A method for updating a post-disaster road recovery graph network based on fusion of heterogeneous disaster situation perception data

Abstract

The application relates to the technical field of emergency management, and discloses a post-disaster road recovery graph network updating method fusing heterogeneous disaster situation sensing data, which comprises the following steps: converting the obtained heterogeneous data into basic probability distribution of observation evidence; predicting the physical field state based on a physical process simulator to generate basic probability distribution of prediction evidence; fusing the two to obtain posterior basic probability distribution; deciding and updating the road graph network state according to the posterior basic probability distribution; and feeding back the updated state to the simulator as a dynamic boundary condition to form bidirectional coupling of physical simulation and road network state. Through the construction of the bidirectional coupling simulation assimilation framework, the technical problem of the separation of physical simulation and infrastructure state in the traditional method is solved, the uncertainty of data is effectively processed by using the evidence theory, and the authenticity of disaster situation evolution prediction and the reliability of road state updating are improved.

Description

Technical Field

[0001] This invention relates to the field of emergency management technology, specifically to a method for updating a post-disaster road recovery map network by integrating heterogeneous disaster perception data. Background Technology

[0002] Following a disaster, timely and accurate access to road network conditions is crucial for emergency response command, resource allocation, and ensuring the safety of vital transportation routes. Currently, updating post-disaster road networks primarily relies on various information acquisition methods, including remote sensing image interpretation, drone aerial photography analysis, and on-site personnel reporting.

[0003] However, these disaster perception data are inherently multi-source, heterogeneous, and vary in spatiotemporal scales, and often contain uncertainties and even conflicts. Existing data fusion methods often face challenges in processing such uncertain and conflicting information, making it difficult to form a unified and reliable situational awareness, which in turn affects the accuracy of road network status decisions.

[0004] Furthermore, to predict the development of disasters, some existing technologies incorporate physical process models such as floods and landslides for simulation. However, these simulations are typically unidirectional, treating the road network as a static disaster-bearing entity. This approach ignores the dynamic reactions of road infrastructure (such as roadbeds and bridges) to the disaster evolution process. For example, roadbeds may impede flood diffusion, while bridge collapses may trigger upstream backflow, leading to discrepancies between simulation results and the real physical world.

[0005] Furthermore, when faced with a high degree of uncertainty in situational awareness due to the lack of perception data in key areas, most existing technical solutions lack proactive cognition and planning capabilities. They are unable to autonomously identify the areas with the highest information value, nor can they provide optimal mission planning suggestions for limited reconnaissance resources (such as drones), demonstrating a passive dependence on data input.

[0006] Therefore, this invention proposes a method for updating a post-disaster road recovery map network by integrating heterogeneous disaster perception data, in order to address the shortcomings of existing technologies. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method for updating post-disaster road recovery maps by fusing heterogeneous disaster perception data. This method solves the problems of insufficient ability to fuse multi-source heterogeneous data, failure to fully reflect the dynamic interaction between infrastructure and the disaster environment, and lack of proactive cognition and planning capabilities when information is incomplete.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] The first aspect of this invention provides a method for updating a post-disaster road recovery map network by fusing heterogeneous disaster sentiment data. This method constructs a bidirectional coupled cyclic framework consisting of physical process simulation and uncertainty data assimilation. The method first acquires heterogeneous disaster sentiment data from different sources and modalities, and then transforms these raw data into a unified mathematical framework based on Dempster-Shafer evidence theory, forming structured observational evidence. This transformation process represents each independent observation as a basic probability allocation function (BPA), denoted as m, on a predefined judgment framework Θ. This function satisfies the following conditions:

[0010]

[0011] Where: A is a subset of the judgment frame Θ. m(A) represents the degree of confidence that the evidence precisely supports the proposition that the true state belongs to set A; The trust level of the empty set;

[0012] Meanwhile, this method is based on one or more physical process simulators. Based on the state of the physical state grid and road map network at the previous moment, it extrapolates the physical field state at the current moment. The prediction result is also transformed into a basic probability allocation function of prediction evidence.

[0013] Subsequently, the method employs the Dempster combination rule to fuse the observed evidence and the predicted evidence to obtain a posterior basic probability assignment function that incorporates multi-source information and quantifies uncertainty. The calculation process of the Dempster combination rule is as follows:

[0014]

[0015] Where: A, B, and C are all subsets of the judgment frame Θ; m pred (B) represents the confidence level of the predicted evidence supporting the proposition's true state as belonging to B; m obs (C) represents the confidence level that the observed evidence supports the proposition's true state belonging to C; K is the conflict coefficient, used to quantify the degree of contradiction between two sources of evidence, and its calculation formula is:

[0016]

[0017] After obtaining the posterior basic probability assignment function, to arrive at a definite state conclusion, this method further calculates the confidence level Bel(H) and likelihood Pl(H) of the single hypothesis H in the decision framework, and makes the final state decision according to the preset decision rules. The calculation process of confidence level and likelihood is as follows:

[0018]

[0019] Wherein: H k It is a single hypothesis in the judgment frame Θ; m post B is the posterior basic probability assignment function to be decided; B is a subset of the decision frame Θ.

[0020] This method updates the final state determined by the decision to the dynamic attributes of the road map network. The updated road map network state will be used as a dynamic boundary condition and fed back to the physical process simulator to constrain and correct the physical process prediction of the next time step. This mechanism constitutes a two-way coupling between the physical environment simulation and the state of the engineering facilities, enabling the simulation process to accurately reflect the dynamic interaction between the two.

[0021]

[0022] Where: a represents a specific perceptual action; U(m) current ) represents the current system uncertainty before taking action a; z represents the possible observation result after taking action a; p(z|a) is the probability distribution of the observation result z under the condition of taking action a, which can be estimated based on the current cognitive state of the system; U(m post (z) represents the posterior uncertainty of the system state after a new observation z is obtained; This represents the expectation of all possible observations z.

