Artificial intelligence-based emergency power supply three-dimensional simulation method and system

Abstract

The present application relates to the technical field of emergency power supply simulation, and particularly relates to an emergency power supply three-dimensional simulation method and system based on artificial intelligence. A three-dimensional simulation model is constructed by acquiring topology structure data, equipment operation state data and space environment data, state evolution data is generated by time sequence feature extraction and evolution reasoning on the equipment operation state data and the model is dynamically updated, a power supply scheme candidate set is obtained by simulating and calculating the power supply path based on power supply constraint rules and load distribution strategies, a knowledge structure associated with historical fault events is established and potential risk nodes and risk propagation paths are identified by association reasoning to perform risk assessment to generate an optimal power supply scheme, and finally the optimal scheme execution process is visualized and rendered in the three-dimensional simulation model and user interaction instructions are collected to adaptively adjust the rules and strategies. The present application improves the accuracy and dynamic adaptability of the emergency power supply scheme.

Description

Technical Field

[0001] This invention relates to the field of emergency power supply simulation technology, and in particular to an artificial intelligence-based three-dimensional simulation method and system for emergency power supply. Background Technology

[0002] In the field of emergency power supply system simulation, existing technologies mainly rely on two-dimensional planar diagrams or static three-dimensional models for power supply scheme display and evaluation. The conventional approach is for technicians to first collect basic data such as electrical wiring diagrams and equipment parameter tables for the emergency power supply system. They then manually or with the aid of software, draw a two-dimensional topology diagram, marking equipment locations, line connections, and key parameters. Based on this, and combining preset power supply constraints (such as line capacity and equipment rated power) and fixed load allocation strategies, they deduce the power supply path through manual experience or simple algorithms, and output several candidate schemes. These schemes are usually presented in the form of tables or two-dimensional flowcharts for management personnel to use as a reference for decision-making. Some advanced schemes introduce three-dimensional modeling tools to construct static three-dimensional scenes based on on-site survey data to display equipment layout and environmental overview; however, the model content is fixed and does not include dynamic operational information.

[0003] Existing conventional practices have significant drawbacks. On the one hand, static models cannot reflect the time-varying characteristics of equipment states. For example, key parameters that evolve over time, such as battery power degradation and generator load fluctuations, are not included in the simulation process, leading to deviations between the generated power supply scheme and the actual operating scenario, making it difficult to cope with sudden failures or load changes. On the other hand, scheme evaluation relies heavily on human experience and lacks systematic identification of potential risks. For instance, the cascading failure paths that may be triggered by the failure of a single piece of equipment cannot be automatically predicted, making it difficult to guarantee the safety and robustness of the final scheme. Furthermore, two-dimensional or static three-dimensional displays have poor interactivity, preventing users from intuitively participating in scheme adjustments and hindering the flexibility and efficiency of emergency decision-making. Summary of the Invention

[0004] This invention provides an artificial intelligence-based three-dimensional simulation method and system for emergency power supply, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides an artificial intelligence-based three-dimensional simulation method for emergency power supply, comprising: Acquire topology data, equipment operating status data, and spatial environment data of the emergency power supply system, and construct a three-dimensional simulation model of the emergency power supply system based on the topology data, equipment operating status data, and spatial environment data; The device operation status data is subjected to time-series feature extraction and evolutionary reasoning to generate state evolution data of each device node in the future time period, and the state evolution data is mapped to the corresponding device node in the three-dimensional simulation model, and the three-dimensional simulation model is dynamically updated. Based on power supply constraint rules and load allocation strategies, the power supply paths in the three-dimensional simulation model are simulated and calculated to obtain a set of candidate power supply schemes; A knowledge structure containing the correlation between the state evolution data and historical fault events is established. Through correlation reasoning, potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set are identified in the knowledge structure. Based on the potential risk nodes and the risk propagation paths, the power supply scheme candidate set is risk-assessed to generate the optimal power supply scheme. The execution process of the optimal power supply scheme is visualized and rendered in the three-dimensional simulation model, and the user's interactive operation commands for the execution process are collected. The power supply constraint rules and the load allocation strategy are adaptively adjusted using the interactive operation commands.

[0006] The process involves extracting temporal features and performing evolutionary reasoning on the equipment operating status data to generate state evolution data for each equipment node within a future time period. This state evolution data is then mapped to the corresponding equipment nodes in the 3D simulation model. Dynamic updates to the 3D simulation model include: The equipment operation status data is decomposed into multi-scale time series, and the long-term trend component, periodic component and random disturbance component are analyzed and recombined. The baseline evolution direction of each equipment node is projected and mapped according to the long-term trend component. Based on the periodic component identification device's cyclic mode, and by coupling the cyclic mode with the reference evolution direction, a steady-state evolution trajectory of the device node is generated. A disturbance propagation network is constructed using the random disturbance components. The transmission relationship and influence weight of the random disturbance components between device nodes are decoupled from their source. The steady-state evolution trajectory is adaptively fused based on the transmission relationship and influence weight to obtain a dynamic evolution trajectory that considers the coupling effect between nodes. The state parameters of the device nodes at each moment in the dynamic evolution trajectory are aggregated to form state evolution data, and a spatial mapping relationship is established between the topological location information of the device nodes and the corresponding device nodes in the three-dimensional simulation model. Based on the spatial mapping relationship, the state parameters and visual representations of the corresponding device nodes are updated in chronological order in the three-dimensional simulation model to achieve dynamic updating of the three-dimensional simulation model.

[0007] Based on power supply constraint rules and load allocation strategies, the power supply paths in the three-dimensional simulation model are simulated and calculated to obtain a candidate set of power supply schemes, including: The power supply constraint rules are transformed into a multi-dimensional constraint space, and a feasible domain boundary for the power supply path is established in the multi-dimensional constraint space. The topology of the connection path between the power supply node and the load node in the three-dimensional simulation model is filtered through the feasible domain boundary to generate an initial set of feasible paths. For each path in the initial feasible path set, a load-path mapping matrix is ​​constructed based on the load allocation strategy, and the power flow distribution of each path under different load allocation modes is calculated using the mapping matrix. The power flow distribution is substituted into the multidimensional constraint space for constraint verification, the path state points that trigger the constraint boundary are identified, and the path state points are adaptively adjusted so that the power flow distribution returns to the interior of the feasible domain boundary. Based on the regulated power flow distribution, the resource utilization efficiency index and safety margin index of each power supply path under the load allocation strategy are calculated, and the resource utilization efficiency index and safety margin index are weighted and fused to obtain the comprehensive evaluation value of each power supply path. Based on the comprehensive evaluation value, the initial feasible path set is optimized and sorted, and power supply paths that meet the evaluation threshold and their corresponding load allocation schemes are selected and combined to form a power supply scheme candidate set.

