AI-driven infrastructure risk operation optimization management system

Abstract

This invention discloses an AI-driven infrastructure risk operation optimization management system, belonging to the fields of computer data processing and business management. It includes a multimodal causal twin construction module, which integrates on-site multimodal data streams to construct a dynamic spatiotemporal causal graph; a risk evolution inference module, which performs counterfactual simulation based on the causal graph to construct a forward-looking risk model; a collaborative configuration optimization module, which solves for the optimal collaborative defense strategy based on the risk model; an instruction parsing and digital prescription generation module, which parses the defense strategy into a digital prescription for specific risk scenarios; and an intervention effectiveness attribution and evolution correction module, which performs attribution analysis based on the execution effect of the digital prescription and adaptively updates the causal graph. This invention employs a comprehensive method of constructing a dynamic causal graph for risk inference, coupling resource constraints for collaborative optimization, and closed-loop feedback to correct the model, enabling proactive prediction, precise intervention, and continuous learning optimization of infrastructure risks.

Description

Technical Field

[0001] This invention relates to the fields of computer data processing and business management, and in particular to an AI-driven infrastructure risk operation optimization management system. Background Technology

[0002] Large-scale infrastructure construction, as a crucial pillar of the national economy, involves complex planning, scheduling, operation, and risk control in its project management. In the wave of digital transformation, project management is increasingly relying on data-driven decision support systems, aiming to optimize resource allocation, improve operational efficiency, and ensure the safety of personnel and assets through the analysis of massive amounts of data. These systems typically integrate multiple technologies such as business intelligence, data mining, and operations management to achieve refined management throughout the entire project lifecycle.

[0003] In existing technologies, risk management for infrastructure projects typically relies on isolated information systems. For example, building information models (BIMs) are used for visual management, IoT sensors collect equipment status data, or environmental monitoring systems issue weather warnings. At the data analysis level, machine learning models based on historical data statistics are often used to predict the probability of a single risk event, or fixed threshold rules are set for alerts. At the decision support level, risk warning is usually a separate process from resource scheduling and work arrangements, relying on the experience of managers for manual coordination and instruction.

[0004] However, the aforementioned existing technical solutions have significant drawbacks. First, information silos are formed between various data systems, failing to effectively reveal the dynamic and interdependent relationships between different physical entities, resulting in blind spots in risk perception. Second, statistically based prediction models struggle to explain their predictions and cannot predict rare but impactful cascading risk events. Finally, the separation of risk warning and operational decision-making leads to response delays, frequent resource allocation conflicts, and a lack of evaluation and feedback on the effectiveness of intervention measures, preventing the management system from continuously improving through practice and making it difficult to adapt to complex and ever-changing construction site environments. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an AI-driven infrastructure risk operation optimization management system. It employs a comprehensive approach that involves constructing a dynamic causal graph for risk projection, coupling resource constraints for collaborative optimization, and using closed-loop feedback to correct the model. This approach enables proactive prediction, precise intervention, and continuous learning and optimization of infrastructure risks.

[0006] The above objectives can be achieved through the following approach:

[0007] The AI-driven infrastructure risk operation optimization management system includes a multimodal causal twin construction module for real-time fusion of multimodal data streams from the infrastructure site. These multimodal data streams include building information model (BIM) data, IoT sensor array data, environmental monitoring data, and personnel physiological data. The module performs causal network deduction to extrapolate nonlinear causal dependencies between physical entities, constructing a dynamic spatiotemporal causal graph. A risk evolution deduction module, based on the dynamic spatiotemporal causal graph, applies virtual perturbations to key nodes for counterfactual simulation, deducing the risk evolution path and cascading effects triggered by single-point risks, and constructing a forward-looking risk model. A collaborative configuration optimization module uses the forward-looking risk model as a constraint and couples it with the dynamic spatiotemporal causal graph. The system analyzes the resource dependencies of the dynamic spatiotemporal causal graph, constructs a dynamic programming model, solves for the optimal collaborative defense strategy under different resource constraints, and outputs a time-varying risk intervention strategy matrix. A command parsing and digital prescription generation module parses the time-varying risk intervention strategy matrix into command paradigms and, combined with real-time state snapshots of the dynamic spatiotemporal causal graph, generates operational digital prescriptions for specific dynamic risk scenarios. An intervention effectiveness attribution and evolution correction module continuously monitors the evolution of the on-site state after applying the operational digital prescriptions, compares the actual evolution results with the prediction results of the forward-looking risk model, calculates the effectiveness deviation of the intervention measures, performs attribution analysis, and adaptively updates the causal link weights in the dynamic spatiotemporal causal graph based on attribution confidence.

[0008] Optionally, the multimodal causal twin construction module includes: a data spatiotemporal alignment unit, used to map and synchronize the timestamps and spatial coordinates in the IoT sensor array data, the environmental monitoring data, and the personnel physiological characteristic data using the building information model data as a three-dimensional spatial reference, and output a multimodal dataset; a cross-modal latent feature extraction unit, used to learn and extract the intrinsic state and interaction behavior of physical entities based on the multimodal dataset, and obtain latent feature vectors; a causal structure learning unit, used to perform temporal analysis on the latent feature vectors, quantify nonlinear causal dependencies, and search and determine the directed causal links and nonlinear function relationships of each latent feature under the condition of satisfying noncyclic constraints; and a graph instantiation and weighting unit, used to instantiate and generate a dynamic spatiotemporal causal graph using the physical entities in the building information model data as nodes, the directed causal links as edges, and the influence intensity of the nonlinear function relationships as weights.

[0009] Optionally, the risk evolution simulation module includes: a source point location and disturbance set generation unit, used to analyze the topological structure of the dynamic spatiotemporal causal graph, calculate the causal out-degree and influence weight of each node, locate the risk source point based on the calculation results, and generate a parameterized virtual disturbance set around the risk source point; a causal chain simulation and cascade effect simulation unit, used to apply the parameters in the parameterized virtual disturbance set to the risk source point one by one, perform discrete-time iterative simulation in the dynamic spatiotemporal causal graph, simulate multiple risk evolution paths triggered by a single point of risk and their cross-influence, and obtain the simulated evolution path; and a forward-looking risk field construction unit, used to statistically aggregate the simulated evolution path, construct a risk propagation probability field in a four-dimensional spatiotemporal coordinate system, quantify the probability that any future spatiotemporal point will be affected by the risk of a specific source point, and use it as a forward-looking risk model.