[0023] In another specific implementation, the method also possesses robust self-evaluation capabilities. This is achieved through a digital copy S of the current disaster situation. real In (t), a virtual disaster event V is injected. event To generate an adversarial simulation scenario S sim :

[0024]

[0025] Wherein: S sim It is a generated high-risk adversarial simulation scenario; S real (t) is a digital copy of the actual physical state at the current time t; V event It is a parameterized virtual disaster event; The symbol represents a superposition operator of physical effects.

[0026] The system performs a complete simulation under this adversarial scenario and quantitatively evaluates the road network performance based on the system robustness metric formula R. A specific calculation method for R is as follows:

[0027]

[0028] Where: R is the system robustness measure; Vcritical It is a predefined set of key nodes; d real (v i ,v j (This refers to the critical node v before the virtual disaster injection.) i and v j The shortest travel distance or time between them on the road network; d sim (v i ,v j ) is the key node v after the adversarial scenario simulation. i and v j The shortest travel distance or time between two points on the damaged road network; if the two points are no longer connected, this value is infinity.

[0029] A second aspect of the present invention provides a post-disaster road recovery map network update system that integrates heterogeneous disaster sentiment data, the system comprising:

[0030] The system comprises a data interface and preprocessing unit, a core processing unit, and an application and publishing unit. The data interface and preprocessing unit performs the function of acquiring heterogeneous disaster sentiment data and transforming it into a basic probability allocation function for observational evidence, as described in the aforementioned method. The core processing unit is the core computational carrier for executing the method, performing physical process prediction, evidence fusion, and achieving bidirectional coupling feedback. The application and publishing unit performs state decision-making, updates the dynamic attributes of the roadmap network, and publishes the results externally. In one specific implementation, the core processing unit is also equipped with an active perception and causal tracing module and a system robustness self-evaluation module to realize the active perception and robustness self-evaluation functions described in the aforementioned method.

[0031] This invention provides a method for updating a post-disaster road recovery map network by integrating heterogeneous disaster perception data.

[0032] It has the following beneficial effects:

[0033] 1. This invention establishes a bidirectional coupling between the physical simulation and the road network state by using the updated road network state as a dynamic boundary condition and inputting it back into the physical process simulator. This overcomes the shortcomings of traditional one-way simulations in failing to reflect the dynamic impact of infrastructure damage on the disaster process, such as roadbed obstruction or bridge collapse altering water flow, enabling the digital twin model to more realistically reflect the complex interaction between disasters and engineering facilities.

[0034] 2. This invention adopts the Dempster-Shafer evidence theory to unify raw data from different sources and modalities into a basic probability allocation function that can be mathematically calculated. This framework can explicitly quantify and process the uncertainty and conflict in the information. Compared with traditional fusion methods, this invention can obtain more robust and reliable fusion results when dealing with incomplete information or highly conflicting evidence, thereby improving the accuracy of the final road map network state decision.

[0035] 3. This invention introduces a proactive sensing and causal tracing mechanism. When the system's perception of the state of key areas is highly uncertain, it can automatically trace back to the root cause of the uncertainty and, based on information value assessment, quantitatively calculate the information gain that different sensing tasks can bring to the system. This transforms the system from a passive data receiver into an intelligent agent capable of proactively planning information acquisition, thereby prioritizing the deployment of limited sensing resources to the most valuable areas and significantly improving the efficiency and targeting of emergency response. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0037] Figure 2 This is a block diagram of the system hardware implementation of the present invention;

[0038] Figure 3 This is a schematic diagram illustrating the coupling between the road graph network and the multi-layer physical state grid of the present invention;

[0039] Figure 4 This is a flowchart illustrating the transformation of heterogeneous data into physical evidence according to the present invention.

[0040] The system comprises: 10. Data interface and preprocessing unit; 20. Coupled model construction unit; 30. Core processing unit; 31. Physical evidence conversion module; 32. Two-way coupled simulation assimilation module; 33. Active perception and causal tracing module; 34. System robustness self-evaluation module; and 40. Application and release unit. Detailed Implementation

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] See attached document Figure 1 , Figure 1This is a schematic diagram of the overall system architecture according to an embodiment of the present invention. The present invention provides a network update system for post-disaster road recovery maps that integrates heterogeneous disaster information perception data. At the hardware level, the system can be deployed on one or more servers, and the servers include conventional computing devices such as processors, memory, and communication interfaces.

[0043] The system of the present invention logically includes: a data interface and preprocessing unit 10, a coupling model construction unit 20, a core processing unit 30, and an application and publishing unit 40.

[0044] The data interface and preprocessing unit 10 are configured to acquire disaster perception data from multiple channels, including but not limited to: satellite remote sensing images, drone aerial photography data, video streams collected by ground vehicles, official announcements issued by traffic management departments, and public observation information reported through social media or dedicated apps.

[0045] Coupled model building unit 20 is configured to establish a basic digital twin model of the disaster area, as shown in the appendix. Figure 3 This unit constructs a road map network model G and constructs the geographic and physical attributes of the region into a multi-layer physical state grid P. This unit further establishes a bidirectional spatial mapping relationship between the road map network model G and the multi-layer physical state grid P, so that the geometric entities of the road network can be accurately associated with the physical environment state they carry.

[0046] The core processing unit 30 is the computing center for executing the core method of the present invention. This unit specifically includes: a physical evidence conversion module 31, a bidirectional coupled simulation assimilation module 32, an active perception and causal tracing module 33, and a system robustness self-evaluation module 34.

[0047] The physical evidence conversion module 31 connects the data interface and the preprocessing unit 10. Its function is to convert the received raw data of different modalities into a unified mathematical expression based on the Dempster-Shafer evidence theory, namely the basic probability allocation function. The specific calculation of this function will be described in detail in subsequent chapters.

[0048] The bidirectionally coupled simulation assimilation module 32 is the main loop engine of the system. On the one hand, this module drives the disaster model (such as the flood evolution model) in the physical state grid P to predict the state of the next time step; on the other hand, it fuses the observational evidence generated by the physical evidence transformation module 31 with the predictive evidence of the model. This fusion process is calculated using the Dempster combination rule formula. A key technology is that this module also uses the updated road map network state as a dynamic boundary condition, which is then input back into the physical model, thereby realizing the dynamic influence of road facilities on the disaster evolution process.