[0008] For each path in the initial feasible path set, a load-path mapping matrix is ​​constructed based on the load allocation strategy, and the power flow distribution of each path under different load allocation modes is calculated using the mapping matrix, including: The load demand vector and path capacity vector are projected backward from the load allocation strategy, and a mapping relationship matrix representing the allocation weight between load nodes and power supply paths is constructed based on the coupling relationship between the load demand vector and the path capacity vector. The mapping relationship matrix is ​​decoupled by feature to obtain the load space basis vector and the path space basis vector. Multiple load allocation patterns are generated by multi-order coupling and recombination of the load space basis vector and the path space basis vector. For each load distribution mode, the distribution weights in the mapping matrix are calculated and transformed into the actual load carrying capacity of each power supply path. The actual load carrying capacity is then integrated and excited into the electrical topology network corresponding to the initial feasible path set. The power flow distribution on each path is obtained by solving the power flow equation. Cross-validate the power flow distribution with the allocation weights in the mapping matrix, invert and lock out the path nodes where the power distribution deviates, and dynamically coordinate their allocation weights until the power flow distribution and the allocation weights reach a convergence state, thus obtaining the power flow distribution of each path under different load allocation modes.

[0009] By using associative reasoning to identify potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set within the knowledge structure, and by performing a risk assessment on the power supply scheme candidate set based on the potential risk nodes and risk propagation paths, the optimal power supply scheme is generated, including: The network topology feature vectors of each candidate scheme in the power supply scheme candidate set are input into the knowledge structure, and a risk reasoning graph is constructed in the knowledge structure. By performing correlation reasoning in the risk reasoning graph, potential risk nodes in each candidate scheme that match historical failure modes are identified. For the potential risk node, the risk causal chain associated with it is traced in the risk reasoning graph, and the node propagation relationship in the risk causal chain is converted into a risk propagation intensity matrix. By decoupling the risk propagation intensity matrix by features, the risk propagation path starting from the potential risk node is obtained. The potential risk nodes and the risk propagation paths are integrated and incentivized into the power supply network topology corresponding to each candidate scheme. By performing multi-level coupling and reorganization in the power supply network topology, the diffusion pattern and cascading impact range of the risk in the network are deduced. Based on the aforementioned diffusion trend and the cascading impact range, a comprehensive risk assessment function is constructed that includes risk coverage and risk propagation rate. The comprehensive risk assessment function is then used to quantitatively evaluate each candidate scheme in the power supply scheme candidate set to obtain the risk assessment value of each candidate scheme. The candidate schemes in the power supply scheme candidate set are ranked and selected according to the risk assessment value, and the candidate scheme with the lowest risk assessment value is determined as the optimal power supply scheme.

[0010] The potential risk nodes and the risk propagation paths are fused and incentivized into the power supply network topology corresponding to each candidate scheme. Through multi-level coupling and reorganization within the power supply network topology, the diffusion pattern and cascading impact range of the risk in the network are deduced, including: The node state tensor of potential risk nodes and the propagation strength tensor of risk propagation paths are coupled and superimposed into the topology matrix of the power supply network topology corresponding to each candidate scheme, forming a risk-incentivized topology that carries risk information. The coupling relationships between nodes in the risk incentive topology are decomposed hierarchically to obtain direct coupling subgraphs and indirect coupling subgraphs. Based on the topological transitivity between the direct coupling subgraphs and indirect coupling subgraphs, a multi-order coupling recombination tensor is constructed. The risk state in the risk-incentive topology is propagated in a temporal iterative manner by the multi-order coupled recombination tensor to obtain a dynamic evolution sequence of risk radiating outward along the topological path, and the spatial diffusion pattern of risk is extracted from the dynamic evolution sequence. Based on the aforementioned spatial diffusion trend, the node links where risk transmission occurs are traced back and located. By performing topological closure operations on the node links, the set of network nodes affected by the risk through cascading effects is delineated, thus obtaining the cascading impact range of the risk.

[0011] The execution process of the optimal power supply scheme is visualized and rendered in the three-dimensional simulation model, and user interaction commands for the execution process are collected. The power supply constraint rules and load allocation strategy are adaptively adjusted using these interaction commands, including: The power supply path topology and time-series power flow data in the optimal power supply scheme are mapped to the rendering space of the three-dimensional simulation model, and a dynamic visualization scene reflecting the execution process of the optimal power supply scheme is generated through multi-level visual encoding. An interactive perception layer is deployed in the dynamic visualization scenario. The interactive perception layer captures the user's interactive operation behavior for power supply nodes, power supply paths and load distribution status in real time. The interactive operation behavior is subjected to intent recognition and semantic parsing to generate interactive operation instructions containing adjustment targets and adjustment parameters. The user's intention to modify the power supply constraints and load allocation weights is projected backward from the interactive operation instructions, and the modification intention is converted into constraint rule adjustment vector and strategy weight adjustment vector. The constraint rule adjustment vector is fused and excited into the constraint space boundary of the power supply constraint rule, and the strategy weight adjustment vector is fused and excited into the weight allocation matrix of the load allocation strategy. The power supply constraint rule and the load allocation strategy are adaptively adjusted through dynamic coordination.

[0012] A second aspect of this invention provides an artificial intelligence-based three-dimensional simulation system for emergency power supply, comprising: The modeling unit is used to acquire topology data, equipment operating status data, and spatial environment data of the emergency power supply system, and to construct a three-dimensional simulation model of the emergency power supply system based on the topology data, equipment operating status data, and spatial environment data. An evolution unit is used to extract time-series features and perform evolutionary reasoning on the equipment operation status data, generate state evolution data of each equipment node in the future time period, map the state evolution data to the corresponding equipment node in the three-dimensional simulation model, and dynamically update the three-dimensional simulation model. The simulation unit is used to perform simulation calculations on the power supply path in the three-dimensional simulation model based on power supply constraint rules and load distribution strategies, and to obtain a set of candidate power supply schemes; The evaluation unit is used to establish a knowledge structure containing the correlation between the state evolution data and historical fault events, identify potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set through correlation reasoning in the knowledge structure, and perform risk assessment on the power supply scheme candidate set based on the potential risk nodes and the risk propagation paths to generate the optimal power supply scheme. The adjustment unit is used to visualize and render the execution process of the optimal power supply scheme in the three-dimensional simulation model, collect user interaction operation commands for the execution process, and adaptively adjust the power supply constraint rules and the load allocation strategy using the interaction operation commands.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] Compared with existing technologies, this solution achieves several groundbreaking and beneficial effects in the field of emergency power supply simulation. By simultaneously acquiring topology, equipment status, and spatial environment data and constructing a three-dimensional simulation model, it can realistically reflect the physical layout and operating environment of complex power supply systems, providing a high-fidelity digital twin foundation for subsequent analysis and significantly improving the accuracy and reliability of simulation results.

[0016] By extracting time-series features and performing evolutionary reasoning on equipment operating status data, future state evolution data is generated and the 3D model is dynamically updated. This gives the simulation process a forward-looking and dynamically adaptable time dimension, enabling it to predict equipment state change trends and avoid decision-making lag problems caused by static simulation. Based on power supply constraint rules and load allocation strategies, power supply paths are simulated and calculated to generate a candidate set of power supply schemes, providing decision-makers with a multi-path selection space and significantly improving the flexibility and robustness of emergency response.

