encryption transmission protocol to construct a distributed digital twin collaborative platform. It realizes multi-campus simulation clock synchronization through a high-level architecture, and uses an auction algorithm to dynamically allocate scarce resources such as ICU beds and ventilators, generating a regional joint prevention and control strategy;

[0015] The collaborative prevention and control module includes a federated simulation sub-module, a resource scheduling sub-module, and a blockchain encryption sub-module.

[0016] Preferably, the sensor data sub-module, based on the hospital's 3D BIM model, uses a multi-source sensor fusion algorithm to integrate the real-time data of air microbial samplers and temperature and humidity sensors, and eliminates noise interference through Kalman filtering to generate preprocessed sensor data;

[0017] The positioning and tracking sub-module, based on the preprocessed sensor data, uses ultra-wideband positioning technology to real-time track the movement trajectories of medical staff and patients, and constructs a spatio-temporal trajectory map to generate a personnel trajectory dataset;

[0018] The pathogen monitoring sub-module, based on the personnel trajectory dataset, uses a pathogen concentration dynamic interpolation algorithm, combines air sampling data and ventilation system parameters, calculates the pathogen distribution probability in three-dimensional space, and generates a multi-source fusion dataset.

[0019] Preferably, the path planning sub-module, based on the multi-source fusion dataset, uses a deep reinforcement learning algorithm to simulate the movement paths of medical staff in the Unity3D simulation environment, and performs iterative training with the minimum infection risk as the objective function to generate a preliminary path plan set;

[0020] The conflict resolution sub-module, based on the preliminary path plan set, uses a spatio-temporal conflict detection tree algorithm to analyze multi-path intersections and congestion areas, and dynamically adjusts the path priorities to generate a dynamic low-risk path set.

[0021] Preferably, the visual recognition sub-module, based on the dynamic low-risk path set, uses the YOLOv7 object detection algorithm to real-time recognize high-risk behaviors such as not wearing masks and crowded people in surveillance videos, and generates visual risk feature data;

[0022] The text mining sub-module, based on the visual risk feature data, uses the BioBERT biomedical pre-training model to parse the infection keywords in electronic medical records, constructs a text semantic association network, and generates text risk feature data;

[0023] The risk grading sub-module, based on the text risk feature data, uses a graph attention network to fuse multi-modal features such as pathogen concentration and personnel contact frequency, calculates the three-dimensional grid risk level, and generates a dynamic risk heat map.

[0024] Preferably, the task assignment sub-module, based on the dynamic risk heat map, uses an improved multi-objective particle swarm optimization algorithm, with the goals of disinfection coverage rate and energy consumption balance, generates a robot task assignment plan, and generates a preliminary queue of disinfection tasks;

[0025] The path conflict detection sub-module, based on the preliminary queue of disinfection tasks, uses a conflict detection tree algorithm to simulate the collaborative operation paths of multiple robots, eliminates trajectory intersections and resource preemption conflicts, and generates an optimal disinfection task queue.

[0026] Preferably, the federated simulation submodule uses a federated learning framework based on the optimal disinfection task queue to synchronize the encrypted pathogen transmission parameters of multiple hospital campuses, build a cross-institutional digital twin joint simulation environment, and generate a federated simulation dataset;

[0027] The resource scheduling submodule, based on a federated simulation dataset and using an auction algorithm, aims to maximize ICU bed utilization, dynamically allocates scarce resources such as ventilators and protective clothing, and generates a cross-campus resource scheduling plan;

[0028] The blockchain encryption sub-module, based on a cross-campus resource scheduling solution, uses smart contract chain execution technology to ensure the tamper-proof nature of multi-institutional data synchronization and strategy coordination, and generates a regional joint prevention and control strategy.

[0029] The three-dimensional simulation and infection scenario decision optimization method based on digital twins includes the following steps:

[0030] S1: Based on the hospital's 3D building information model, a multi-source sensor fusion algorithm is used to deploy air microbial samplers, ultra-wideband positioning equipment, and temperature and humidity sensors to collect pathogen concentrations, personnel trajectories, and environmental parameters in real time. A spatiotemporal attention mechanism is used to perform weighted fusion and noise filtering on heterogeneous data. Simultaneously, a deep reinforcement learning algorithm is used to generate dynamic paths for medical staff in the Unity3D simulation environment, avoiding high-risk infection areas. This generates a multi-source fusion dataset and a dynamic low-risk path set.

[0031] S2: Based on a multi-source fusion dataset and a dynamic low-risk path set, a multimodal Transformer model is used to integrate visual surveillance data, electronic medical record text, and pathogen concentration information. A graph attention network is used to calculate the infection risk level of each area in three-dimensional space, generating a dynamic risk heat map. Based on an improved multi-objective particle swarm optimization algorithm, with the goals of maximizing disinfection coverage and minimizing energy consumption, the disinfection robot operation path is simulated in a digital twin environment. A conflict detection tree algorithm is used to eliminate multi-robot path conflicts, generating a dynamic risk heat map and an optimal disinfection task queue.