[0049] The active perception and causal tracing module 33 is used to realize the cognitive function of the system. When there is high uncertainty or high conflict in the output of the bidirectional coupled simulation assimilation module 32, this module is activated. Using the causal relationship built into the physical model, it traces back to the root of uncertainty and calculates the information gain that different perception tasks (such as instructing the UAV to fly to a specific area) can bring to the system according to the information value assessment formula, and finally generates a priority list of perception tasks.

[0050] The system robustness self-assessment module 34 is used to stress test the system itself and existing emergency plans. This module generates an adversarial simulation scenario by injecting a virtual, higher-intensity disaster event into the current real disaster digital twin. Subsequently, the module drives the entire core processing unit 30 to perform simulations under this extreme scenario and quantitatively assesses the vulnerability of the road network and the cognitive shortcomings of the system according to the system robustness measurement formula.

[0051] The application and release unit 40 is connected to the core processing unit 30. Its function is to receive the final processing results and transform them into application products for different users. First, the unit makes a final state decision on the fused uncertainty information according to the confidence and likelihood calculation formulas, and generates an updated road map network. Then, the unit releases the updated graph network, active perception task suggestions, and system robustness assessment report to the emergency command center, rescue teams, or the general public through a visualization interface or data interface.

[0052] See attached document Figure 2 , Figure 2 This is a system hardware implementation block diagram according to an embodiment of the present invention. The system provided by the present invention can be deployed on one or more computing devices at the hardware level to form a centralized or distributed computing environment. The computing device can be a physical server, a virtual machine, or a containerized instance on a cloud platform.

[0053] S211. In a specific embodiment, the hardware deployment of the system includes a central processing server, at least one data acquisition terminal, and at least one user interaction terminal, which communicate with each other via a network, which may be the Internet, a local area network, a mobile communication network (e.g., 5G), or a combination thereof.

[0054] S212. The central processing server is the physical carrier that executes the core algorithm of this invention. Specifically, it includes:

[0055] One or more processing units are included. Each processing unit is a core component for performing numerical calculations and logic control. In one embodiment, the processing unit can be implemented by a general-purpose central processing unit (CPU) and a graphics processing unit (GPU) working together. The CPU is mainly responsible for performing overall system logic control, data scheduling, and serial computing tasks. The GPU, with its parallel computing capabilities, is mainly responsible for performing large-scale numerical calculations in physical process simulations and inference tasks for machine learning models such as semantic segmentation of remote sensing images. In another embodiment, to pursue ultimate computing performance, the processing unit can also be implemented by an application-specific integrated circuit (ASIC) or a field-programmable gate array (FPGA), embedding the core algorithm in the hardware logic. The above computing hardware can be used alone or in combination.

[0056] The storage unit is used to store program instructions and computational data, and may specifically include high-speed random access memory (RAM) and non-volatile memory. The RAM is used to store the operating system, application programs, and intermediate data that needs to be accessed at high speed by the processing unit during system operation, such as the dynamic attributes of the road map network model G and the real-time state of the physical state grid P at the current moment. The non-volatile memory, such as solid-state drive (SSD) or hard disk drive (HDD), is used for long-term storage of the program code of this invention, basic geographic information data, digital elevation model (DEM), pre-trained machine learning model, and historical disaster data.

[0057] Communication interfaces, such as Ethernet cards or wireless network interfaces, are used to enable data exchange between the central processing server and data acquisition terminals and user interaction terminals.

[0058] S213. The data acquisition terminal is the data source of the system. This terminal is a general collection of functions of various physical devices. Specifically, the data acquisition terminal can be a remote sensing satellite in orbit, a drone performing aerial photography missions, a ground patrol vehicle equipped with a camera, a fixed sensor deployed in a key location (such as a water level gauge or traffic camera), or a mobile smart terminal (such as a smartphone) with a specific application installed. These terminals acquire raw disaster data through their own sensors and transmit the data to the central processing server through the network.

[0059] S214. The user interaction terminal is the interface for displaying and interacting with the system's results. This terminal can be a large-screen display system deployed in the emergency command center, a desktop workstation used by emergency management personnel, or a tablet computer or mobile phone held by on-site rescue personnel. Users can use this terminal to view the real-time updated road map network status, receive proactive perception task suggestions generated by the system, and analyze the robustness assessment report output by the system. They can also issue instructions to the system or perform manual calibration.

[0060] See attached document Figure 3 , Figure 3 This is a schematic diagram illustrating the coupling of a road map network and a multi-layer physical state mesh according to an embodiment of the present invention. The basic model construction in this invention aims to provide an accurate and information-complete digital twin foundation for subsequent physical simulations and data assimilation.

[0061] S221. Construct a road network model. This process begins with acquiring basic road network data for the target area. The data source can be publicly available geographic information data (such as OpenStreetMap), GIS data provided by government surveying and mapping departments, or commercial navigation map data. The system will perform topology verification and processing on the acquired raw road network data and formalize it into an attribute graph network G = (V, E).

[0062] S222. In this graph network G, V represents the set of nodes, where each node v can be a road intersection, bridge, tunnel entrance, or other key point with specific engineering significance; E represents the set of edges, where each edge e represents a specific road segment connecting two nodes.

[0063] S223. Assign a multi-dimensional attribute set to each road segment e and key node v. This attribute set is divided into static attributes and dynamic attributes. Static attributes are their inherent physical characteristics that do not change during the disaster process. Specifically, they include: a unique identifier ID, administrative level, number of lanes, design capacity, road surface material, and a three-dimensional spatial coordinate sequence containing longitude, latitude, and elevation. Dynamic attributes are the states that the system needs to evaluate and update in real time. Their initial values ​​are set to default or unknown states. Specifically, they include: traffic status (e.g., smooth, congested, interrupted), physical causes of disaster (e.g., flooding, landslide, structural damage), and a measure of uncertainty in the current state assessment.