[0017] A knowledge structure is established that includes correlations between state evolution data and historical failure events. Through correlational reasoning, potential risk nodes and risk propagation paths in candidate solutions are identified, enabling a comprehensive risk assessment of power supply schemes. This allows for the early elimination of high-risk schemes, effectively reducing the probability of failure and the extent of losses during actual implementation. The optimal power supply scheme is generated by combining risk analysis, ensuring that emergency power supply strategies achieve the best balance between safety and efficiency. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the artificial intelligence-based three-dimensional simulation method for emergency power supply according to an embodiment of the present invention. Figure 2 This is a system architecture diagram of an embodiment of the present invention; Figure 3 This is a flowchart illustrating the simulation entity modeling and scene construction process in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the data storage and management process in an embodiment of the present invention. Figure 5 This is a flowchart of the artificial intelligence engine according to an embodiment of the present invention; Figure 6 This is a flowchart illustrating the functional simulation and verification process of an embodiment of the present invention. Figure 7 This is a flowchart of the interactive and visualization output layer in an embodiment of the present invention. Detailed Implementation

[0019] 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, 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.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 This is a flowchart illustrating the artificial intelligence-based three-dimensional simulation method for emergency power supply according to an embodiment of the present invention.

[0022] Artificial intelligence-based 3D simulation methods for emergency power supply include: Acquire topology data, equipment operating status data, and spatial environment data of the emergency power supply system, and construct a three-dimensional simulation model of the emergency power supply system based on the topology data, equipment operating status data, and spatial environment data; The device operation status data is subjected to time-series feature extraction and evolutionary reasoning to generate state evolution data of each device node in the future time period, and the state evolution data is mapped to the corresponding device node in the three-dimensional simulation model, and the three-dimensional simulation model is dynamically updated. Based on power supply constraint rules and load allocation strategies, the power supply paths in the three-dimensional simulation model are simulated and calculated to obtain a set of candidate power supply schemes; A knowledge structure containing the correlation between the state evolution data and historical fault events is established. Through correlation reasoning, potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set are identified in the knowledge structure. Based on the potential risk nodes and the risk propagation paths, the power supply scheme candidate set is risk-assessed to generate the optimal power supply scheme. The execution process of the optimal power supply scheme is visualized and rendered in the three-dimensional simulation model, and the user's interactive operation commands for the execution process are collected. The power supply constraint rules and the load allocation strategy are adaptively adjusted using the interactive operation commands.

[0023] In one optional implementation, the device operating status data is subjected to time-series feature extraction and evolutionary reasoning to generate state evolution data of each device node in a future time period, and the state evolution data is mapped to the corresponding device node in the three-dimensional simulation model. Dynamic updates to the three-dimensional simulation model include: The equipment operation status data is decomposed into multi-scale time series, and the long-term trend component, periodic component and random disturbance component are analyzed and recombined. The baseline evolution direction of each equipment node is projected and mapped according to the long-term trend component. Based on the periodic component identification device's cyclic mode, and by coupling the cyclic mode with the reference evolution direction, a steady-state evolution trajectory of the device node is generated. A disturbance propagation network is constructed using the random disturbance components. The transmission relationship and influence weight of the random disturbance components between device nodes are decoupled from their source. The steady-state evolution trajectory is adaptively fused based on the transmission relationship and influence weight to obtain a dynamic evolution trajectory that considers the coupling effect between nodes. The state parameters of the device nodes at each moment in the dynamic evolution trajectory are aggregated to form state evolution data, and a spatial mapping relationship is established between the topological location information of the device nodes and the corresponding device nodes in the three-dimensional simulation model. Based on the spatial mapping relationship, the state parameters and visual representations of the corresponding device nodes are updated in chronological order in the three-dimensional simulation model to achieve dynamic updating of the three-dimensional simulation model.

[0024] When performing multi-scale time-series decomposition on equipment operating status data, methods such as empirical mode decomposition or wavelet multi-resolution analysis are used to decompose the collected time-series signals of voltage, current, power, and temperature of each equipment node into three types of components: long-term trend components reflecting the long-term degradation or load growth trend of the equipment, periodic components reflecting daily or weekly cycles, and random disturbance components introduced by sudden disturbances or measurement noise. The long-term trend component is extracted from the original time series through low-frequency filtering or polynomial fitting. Based on the slope and curvature information of this component, a reference evolution direction is projected onto each equipment node, which represents the macroscopic change trend of the node's state parameters within a future time window. The reference evolution direction is stored in vector form, and its components correspond to the first derivative estimates of each state parameter on the time axis, serving as the basic reference for subsequent trajectory generation.

[0025] The identification of periodic components involves determining the main period through spectral analysis or autocorrelation function, and then extracting the amplitude and phase information of each periodic mode using Fast Fourier Transform or Hilbert-Huang Transform to form a cyclic mode description of equipment operation. When coupling the cyclic mode with the baseline evolution direction, the predicted values ​​of the periodic components are superimposed on the trend prediction values ​​given by the baseline evolution direction, thus modeling the steady-state evolution trajectory of the equipment node under normal operating conditions. The steady-state evolution trajectory describes the expected sequence of changes in the equipment node state parameters over time without abnormal disturbances, providing a stable baseline for subsequent random disturbance corrections.

[0026] The processing of random disturbance components is the most critical step in the entire time-series evolution inference. After extraction, the random disturbance components are not processed independently for each device node. Instead, a disturbance propagation network is constructed to characterize the transmission relationship of disturbances between device nodes. The disturbance propagation network uses device nodes as vertices and electrical connections or physical coupling relationships between nodes as edges. Transmission coefficients are attached to the edges to quantify the impact weight of disturbance propagation from one node to another. The calculation of the impact weights comprehensively considers the electrical distance between nodes, impedance ratios, and the actual propagation intensity in historical disturbance events. By using methods such as Granger causality tests or mutual information analysis, the causal transmission relationships between nodes are decoupled from historical data, ensuring that the edge weight settings of the disturbance propagation network are based on data-driven principles rather than simply relying on prior topological assumptions.

[0027] After constructing the disturbance propagation network and calibrating the influence weights, the steady-state evolution trajectory is adaptively fused based on the transmission relationship and influence weights. Specifically, for each target device node, its own random disturbance component is weighted and superimposed with the propagated disturbance component from adjacent nodes after weight attenuation, forming a comprehensive disturbance correction term for that node. This correction term is then superimposed onto the steady-state evolution trajectory to obtain a dynamic evolution trajectory considering the coupling effect between nodes. The dynamic evolution trajectory stores the predicted state parameter values ​​of each device node in discrete time steps in the time dimension. The time step size can be configured to minutes or hours according to the simulation accuracy requirements to adapt to the simulation timeliness requirements of different emergency power supply scenarios.