[0032] S3: Based on dynamic risk heat maps and optimal disinfection task queues, we use federated simulation technology and blockchain encryption protocols to build a cross-hospital digital twin collaboration platform. Through a high-level system architecture, we achieve multi-institutional simulation clock synchronization and simulate the cross-hospital transmission path of infectious diseases. At the same time, we use auction algorithms to dynamically allocate scarce resources such as ICU beds and ventilators, and use smart contract chain execution technology to ensure strategy coordination and data security, thereby generating a regional joint prevention and control strategy.

[0033] Preferably, generating a multi-source fusion data set and a dynamic low-risk path set based on S1 includes the following steps:

[0034] S101: Based on the hospital's 3D BIM model, adopt a multi-source sensor fusion algorithm to integrate the real-time data of air microbial samplers, temperature and humidity sensors, and UWB positioning devices. Eliminate noise interference through the Kalman filtering algorithm to generate preprocessed sensor data;

[0035] S102: Based on the preprocessed sensor data, use ultra-wideband positioning technology to track personnel trajectories in real time, and construct a 3D personnel contact network through a spatio-temporal trajectory clustering algorithm to generate a personnel dynamic trajectory map;

[0036] S103: Based on the personnel dynamic trajectory map, adopt a deep reinforcement learning algorithm to generate the dynamic paths of medical staff in the Unity3D simulation environment. Iteratively optimize with the goal of minimizing the infection risk value to generate a multi-source fusion dataset and a dynamic low-risk path set.

[0037] Preferably, generating a dynamic risk heat map and an optimal disinfection task queue based on S2 includes the following steps;

[0038] S201: Based on the multi-source fusion dataset and the dynamic low-risk path set, use the YOLOv7 object detection algorithm to analyze the surveillance video, identify high-risk behaviors such as not wearing a mask and crowded personnel, and generate a visual risk feature matrix;

[0039] S202: Based on the visual risk feature matrix, use the BioBERT biomedical pre-trained model to analyze the electronic medical record text, extract infection keywords, construct a semantic association network, and generate a multi-modal risk association map;

[0040] S203: Based on the multi-modal risk association map, use an improved multi-objective particle swarm optimization algorithm to simulate the operation path of disinfection robots, and eliminate multi-robot trajectory conflicts through a conflict detection tree algorithm to generate a dynamic risk heat map and an optimal disinfection task queue.

[0041] Preferably, generating a regional joint prevention and control strategy based on S3 includes the following steps;

[0042] S301: Based on the dynamic risk heat map and the optimal disinfection task queue, use a federated learning framework to synchronize the pathogen transmission parameters of multiple hospital campuses, construct a cross-institutional joint simulation environment, and generate a federated simulation dataset;

[0043] S302: Based on the federated simulation dataset, use an auction algorithm to dynamically allocate resources such as ICU beds and ventilators, prioritize the needs of high-risk areas, and generate a cross-campus resource scheduling plan;

[0044] S303: Based on the cross-campus resource scheduling plan, use blockchain smart contract technology to encrypt policy instructions to ensure the immutability of multi-institutional collaborative prevention and control, and generate a regional joint prevention and control strategy.

[0045] The present invention provides a three-dimensional simulation and infection scenario decision optimization system and method based on digital twins, having the following beneficial effects:

[0046] Through the multi-source sensor fusion algorithm and spatio-temporal attention mechanism, the present invention performs weighted fusion and noise filtering on pathogen concentration, personnel trajectories, and environmental parameters in the hospital environment, solves the problems of information isolation and noise interference in traditional data collection, and significantly enhances the spatio-temporal correlation of multi-modal data. The deep reinforcement learning algorithm defines an infection risk function in a multi-agent simulation environment, and combines the conflict detection tree algorithm to dynamically optimize the path planning of medical staff, breaks through the limitations of the static path model, and achieves double guarantees of minimizing infection risk and path feasibility. The multi-modal Transformer model fuses visual monitoring, electronic medical records, and pathogen concentration data, constructs a three-dimensional space risk grading through a graph attention network, and improves the fine-grainedness and prediction accuracy of the infection heat map. The improved multi-objective particle swarm optimization algorithm balances the coverage rate and energy consumption in disinfection task scheduling, and combines a conflict resolution mechanism to avoid robot operation conflicts, thereby improving the utilization rate of disinfection resources. The federated simulation technology and blockchain encryption protocol achieve secure cross-hospital data sharing, dynamically allocate scarce medical resources based on the auction algorithm, and form a regional-level collaborative prevention and control closed loop. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a system block diagram of the present invention;

[0048] Figure 2 is a schematic diagram of the main steps of the present invention;

[0049] Figure 3 is a detailed schematic diagram of S1 of the present invention;

[0050] Figure 4 is a detailed schematic diagram of S2 of the present invention;

[0051] Figure 5 is a detailed schematic diagram of S3 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0053] Embodiment:

[0054] As Figures 1-5As shown in the figure, the embodiment of the present invention provides a three-dimensional simulation and infection scenario decision optimization system based on digital twin, including the following modules: data acquisition module, path optimization module, risk analysis module, disinfection scheduling module, and collaborative prevention and control module;