[0064] S224. Construct and represent a multi-layer physical state grid. The system creates a discretized, regular or irregular geographic grid P in the same geographic coordinate system as the road map network G; this grid covers the entire target area, with each grid cell p... j It corresponds to a specific geographical block in space; the grid is a multi-layered data structure, with each layer representing a physical attribute or disaster-related field variable.

[0065] S225, the specific hierarchy of the physical state grid P may include:

[0066] Topography layer: stores each grid cell p j The average elevation, slope, aspect, and other information are obtained directly from a high-precision digital elevation model (DEM).

[0067] Hydrogeological layer: stores each grid cell p jInformation such as soil type, permeability coefficient, land use type, and distance from the river is used to support the simulation of the evolution of floods or waterlogging.

[0068] Infrastructure vulnerability layer: For layers containing critical infrastructure (such as bridge nodes v) k ) mesh element p j This layer stores the structural health status of the infrastructure, its design disaster resistance threshold, or vulnerability curve parameters, which are used to determine its probability of failure under disaster impact.

[0069] S226. Implement the coupling and association between the graph network and the physical mesh. To enable information exchange between the two heterogeneous models, the system must establish a bidirectional spatial mapping relationship between them. This mapping relationship is achieved by pre-computing and storing indexes.

[0070] S227. Specifically, the system performs the following operations:

[0071] Establishing the forward mapping index: The system traverses each road segment e in the road graph network G. i Road segment e is determined through spatial geometric intersection operations. i The set of all physical mesh elements traversed by the three-dimensional geometric line {p j The system will assign this correspondence (e) i →{p j The forward mapping stores the road segment's attributes (e.g., roadbed height) in a lookup table or spatial index (such as an R-tree). This forward mapping allows the attributes of the road segment (e.g., roadbed height) to be quickly passed to its physical environment, thereby affecting the calculation of the physical model.

[0072] Building the reverse mapping index: The system traverses each physical grid cell p j Similarly, through spatial querying, the set {e} of all road segments passing through this unit is determined. i}; The system will assign this correspondence (p j →{e i}) is stored in the index, and this reverse mapping allows for changes in the physical environment (e.g., mesh cell p) to be stored. j An increase in water depth can be quickly transmitted to the relevant road entities to trigger updates to their dynamic properties.

[0073] See attached document Figure 4 , Figure 4 This is a flowchart illustrating the transformation of heterogeneous data into physical evidence according to an embodiment of the present invention. The physical evidence transformation module 31 in the present invention functions to standardize and quantify the uncertainty of the original disaster perception data received from the data interface and the preprocessing unit 10, which are of diverse sources and in different formats, to form a mathematical object that can be uniformly calculated by subsequent processing modules.

[0074] S231. Before data transformation, the system first establishes a judgment framework Θ for the specific state to be evaluated. The judgment framework Θ is a set of N mutually exclusive and complete hypotheses, Θ = {H1, H2, ..., H...}. n For example, when assessing the traffic status of a certain road segment e, a specific judgment framework can be set as: Θ traffic ={Smooth, Congested, Disrupted}; For the structural status of critical infrastructure nodes such as bridges, another judgment framework can be set: Θ structure = {intact, damaged, collapsed}.

[0075] S232. After establishing the judgment framework, the system transforms each independent raw observation data into a basic probability assignment function (BPA), also known as the Mass function, denoted as m(·), on the judgment framework Θ. This function assigns a confidence value between 0 and 1 to each subset A in the power set 2^Θ of the judgment framework Θ. This function must satisfy the following condition:

[0076]

[0077] Where: A is a subset of the judgment frame Θ. m(A) represents the degree of confidence that the evidence precisely supports the proposition that the true state belongs to set A; The confidence level is an empty set, and its value is always 0; in particular, m(Θ) represents the uncertainty or ignorance in the evidence that is not assigned to any particular hypothesis, and it represents a confidence measure that makes it impossible to make a more accurate judgment due to insufficient information.

[0078] S233. Specific implementation methods for generating the BPA function for input data of different modalities include:

[0079] For data derived from remote sensing imagery or UAV video: the system first employs a pre-trained deep learning semantic segmentation model (e.g., U-Net or its variants) to segment the image corresponding to a road segment e or physical grid cell p. j The system performs pixel-level classification of the region, and outputs the probability that each pixel belongs to the category of water, road, landslide deposits, etc. The system aggregates the pixel classification probabilities in the region. If the average probability of water is 0.9, a BPA can be generated: m({interruption}) = 0.9 × c, where c is the confidence coefficient related to the sensor type and imaging conditions, and the remaining confidence 1 - 0.9 × c is assigned to m(Θ), i.e., m({smooth, congested, interrupted}), which represents the uncertainty in the model and observation process.

[0080] For data derived from text information: The system uses a natural language processing model to perform entity recognition and relation extraction on the text, extracting key information such as the location, type, and extent of the disaster. The system maintains a source reliability database and assigns a reliability value r based on the information source (e.g., official announcements, certified users, anonymous users). If an official announcement clearly states that a certain bridge is interrupted, a BPA can be generated: m({interruption}) = r, and m(Θ) = 1 - r, where r can take a value close to 1 for official announcements, such as 0.95.

[0081] For data originating from physical sensors (such as water level gauges): the system compares the sensor readings with preset physical thresholds. For example, if a disaster-causing water depth threshold of 0.5 meters is set for a certain road segment e, and the water level gauge reading near that road segment is 0.8 meters, the system generates a BPA based on the sensor's accuracy indicators; for example, for a high-precision sensor, a BPA can be generated as follows: m({interruption}) = 0.99, m(Θ) = 0.01.

[0082] Through the above steps, all the messy, multimodal input data is unified under the framework of evidence theory, forming structured evidence with quantifiable uncertainty that can be mathematically calculated, providing standardized input for subsequent evidence fusion and state updates.