[0028] After the dynamic evolution trajectory is generated, the device node state parameters at each time point are aggregated according to node number and timestamp to form a structured state evolution dataset. Each record in the state evolution dataset contains a unique identifier for the device node, a timestamp, the predicted value of each state parameter, and a prediction confidence interval. The confidence interval is estimated by the statistical distribution characteristics of the random disturbance component and is used to assist decision-makers in judging the reliability of the evolution prediction in the form of uncertainty visualization in the three-dimensional simulation model.

[0029] When establishing spatial mapping relationships, the topological location information of device nodes (including the logical coordinates of nodes in the power supply network and their corresponding physical geographic coordinates) is matched with the three-dimensional spatial coordinates of each device node in the 3D simulation model. The matching process uses the node's unique identifier as the primary key for precise mapping. For cases with positional deviations, a coordinate transformation matrix is ​​used to map the logical topological coordinates to the 3D scene coordinate system, ensuring that state evolution data can be accurately positioned on the corresponding device nodes in the 3D simulation model, avoiding the problem of misaligned state parameter rendering due to coordinate system inconsistencies.

[0030] Based on the spatial mapping relationship, the state parameters and visual representations of the corresponding device nodes are updated frame by frame in chronological order within the 3D simulation model. Updates to state parameters include updating key indicators such as voltage amplitude, current load factor, and device health index. Updates to visual representations are achieved through rendering techniques such as color mapping, geometric deformation, or dynamic annotation. For example, device nodes with excessively high load factors are rendered as a red warning color, nodes with health indices falling below a threshold are given a flashing effect, and nodes within the normal operating range are displayed stably in green. The dynamic updates of the 3D simulation model are driven by an asynchronous refresh mechanism. Whenever new time step state evolution data is ready, the rendering update of the corresponding node is triggered, ensuring the temporal synchronization and consistency between the simulation image and the state evolution data. This provides a real-time and accurate 3D simulation foundation for subsequent power supply path simulation calculations and risk assessments.

[0031] In one optional implementation, based on power supply constraint rules and load allocation strategies, the power supply paths in the three-dimensional simulation model are simulated and calculated to obtain a candidate set of power supply schemes, including: The power supply constraint rules are transformed into a multi-dimensional constraint space, and a feasible domain boundary for the power supply path is established in the multi-dimensional constraint space. The topology of the connection path between the power supply node and the load node in the three-dimensional simulation model is filtered through the feasible domain boundary to generate an initial set of feasible paths. For each path in the initial feasible path set, a load-path mapping matrix is ​​constructed based on the load allocation strategy, and the power flow distribution of each path under different load allocation modes is calculated using the mapping matrix. The power flow distribution is substituted into the multidimensional constraint space for constraint verification, the path state points that trigger the constraint boundary are identified, and the path state points are adaptively adjusted so that the power flow distribution returns to the interior of the feasible domain boundary. Based on the regulated power flow distribution, the resource utilization efficiency index and safety margin index of each power supply path under the load allocation strategy are calculated, and the resource utilization efficiency index and safety margin index are weighted and fused to obtain the comprehensive evaluation value of each power supply path. Based on the comprehensive evaluation value, the initial feasible path set is optimized and sorted, and power supply paths that meet the evaluation threshold and their corresponding load allocation schemes are selected and combined to form a power supply scheme candidate set.

[0032] When transforming power supply constraints into a multi-dimensional constraint space, it is necessary to unify the mathematical expression of various rules, such as voltage deviation constraints, current capacity constraints, power balance constraints, and line thermal stability constraints. Each type of constraint corresponds to a dimension in the multi-dimensional constraint space, and the upper and lower bounds of each dimension jointly determine the feasible region boundary of the power supply path. Mathematically, the feasible region boundary can be represented as the intersection of a set of inequality constraints. By substituting each connection path between all power supply nodes and load nodes in the 3D simulation model into this set of inequalities, paths that exceed the bounds in any dimension are eliminated, and paths that satisfy the constraints in all dimensions are retained, forming an initial set of feasible paths. During the topology screening process, the physical reachability of the paths needs to be verified by combining the spatial positional relationships and electrical connection relationships of the equipment nodes in the 3D simulation model, ensuring that each path in the initial set of feasible paths has physical connectivity under the actual topology structure of the simulation model.

[0033] For each path in the initial feasible path set, a load-path mapping matrix is ​​constructed based on the load allocation strategy. Let the initial feasible path set contain... There are [number] paths in the system. For each load node, the mapping matrix is... for 3D matrix, matrix elements Indicates the first The path undertakes the first The allocation ratio of power supply tasks to each load node is determined to satisfy the needs of each load node. The normalization condition is that the sum of all path allocation ratios equals 1. Load allocation strategies can generate different mapping matrix instances based on various modes such as equal allocation, allocation by capacity ratio, and allocation by inverse path impedance, thus covering power flow distribution scenarios under multiple load allocation modes. By multiplying the mapping matrix with the real-time power demand vector of each load node, the active and reactive power carried by each path under each load allocation mode are calculated, forming a power flow distribution result set.

[0034] When power flow distribution is substituted into the multi-dimensional constraint space for constraint verification, each path's power flow under each allocation mode is examined to see if it touches the feasible region boundary. When the power flow of a path under a specific load allocation mode exceeds the upper limit of the current capacity constraint, or causes the node voltage deviation to exceed the allowable range, the corresponding state point of that path under that allocation mode is identified as a path state point that triggers the constraint boundary. For adaptive control of these path state points, a constraint repair strategy is used to locally adjust the load allocation ratio: for paths exceeding the current limit, the load allocation ratio they bear is reduced, and the excess is transferred to other paths with remaining capacity; for nodes exceeding the voltage deviation limit, the power injection amount of the paths connected to that node is adjusted to restore the node voltage to within the constraint range. The control process proceeds iteratively, with constraint verification re-executed after each iteration until all path state points return to within the feasible region boundary, or until the preset maximum number of iterations is reached. If, after the maximum number of iterations, there are still unrepairable over-limit state points, the corresponding path and load allocation mode combination are removed from the candidate set.

[0035] Based on the regulated power flow distribution, the resource utilization efficiency index and safety margin index of each power supply path under the load allocation strategy are calculated. Resource utilization efficiency index Reflecting the The ratio of the actual carrying capacity of a route to its rated capacity is used to measure the adequacy of route utilization. The calculation method is as follows: ,in Post-regulation path Actual apparent power For path Rated apparent power capacity. Safety margin specifications. Reflecting the relative distance of the path from the constraint boundary at the current power flow level, the calculation method is as follows: That is, the proportion of the rated capacity that has not yet been used. A higher value indicates a more ample safety margin for the power supply path in response to sudden load fluctuations. The resource utilization efficiency index and the safety margin index are weighted and fused to obtain a comprehensive evaluation value for each power supply path. The calculation method is as follows ,in This is a weighting coefficient for resource utilization efficiency. For the safety margin, the weighting coefficient is... The weighting coefficient can be configured according to the priority strategy of the emergency power supply scenario: in scenarios where power resources are scarce and it is necessary to maximize utilization, the weighting coefficient should be appropriately increased. In scenarios with high security requirements and where sufficient redundancy needs to be maintained, appropriately increase... .