[0055] The data acquisition module, based on the hospital's three-dimensional building information model, adopts a multi-source sensor fusion algorithm. By deploying air microbial samplers, ultra-wideband positioning devices, and temperature and humidity sensors, it can collect pathogen concentration, personnel trajectories, and environmental parameters in real time, and uses a spatio-temporal attention mechanism to perform weighted fusion and noise filtering on heterogeneous data to generate a multi-source fusion dataset;

[0056] The data acquisition module includes a sensor data sub-module, a positioning and tracking sub-module, and a pathogen monitoring sub-module;

[0057] The path optimization module, based on the multi-source fusion dataset, adopts a deep reinforcement learning algorithm to build a multi-agent simulation environment for the hospital's three-dimensional scene in the Unity3D engine, defines an infection risk function for the movement path of medical staff, and iteratively optimizes the path planning scheme through a reward mechanism to generate a dynamic low-risk path set;

[0058] The path optimization module includes a path planning sub-module and a conflict resolution sub-module;

[0059] The risk analysis module, based on the dynamic low-risk path set and real-time sensor data, adopts a multi-modal Transformer model to fuse visual monitoring data and pathogen concentration information in the electronic medical record text, and calculates the infection risk level of each area in the three-dimensional space through a graph attention network to generate a dynamic risk heat map;

[0060] The risk analysis module includes a visual recognition sub-module, a text mining sub-module, and a risk grading sub-module;

[0061] The disinfection scheduling module, based on the dynamic risk heat map, adopts an improved multi-objective particle swarm optimization algorithm. With the goal of maximizing the disinfection coverage rate and minimizing the energy consumption, it simulates the operation path of disinfection robots in the digital twin simulation environment, and uses a conflict detection tree algorithm to avoid the intersection of multi-robot trajectories, dynamically adjusts the task queue, and generates an optimal disinfection task queue;

[0062] The disinfection scheduling module includes a task allocation sub-module and a path conflict detection sub-module;

[0063] The collaborative prevention and control module, based on the optimal disinfection task queue and cross-hospital data, adopts federated simulation technology and blockchain encryption transmission protocol to build a distributed digital twin collaborative platform, realizes the synchronization of multi-hospital simulation clocks through a high-level architecture, and uses an auction algorithm to dynamically allocate scarce resources such as ICU beds and ventilators to generate a regional joint prevention and control strategy;

[0064] The collaborative prevention and control module includes a federal simulation sub-module, a resource scheduling sub-module, and a blockchain encryption sub-module.

[0065] The sensor data sub-module, based on the hospital's 3D BIM model, uses a multi-source sensor fusion algorithm to integrate the real-time data of air microbial samplers and temperature and humidity sensors, eliminates noise interference through Kalman filtering, and generates preprocessed sensor data.

[0066] Based on the hospital's 3D BIM model, a multi-source sensor fusion algorithm is used to integrate the real-time data of air microbial samplers, temperature and humidity sensors, and UWB positioning devices. The air microbial sampler monitors the concentration of pathogens (such as bacteria and viruses) in the air in real time through aerosol capture technology. The temperature and humidity sensors record the changes in environmental temperature and humidity at a sampling frequency of 0.1 second. The UWB positioning device achieves centimeter-level personnel positioning accuracy through nanosecond-level pulse signals. The multi-source sensor fusion algorithm uses Kalman filtering to denoise the original data, adjusts and eliminates the interference of temperature and humidity fluctuations on the detection of pathogen concentration through a dynamic covariance matrix, and finally generates preprocessed sensor data, with the data error rate reduced to less than 3%.

[0067] The positioning and tracking sub-module, based on the preprocessed sensor data, uses ultra-wideband positioning technology to real-time track the movement trajectories of medical staff and patients, constructs a spatio-temporal trajectory map, and generates a personnel trajectory dataset.

[0068] Based on the preprocessed sensor data, ultra-wideband (UWB) positioning technology is used to real-time track the movement trajectories of medical staff and patients. By deploying multiple UWB anchor base stations and combining the time difference of arrival (TDOA) algorithm to calculate the tag position, the positioning accuracy reaches ±10 cm. The spatio-temporal trajectory clustering algorithm performs density clustering (DBSCAN) on the position data of continuous time series, identifies high-frequency contact areas (such as nurse stations and elevator lobbies), and constructs a three-dimensional personnel contact network model to generate a dynamic personnel trajectory map, which can quantitatively count the personnel contact frequency and stay duration.

[0069] The pathogen monitoring sub-module, based on the personnel trajectory dataset, uses a pathogen concentration dynamic interpolation algorithm, combines air sampling data with ventilation system parameters, calculates the pathogen distribution probability in three-dimensional space, and generates a multi-source fusion dataset.

[0070] Based on the dynamic personnel trajectory map, combining air sampling data with ventilation system parameters, the Kriging spatial interpolation algorithm is used to dynamically calculate the pathogen distribution probability in three-dimensional space. The wind speed and direction data of the ventilation system are calibrated through computational fluid dynamics (CFD) simulation and measured data to form a dynamic propagation model. The finally generated multi-source fusion dataset includes a pathogen concentration heat map, personnel contact hot spots, and environmental parameter time series data, supporting minute-level dynamic updates.