[0083] The bidirectionally coupled simulation assimilation module 32 in this invention is the core computing engine of the system. It continuously updates the system’s understanding of the state of the disaster environment through a loop that iterates at discrete time steps Δt.

[0084] S234. At the beginning of each time step t, the module first executes a physical process prediction based on dynamic boundaries. The system calls one or more physical process simulators, such as a flood evolution model based on cellular automata or a landslide dynamics model based on the finite element method. The simulator, based on the state of the physical state grid P at time t-1 and the updated state of the road graph network G at time t-1, extrapolates and predicts the physical field state at time t. This prediction result is then transformed into a basic probability assignment function (BPA), denoted as m. pred This serves as prior verification evidence generated within the system.

[0085] S235. A key technical feature of this process is the implementation of a two-way coupling mechanism. Specifically, the properties of the road map network G at time t-1 are used as dynamic boundary conditions or model parameters of the physical simulator. In one embodiment, if the state of a road segment e is updated to have an intact roadbed and its elevation is higher than the surrounding terrain, then in the flood evolution model, the grid cell corresponding to that road segment will be assigned an extremely low permeability or regarded as a temporary dam, thereby hindering or guiding the water flow. In another embodiment, if the state of a bridge node v is updated to structural collapse, then in the hydrodynamic model, the grid cell corresponding to that node will be set as an obstacle, which may lead to upstream backflow and changes in downstream flow velocity. This reverse guidance enables the simulation process to reflect the dynamic interaction between human engineering facilities and natural disasters.

[0086] S236. The module performs data assimilation based on evidence theory. This step aims to mathematically fuse external, real-time observational evidence with internal physical prediction evidence. The system first aggregates all new observational evidence (BPA) received from the physical evidence transformation module 31 within time step t, denoted as m. obs Subsequently, the Dempster combination rule was used to predict the evidence m. pred With comprehensive observational evidence m obs The fusion is performed to calculate the posterior BPA at time t, denoted as m. post The formula for calculating Dempster's combination rule is:

[0087]

[0088] Where: A, B, and C are all subsets of the judgment frame Θ; m pred (B) represents the confidence level of the predicted evidence supporting the proposition's true state as belonging to B; m obs (C) represents the confidence level that the observed evidence supports the proposition's true state belonging to C; K is the conflict coefficient, used to quantify the degree of contradiction between two sources of evidence, and its calculation formula is:

[0089]

[0090] A K value close to 1 indicates a serious conflict between prediction and observation, which can serve as a signal to trigger anomaly detection or human intervention.

[0091] S237. Before performing evidence fusion, the module may selectively perform a physical simulation-based cross-modal self-consistency verification step to enhance the robustness of the system when a new observational evidence m is received. obsEspecially when there is a high degree of conflict between the observed evidence and the existing state or other evidence, the system does not immediately integrate it. The system takes the physical phenomenon described by the evidence (e.g., a landslide at point A) as a hypothesis and inputs it into the physics simulator. The simulator verifies whether the phenomenon is consistent with the currently known upstream conditions and physical laws. For example, for a report of a landslide at point A, the system checks the geological parameters, slope, and current soil moisture content of the grid cell where point A is located, and simulates the probability of a landslide occurring under these conditions. If the simulation results show that the landslide is highly probable, the observed evidence is considered consistent with the physical world and participates in the integration with a high degree of trust. Conversely, if the simulation results show that a landslide is almost impossible under the current physical conditions, the observed evidence is considered anomalous or erroneous, and the system will reduce its weight in the integration or isolate it and issue an alert to the user. This verification step uses the physical model as the highest arbiter, effectively filtering out erroneous information and false reports, and avoiding contamination of the overall system state perception.

[0092] The active perception and causal tracing module 33 in this invention enables the system to have cognitive ability, and when faced with uncertainty, it can autonomously and purposefully plan information acquisition tasks, thereby maximizing the utility of limited perception resources (such as drones and patrol personnel).

[0093] S238. The operation of this module is based on the quantification of system uncertainty. The bidirectionally coupled simulation assimilation module 32 outputs the posterior basic probability assignment function (BPA)m. post Then, the module first calculates the uncertainty of the state for each critical road segment e or region p; in one embodiment, this uncertainty U(m) can be quantified by a formula based on Hartley entropy:

[0094]

[0095] Where: m is the basic probability assignment function to be evaluated, for example m post A is a subset of the judgment frame Θ; m(A) is the confidence assignment of BPA on subset A; |A| is the number of basic assumptions contained in subset A.

[0096] The larger the calculated result of this formula, the more ambiguous the system's understanding of the state. When the calculated uncertainty value U(m) exceeds a preset threshold τ... u If the current information is insufficient to make a reliable judgment, the system will automatically trigger the subsequent causal tracing and perception planning process.

[0097] S239. The module executes a causal chain reverse tracing algorithm. This algorithm aims to answer the question, "What caused the high uncertainty of the current node?" Utilizing the causal logic inherent in the physical process simulator itself, it traces back from the downstream node (result) with high uncertainty to the upstream physical state or input data (cause) most likely to cause this uncertainty. In one specific implementation, this tracing can be achieved by calculating the sensitivity of the physical simulator model's state transition function. The system analyzes the partial derivatives of the posterior state with respect to each upstream input parameter or boundary condition; the upstream factor with the larger the partial derivative value is considered the key causal source. In another implementation, if the physical model can be represented as a causal graph, the parent node can be identified by executing a reverse search algorithm on the causal graph. For example, if a certain road segment e... k The interruption status is highly uncertain, and the tracing algorithm may find that the direct cause is the upstream physical grid cell p. j The water depth is uncertain, while p j The uncertainty in water depth is due to the lack of data from upstream hydrological stations.