[0036] Based on the comprehensive evaluation value When optimizing and ranking the initial set of feasible paths, all paths that have passed constraint checks and adaptive adjustments are arranged from highest to lowest based on their comprehensive evaluation scores. An evaluation threshold is set. Only retain comprehensive evaluation values ​​that are not lower than The selected paths proceed to the subsequent candidate set construction stage. For each selected path, its corresponding optimal load allocation scheme is extracted, which is the mapping matrix instance corresponding to the allocation mode with the highest comprehensive evaluation value among all regulated and verified load allocation modes. The selected paths are combined with their corresponding optimal load allocation schemes to form a path-scheme pair, and multiple path-scheme pairs together constitute the power supply scheme candidate set. Each candidate scheme in the candidate set contains a clear power supply path topology, the power allocation ratio of each load node, the power flow distribution result of the path, and the corresponding comprehensive evaluation value, providing complete scheme input for subsequent knowledge structure association reasoning and risk assessment.

[0037] In practical engineering applications, the initial feasible path set is often large in size and has a large number of paths. The product of the number of load distribution patterns determines the computational complexity of constraint verification. To improve simulation efficiency, a graph search-based pruning strategy can be introduced during the topology selection stage, prioritizing connected paths with short electrical distances and low path impedance, thus compressing the initial feasible path set. During the constraint verification stage, parallel computation is employed to simultaneously verify multiple distribution patterns for multiple paths, shortening the iteration convergence time. Through this simulation process, the final power supply scheme candidate set, while satisfying multi-dimensional constraints, comprehensively considers both resource utilization efficiency and safety margin, providing high-quality candidate input for subsequent optimal power supply scheme identification and risk assessment.

[0038] In one optional implementation, for each path in the initial feasible path set, a load-path mapping matrix is ​​constructed according to the load allocation strategy, and the power flow distribution of each path under different load allocation modes is calculated using the mapping matrix, including: The load demand vector and path capacity vector are projected backward from the load allocation strategy, and a mapping relationship matrix representing the allocation weight between load nodes and power supply paths is constructed based on the coupling relationship between the load demand vector and the path capacity vector. The mapping relationship matrix is ​​decoupled by feature to obtain the load space basis vector and the path space basis vector. Multiple load allocation patterns are generated by multi-order coupling and recombination of the load space basis vector and the path space basis vector. For each load distribution mode, the distribution weights in the mapping matrix are calculated and transformed into the actual load carrying capacity of each power supply path. The actual load carrying capacity is then integrated and excited into the electrical topology network corresponding to the initial feasible path set. The power flow distribution on each path is obtained by solving the power flow equation. Cross-validate the power flow distribution with the allocation weights in the mapping matrix, invert and lock out the path nodes where the power distribution deviates, and dynamically coordinate their allocation weights until the power flow distribution and the allocation weights reach a convergence state, thus obtaining the power flow distribution of each path under different load allocation modes.

[0039] After obtaining the initial set of feasible paths, a quantified allocation relationship between load and path needs to be established for each path in the set. The load allocation strategy includes information such as the priority weight of each load node, the rated power demand, and the upper limit of the path's transmission capacity. The load demand vector is then projected inversely from the load allocation strategy. and path capacity vector ,in The Each component Indicates the first Active power demand of each load node The Each component Indicates the first The maximum power capacity that a path can carry. The so-called "reverse projection" refers to reversing the process of parsing the demand parameters and path parameters of each node from the existing load allocation strategy description, rather than directly issuing scheduling instructions downwards. This process requires combining the node connection relationships in the topology data to complete parameter reconstruction.

[0040] Based on load demand vector With path capacity vector The coupling relationships between them are used to construct a mapping matrix. The rows of the matrix correspond to the paths in the initial feasible path set, and the columns correspond to the load nodes in the system. Matrix elements... Indicates the first The path undertakes the first The allocation ratio of power supply tasks for each load node, with a value range of [value missing]. Furthermore, for the same load node, the sum of the allocation ratios of all paths satisfies the normalization constraint, i.e. ,in This represents the total number of paths in the initial feasible path set. This represents the total number of load nodes in the system. The initial value of the matrix is ​​determined by the priority weight and path capacity ratio in the load allocation strategy. Paths with larger capacities receive higher proportional weights during initial allocation, thus ensuring that the initial mapping matrix has a reasonable physical meaning.

[0041] For the mapping relation matrix The purpose of feature decoupling is to separate the coupling information between the load space and the path space, in order to generate diverse load allocation patterns. Performing singular value decomposition yields a left singular matrix, a singular value diagonal matrix, and a right singular matrix. The column vectors of the left singular matrix form the path space basis vector set. The column vectors of the right singular matrix form the basis vector set of the load space. , Let be the rank of the matrix. The path space basis vectors describe the main directions of change in the allocation patterns for different paths, while the load space basis vectors describe the main response patterns of different load nodes when allocated. By performing multi-order coupling and recombination on these two sets of basis vectors—that is, selecting basis vector components of different orders and resynthesizing the allocation matrix according to different linear combination coefficients—multiple load allocation patterns are generated. Each load allocation pattern corresponds to a specific allocation weight configuration scheme. Different patterns differ in load-bearing ratios, thus covering various possibilities from balanced allocation to centralized allocation, providing sufficient input samples for subsequent power flow simulation calculations.

[0042] For each load allocation pattern, the allocation weights in the mapping matrix are calculated and transformed into the actual load capacity of each power supply path. Specifically, for the first... The actual load capacity of the path. This is obtained by weighted summation of the allocation ratio of all load nodes and their corresponding load demands under the current allocation mode for this path. ,in For the first Active power demand of each load node. Actual load capacity of each path. The active power flow is injected as an excitation into the electrical topology network corresponding to the initial feasible path set. This electrical topology network consists of the node admittance matrix and branch parameters extracted from the 3D simulation model, including the voltage magnitude and phase angle of each node, as well as the resistance and reactance parameters of the branches. By solving the power flow equations, the active power flow on each path is obtained. and reactive power flow The power flow equations are solved using the Newton-Raphson iterative method, with nodal power imbalance as the convergence criterion. Convergence occurs when the power imbalance at all nodes is below a preset threshold. When the power flow calculation is considered to have converged, the power flow distribution results for each path are output.

[0043] Cross-validating the power flow distribution obtained from power flow calculations with the allocation weights in the mapping matrix is ​​a crucial step in ensuring the physical consistency of the allocation scheme. The core logic of cross-validation lies in: if the actual power flow of a certain path... If there is a deviation between the power distribution at a path node and the power it is expected to receive according to its assigned weights, it indicates that the weighting at that path node is unreasonable and requires dynamic coordination. Define a power distribution deviation index. ,when Exceeding the deviation tolerance threshold When a path node is identified as a power distribution deviation node, its weight allocation is adjusted according to the direction of deviation: if the actual power flow is higher than the expected capacity, the allocation ratio for high-demand load nodes on that path is appropriately reduced; if the actual power flow is lower than the expected capacity, the corresponding allocation ratio is appropriately increased. The weight adjustment magnitude is related to the deviation amount. Proportional to the adaptive learning rate, the step size is adjusted based on the adaptive learning rate. control, In the early stages of iteration, a larger value is used to accelerate the convergence speed, and the value is gradually reduced as the number of iterations increases to ensure convergence stability.