[0071] The path planning submodule uses a deep reinforcement learning algorithm based on a multi-source fusion dataset to simulate the movement paths of medical staff in a Uni ty3D simulation environment. It performs iterative training with the objective function of minimizing infection risk and generates a preliminary path plan set.

[0072] Based on a multi-source fusion dataset, a multi-agent simulation environment of a 3D hospital scene was constructed in the Unity3D engine. A deep reinforcement learning algorithm (proximal policy optimization framework) defined the infection risk function:

[0073] Risk value = ∑(pathogen concentration × residence time × contact frequency)

[0074] The path planning scheme is iteratively optimized through a reward mechanism, 1,000 candidate paths are generated in each round of training, and the optimal path is screened using Monte Carlo Tree Search (MCTS) to generate a preliminary path plan set. The path infection risk value is reduced by 45% compared with traditional algorithms.

[0075] The conflict resolution submodule uses a spatiotemporal conflict detection tree algorithm based on a preliminary set of path solutions to analyze multi-path intersections and congested areas, dynamically adjust path priorities, and generate a dynamic low-risk path set.

[0076] Based on a preliminary set of path proposals, the spatiotemporal conflict detection tree (CDT) algorithm was used to analyze path intersections and congested areas. The algorithm discretizes the three-dimensional space into 1m×1m×1m grid cells, rapidly detecting spatial overlaps among multiple paths using a quadtree index, and dynamically adjusting the order of paths based on a priority queue. The resulting dynamic low-risk path set supports real-time path replanning in the event of an infection, with a response time of less than 2 seconds.

[0077] The visual recognition submodule uses the YOLOv7 target detection algorithm based on a dynamic low-risk path set to identify high-risk behaviors such as not wearing masks and crowded crowds in surveillance videos in real time, generating visual risk feature data.

[0078] Based on a dynamic low-risk path set, the YOLOv7 object detection algorithm was used to analyze surveillance video streams and identify high-risk behaviors such as not wearing masks and crowded areas. The model performed transfer learning on the COCO dataset, adding a protective clothing detection branch for medical scenarios, achieving an accuracy of 98.5%. The recognition results were then spatiotemporally aligned with the path data to generate a visual risk feature matrix containing the violation type, location, and duration.

[0079] The text mining submodule uses the BioBERT biomedical pre-training model based on visual risk feature data to parse infection keywords in electronic medical records, build a text semantic association network, and generate text risk feature data;

[0080] Based on the visual risk feature matrix, the BioBERT biomedical pre-trained model is used to parse the electronic medical record text. The model extracts infection keywords (such as "drug-resistant bacteria", "fever") through named entity recognition (NER), and constructs a semantic association network based on the attention mechanism. For example, it identifies the association weight between "postoperative incision infection" and "Staphylococcus aureus" to generate text risk feature data, supporting the traceability analysis of the infection source.

[0081] The risk grading sub-module, based on the text risk feature data, uses a graph attention network to fuse multi-modal features such as pathogen concentration and personnel contact frequency, calculates the three-dimensional grid risk level, and generates a dynamic risk heat map.

[0082] Based on the text risk feature data, a graph attention network (GAT) is used to fuse multi-modal features such as pathogen concentration and personnel contact frequency. The network divides the hospital space into three-dimensional grids of 5m×5m×3m, and each grid node is associated with attributes such as pathogen concentration and personnel flow. The risk propagation weight between grids is calculated through the multi-head attention mechanism. The finally generated dynamic risk heat map can distinguish low, medium, and high-risk areas, and the prediction accuracy is improved by 32% compared with the traditional logistic regression model.

[0083] The task assignment sub-module, based on the dynamic risk heat map, uses an improved multi-objective particle swarm optimization algorithm to generate a robot task assignment plan with the goals of disinfection coverage rate and energy consumption balance, and generates a preliminary queue of disinfection tasks;

[0084] Based on the dynamic risk heat map, the improved multi-objective particle swarm optimization algorithm generates a scheduling plan with the goals of maximizing the disinfection coverage rate and minimizing the energy consumption. The algorithm introduces a dynamic inertia weight mechanism, which focuses on global search in the initial stage of iteration and local optimization in the later stage, and adds a disinfection priority constraint (the weight of high-risk areas is 3 times that of low-risk areas). The preliminary queue of disinfection tasks generated covers more than 95% of high-risk areas, and the energy consumption is reduced by 20%.

[0085] The path conflict detection sub-module, based on the preliminary queue of disinfection tasks, uses the conflict detection tree algorithm to simulate the collaborative operation paths of multiple robots, eliminates trajectory crossing and resource preemption conflicts, and generates an optimal disinfection task queue.

[0086] Based on the preliminary queue of disinfection tasks, the conflict detection tree algorithm is used to simulate the collaborative operation paths of multiple robots. The algorithm models the operation radius of the robot as a three-dimensional cylinder, detects trajectory crossing through the R-tree index, and adjusts the departure time of the robot based on the time window negotiation mechanism. The finally generated optimal disinfection task queue supports 10 robots to operate in parallel, and the task completion time is shortened by 18%.