[0010] Optionally, the parameterized virtual perturbation set includes: a state offset vector set and a propagation failure operator set, wherein: the state offset vector set is used to simulate extreme deterioration scenarios of the physical entity's own attribute parameters in a statistical sense; the propagation failure operator set is used to simulate the local interruption or nonlinear distortion of the causal relationship transmission path between entities in the dynamic spatiotemporal causal graph.

[0011] Optionally, the system further includes: matching the latent feature vector with the simulation evolution path, calculating the distance between the initial state and key feature points, evaluating the convergence degree of the current infrastructure site state to each risk evolution path, and generating a path activation confidence vector.

[0012] Optionally, the collaborative configuration optimization module includes: a multi-dimensional constraint space construction unit, used to transform the risk propagation probability field of the forward-looking risk model into the state transition cost of the dynamic programming model, and extract the logical dependency relationship between resource scheduling and task execution from the dynamic spatiotemporal causal graph as state constraints, jointly defining the solution space; a strategy iteration solution unit, used to recursively solve the state action sequence that can accumulate risk costs under the condition of satisfying the state constraints within the solution space, to obtain the optimal collaborative defense strategy; and a strategy matrix generation unit, used to map and discretize the optimal collaborative defense strategy in the time dimension to generate a time-varying risk intervention strategy matrix.

[0013] Optionally, the instruction parsing and digital prescription generation module includes: a strategy matrix parsing unit, used to decode the time-varying risk intervention strategy matrix into structured intervention rule primitives; a dynamic risk scenario matching unit, used to identify dynamic risk scenarios using the path activation confidence vector, and match the optimal intervention rule from the structured intervention rule primitives; and a digital prescription generation unit, used to call the real-time state snapshot of the dynamic spatiotemporal causal graph to instantiate the parameters in the optimal intervention rule, and encapsulate them into a standardized data package as an operational digital prescription and send it to the field execution terminal.

[0014] Optionally, the standardized data package includes: a set of operation instructions, risk context data, and performance verification parameters, wherein: the set of operation instructions is used to define intervention actions, parameter thresholds, and execution sequences for the target physical entity; the risk context data is used to encapsulate the activated risk evolution path, the real-time status of the risk source node, and the expected loss assessment after intervention failure; and the performance verification parameters are used to specify the monitoring indicators and state transition targets that must be met for successful intervention.

[0015] Optionally, the intervention efficacy attribution and evolution correction module includes: an efficacy monitoring and deviation calculation unit, used to continuously track the on-site data stream after the digital prescription for the operation is issued, and compare the actual monitoring indicators with the state transition targets predicted by the prospective risk model to quantify the efficacy deviation vector; a causal attribution and confidence assessment unit, used to perform contribution tracing analysis on the dynamic spatiotemporal causal graph based on the efficacy deviation vector, identify and calculate causal links, and generate attribution confidence for each identified causal link; and a graph weight update unit, used to update the parameters of the causal links with the attribution confidence as weights, and incorporate new observational evidence while retaining historical knowledge to achieve adaptive evolution of the dynamic spatiotemporal causal graph.

[0016] Based on the same inventive concept, this invention also provides an AI-driven method for optimizing and managing infrastructure risk operations. The method includes: real-time fusion of multimodal data streams from the infrastructure site, including building information model data, IoT sensor array data, environmental monitoring data, and personnel physiological data; performing causal network deduction to extrapolate nonlinear causal dependencies between physical entities and constructing a dynamic spatiotemporal causal graph; based on the dynamic spatiotemporal causal graph, applying virtual perturbations to key nodes to conduct counterfactual simulations, deducing the risk evolution path and cascading effects caused by single-point risks, and constructing a forward-looking risk model; using the forward-looking risk model as a constraint condition, and... By coupling the resource dependencies of the dynamic spatiotemporal causal graph, a dynamic programming model is constructed to solve the optimal collaborative defense strategy under different resource constraints, and a time-varying risk intervention strategy matrix is ​​output. The time-varying risk intervention strategy matrix is ​​parsed into an instruction paradigm, and combined with the real-time state snapshot of the dynamic spatiotemporal causal graph, a digital prescription for operation is generated for specific dynamic risk scenarios. The evolution of the field state after applying the digital prescription for operation is continuously monitored, and the actual evolution results are compared with the prediction results of the prospective risk model. The effectiveness deviation of the intervention measures is calculated and attribution analysis is performed. The causal link weights in the dynamic spatiotemporal causal graph are adaptively updated based on the attribution confidence.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] 1. This invention achieves a fundamental understanding of infrastructure risks by deeply integrating multi-source heterogeneous data and constructing a dynamic spatiotemporal causal graph. It transforms risk management from a traditional passive response model based on surface-level phenomena to a proactive prediction model based on deep causal mechanisms. This approach can identify complex chain reactions triggered by minor events that are difficult to detect using traditional statistical methods, thus providing accurate early warnings at the nascent stage of risks and improving the predictability and accuracy of risk prevention and control from the source.

[0019] 2. This invention integrates risk evolution simulation with resource collaborative allocation optimization. By using a forward-looking risk model as a constraint in dynamic programming, it ensures that the generated intervention strategy is not only theoretically optimal but also takes into account limited human and material resources in practice. This transforms the decision-making scheme from an isolated risk alert into a directly executable, globally coordinated operational instruction, thereby solving the problem of the disconnect between risk management and operational scheduling, and greatly improving emergency response speed and resource utilization efficiency.

[0020] 3. This invention constructs a complete feedback loop from decision execution to model correction. By analyzing the deviation between the actual effects of intervention measures and the predictive model, and using causal attribution techniques to accurately pinpoint the shortcomings in the model, it achieves adaptive learning and evolution of the core causal graph. This enables continuous learning from practice, constantly improving its cognitive depth and decision-making quality for specific project environments, overcoming the fundamental defects of traditional models that are rigid and unable to adapt to dynamic changes, and ensuring long-term effectiveness and reliability.