[0044] After each weight adjustment, the actual load capacity of each path is recalculated, the power flow equations are solved again, the power flow distribution is updated, and cross-validation is performed again. This process is iterated until the power distribution deviation index of all paths is met. All are below the deviation tolerance threshold If the normalization constraints of each column in the mapping matrix are still satisfied, then the power flow distribution and allocation weights are considered to have reached a convergence state. The converged power flow distribution result is the final power flow distribution of each path under the current load allocation mode. The convergence results corresponding to all load allocation modes are summarized to form a complete power flow distribution dataset for each path under different load allocation modes, which is used for the evaluation and screening of subsequent power supply scheme candidate sets. This cross-validation and dynamic coordination mechanism effectively avoids power over-limit or under-load problems caused by improper initial allocation weight settings, and improves the reliability of simulation calculation results.

[0045] In one optional implementation, identifying potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set through associative reasoning within the knowledge structure, and performing a risk assessment on the power supply scheme candidate set based on the potential risk nodes and the risk propagation paths to generate the optimal power supply scheme includes: The network topology feature vectors of each candidate scheme in the power supply scheme candidate set are input into the knowledge structure, and a risk reasoning graph is constructed in the knowledge structure. By performing correlation reasoning in the risk reasoning graph, potential risk nodes in each candidate scheme that match historical failure modes are identified. For the potential risk node, the risk causal chain associated with it is traced in the risk reasoning graph, and the node propagation relationship in the risk causal chain is converted into a risk propagation intensity matrix. By decoupling the risk propagation intensity matrix by features, the risk propagation path starting from the potential risk node is obtained. The potential risk nodes and the risk propagation paths are integrated and incentivized into the power supply network topology corresponding to each candidate scheme. By performing multi-level coupling and reorganization in the power supply network topology, the diffusion pattern and cascading impact range of the risk in the network are deduced. Based on the aforementioned diffusion trend and the cascading impact range, a comprehensive risk assessment function is constructed that includes risk coverage and risk propagation rate. The comprehensive risk assessment function is then used to quantitatively evaluate each candidate scheme in the power supply scheme candidate set to obtain the risk assessment value of each candidate scheme. The candidate schemes in the power supply scheme candidate set are ranked and selected according to the risk assessment value, and the candidate scheme with the lowest risk assessment value is determined as the optimal power supply scheme.

[0046] After inputting the network topology feature vectors of each candidate power supply scheme into the constructed knowledge structure, features such as node connection relationships, branch impedance distribution, equipment rated parameters, and current operating status are extracted for each candidate scheme to form a structured topology feature vector representation. Within the knowledge structure, a risk reasoning graph is constructed based on historical fault events and their associated equipment state evolution data. Each node in the risk reasoning graph corresponds to a device node or functional unit in the power supply network, and directed edges between nodes represent electrical coupling relationships or fault propagation relationships between devices. The topology feature vectors of each candidate scheme are matched with the historical fault patterns in the risk reasoning graph for similarity. Through an association reasoning mechanism, subgraph structures in the risk reasoning graph that highly match the current topology features are traversed to identify device nodes in each candidate scheme that match the historical fault patterns. These nodes are marked as potential risk nodes. During the matching process, a graph isomorphism comparison method is used to quantify the structural similarity between the candidate scheme's topology subgraph and the historical fault subgraph in the knowledge structure. Nodes with similarity exceeding a preset threshold are included in the potential risk node set.

[0047] For identified potential risk nodes, causal chain tracing is performed along directed edges in the risk reasoning graph to trace the complete causal transmission path from the initial trigger node to downstream affected nodes, forming a risk causal chain. The risk causal chain describes the complete propagation sequence of a fault starting from the origin node, passing through intermediate transmission nodes, and ultimately affecting the terminal load node or critical equipment node. The propagation relationships between nodes in the risk causal chain are quantitatively calculated, using the coupling strength between nodes, historical fault propagation probability, and current equipment state deviation as comprehensive inputs to construct a risk propagation strength matrix. , where matrix elements Indicates risk from nodes propagation to nodes The intensity value, The value of is determined by comprehensively considering the degree of electrical coupling and the frequency of historical fault associations. This is applied to the risk propagation strength matrix. Perform feature decoupling by... Singular value decomposition or eigenvalue decomposition is performed to extract the main feature directions, separating high-intensity propagation paths from the overall matrix to obtain a set of risk propagation paths originating from each potential risk node. During feature decoupling, principal components with larger eigenvalues ​​are retained, while weakly coupled paths with smaller contributions are discarded, thus ensuring that the identified risk propagation paths have high representativeness and accuracy.

[0048] The identified potential risk nodes and risk propagation paths are integrated into the power supply network topology corresponding to each candidate scheme. Specifically, in the power supply network topology, the state of the potential risk nodes is set to an active state, and simulated fault excitation signals are injected along the risk propagation path to trigger the response calculation of each node in the network topology. Through multi-order coupling and reorganization in the power supply network topology, the process of risk propagation in the network is simulated. Multi-order coupling and reorganization means that in the first-order calculation, only the direct adjacent nodes of the potential risk node are considered; in the second-order calculation, the state changes of the first-order affected nodes are used as new excitation inputs to further calculate the responses of their adjacent nodes; and so on, until the diffusion effect tends to converge or reaches the preset maximum diffusion order. Each stage of the coupling and reorganization process is based on real-time status data of the current network topology, ensuring that the simulation results accurately reflect the dynamic diffusion behavior of risks in the network. After multiple stages of coupling and reorganization, the diffusion pattern of risks in the entire power supply network is simulated, including the degree of impact on each node at different diffusion stages, as well as the final cascading impact range, i.e., the set of equipment nodes affected by the risk and the corresponding load loss estimate.

[0049] Based on the diffusion pattern and the scope of cascading effects, a comprehensive risk assessment function is constructed. The comprehensive risk assessment function includes two core indicators: risk coverage. and risk propagation rate Risk coverage Defined as the ratio of the number of critical nodes affected within the cascading impact range to the total number of critical nodes in the power supply network, it is used to measure the breadth of risk propagation; risk propagation rate. Defined as the reciprocal of the average diffusion order a risk undergoes from a potential risk node to a steady state, it is used to measure the speed of risk diffusion. A higher value indicates a faster spread of risk and a greater difficulty in containing the harm in a timely manner. (Comprehensive risk assessment function) Expressed as: ; in, This is the weighting coefficient for risk coverage. This is a weighting coefficient for the risk propagation rate. and The value of is configured based on the level of concern regarding the breadth and speed of diffusion in emergency power supply scenarios, with the sum of the two being 1. For each candidate scheme in the power supply scheme candidate set, its corresponding is substituted... and Calculated risk assessment value , The lower the value, the lower the overall risk level of the candidate solution in terms of both the breadth of risk coverage and the speed of risk propagation, and the better the corresponding power supply solution is in terms of risk resistance capability.