[0098] S240. After identifying the potential causes of uncertainty (e.g., data missing from multiple upstream areas), the module further performs task generation based on information value assessment. This step aims to select the task from multiple possible sensing tasks that is expected to minimize the overall uncertainty of the system. The system generates a task for each candidate sensing action a (e.g., dispatching a drone to survey area p). j ) Calculate its information value IV(a); in one embodiment, the information value is defined as the expected reduction in the uncertainty of the system state after the action is performed, and is calculated using the following formula:

[0099]

[0100] Where: a represents a specific perceptual action; U(m) current ) represents the current system uncertainty before taking action a; z represents the possible observation result after taking action a; p(z|a) is the probability distribution of the observation result z under the condition of taking action a, which can be estimated based on the current cognitive state of the system; U(m post (z) represents the posterior uncertainty of the system state after a new observation z is obtained; This represents the expectation of all possible observations z.

[0101] S241. The system will traverse all feasible sensing actions and calculate their respective information value IV(a). Finally, the module will output a list of proactive sensing tasks sorted from high to low information value. This list will explicitly recommend which areas decision-makers or automated sensing devices (such as drone swarms) should prioritize for intelligence reconnaissance, thereby addressing the key bottlenecks in the current system's cognition in the most efficient way.

[0102] The system robustness self-assessment module 34 in this invention is designed to go beyond passive response to the current real disaster situation. By actively constructing and simulating extreme pressure scenarios, it conducts forward-looking stress tests and optimizations on the system's own cognitive capabilities and the emergency plans it serves.

[0103] S242. The operation of this module begins with the generation of an adversarial scenario. The system first creates a real physical state S at the current time t in the storage unit. real (t) is a complete digital copy. Subsequently, the system selects one or more virtual disaster events V from a pre-set high-risk event library. event The physical effects are then injected into this digital copy to generate an adversarial simulation scenario S for deduction. sim This process can be represented by the following formula:

[0104]

[0105] Wherein: S sim It is a generated high-risk adversarial simulation scenario; S real (t) is a digital copy of the actual physical state at the current time t; V event It is a parameterized virtual disaster event, whose parameters can define the type of event (such as upstream dam breach, strong aftershock, failure of critical communication base stations), location, intensity and time of occurrence; The symbol represents a superposition operator of physical effects, rather than a simple numerical addition; it signifies the superposition of virtual events V. event The resulting physical processes (such as massive floods caused by breaches) are applied as new source terms or boundary conditions to the real state S. real Above (t), thus initiating a new and more intense disaster evolution process.

[0106] S243, Generate adversarial scenarios S simThen, the module uses the scenario as the initial condition to fully drive the bidirectionally coupled simulation assimilation module 32 and active perception and causal tracing module 33 in the core processing unit 30 to perform a complete and accelerated future extrapolation. During this extrapolation, the system simulates the chain reaction of the disaster and the damage process of the road map network. It also simulates how the system's own perception planning module will deal with this extreme situation, such as what new uncertain areas it will identify and what kind of perception tasks it will plan.

[0107] S244. After the simulation, the module performs a quantitative evaluation of the system's robustness in adversarial scenarios. In one embodiment, robustness R is mainly measured from the perspective of the ability to maintain road network connectivity, and its calculation formula can be:

[0108]

[0109] Where: R is a measure of system robustness, the closer to 1, the stronger the robustness; V critical It is a predefined set of key nodes, such as all hospitals, shelters, and supply warehouses within the area; d real (v i ,v j (This refers to the critical node v before the virtual disaster injection.) i and v j The shortest travel distance or time between them on the road network; d sim (v i ,v j ) is the key node v after the adversarial scenario simulation. i and v j The shortest travel distance or time between two points on the damaged road network; if the two points are no longer connected, this value is infinity.

[0110] S245. The above robustness measurement formula is only one specific implementation method; in other implementation methods, the robustness measurement can also be based on other indicators, such as the number of critical nodes that are isolated, the expected time increment for emergency rescue forces to reach the designated area, or the uncertainty of the system's final understanding of the overall state of the road network at the end of the simulation.

[0111] S246. This module will conduct a comprehensive analysis of the quantitative assessment result R and the systemic risks identified during the simulation process (e.g., the vulnerability of a single node in an evacuation route, or a communication coverage blind spot in a certain area), automatically generate a system vulnerability report and emergency plan optimization suggestions, and submit them to the application and release unit 40 to provide decision support for decision-makers to anticipate and avoid potential risks before a real disaster occurs.

[0112] The road map network status decision and update module belongs to the application and publishing unit 40. Its function is to transform the uncertain posterior cognitive results output by the core processing unit 30 into clear, executable, and publishable road network status information.

[0113] S247. The execution of this module begins with the state decision algorithm. After each time step of the bidirectionally coupled simulation assimilation module 32, the system generates a posterior basic probability assignment function (BPA)m for each road segment e or key node v. post This function contains the confidence distribution of all possibilities for the target state. In order to obtain a unique and definite state conclusion, this module needs to apply a decision criterion.

[0114] S248. In one embodiment, the decision criterion is based on each individual hypothesis H. k (For example, H) k ={Interruption}) Confidence (Belief) and likelihood (Plausibility) calculation, confidence (Belief, Plausibility) k This indicates that all evidence fully supports hypothesis H. k The sum of trust levels is a lower bound or conservative estimate, and the likelihood Pl(H) k ) represents all that are not related to hypothesis H k The sum of the confidence levels of contradictory evidence is an upper limit or an optimistic estimate, and its calculation formulas are as follows:

[0115]

[0116] Wherein: H k It is a single hypothesis in the judgment frame Θ; m post B is the posterior basic probability assignment function to be decided; B is a subset of the decision frame Θ.

[0117] S249. After calculating the confidence and likelihood of all individual hypotheses, the system uses a preset decision rule to select the final state. In a specific implementation, the system selects the state that makes the confidence level Bel(H) equal to the likelihood level Bel(H). k The biggest hypothesis H k As the final decision outcome; in another implementation, to comprehensively consider the support and consistency of evidence, the system may also choose to make the average of the confidence and likelihood (Bel(H)) the result. k )+Pl(H k The largest hypothesis H is )) / 2 k As a result of the decision-making process, the selection of the aforementioned decision-making rules can be configured according to the application's risk preferences.