[0021] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a framework diagram of the AI-driven infrastructure risk operation optimization management system according to an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the infrastructure risk operation optimization management system according to an embodiment of the present invention.

[0025] Figure 3 This is a schematic diagram of the dynamic spatiotemporal causal graph structure according to an embodiment of the present invention.

[0026] Figure 4 This is a time-series probability distribution cloud and rain diagram of risk evolution extrapolation in an embodiment of the present invention.

[0027] Figure 5 This is a schematic diagram illustrating the optimized configuration of the collaborative defense strategy in an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of 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, 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.

[0029] Reference Figure 1 One embodiment of the present invention proposes an infrastructure risk operation optimization management system, which adopts a comprehensive approach that uses a dynamic causal graph to perform risk extrapolation, coupled resource constraints for collaborative optimization, and a closed-loop feedback correction model. This approach enables proactive prediction, precise intervention, and continuous learning optimization of infrastructure risks.

[0030] like Figure 2 As shown, the system in this embodiment specifically includes:

[0031] A multimodal causal twin construction module is used to fuse multimodal data streams from infrastructure sites in real time. The multimodal data streams include building information model data, IoT sensor array data, environmental monitoring data, and personnel physiological data. The module performs causal network deduction to extrapolate nonlinear causal dependencies between physical entities and constructs a dynamic spatiotemporal causal graph.

[0032] The risk evolution simulation module is used to apply virtual perturbations to key nodes based on the dynamic spatiotemporal causal graph to conduct counterfactual simulations, deduce the risk evolution path and cascading effects caused by single-point risks, and construct a forward-looking risk model.

[0033] The collaborative configuration optimization module is used to take the forward-looking risk model as a constraint and couple the resource dependency relationship of the dynamic spatiotemporal causal graph to construct a dynamic programming model, solve the optimal collaborative defense strategy under different resource constraints, and output a time-varying risk intervention strategy matrix.

[0034] The instruction parsing and digital prescription generation module is used to parse the time-varying risk intervention strategy matrix into an instruction paradigm, and combine it with the real-time status snapshot of the dynamic spatiotemporal causal graph to generate an operational digital prescription for a specific dynamic risk scenario.

[0035] The intervention efficacy attribution and evolution correction module is used to continuously monitor the evolution of the field status after applying the digital prescription, compare the actual evolution results with the prediction results of the prospective risk model, calculate the efficacy deviation of the intervention measures and perform attribution analysis, and adaptively update the causal link weights in the dynamic spatiotemporal causal graph based on the attribution confidence.

[0036] By employing a comprehensive approach that combines risk projection through the construction of dynamic causal graphs, collaborative optimization by coupling resource constraints, and closed-loop feedback correction models, it is possible to proactively anticipate, precisely intervene in, and continuously learn and optimize infrastructure risks.

[0037] Optionally, the multimodal causal twin building module includes:

[0038] The data spatiotemporal alignment unit is used to map and synchronize the timestamps and spatial coordinates in the IoT sensor array data, the environmental monitoring data, and the personnel physiological characteristic data with the building information model data as a three-dimensional spatial reference, and output a multimodal dataset.

[0039] Specifically, the data spatiotemporal alignment unit first loads Building Information Modeling (BIM) data. The BIM model here is a comprehensive 3D model containing building geometry, spatial relationships, geographic information, and component attributes, serving as a unified spatial reference for all subsequent data. For each data point with timestamps and original coordinates collected from Internet of Things (IoT) sensor arrays, environmental monitoring stations, and wearable devices, this unit invokes a coordinate transformation function to map its spatial coordinates to the global coordinate system of the BIM model. Simultaneously, the timestamps of all data are synchronized with the central server via Network Time Protocol (NTP) to ensure consistency of the time reference. After mapping and synchronization, data streams from different sources are integrated into a structured multimodal dataset under a unified spatiotemporal reference, providing high-quality input for subsequent feature extraction.

[0040] The cross-modal latent feature extraction unit is used to learn and extract the intrinsic state and interaction behavior of physical entities based on the multimodal dataset to obtain latent feature vectors;

[0041] Specifically, this cross-modal latent feature extraction unit employs a Variational Auto-Encoder (VAE) to process multimodal datasets. The VAE model maps high-dimensional input data to a probability distribution in a low-dimensional latent space through an encoder network, and then reconstructs the original data from the latent space through a decoder network. The model's training objective is to maximize the Evidence Lower Bound (ELBO), and its optimization objective function is:

[0042] ,

[0043] In the formula, The input is a multimodal data sample; These are latent feature vectors; The prior distribution of the latent feature vectors is usually a standard normal distribution; It is determined by the parameter as The posterior distribution learned by the encoder network; It is determined by the parameter as The conditional distribution learned by the decoder network; This is the reconstruction loss term, used to measure the similarity between the generated data and the original data; KL divergence (KL divergence) measures the difference between the posterior and prior distributions. By optimizing this objective function, the latent feature vector $z$ learned by this unit can capture the core, decoupled features that characterize the intrinsic state and interaction behavior of physical entities.

[0044] The causal structure learning unit is used to perform time-series analysis on the latent feature vectors, quantify nonlinear causal dependencies, and search and determine the directed causal links and nonlinear function relationships of each latent feature under the condition of satisfying noncyclic constraints.

[0045] Specifically, this causal structure learning unit receives a time-ordered sequence of latent feature vectors output by the previous unit. To identify the causal structure among these temporal features, this unit applies a structure discovery algorithm based on continuous optimization. This algorithm parameterizes the adjacency matrix of the causal graph and constructs a loss function that includes a data fit goodness-of-fit term and a graph structure sparsity regularization term. A key constraint is the non-cyclic constraint, ensuring that there are no directed loops in the learned causal graph. For example, this constraint can be achieved using a penalty function whose value increases sharply when the trace of a power of the adjacency matrix is ​​not zero. During training, this unit iteratively solves the problem using optimization methods such as gradient descent, ultimately obtaining an adjacency matrix that describes the directed causal links between the latent features and their corresponding nonlinear functional relationships.

[0046] The graph instantiation and weighting unit is used to instantiate and generate a dynamic spatiotemporal causal graph using the physical entities of the building information model data as nodes, the directed causal links as edges, and the influence intensity of the nonlinear function relationship as weights.