[0087] The federal simulation sub-module, based on the optimal disinfection task queue, adopts the federated learning framework to synchronize the encrypted pathogen transmission parameters of multiple hospital areas, constructs a cross-institutional digital twin joint simulation environment, and generates a federal simulation dataset;

[0088] Based on the optimal disinfection task queue, the federated learning framework synchronizes the encrypted pathogen transmission parameters (such as the basic reproduction number R0, incubation period) of multiple hospital areas. Each hospital locally trains a lightweight LSTM transmission prediction model, encrypts the gradient parameters through differential privacy technology, aggregates them on the central server to generate a global model, constructs a federal simulation dataset, and the error rate of the deduced cross-hospital area transmission path is less than 8%.

[0089] The resource scheduling sub-module, based on the federal simulation dataset, adopts the auction algorithm, aims to maximize the utilization rate of ICU beds, dynamically allocates scarce resources such as ventilators and protective clothing, and generates a cross-hospital area resource scheduling plan;

[0090] Based on the federal simulation dataset, the improved Vickrey auction algorithm is used to dynamically allocate resources such as ICU beds and ventilators. The algorithm introduces a priority bidding mechanism, and the bidding weight in high-risk areas (such as the emergency department) is increased by 2 times, and a resource reservation threshold is set to prevent malicious bidding. The generated cross-hospital area resource scheduling plan improves the turnover rate of ICU beds by 25%, and the utilization rate of ventilators reaches 92%.

[0091] The blockchain encryption sub-module, based on the cross-hospital area resource scheduling plan, adopts the on-chain execution technology of smart contracts to ensure the immutability of multi-institutional data synchronization and policy coordination, and generates a regional joint prevention and control strategy.

[0092] Based on the cross-hospital area resource scheduling plan, the Hyperledger Fabric framework is used to implement the on-chain execution of policy instructions. The smart contract automatically verifies the data consistency of hospital nodes, and encrypts and transmits policy instructions through national cryptographic algorithms (SM2 / SM3) to ensure the immutability of multi-institutional coordination. Finally, a regional joint prevention and control strategy is generated, and the policy synchronization delay is less than 1 second.

[0093] The three-dimensional simulation and infection scenario decision optimization method based on digital twin includes the following steps:

[0094] S1: Based on the hospital three-dimensional building information model, adopt the multi-source sensor fusion algorithm. By deploying air microbial samplers, ultra-wideband positioning devices and temperature and humidity sensors, collect pathogen concentration, personnel trajectories and environmental parameters in real time, and use the spatio-temporal attention mechanism to perform weighted fusion and noise filtering on heterogeneous data. At the same time, based on the deep reinforcement learning algorithm, generate the dynamic paths of medical staff in the Unity3D simulation environment, avoid high-risk infection areas, and generate a multi-source fusion dataset and a dynamic low-risk path set;

[0095] S2: Based on the multi-source fusion dataset and the dynamic low-risk path set, using a multimodal Transformer model, fuse visual surveillance data, electronic medical record texts, and pathogen concentration information. Calculate the infection risk levels of each region in the three-dimensional space through a graph attention network, generate a dynamic risk heat map. Based on the improved multi-objective particle swarm optimization algorithm, with the goals of maximizing the disinfection coverage rate and minimizing the energy consumption, simulate the operation path of the disinfection robot in the digital twin environment, and eliminate the multi-robot path conflicts through the conflict detection tree algorithm, generating a dynamic risk heat map and an optimal disinfection task queue;

[0096] S3: Based on the dynamic risk heat map and the optimal disinfection task queue, adopt the federated simulation technology and the blockchain encryption protocol to build a cross-hospital digital twin collaborative platform. Achieve multi-institution simulation clock synchronization through the high-level architecture, simulate the cross-hospital transmission path of infectious diseases. At the same time, use the auction algorithm to dynamically allocate scarce resources such as ICU beds and ventilators, and ensure policy coordination and data security through the on-chain execution technology of smart contracts, generating a regional joint prevention and control strategy.

[0097] Generating the multi-source fusion dataset and the dynamic low-risk path set in S1 includes the following steps;

[0098] S101: Based on the hospital's three-dimensional BIM model, adopt the multi-source sensor fusion algorithm to integrate the real-time data of air microbial samplers, temperature and humidity sensors, and UWB positioning devices. Eliminate noise interference through the Kalman filter algorithm to generate preprocessed sensor data;

[0099] The air microbial sampler uses the centrifugal aerosol enrichment technology with a capture efficiency of 99.9%. Combined with the dew point temperature compensation algorithm of the temperature and humidity sensor, it eliminates the misdetection of pathogen concentration in high-humidity environments. The Kalman filter performs time series alignment on multi-source data through the state space model to eliminate the multipath interference of UWB signals.