[0118] S250. After making a state decision, the module performs an attribute update operation, and the system writes the final state (e.g., interruption) into the corresponding road segment e in the road graph network model G. i dynamic attribute set In addition to the status itself, this update may also include other accompanying information, specifically:

[0119] Physical causes of disasters: Through analysis of m post The source of the data is determined, identifying which physical process (such as flood simulation or landslide risk assessment) contributes most to the final decision, and the cause (such as inundation or landslide deposition) is also recorded.

[0120] Uncertainty measurement: the measurement of m on which this decision is based post Calculated uncertainty value U(m) post Stored in an attribute set, providing users with a quantitative reference for the reliability of the decision;

[0121] Update timestamp: Records the precise time of this attribute update.

[0122] S251. After completing the attribute update of the internal data model, the module finally performs a publishing operation. This operation aims to provide the updated road map network information to different user terminals in one or more forms. Specific publishing methods include:

[0123] Generate structured data files: Export the updated road map network G and all its attributes into standard format files commonly used by Geographic Information Systems (GIS), such as GeoJSON, Shapefile, or GML, for other professional systems to call and analyze;

[0124] Provide an Application Programming Interface (API): Expose an API endpoint via the network in the form of a web service; authorized users or systems can use this API to query the latest status of any road segment in real time.

[0125] Visual rendering: The updated road network status is rendered on the electronic map to form an intuitive disaster map. This map can be stratified and customized according to user type; for example, an advanced view containing all details and uncertainties can be shown to emergency commanders, while a simplified version containing only access and disruption information can be released to the public.

[0126] To illustrate the technical solution of the present invention in a concrete way, the following will take an urban flood disaster scenario as an example to describe the entire process of the system of the present invention from receiving initial information to publishing the final road network status.

[0127] S301, Scene Initialization: Due to a forecast of a severe rainstorm in a certain city, the system of this invention is activated. (Refer to the appendix...) Figure 1With appendix Figure 2 The system is deployed on the city's emergency command center server. Before the disaster, the coupled model construction unit 20 has constructed a road map network model G containing thousands of nodes and road segments, as well as a multi-layer physical state grid P covering the entire city with a spatial resolution of 10 meters, based on the city's basic geographic information, as shown in the attached figure. Figure 3 As shown. This grid P already contains static data such as the digital elevation model (DEM), the distribution of urban drainage pipe network, and the location of underpasses. At this moment, the dynamic attribute traffic status of all road segments in the graph network G is unobstructed.

[0128] S302, Disaster Development and Data Input: As the rainstorm begins, the data interface and preprocessing unit 10 start receiving real-time data from multiple data acquisition terminals. For example:

[0129] Rain gauges deployed by the city's meteorological department transmit real-time rainfall data for each area;

[0130] The water level sensor deployed at the A underpass interchange is transmitting readings, and the value is continuously rising.

[0131] Posts from citizens appeared on social media: "Main Road B is severely flooded, and traffic is slow," accompanied by photos of the scene.

[0132] S303, Evidence Conversion: The physical evidence conversion module 31 processes the aforementioned heterogeneous data, as shown in the appendix. Figure 4 As shown.

[0133] For the water level sensor reading at Interchange A, when it exceeds the preset disaster threshold of 0.5 meters, the system generates a BPA based on the sensor's reliability: m obs1 ({Interruption}) = 0.98,m obs1 (Θ) = 0.02.

[0134] For citizen-reported information from Main Road B, the system uses natural language processing and image recognition, combined with the historical reliability of the information source, to generate a BPA: m obs2 ({Interruption}) = 0.8,m obs1 (Θ) = 0.2, where the confidence level is assigned to a subset containing two possibilities to express the imprecision of the information.

[0135] S304, Simulated Assimilation Loop (First Round): The bidirectionally coupled simulated assimilation module 32 begins its first iteration.

[0136] Prediction: The urban flooding physical model within the system predicts the city's waterlogging distribution 15 minutes later based on input real-time rainfall data and DEM (Density Earth Scale), generating predictive evidence covering the entire road network. predThe forecast indicates that both Interchange A and Main Road B have a high risk of flooding.

[0137] Assimilation: For interchange A, the system uses the Dempster combination rule formula, which incorporates the predictive evidence m. pred and observational evidence m obs1 Due to the extremely high reliability of observational evidence, the fused posterior BPAm post In the middle, the confidence level for {interruption} will be very close to 1; for main road B, the system integrates predictive evidence m pred and observational evidence m obs2 Similarly, an updated posterior BPA is obtained.

[0138] S305, State Update and Bidirectional Coupling (Second Round):

[0139] Decision and Update: The road map network state decision and update module makes a decision on the posterior BPA of interchange A based on the confidence and likelihood calculation formulas, and its state is explicitly updated to interrupted.

[0140] Prediction under bidirectional coupling: In the prediction stage of the second time step, since the state of interchange A is interrupted, the physical model marks it as blocked in the corresponding physical grid cell P; this will cause the water flow predicted by the model to bypass interchange A and converge upstream and in the lower-lying area C, thus significantly increasing the model's prediction of flood risk in area C.

[0141] S306. Active perception and cognitive intervention: At this time, for the newly emerging high-risk area C, there is no direct observation data in the system, only the model's prediction; therefore, the uncertainty value U(m) of its posterior BPA is very high, and the active perception and causal tracing module 33 is triggered.

[0142] Tracing the source: The module traces the source in reverse through the causal chain and confirms that the high uncertainty in region C originates from the model extrapolation after the change in the state of interchange A, and that there is a lack of local observation data for verification.

[0143] Task generation: Based on the information value assessment formula, the module calculates that the information value of the perception action of dispatching a drone to area C for aerial reconnaissance is far higher than that of other actions (such as retrieving traffic cameras from further away).

[0144] Release: The system sends a high-priority suggestion to the user interaction terminal of the emergency command center through the application and release unit 40: There is a high risk of waterlogging in area C, and it is recommended to immediately dispatch a drone to verify.