[0047] Specifically, the graph instantiation and weighting unit first extracts explicit physical entities from the BIM data, such as a specific tower crane, a load-bearing wall, or a high-risk work area, and uses these as nodes in the graph. Then, based on the adjacency matrix output by the causal structure learning unit, the unit establishes directed edges between the corresponding nodes, representing directed causal links between them. The edge weights, i.e., the influence strength of the nonlinear functional relationship, are determined by calculating the expected partial derivative of the latent eigenvector of the dependent variable with respect to the latent eigenvector of the independent variable. This influence strength value quantifies the expected impact of a small change in one node on another node. Through these steps, the unit ultimately instantiates and generates a weighted, directed, dynamic spatiotemporal causal graph that dynamically reflects the complex causal relationships between various elements on the infrastructure site, such as... Figure 3As shown, this is a schematic diagram of the string diagram visualization of the dynamic spatiotemporal causal graph in this invention. The outer arc segments represent physical entity nodes, and the thickness of the inner strings represents the intensity of causal influence between nodes.

[0048] Optionally, the risk evolution simulation module includes:

[0049] The source point location and disturbance set generation unit is used to analyze the topological structure of the dynamic spatiotemporal causal graph, calculate the causal out-degree and influence weight of each node, locate the risk source point based on the calculation results, and generate a parameterized virtual disturbance set around the risk source point.

[0050] Specifically, the risk source localization and disturbance set generation unit first receives a dynamic spatiotemporal causal graph generated by the multimodal causal twin construction module. To locate risk sources, this unit analyzes the graph's topology and calculates two core indicators for each node: causal out-degree (the number of directed edges from that node to other nodes) and influence weight (the sum of all out-edge weights for that node). Subsequently, the unit uses a weighted model to calculate the criticality score of each node. This model uses weighted coefficients to weight and sum the node's causal out-degree and influence weight. These weighted coefficients can be configured based on historical data analysis or an expert knowledge base to balance the node's influence range and depth. After calculation, the unit compares each node's criticality score with a criticality threshold. Nodes with scores exceeding this threshold are identified as risk sources. Finally, based on the attributes of the physical entities represented by the identified risk sources and historical anomaly data, the unit generates a set of parameterized virtual disturbances to simulate potential extreme conditions.

[0051] The causal chain deduction and cascade effect simulation unit is used to apply the parameters of the parameterized virtual disturbance set to the risk source point one by one, perform discrete-time iterative deduction in the dynamic spatiotemporal causal graph, simulate multiple risk evolution paths triggered by a single point of risk and their cross-influence, and obtain the simulated evolution path.

[0052] Specifically, the causal chain deduction and cascading effect simulation unit initiates a counterfactual simulation process. This unit extracts disturbance parameters from the parameterized virtual disturbance set generated by the previous unit and applies them one by one to the corresponding risk source points as initial conditions for the simulation. Driven by discrete time steps, the unit iteratively calculates and updates the future states of downstream nodes affected by the disturbed nodes based on the nonlinear functional relationships defined in the dynamic spatiotemporal causal graph. For example, at time t+1, the state of a node is a function of its state at time t and the states of all upstream nodes pointing to it. This deduction process propagates continuously along the causal links of the graph until the risk effect dissipates in the network or the simulation deadline is reached. The record of this single complete deduction process constitutes a simulation evolution path. By traversing and simulating all parameters in the parameterized virtual disturbance set, this unit can generate a massive number of simulation evolution paths. These path sets comprehensively reveal the various risk evolution processes and complex cascading effects that may be triggered by a single point of risk.

[0053] The forward-looking risk field construction unit is used to statistically aggregate the simulation evolution path, construct a risk propagation probability field in a four-dimensional spatiotemporal coordinate system, quantify the probability that any future spatiotemporal point will be affected by the risk of a specific source point, and serve as a forward-looking risk model.

[0054] Specifically, this forward-looking risk field construction unit statistically aggregates all simulation evolution paths generated by the previous unit. Within a four-dimensional spatiotemporal coordinate system containing three-dimensional spatial coordinates and one-dimensional time coordinates, it counts the total frequency of risk state transmission and occurrence within each four-dimensional spatiotemporal cell across all simulations. The risk probability at any spatiotemporal point is determined by dividing the total number of simulation paths where the risk occurs at that point by the total number of simulations. By calculating across the entire coordinate system, this unit ultimately constructs a dynamic, quantified risk propagation probability field. This risk propagation probability field visually demonstrates the distribution density and diffusion trend of risks triggered by specific risk sources in future time and space dimensions, and is ultimately encapsulated as a forward-looking risk model for subsequent collaborative configuration optimization modules to call, such as... Figure 4 As shown, the time-series probability distribution cloud map of risk evolution in this invention demonstrates the dynamic evolution of the probability distribution of key risk indicators over time.

[0055] Optionally, the parameterized virtual perturbation set includes: a set of state offset vectors and a set of propagation failure operators, wherein:

[0056] The state offset vector set is used to simulate extreme deterioration scenarios of the physical entity's own attribute parameters in a statistical sense.

[0057] Specifically, for the state offset vector set, the generation process first identifies the key state parameters of the physical entity corresponding to a certain risk source point, such as the operating temperature of equipment, the strain value of the structure, or the physiological fatigue index of personnel, and forms a baseline state vector based on its normal operation data. Subsequently, based on industry safety standards or historical extreme failure data, one or more target state vectors under deterioration scenarios are defined. The state offset vector is the difference between the target state vector and the baseline state vector, and its calculation formula is:

[0058] ,

[0059] In the formula, This is the state offset vector; This represents the target state vector under the deteriorating scenario. The baseline state vector is used. Multiple state offset vectors, representing different directions and degrees of deterioration, are generated to form a state offset vector set. During simulation, the instantaneous deterioration of an entity's properties is simulated by superimposing a state offset vector onto the current state of a risk source point.

[0060] The set of transmission failure operators is used to simulate the local interruption or nonlinear distortion of the causal relationship transmission path between entities in the dynamic spatiotemporal causal graph.