[0100] S102: Based on the preprocessed sensor data, adopt the ultra-wideband positioning technology to real-time track the personnel trajectories, and construct a three-dimensional personnel contact network through the spatio-temporal trajectory clustering algorithm to generate a personnel dynamic trajectory map;

[0101] The spatio-temporal trajectory clustering algorithm introduces the dynamic time warping to measure the trajectory similarity, identifies the high-frequency round-trip paths of medical staff (such as from the pharmacy to the ward), and constructs an undirected weighted graph of the contact network. The edge weight is the number of contacts, and the node attributes include the personnel roles (doctors, patients).

[0102] S103: Based on the personnel dynamic trajectory map, adopt the deep reinforcement learning algorithm to generate the dynamic paths of medical staff in the Unity3D simulation environment, and iteratively optimize with the goal of minimizing the infection risk value to generate the multi-source fusion dataset and the dynamic low-risk path set.

[0103] Deep reinforcement learning simulated the coordinated movement of 200 medical staff in Unity3D, accelerated training through the asynchronous advantage actor-critic (A3C) framework, and converged to a stable strategy after 100,000 iterations, reducing the path conflict rate from 15% to 3%.

[0104] Generating a dynamic risk heat map and optimal disinfection task queue based on S2 includes the following steps:

[0105] S201: Based on a multi-source fusion dataset and a dynamic low-risk path set, the YOLOv7 object detection algorithm is used to analyze surveillance videos, identify high-risk behaviors such as not wearing masks and crowded crowds, and generate a visual risk feature matrix;

[0106] The YOLOv7 model is deployed on an NVIDIA Jetson edge computing device, accelerating inference at 30 FPS using TensorRT. Violation detection results are fused with UWB positioning data to mark the 3D coordinates of the violation (e.g., "Not wearing a mask at 12:00 PM in the corridor on the third floor").

[0107] S202: Based on the visual risk feature matrix, the BioBERT biomedical pre-trained model is used to parse electronic medical record text, extract infection keywords, construct a semantic association network, and generate a multimodal risk association map;

[0108] The BioBERT model was fine-tuned on the MIMI C-Ⅲ medical dataset. Through the bidirectional Transformer encoder, it extracted the logical chain of "fever → positive blood culture → drug-resistant bacterial infection" and constructed a causal reasoning network for infection transmission.

[0109] S203: Based on the multimodal risk association map, an improved multi-objective particle swarm optimization algorithm is used to simulate the disinfection robot's operation path, and the conflict detection tree algorithm is used to eliminate multi-robot trajectory conflicts to generate a dynamic risk heat map and the optimal disinfection task queue.

[0110] The improved particle swarm optimization introduces the Pareto front screening mechanism to balance disinfection coverage (Goal 1) and energy consumption (Goal 2), generate a non-dominated solution set for decision makers to choose, and support dynamic adjustment of preference weights (such as focusing on coverage during an epidemic outbreak).

[0111] Generating a regional joint prevention and control strategy based on S3 includes the following steps:

[0112] S301: Based on dynamic risk heat maps and optimal disinfection task queues, a federated learning framework is used to synchronize pathogen transmission parameters across multiple hospital campuses, build a cross-institutional joint simulation environment, and generate a federated simulation dataset.

[0113] Federated learning uses the FedAvg algorithm to aggregate model parameters. L2 regularization is introduced during local training in each hospital to prevent overfitting. The central server improves model robustness through elastic averaging, and the deduction results are visualized to display the cross-hospital transmission chain.

[0114] S302: Based on the federated simulation dataset, an auction algorithm is used to dynamically allocate ICU beds, ventilators, and other resources, prioritizing the needs of high-risk areas and generating a cross-campus resource scheduling plan.

[0115] The auction algorithm designs a secondary price sealed bidding rule. Hospital nodes submit resource requirements and priority scores, and the smart contract automatically calculates the optimal allocation plan to prevent malicious bidding and ensure resource supply in high-risk areas.

[0116] S303: Based on the cross-campus resource scheduling plan, blockchain smart contract technology is used to encrypt policy instructions to ensure the tamper-proofness of multi-institutional collaborative prevention and control, and generate a regional joint prevention and control strategy.

[0117] The blockchain nodes adopt the PBFT consensus mechanism, which can tolerate 1 / 3 Byzantine node failures. The smart contract verifies data integrity through the Merkle tree to ensure the tamper-proof and traceability of the joint prevention and control strategy during transmission and execution.