[0050] All candidate schemes in the power supply scheme candidate set are evaluated according to risk assessment values. The risk assessment value is ranked from lowest to highest and selected based on the highest ranking. The candidate with the lowest risk assessment value was determined as the optimal power supply scheme. During the ranking process, if multiple candidate schemes have different risk assessment values... The difference is very small (the difference is less than the preset discrimination threshold). Then, auxiliary evaluation indicators are introduced to further screen these schemes. These auxiliary evaluation indicators may include the bus loss of the power supply path and the degree of power supply continuity guarantee for critical loads, to ensure that the final optimal power supply scheme achieves a reasonable balance between risk control and power supply efficiency. After the optimal power supply scheme is determined, it is transferred to the subsequent 3D simulation model for visualization rendering, allowing users to interactively review and adjust the operation.

[0051] In one optional implementation, the potential risk nodes and the risk propagation paths are fused and incentivized into the power supply network topology corresponding to each candidate scheme. Then, through multi-level coupling and reorganization within the power supply network topology, the spread of risk in the network and the scope of its cascading impact are deduced, including: The node state tensor of potential risk nodes and the propagation strength tensor of risk propagation paths are coupled and superimposed into the topology matrix of the power supply network topology corresponding to each candidate scheme, forming a risk-incentivized topology that carries risk information. The coupling relationships between nodes in the risk incentive topology are decomposed hierarchically to obtain direct coupling subgraphs and indirect coupling subgraphs. Based on the topological transitivity between the direct coupling subgraphs and indirect coupling subgraphs, a multi-order coupling recombination tensor is constructed. The risk state in the risk-incentive topology is propagated in a temporal iterative manner by the multi-order coupled recombination tensor to obtain a dynamic evolution sequence of risk radiating outward along the topological path, and the spatial diffusion pattern of risk is extracted from the dynamic evolution sequence. Based on the aforementioned spatial diffusion trend, the node links where risk transmission occurs are traced back and located. By performing topological closure operations on the node links, the set of network nodes affected by the risk through cascading effects is delineated, thus obtaining the cascading impact range of the risk.

[0052] After obtaining the power supply network topology corresponding to each candidate scheme, the node state tensor of potential risk nodes and the propagation strength tensor of risk propagation paths are coupled and superimposed into the corresponding topology matrix. Specifically, the topology matrix of the power supply network is constructed based on node adjacency relationships, and each element in the matrix describes the connectivity state and line parameters between nodes. The node state tensor of potential risk nodes includes multi-dimensional risk attributes such as the node's current failure probability, load exceedance degree, and equipment aging coefficient; the propagation strength tensor of risk propagation paths quantifies the attenuation coefficient and coupling gain when the risk propagates along each path. These two types of tensors are injected into the corresponding positions of the topology matrix in an element-wise superposition manner, transforming the original pure topology matrix into a risk-incentivized topology that simultaneously carries network connectivity information and risk incentive information. This fusion process ensures that the network structure used in subsequent simulations not only reflects physical connectivity relationships but also embeds the spatial distribution and propagation tendency of risk sources, providing accurate initial incentive conditions for multi-order coupling reorganization.

[0053] When performing a hierarchical decomposition of the coupling relationships between nodes in a risk incentive topology, a direct coupling subgraph is constructed based on the first-order adjacency relationships where nodes are directly connected. This subgraph retains only node pairs with direct physical connections and their corresponding coupling weights. Building upon this, second-order and higher-order indirect connectivity relationships are extracted progressively through matrix exponentiation operations. Node pairs that can only influence each other via intermediate nodes are categorized into the indirect coupling subgraph. The indirect coupling subgraph reflects the potential transmission paths through which risk can still have an impact after traversing multiple intermediate nodes. Based on the topological transitivity between the direct and indirect coupling subgraphs, i.e., if nodes... With nodes Direct coupling, node With nodes Direct coupling, then and There exists a second-order indirect coupling relationship, and so on, to construct multi-order coupled recombinant tensors. . It is a three-dimensional tensor, whose first dimension corresponds to the coupling order. ( ,in (where the maximum diffusion order is defined), the second and third dimensions correspond to the source and target nodes in the network, respectively, and the tensor elements are... Indicates the first Nodes under first-order coupling conditions For nodes The intensity of the influence. By weighting and superimposing the coupling subgraphs of each order, It fully characterizes the comprehensive coupling capability of risk incentive topology, which starts from any risk source node, passes through intermediate nodes of any order, and is transmitted to the target node.

[0054] Reconstructing Tensors Using Multi-Order Coupling When performing temporal iterative propagation of risk states in a risk-incentive topology, the initial risk incentive value of each potential risk node is used as the starting point for iteration. In each time step, the current risk state vector is compared with... Perform tensor shrinkage to obtain the cumulative risk value of each node in the next time step. Let the... The risk state vector of each node in the network at each time step is: Then the first The risk status update at each time step satisfies ,in for In the Matrix on a slice of order, For the first The attenuation weighting coefficient of the first-order coupling, Let be the initial risk incentive vector. A nonlinear activation function is used to constrain the risk value within a reasonable range. The iterative process continues until the norm of the difference between the risk state vectors of two adjacent time steps falls below the convergence criterion. This continues until a preset maximum number of iterations is reached. The risk state vectors at each time step are arranged sequentially to obtain a dynamic evolution sequence of risk radiating outward along the topological path. From this dynamic evolution sequence, by spatial interpolating and extracting isosurfaces from the gradient of risk value changes of each node over time, the diffusion pattern of risk in cyberspace can be inverted, that is, the directionality, rate distribution, and spatial location of high-risk clusters can be identified.

[0055] Based on the spatial diffusion pattern, when tracing back to locate the node link where risk transmission occurs, starting from the last time step of the dynamic evolution sequence, the process traces backward along the risk value gradient direction, recording the points in each time step where the risk value exceeds the preset transmission activation threshold. The nodes and their interconnections are defined as the nodes and the edges that connect them. The subgraph formed by these nodes and edges is defined as the risk transmission node link. . It precisely describes the complete set of paths through which risk propagates from the initial incentive node, passing through intermediate nodes and along which routes to the distant node. After obtaining... Then, perform topological closure operation on it: All nodes in the set form a seed set, and each node in the seed set will have a direct connection to any node in the original power supply network topology, or... All nodes reachable within order 1 are included in the closure set. The topological closure operation is implemented through breadth-first expansion. In each round of expansion, all first-order neighbors of the current nodes are added to the set. This process is repeated until the set stops growing or the expansion order is reached. The final set of closures represents the set of network nodes affected by the risk through cascading effects, constituting the cascading impact range of the risk. . The nodes in the diagram represent all power supply nodes that will be affected by a chain reaction if a potentially risky node fails under the current candidate power supply scheme. This provides a precise risk boundary basis for generating the optimal power supply scheme in the future.