[0061] Specifically, the set of conduction failure operators focuses on the "edges" in the dynamic spatiotemporal causal graph, i.e., the causal transmission relationships between physical entities. A conduction failure operator is defined as a mathematical transformation acting on the causal link transfer function. For example, one operator simulates a local interruption of a causal path by temporarily setting the influence intensity weight corresponding to that path to zero during simulation iterations. Another operator simulates nonlinear distortion by acting on the original causal transfer function through a distortion function, thereby simulating complex situations such as unexpected amplification, attenuation, or delay of signals during transmission. These operators are selected from an operator library based on common failure modes, collectively forming the set of conduction failure operators.

[0062] Optionally, the system further includes:

[0063] The hidden feature vector is matched with the simulation evolution path, the distance between the initial state and the key feature points is calculated, the convergence degree of the current infrastructure site state to each risk evolution path is evaluated, and a path activation confidence vector is generated.

[0064] Specifically, to dynamically assess which simulated risk path the current infrastructure site is trending towards, this process first acquires latent feature vectors representing the current site's true state, generated in real-time by the cross-modal latent feature extraction unit. Simultaneously, the process calls upon a set of simulation evolution paths generated by the causal chain deduction and cascade effect simulation unit. For each simulation evolution path, the process extracts the latent feature vector of its initial state and the latent feature vectors of several key inflection points along the path. Subsequently, the process calculates the Euclidean distance between the current state's real-time latent feature vector and the initial state vector of each simulation path, as well as the vector of the nearest key inflection point. These two distance values ​​are weighted and summed using a weighting factor to obtain a comprehensive path convergence index. The smaller this index, the closer the current state is to the simulated evolution path. To transform the convergence of all paths into standardized probability values, the process finally uses the Softmax function for normalization, thereby generating a path activation confidence vector, calculated as follows:

[0065] ,

[0066] In the formula, For the first Activation confidence of each simulated evolution path; For the current real-time status and the first The path convergence of the simulation evolution path; This represents the path convergence between the current real-time state and all alternative simulation evolution paths; This represents summing over all paths; This is a temperature hyperparameter, derived from cross-validation, used to adjust the smoothness of the confidence probability distribution. The smaller the value, the sharper the probability distribution. In the final generated path activation confidence vector, the value of each element represents the confidence that the current infrastructure site status is developing along the corresponding risk evolution path.

[0067] Optionally, the collaborative configuration optimization module includes:

[0068] A multidimensional constraint space construction unit is used to transform the risk propagation probability field of the forward-looking risk model into the state transition cost of the dynamic programming model, and extract the logical dependency relationship between resource scheduling and task execution from the dynamic spatiotemporal causal graph as state constraints, together defining the solution space.

[0069] Specifically, this multidimensional constraint space building unit is responsible for defining a precise mathematical environment for subsequent optimization solutions. This unit performs two key operations: First, it receives the risk propagation probability field output by the prospective risk model and transforms the probability values ​​in this field into state transition costs in the dynamic programming model. Specifically, when taking an action to transition to the next state in a given state, if the spatiotemporal location of the next state corresponds to a higher risk probability, the cost of that state transition increases accordingly. Second, this unit analyzes the dynamic spatiotemporal causal graph, extracting the resource mutual exclusion relationships between physical entities and the logical order of task execution. For example, a specific device can only execute one task at a time, or a task can only begin after another task has been completed. These logical dependencies are formalized as a set of state constraints, which together constitute the boundary conditions of the solution space.

[0070] The strategy iteration solution unit is used to recursively solve the state action sequence that accumulates risk cost under the condition of satisfying the state constraints within the solution space, so as to obtain the optimal cooperative defense strategy.

[0071] Specifically, this strategy iterative solution unit solves the problem within the multidimensional constraint space defined by the previous unit using a dynamic programming algorithm based on value iteration or strategy iteration. The goal of this unit is to find an optimal policy, i.e., a complete mapping that guides the choice of action in any possible state. The solution process is recursive, starting from the final target state and working backward to calculate the optimal value function for each intermediate state. In each iteration, the unit evaluates the expected cumulative risk cost of taking all possible actions in a given state and selects the action that minimizes this cost. This process iterates repeatedly until the value functions of all states converge; the final converged policy is the optimal cooperative defense policy.

[0072] The strategy matrix generation unit is used to map and discretize the optimal collaborative defense strategy in the time dimension to generate a time-varying risk intervention strategy matrix.

[0073] Specifically, this strategy matrix generation unit is responsible for transforming the state-based optimal collaborative defense strategy output by the previous unit into a directly applicable format associated with the project timeline. This unit discretizes the total project duration into a series of time steps. At each time step, the unit predicts the most likely state based on the construction plan and retrieves the optimal intervention action corresponding to that state from the optimal collaborative defense strategy. By arranging and combining the optimal actions corresponding to each time step, the unit ultimately generates a two-dimensional or higher-dimensional matrix. The dimensions of this matrix can include time, risk type, intervention target, etc., and the elements in the matrix are the specific intervention instructions or parameters, thus forming a time-varying risk intervention strategy matrix, such as... Figure 5 The diagram shown is an optimized configuration of the collaborative defense strategy in this invention. The triangular matrix on the left shows the collaborative or conflicting relationships between different intervention measures, while the bipartite diagram on the right shows the defensive effectiveness of each intervention measure against different risk evolution paths.

[0074] Optionally, the instruction parsing and digital prescription generation module includes:

[0075] The strategy matrix parsing unit is used to decode the time-varying risk intervention strategy matrix into structured intervention rule primitives;

[0076] Specifically, the strategy matrix parsing unit receives a time-varying risk intervention strategy matrix generated by the collaborative configuration optimization module. This matrix is ​​a high-dimensional data structure whose internal elements encode the optimal intervention actions under different times and risk scenarios. The unit uses a decoder to parse the matrix, translating the strategy information contained within into a set of logically clear, machine-readable structured intervention rule primitives. Each primitive defines an atomic-level intervention action framework; for example, a primitive might be defined as: "When [risk scenario type] is activated, perform [operation type] on [target entity type], adjusting [key parameters] to [target threshold]."

[0077] The dynamic risk scenario matching unit is used to identify dynamic risk scenarios by activating the confidence vector with the path, and to match the optimal intervention rule from the structured intervention rule primitives.