[0118] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional simulation and infection scenario decision optimization system based on digital twins, characterized in that, It includes the following modules: data acquisition module, path optimization module, risk analysis module, disinfection scheduling module, and collaborative prevention and control module; The data acquisition module, based on the hospital's three-dimensional building information model, adopts a multi-source sensor fusion algorithm. By deploying air microbial samplers, ultra-wideband positioning devices, and temperature and humidity sensors, it can collect pathogen concentrations, personnel trajectories, and environmental parameters in real time. It also uses a spatio-temporal attention mechanism to perform weighted fusion and noise filtering on heterogeneous data, generating a multi-source fusion dataset; The data acquisition module includes a sensor data sub-module, a positioning and tracking sub-module, and a pathogen monitoring sub-module; The path optimization module, based on the multi-source fusion dataset, adopts a deep reinforcement learning algorithm to construct a multi-agent simulation environment for the hospital's three-dimensional scene in the Unity3D engine. It defines an infection risk function for the movement paths of medical staff, and iteratively optimizes the path planning scheme through a reward mechanism to generate a set of dynamic low-risk paths; The path optimization module includes a path planning sub-module and a conflict resolution sub-module; The risk analysis module, based on the set of dynamic low-risk paths and real-time sensor data, adopts a multi-modal Transformer model to fuse visual monitoring data and pathogen concentration information in electronic medical record texts. It calculates the infection risk levels of each area in the three-dimensional space through a graph attention network, generating a dynamic risk heat map; The risk analysis module includes a visual recognition sub-module, a text mining sub-module, and a risk grading sub-module; The disinfection scheduling module, based on the dynamic risk heat map, adopts an improved multi-objective particle swarm optimization algorithm. With the goals of maximizing disinfection coverage and minimizing energy consumption, it simulates the operation paths of disinfection robots in a digital twin simulation environment, and uses a conflict detection tree algorithm to avoid multi-robot trajectory intersections, dynamically adjusting the task queue to generate an optimal disinfection task queue; The disinfection scheduling module includes a task allocation sub-module and a path conflict detection sub-module; The collaborative prevention and control module, based on the optimal disinfection task queue and cross-hospital data, adopts federated simulation technology and blockchain encryption transmission protocol to construct a distributed digital twin collaborative platform. It realizes multi-hospital simulation clock synchronization through a high-level architecture, and uses an auction algorithm to dynamically allocate scarce resources such as ICU beds and ventilators, generating a regional joint prevention and control strategy; The collaborative prevention and control module includes a federated simulation sub-module, a resource scheduling sub-module, and a blockchain encryption sub-module.

2. The three-dimensional simulation and infection scenario decision optimization system and method based on digital twin according to claim 1, wherein: The sensor data sub-module, based on the hospital's three-dimensional BIM model, adopts a multi-source sensor fusion algorithm to integrate the real-time data of air microbial samplers and temperature and humidity sensors, and eliminates noise interference through Kalman filtering to generate preprocessed sensor data; The positioning and tracking sub-module, based on the preprocessed sensor data, adopts ultra-wideband positioning technology to track the movement trajectories of medical staff and patients in real time, and constructs a spatio-temporal trajectory map to generate a personnel trajectory dataset; The pathogen monitoring sub-module, based on the personnel trajectory dataset, adopts a pathogen concentration dynamic interpolation algorithm, combines air sampling data with ventilation system parameters, calculates the pathogen distribution probability in the three-dimensional space, and generates a multi-source fusion dataset.

3. The three-dimensional simulation and infection scenario decision optimization system and method based on digital twin according to claim 1, characterized in that: The path planning sub-module, based on the multi-source fusion dataset, uses the deep reinforcement learning algorithm to simulate the movement path of medical staff in the Unity3D simulation environment, and conducts iterative training with the minimum infection risk as the objective function to generate a preliminary path plan set; The conflict resolution sub-module, based on the preliminary path plan set, uses the spatio-temporal conflict detection tree algorithm to analyze multi-path intersections and congestion areas, dynamically adjusts the path priorities, and generates a dynamic low-risk path set.

4. The three-dimensional simulation and infection scenario decision optimization system and method based on digital twin according to claim 1, wherein: The visual recognition sub-module, based on the dynamic low-risk path set, uses the YOLOv7 object detection algorithm to real-time recognize high-risk behaviors such as not wearing masks and crowded people in the surveillance video, and generates visual risk feature data; The text mining sub-module, based on the visual risk feature data, uses the BioBERT biomedical pre-trained model to parse the infection keywords in the electronic medical records, constructs a text semantic association network, and generates text risk feature data; The risk grading sub-module, based on the text risk feature data, uses the graph attention network to fuse multi-modal features such as pathogen concentration and personnel contact frequency, calculates the three-dimensional grid risk level, and generates a dynamic risk heat map.

5. The three-dimensional simulation and infection scenario decision optimization system and method based on digital twin according to claim 1, characterized in that: The task allocation sub-module, based on the dynamic risk heat map, uses the improved multi-objective particle swarm optimization algorithm, with the goals of disinfection coverage rate and energy consumption balance, generates a robot task allocation plan, and generates a preliminary queue of disinfection tasks; The path conflict detection sub-module, based on the preliminary queue of disinfection tasks, uses the conflict detection tree algorithm to simulate the collaborative operation paths of multiple robots, eliminates trajectory intersections and resource preemption conflicts, and generates an optimal disinfection task queue.