[0145] S307 Robustness Assessment: During emergency response, the commander needs to consider a worst-case scenario: the D drainage pumping station in the city center fails due to a power outage; the commander triggers the system's robustness self-assessment module 34 through the user interaction terminal; the system uses a virtual adversarial scenario generation formula to inject a virtual event of the D pumping station failure into the current disaster digital twin copy; the system performs a rapid simulation and calculates through the system robustness measurement formula that, under this situation, a critical lifeline channel E leading to the city center hospital will be interrupted within 25 minutes; the system then generates a report reminding the commander to prepare a backup passage plan or emergency rescue force for channel E;

[0146] S308. Continuous Updates and Releases: Throughout the disaster, the cycle from S304 to S306 continues. The system continuously integrates newly arriving data (including video streams confirming water accumulation transmitted by drones from area C) and continuously displays the updated road map network, which includes information such as traffic status, causes of the disaster, and uncertainties, on the electronic map of the emergency command center in real time through the application and release unit 40. It also provides API data services to downstream navigation applications.

[0147] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for updating a post-disaster road recovery map network by integrating heterogeneous disaster perception data, characterized in that, Includes the following steps: S1. Obtain heterogeneous disaster sentiment data and convert the heterogeneous disaster sentiment data into a basic probability allocation function for observational evidence; S2. Based on the physical process simulator, predict the physical field state at the current moment according to the state of the physical state grid and road map network at the previous moment, and generate the basic probability allocation function of the prediction evidence. S3. The basic probability assignment function of the observed evidence and the basic probability assignment function of the predicted evidence are combined and calculated using the Dempster combination rule to handle the conflict between the evidence and obtain the posterior basic probability assignment function. S4. Based on the posterior basic probability allocation function, calculate the confidence and likelihood of each single hypothesis in the decision framework, and select the traffic status of road segments or nodes in the road map network based on the confidence and likelihood according to the preset decision rules, and update the dynamic attributes of the road map network to obtain the updated road map network. S5. The updated state of the road map network is used as the dynamic boundary condition for the physical process simulator to predict the physical process at the next moment. Specifically, this includes setting road sections with intact roadbeds as low permeability or dams in the physical state grid, and / or setting nodes with structural collapse as obstacles in the physical state grid, thereby forming a two-way coupling between the physical process simulation and the state of the road map network.

2. The method for updating a post-disaster road recovery map network by fusing heterogeneous disaster perception data as described in claim 1, characterized in that, Step S1, which involves converting the heterogeneous disaster perception data into a basic probability allocation function of observational evidence, includes at least one of the following: A semantic segmentation model is used to process remote sensing image or video data, and a basic probability allocation function is generated based on the pixel classification probability to obtain the observation evidence. The text information is processed using a natural language processing model, and a basic probability allocation function for the observed evidence is generated based on the extracted key information and the reliability of the information source. The readings of the physical sensors are compared with a preset physical threshold, and a basic probability allocation function for the observation evidence is generated based on the comparison results and the sensor accuracy.

3. The method for updating a post-disaster road recovery map network by fusing heterogeneous disaster perception data as described in claim 1, characterized in that, The method further includes: After obtaining the posterior basic probability assignment function, calculate the uncertainty of the posterior basic probability assignment function; When the uncertainty exceeds a preset threshold, the physical process simulator is used to perform reverse causal chain tracing to determine the upstream physical state or input data that leads to high uncertainty, and an active perception task list is generated based on information value assessment.

4. The method for updating a post-disaster road recovery map network by fusing heterogeneous disaster perception data as described in claim 1, characterized in that, After obtaining the updated road map network in step S4, the method further includes: Inject one or more virtual disaster events into a digital copy of the current disaster state based on an updated road map network to generate adversarial simulation scenarios; Using the adversarial simulation scenario as initial conditions, the deduction of physical process prediction, evidence fusion, and road map network state update is performed; Based on a preset system robustness metric formula, the robustness of the road map network in adversarial simulation scenarios is quantitatively evaluated.

5. The method for updating a post-disaster road recovery map network by fusing heterogeneous disaster perception data as described in claim 1, characterized in that, Before step S4, the method further includes: The basic probability assignment function of the received observation evidence is subjected to cross-modal self-consistency verification based on physical simulation. If the simulation results do not match the observational evidence, the weight of the basic probability allocation function of the observational evidence in the fusion is reduced or the basic probability allocation function of the observational evidence is isolated.

6. The method for updating a post-disaster road recovery map network by fusing heterogeneous disaster perception data as described in claim 3, characterized in that, The steps for generating the proactive sensing task list based on information value assessment include: Calculate information value for multiple candidate sensing actions, whereby the information value is defined as the expected reduction in system state uncertainty after the action is performed; A list of proactive sensing tasks is generated based on the ranking of the information values.

7. The method for updating a post-disaster road recovery map network by fusing heterogeneous disaster perception data as described in claim 4, characterized in that, The system robustness measurement formula measures the robustness of the road map network by calculating the change in the shortest travel distance between a predefined set of key nodes in the adversarial simulation scenario.

8. A post-disaster road recovery map network update system integrating heterogeneous disaster sentiment data, applied to the method described in any one of claims 1-7, characterized in that, The system includes: The data interface and preprocessing unit are used to acquire heterogeneous disaster perception data and convert the heterogeneous disaster perception data into a basic probability allocation function of observation evidence; The core processing unit is used to predict the physical field state at the current moment based on the physical state grid and road map network state at the previous moment, based on the physical process simulator, and generate the basic probability allocation function of the prediction evidence; and to obtain the posterior basic probability allocation function by fusing the basic probability allocation function of the observation evidence and the basic probability allocation function of the prediction evidence. The application and publishing unit is used to determine the traffic status of road segments or nodes in the road map network according to the posterior basic probability allocation function, and update the dynamic attributes of the road map network to obtain the updated road map network. The core processing unit is also used to use the updated state of the road map network as the dynamic boundary condition for the physical process simulator to predict the physical process at the next moment, thus forming a two-way coupling between the physical process simulation and the state of the road map network.

Images