[0078] Specifically, the dynamic risk scenario matching unit acquires the generated path activation confidence vector in real time. This unit identifies the simulation evolution path corresponding to the element with the highest confidence value in the vector as the most likely dynamic risk scenario. Subsequently, using the identified dynamic risk scenario type as an index, the unit searches and matches within the set of structured intervention rule primitives generated by the previous unit. Based on a priority rule base, the unit selects the optimal intervention rule from all matching rules to address the most pressing risk.

[0079] The digital prescription generation unit is used to call the real-time state snapshot of the dynamic spatiotemporal causal graph to instantiate the parameters in the optimal intervention rule, and encapsulate them into a standardized data package as an operational digital prescription and send it to the field execution terminal.

[0080] Specifically, the digital prescription generation unit receives the optimal intervention rule matched by the previous unit. At this point, the rule is still parameterized. To make it fully executable, the unit calls a dynamic spatiotemporal causal graph to obtain real-time state snapshots of physical entities related to placeholders such as "[target entity type]" and "[key parameters]" defined in the rule. For example, it queries the IDs and locations of all personnel within a specific area, or queries the current operating power of a specific device. The unit uses this real-time data to populate and instantiate the parameters in the optimal intervention rule. Finally, the unit encapsulates this fully instantiated intervention rule into a standardized data package, which is the operational digital prescription, and sends it to the designated field execution end through a communication interface, such as the augmented reality glasses of the field management personnel or the automatic control of specific equipment.

[0081] Optionally, the standardized data packet includes: a set of job instructions, risk context data, and performance verification parameters, wherein:

[0082] The set of operation instructions is used to define the intervention actions, parameter thresholds, and execution sequence for the target physical entity;

[0083] Specifically, the task instruction set, as the core execution part of the standardized data package, clearly defines the intervention tasks to be performed in a structured, machine-readable format. The instruction set includes a unique identifier for the target physical entity, the specific type of intervention action (e.g., "reduce equipment operating power," "add temporary structural support," or "evacuate personnel from a designated area"), quantified parameter thresholds (e.g., "reduce power to below 80% of rated value"), and strict execution timing, including the start time of the intervention action, the mandatory completion time, or the allowed response time window.

[0084] The risk context data is used to encapsulate the activated risk evolution path, the real-time status of the risk source node, and the expected loss assessment after intervention failure.

[0085] Specifically, the risk context data aims to provide necessary background information to the on-site execution team to enhance their understanding of instructions. This data encapsulates key information about the risk evolution path that triggered the intervention and was activated, including the risk's source node and propagation route. Simultaneously, it includes a real-time snapshot of the risk source node's status at the time the instruction was generated, revealing the severity and urgency of the problem. Furthermore, the data provides an assessment of expected losses after several pre-failure scenarios, calculated based on a forward-looking risk model, thereby emphasizing the importance of the intervention.

[0086] The effectiveness verification parameters are used to specify the monitoring indicators and state transition objectives that must be met for the intervention to be successful.

[0087] Specifically, the efficacy verification parameters set clear and quantifiable objective standards for subsequent intervention effectiveness evaluation. These parameters specify one or more indicators that need to be monitored after the intervention is implemented, and provide the state transition targets these indicators should achieve after a successful intervention. For example, an efficacy verification parameter can be defined as: within five minutes of executing the "reduce equipment operating power" command, the latent feature vector of the corresponding node of the equipment must migrate to the safe state space. These parameters provide benchmark data for comparison in subsequent intervention efficacy attribution and evolution correction modules.

[0088] Optionally, the intervention efficacy attribution and evolution correction module includes:

[0089] The effect monitoring and deviation calculation unit is used to continuously track the on-site data stream after the digital prescription for operation is issued, and compare the actual monitoring indicators with the state transition targets predicted by the forward-looking risk model to quantify the effectiveness deviation vector.

[0090] Specifically, after the digital prescription for intervention is issued to the field implementation end, the intervention efficacy monitoring and deviation calculation unit is activated. Based on the efficacy verification parameters encapsulated within the digital prescription, this unit continuously tracks the field multimodal data stream corresponding to the specified monitoring indicators, thereby obtaining the actual state evolution results after the intervention is implemented. Subsequently, the unit compares the time series of these actual monitoring indicators with the state transition targets in the efficacy verification parameters, calculates the differences between the two item by item, and finally generates an efficacy deviation vector that quantifies the gap between the actual effect and the expected effect of the intervention.

[0091] The causal attribution and confidence assessment unit is used to perform contribution tracing analysis on the dynamic spatiotemporal causal map based on the efficiency deviation vector, identify and calculate causal links, and generate attribution confidence for each identified causal link.

[0092] Specifically, this causal attribution and confidence assessment unit uses the efficacy deviation vector as the starting point for analysis, performing contribution tracing analysis on a dynamic spatiotemporal causal graph to explore the root causes of the deviation. This analysis process can employ methods such as Shapley value assignment or gradient integration to identify one or more causal links that contribute the most to the final prediction deviation. The unit traces all causal paths from the intervention node to the efficacy monitoring node and calculates the contribution of each causal link on the path to the total deviation. After identifying the causal links to be corrected, the unit also comprehensively evaluates and generates an attribution confidence score for each identified causal link based on the magnitude and frequency of the deviation, as well as the score output by the attribution algorithm itself.

[0093] The graph weight update unit is used to update the parameters of the causal link with the attribution confidence as the weight, and to incorporate new observational evidence while retaining historical knowledge, so as to realize the adaptive evolution of the dynamic spatiotemporal causal graph.

[0094] Specifically, the graph weight update unit updates the dynamic spatiotemporal causal graph based on the causal links to be corrected and their corresponding attribution confidence scores output by the previous unit. This unit uses the attribution confidence score as the learning rate or adjustment weight to perform Bayesian updates on the parameters of the identified causal links. This update process is an incremental learning approach, treating existing parameters of causal links as prior knowledge and the observed biases as new evidence, thereby calculating posterior parameters that incorporate the new evidence. In this way, the unit integrates new observational evidence while retaining historical knowledge, achieving adaptive evolution of the dynamic spatiotemporal causal graph and enabling it to more accurately reflect the physical laws of the real world.