6. The three-dimensional simulation and infection scenario decision optimization system and method based on digital twin according to claim 1, characterized in that Based on: The federated simulation sub-module, based on the optimal disinfection task queue, uses the federated learning framework to synchronize the encrypted pathogen transmission parameters of multiple hospital areas, constructs a cross-institutional digital twin joint simulation environment, and generates a federated simulation dataset; The resource scheduling sub-module, based on the federated simulation dataset, uses the auction algorithm, with the goal of maximizing the ICU bed utilization rate, dynamically allocates scarce resources such as ventilators and protective clothing, and generates a cross-hospital area resource scheduling plan; The blockchain encryption sub-module, based on the cross-hospital area resource scheduling plan, uses the on-chain execution technology of smart contracts to ensure the immutability of multi-institutional data synchronization and policy collaboration, and generates a regional joint prevention and control strategy.

7. Three-dimensional simulation and infection scenario decision optimization method based on digital twin, characterized in that, Including the following steps: S1: Based on the hospital's three-dimensional building information model, using the multi-source sensor fusion algorithm, by deploying air microbial samplers, ultra-wideband positioning devices, and temperature and humidity sensors, real-time collect pathogen concentration, personnel trajectories, and environmental parameters, and use the spatio-temporal attention mechanism to perform weighted fusion and noise filtering on heterogeneous data. At the same time, based on the deep reinforcement learning algorithm, generate the dynamic path of medical staff in the Unity3D simulation environment, avoid high-risk infection areas, and generate a multi-source fusion dataset and a dynamic low-risk path set; S2: Based on the multi-source fusion dataset and the dynamic low-risk path set, use the multi-modal Transformer model to fuse visual surveillance data, electronic medical record texts, and pathogen concentration information. Calculate the infection risk levels of each region in the three-dimensional space through the graph attention network, generate a dynamic risk heat map. Based on the improved multi-objective particle swarm optimization algorithm, with the maximization of disinfection coverage rate and the minimization of energy consumption as the goals, simulate the operation path of the disinfection robot in the digital twin environment, and eliminate the path conflicts of multiple robots through the conflict detection tree algorithm, generating a dynamic risk heat map and an optimal disinfection task queue; S3: Based on the dynamic risk heat map and the optimal disinfection task queue, use the federated simulation technology and the blockchain encryption protocol to build a cross-hospital digital twin collaborative platform. Achieve the synchronization of multi-institution simulation clocks through the high-level architecture, and simulate the cross-hospital transmission path of infectious diseases. At the same time, use the auction algorithm to dynamically allocate scarce resources such as ICU beds and ventilators, and ensure policy coordination and data security through the on-chain execution technology of smart contracts, generating a regional joint prevention and control strategy.

8. The three-dimensional simulation and infection scenario decision optimization method based on digital twin according to claim 7, wherein: Generating the multi-source fusion dataset and the dynamic low-risk path set in S1 includes the following steps; S101: Based on the hospital's three-dimensional BIM model, use the multi-source sensor fusion algorithm to integrate the real-time data of air microbial samplers, temperature and humidity sensors, and UWB positioning devices. Eliminate noise interference through the Kalman filter algorithm to generate preprocessed sensor data; S102: Based on the preprocessed sensor data, use the ultra-wideband positioning technology to track the personnel trajectories in real time, and construct a three-dimensional personnel contact network through the spatio-temporal trajectory clustering algorithm to generate a personnel dynamic trajectory map; S103: Based on the personnel dynamic trajectory map, use the deep reinforcement learning algorithm to generate the dynamic paths of medical staff in the Unity3D simulation environment, and iteratively optimize with the goal of minimizing the infection risk value to generate a multi-source fusion dataset and a dynamic low-risk path set.

9. The three-dimensional simulation and infection scenario decision optimization method based on digital twin according to claim 7, wherein: Generating the dynamic risk heat map and the optimal disinfection task queue in S2 includes the following steps; S201: Based on the multi-source fusion dataset and the dynamic low-risk path set, use the YOLOv7 object detection algorithm to analyze the surveillance video, identify high-risk behaviors such as not wearing masks and crowded people, and generate a visual risk feature matrix; S202: Based on the visual risk feature matrix, use the BioBERT biomedical pre-training model to analyze the electronic medical record texts, extract infection keywords, construct a semantic association network, and generate a multi-modal risk association map; S203: Based on the multi-modal risk association map, use the improved multi-objective particle swarm optimization algorithm to simulate the operation path of the disinfection robot, and eliminate the trajectory conflicts of multiple robots through the conflict detection tree algorithm, generating a dynamic risk heat map and an optimal disinfection task queue.

10. The three-dimensional simulation and infection scenario decision optimization method based on digital twin according to claim 7, wherein: Generating the regional joint prevention and control strategy in S3 includes the following steps; S301: Based on the dynamic risk heat map and the optimal disinfection task queue, use the federated learning framework to synchronize the pathogen transmission parameters of multiple hospitals, construct a cross-institution joint simulation environment, and generate a federated simulation dataset; S302: Based on the federal simulation dataset, use the auction algorithm to dynamically allocate resources such as ICU beds and ventilators, prioritize ensuring the needs of high-risk areas, and generate a cross-hospital resource scheduling plan; S303: Based on the cross-hospital resource scheduling plan, use blockchain smart contract technology to encrypt policy instructions to ensure the immutability of multi-institutional collaborative prevention and control, and generate a regional joint prevention and control strategy.

Images