[0095] Based on the same inventive concept, this invention also provides an AI-driven method for optimizing infrastructure risk operations, the method comprising:

[0096] The system integrates multimodal data streams from the infrastructure construction site in real time. These multimodal data streams include building information model data, IoT sensor array data, environmental monitoring data, and personnel physiological data. The system then performs causal network deduction to extrapolate the nonlinear causal dependencies between various physical entities and constructs a dynamic spatiotemporal causal graph.

[0097] Based on the dynamic spatiotemporal causal graph, virtual perturbations are applied to key nodes to conduct counterfactual simulations, and the risk evolution path and cascading effects caused by single-point risks are deduced to construct a forward-looking risk model.

[0098] Using the aforementioned forward-looking risk model as a constraint and coupling it with the resource dependencies of the dynamic spatiotemporal causal graph, a dynamic programming model is constructed to solve for the optimal collaborative defense strategy under different resource constraints, and output a time-varying risk intervention strategy matrix.

[0099] The time-varying risk intervention strategy matrix is ​​parsed into an instruction paradigm, and combined with the real-time state snapshot of the dynamic spatiotemporal causal graph, a digital prescription for operation is generated for a specific dynamic risk scenario.

[0100] The evolution of the field status after applying the digital prescription is continuously monitored. The actual evolution results are compared with the prediction results of the prospective risk model. The effectiveness deviation of the intervention measures is calculated and attribution analysis is performed. The causal link weights in the dynamic spatiotemporal causal graph are adaptively updated based on the attribution confidence.

[0101] To verify the feasibility of this invention, it was applied to the risk management of high-altitude operations on the P1 main tower of a cross-sea bridge.

[0102] This invention first uses a multimodal causal twin construction module to perform spatiotemporal alignment and feature extraction on multimodal data streams such as BIM, IoT, and human physiology on site, thereby constructing a dynamic spatiotemporal causal graph containing nonlinear causal dependencies between entities.

[0103] Subsequently, the risk evolution simulation module identified the "concrete curing status of main tower P1" as the risk source point based on the map, and applied a parameterized virtual disturbance set simulating abnormal cooling to conduct counterfactual simulation, constructing a forward-looking risk model. This model predicts the risk of structural stress exceeding limits within the next 24 hours. Based on this model, the collaborative configuration optimization module generates a time-varying risk intervention strategy matrix through dynamic programming.

[0104] During project operation, the real-time status was detected to be highly consistent with a pre-simulated risk path, generating an 88% path activation confidence vector. The instruction parsing and digital prescription generation module responded immediately, matching and instantiating the rules in the strategy matrix according to the activation scenario, generating and issuing an operational digital prescription containing the operation instruction set, risk context data, and performance verification parameters. The core instruction was: "Immediately start the heat tracing and insulation system of main tower P1".

[0105] After the intervention, the intervention effectiveness attribution and evolution correction module detected a deviation between the actual strain value and the effectiveness verification parameters. Through contribution source analysis, it was found that the model underestimated the weight of "the impact of high-altitude wind speed on heat dissipation". The module then adaptively updated the weight of this causal link based on the attribution confidence, completing the learning loop.

[0106] Table 1. Examples of data representation for dynamic spatiotemporal causal graph construction and key node identification.

[0107]

[0108] Table 2. Examples of data representation for risk evolution simulation and digital prescription generation

[0109]

[0110] Table 3. Examples of data representation for intervention efficacy attribution and model evolution correction.

[0111]

[0112] As can be seen from the data in Tables 1 to 3 above, the method of this invention can effectively support the intelligent management of infrastructure risks. Table 1 clearly shows that "core tube concrete" was identified as the core risk source through quantitative analysis. Table 2 records how dynamic risks are identified based on high-confidence path activation, and precise intervention instructions are quickly generated. Table 3 specifically demonstrates the closed-loop learning capability of this invention. Through attribution analysis of post-intervention performance deviations, it can locate knowledge deficiencies in the model and perform adaptive corrections, ensuring the continuous evolution and long-term effectiveness of the model.

[0113] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or parameters that can be superimposed in the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. These are conventional technical methods and will not be elaborated further. The electrical connections between the various units described above do not necessarily represent direct or indirect connections; any indirect connection method is applicable to the embodiments of this invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of this invention.

[0114] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. An AI-driven infrastructure risk operation optimization management system, characterized in that, The system includes: A multimodal causal twin construction module is used to fuse multimodal data streams from infrastructure sites in real time. The multimodal data streams include building information model data, IoT sensor array data, environmental monitoring data, and personnel physiological data. The module performs causal network deduction to extrapolate nonlinear causal dependencies between physical entities and constructs a dynamic spatiotemporal causal graph. The risk evolution simulation module is used to perform counterfactual simulations by applying virtual perturbations to key nodes based on the dynamic spatiotemporal causal graph, to deduce the risk evolution path and cascading effects triggered by single-point risks, and to construct a forward-looking risk model. The risk evolution simulation module includes: a source point location and perturbation set generation unit, used to analyze the topological structure of the dynamic spatiotemporal causal graph, calculate the causal out-degree and influence weight of each node, locate the risk source point based on the calculation results, and generate a parameterized virtual perturbation set around the risk source point; a causal chain deduction and cascading effect simulation unit, used to apply the parameters in the parameterized virtual perturbation set one by one to the risk source point, perform discrete-time iterative deduction in the dynamic spatiotemporal causal graph, simulate multiple risk evolution paths triggered by single-point risks and their cross-influences, and obtain the simulated evolution path; and a forward-looking risk field construction unit, used to statistically aggregate the simulated evolution paths, construct a risk propagation probability field in a four-dimensional spatiotemporal coordinate system, quantify the probability that any future spatiotemporal point will be affected by a specific source point risk, and use this as a forward-looking risk model. The collaborative configuration optimization module is used to take the forward-looking risk model as a constraint and couple the resource dependency relationship of the dynamic spatiotemporal causal graph to construct a dynamic programming model, solve the optimal collaborative defense strategy under different resource constraints, and output a time-varying risk intervention strategy matrix. The instruction parsing and digital prescription generation module is used to parse the time-varying risk intervention strategy matrix into an instruction paradigm, and combine it with the real-time status snapshot of the dynamic spatiotemporal causal graph to generate an operational digital prescription for a specific dynamic risk scenario. The intervention efficacy attribution and evolution correction module is used to continuously monitor the evolution of the field state after applying the operational digital prescription, compare the actual evolution results with the prediction results of the prospective risk model, calculate the efficacy deviation of the intervention measures and perform attribution analysis, and adaptively update the causal link weights in the dynamic spatiotemporal causal graph based on the attribution confidence. The intervention efficacy attribution and evolution correction module includes: an effect monitoring and deviation calculation unit, used to continuously track the field data stream after the operational digital prescription is issued, compare the actual monitoring indicators with the state transition targets predicted by the prospective risk model, and quantify the efficacy deviation vector; a causal attribution and confidence assessment unit, used to perform contribution tracing analysis on the dynamic spatiotemporal causal graph based on the efficacy deviation vector, identify and calculate causal links, and generate attribution confidence for each identified causal link; and a graph weight update unit, used to update the parameters of the causal links with the attribution confidence as weights, and incorporate new observational evidence while retaining historical knowledge to achieve adaptive evolution of the dynamic spatiotemporal causal graph.

2. The AI-driven infrastructure risk operation optimization management system according to claim 1, characterized in that, The multimodal causal twin construction module includes: The data spatiotemporal alignment unit is used to map and synchronize the timestamps and spatial coordinates in the IoT sensor array data, the environmental monitoring data, and the personnel physiological characteristic data with the building information model data as a three-dimensional spatial reference, and output a multimodal dataset. The cross-modal latent feature extraction unit is used to learn and extract the intrinsic state and interaction behavior of physical entities based on the multimodal dataset to obtain latent feature vectors; The causal structure learning unit is used to perform time-series analysis on the latent feature vectors, quantify nonlinear causal dependencies, and search and determine the directed causal links and nonlinear function relationships of each latent feature under the condition of satisfying noncyclic constraints. The graph instantiation and weighting unit is used to instantiate and generate a dynamic spatiotemporal causal graph using the physical entities of the building information model data as nodes, the directed causal links as edges, and the influence intensity of the nonlinear function relationship as weights.

3. The AI-driven infrastructure risk operation optimization management system according to claim 1, characterized in that, The parameterized virtual perturbation set includes: a set of state offset vectors and a set of propagation failure operators, wherein: The state offset vector set is used to simulate extreme deterioration scenarios of the physical entity's own attribute parameters in a statistical sense. The set of transmission failure operators is used to simulate the local interruption or nonlinear distortion of the causal relationship transmission path between entities in the dynamic spatiotemporal causal graph.

4. The AI-driven infrastructure risk operation optimization management system according to claim 2, characterized in that, The system also includes: The hidden feature vector is matched with the simulation evolution path, the distance between the initial state and the key feature points is calculated, the convergence degree of the current infrastructure site state to each risk evolution path is evaluated, and a path activation confidence vector is generated.

5. The AI-driven infrastructure risk operation optimization management system according to claim 1, characterized in that, The collaborative configuration optimization module includes: A multidimensional constraint space construction unit is used to transform the risk propagation probability field of the forward-looking risk model into the state transition cost of the dynamic programming model, and extract the logical dependency relationship between resource scheduling and task execution from the dynamic spatiotemporal causal graph as state constraints, together defining the solution space. The strategy iteration solution unit is used to recursively solve the state action sequence that accumulates risk cost under the condition of satisfying the state constraints within the solution space, so as to obtain the optimal cooperative defense strategy. The strategy matrix generation unit is used to map and discretize the optimal collaborative defense strategy in the time dimension to generate a time-varying risk intervention strategy matrix.

6. The AI-driven infrastructure risk operation optimization management system according to claim 4, characterized in that, The instruction parsing and digital prescription generation module includes: The strategy matrix parsing unit is used to decode the time-varying risk intervention strategy matrix into structured intervention rule primitives; The dynamic risk scenario matching unit is used to identify dynamic risk scenarios by activating the confidence vector with the path, and to match the optimal intervention rule from the structured intervention rule primitives. The digital prescription generation unit is used to call the real-time state snapshot of the dynamic spatiotemporal causal graph to instantiate the parameters in the optimal intervention rule, and encapsulate them into a standardized data package as an operational digital prescription and send it to the field execution terminal.

7. The AI-driven infrastructure risk operation optimization management system according to claim 6, characterized in that, The standardized data package includes: a set of work instructions, risk context data, and performance verification parameters, wherein: The set of operation instructions is used to define the intervention actions, parameter thresholds, and execution sequence for the target physical entity; The risk context data is used to encapsulate the activated risk evolution path, the real-time status of the risk source node, and the expected loss assessment after intervention failure. The effectiveness verification parameters are used to specify the monitoring indicators and state transition objectives that must be met for the intervention to be successful.

8. An AI-driven infrastructure risk operation optimization management method, applied to the AI-driven infrastructure risk operation optimization management system as described in any one of claims 1-7, characterized in that, The method includes: The system integrates multimodal data streams from the infrastructure construction site in real time. These multimodal data streams include building information model data, IoT sensor array data, environmental monitoring data, and personnel physiological data. The system then performs causal network deduction to extrapolate the nonlinear causal dependencies between various physical entities and constructs a dynamic spatiotemporal causal graph. Based on the dynamic spatiotemporal causal graph, virtual perturbations are applied to key nodes to conduct counterfactual simulations, and the risk evolution path and cascading effects caused by single-point risks are deduced to construct a forward-looking risk model. Using the aforementioned forward-looking risk model as a constraint and coupling it with the resource dependencies of the dynamic spatiotemporal causal graph, a dynamic programming model is constructed to solve for the optimal collaborative defense strategy under different resource constraints, and output a time-varying risk intervention strategy matrix. The time-varying risk intervention strategy matrix is ​​parsed into an instruction paradigm, and combined with the real-time state snapshot of the dynamic spatiotemporal causal graph, a digital prescription for operation is generated for a specific dynamic risk scenario. The evolution of the field status after applying the digital prescription is continuously monitored. The actual evolution results are compared with the prediction results of the prospective risk model. The effectiveness deviation of the intervention measures is calculated and attribution analysis is performed. The causal link weights in the dynamic spatiotemporal causal graph are adaptively updated based on the attribution confidence.

Images