[0056] The spatial diffusion pattern and cascading impact range derived from the above deductions After feedback to the risk assessment process, the safety of each candidate solution can be quantitatively differentiated. The larger the cascading impact range and the more critical load nodes covered by high-risk areas in the diffusion trend, the higher the risk assessment value of the corresponding candidate solution. Combined with the defined comprehensive risk assessment function, The proportion of the number of nodes to the total number of nodes is used as the risk coverage. The calculation basis is that the risk value in the dynamic evolution sequence exceeds The rate at which the number of nodes increases with time step is used as the risk propagation rate. This provides a quantitative source, enabling risk assessment results to fully reflect the breadth and speed of the actual spread of risk within the network.

[0057] In one optional implementation, the execution process of the optimal power supply scheme is visualized and rendered in the three-dimensional simulation model, and user interaction commands for the execution process are collected. The power supply constraint rules and the load allocation strategy are adaptively adjusted using these interaction commands, including: The power supply path topology and time-series power flow data in the optimal power supply scheme are mapped to the rendering space of the three-dimensional simulation model, and a dynamic visualization scene reflecting the execution process of the optimal power supply scheme is generated through multi-level visual encoding. An interactive perception layer is deployed in the dynamic visualization scenario. The interactive perception layer captures the user's interactive operation behavior for power supply nodes, power supply paths and load distribution status in real time. The interactive operation behavior is subjected to intent recognition and semantic parsing to generate interactive operation instructions containing adjustment targets and adjustment parameters. The user's intention to modify the power supply constraints and load allocation weights is projected backward from the interactive operation instructions, and the modification intention is converted into constraint rule adjustment vector and strategy weight adjustment vector. The constraint rule adjustment vector is fused and excited into the constraint space boundary of the power supply constraint rule, and the strategy weight adjustment vector is fused and excited into the weight allocation matrix of the load allocation strategy. The power supply constraint rule and the load allocation strategy are adaptively adjusted through dynamic coordination.

[0058] In the visualization rendering stage, the power supply path topology and time-series power flow data in the optimal power supply scheme are mapped to the rendering space of the 3D simulation model. The power supply path topology includes spatial attributes such as node coordinates, connection relationships, and equipment types, while the time-series power flow data records the active and reactive power transmission states of each path at different times. After registering these two types of data into the spatial coordinate system of the 3D scene, a dynamic visualization scene is generated through multi-level visual encoding. Specifically, multi-level visual encoding is reflected in: mapping the power transmission magnitude of the path to the thickness of the cable geometry; mapping the load rate range of the path to color gradients (blue for low load, yellow for medium load, and red for high load); mapping the direction and speed of power flow to flowing particle animation; and mapping the current operating margin of equipment nodes to the node halo radius. For key substation equipment nodes, floating labels are also overlaid to display voltage amplitude, frequency deviation, and operating status indicators in real time, ensuring that every change in power supply status during execution provides intuitive visual feedback in the 3D scene. The time-series power flow data drives scene updates frame by frame according to the simulation timeline, enabling the dynamic visualization scene to present a continuous evolution effect aligned with the simulation timeline.

[0059] An interaction perception layer is deployed in the dynamic visualization scenario to capture user actions on power supply nodes, power supply paths, and load allocation status in real time. Based on an event listening mechanism, the interaction perception layer can identify various interaction events such as mouse hover, click selection, drag adjustment, and right-click menu triggering, and records the target object identifier, operation type, and operation parameters corresponding to each interaction event. For user actions in the 3D scene, an intent recognition and semantic parsing process is executed. In the intent recognition stage, the user's operation sequence is input into a pre-trained operation intent classification model. This model takes features such as operation type, target object category, and operation timing interval as input and outputs the operation intent category, such as "increase the power limit of a certain path," "decrease the priority of a certain load node," and "prohibit a certain path from participating in power supply." In the semantic parsing stage, the identified intent category is bound to the attribute information of the target object, extracting the adjustment target (i.e., which path or node) and adjustment parameters (i.e., adjustment amount or adjustment direction), and finally encapsulating them into structured interactive operation instructions. The data structure of the interactive operation command includes four fields: operation intent label, target object index, parameter type identifier, and parameter value. Subsequent processing flows all take this structured command as input.

[0060] The inverse projection method projects the user's intention to modify power supply constraints and load allocation weights from interactive operation commands, and transforms this intention into constraint rule adjustment vectors and strategy weight adjustment vectors. The core idea of ​​inverse projection is that the user's direct operations in the visualization scenario often target specific paths or nodes, while implicitly they represent desired modifications to the boundary parameters of power supply constraint rules or the weights of load allocation strategies. Let the set of constraint parameters for the current power supply constraint rule be represented as a vector. , of which Each component Corresponding to the Boundary values ​​of the constraints; assuming the weight set of the current load allocation strategy can be represented as a vector. , of which Each component Corresponding to the The weights of each load node are assigned. For an interactive operation command, a mapping relationship is established from the operation target to constraint parameters or weight parameters based on its intent label and target object index. If the intent label belongs to a constraint operation (such as modifying the path capacity limit or adjusting the voltage deviation tolerance range), then the constraint rule adjustment vector is calculated and generated. In this process, only the components related to the target object take non-zero adjustment values, while the remaining components are zero. If the intent label belongs to a weighted operation (such as increasing or decreasing the power supply priority of a load node), then the strategy weight adjustment vector is calculated and generated. Only the component corresponding to the target load node takes a non-zero adjustment value. The magnitude of the adjustment is determined by normalizing the operating parameter values ​​to ensure that the adjustment range is proportional to the quantized input of the user operation.

[0061] Adjust the constraint rule vector The fusion of excitations into the constraint space boundary of the power supply constraint rules is specifically achieved by integrating the original constraint parameter vectors. and Updated using a weighted aggregation method, with the aggregation weights... This is used to control the impact of a single user operation on the constraint boundary, preventing drastic changes in the constraint boundary due to user errors. The updated constraint parameter vector. satisfy At the same time, Each component undergoes a physical feasibility check. If a component exceeds the preset safety allowable range, it is truncated to the corresponding upper and lower bounds to ensure that the updated constraint rules remain effective at the physical level. The policy weights are then adjusted in the vector. When integrating incentives into the weighting matrix of the load allocation strategy, firstly... The process is expanded to the corresponding rows or columns of the weight allocation matrix, and then normalization is performed to ensure that the sum of the weights in each row of the weight allocation matrix is ​​1, thus satisfying the probability normalization constraint of load allocation. The dynamic coordination mechanism is triggered after the above two types of updates are completed. The simulation calculation process of the power supply path is re-executed, a new set of power supply scheme candidates is generated based on the updated power supply constraint rules and load allocation strategy, and the risk assessment process is repeated to confirm that the adjusted scheme meets the safety requirements. If the risk assessment value of the new scheme is better than the optimal scheme before adjustment, the new scheme replaces the current optimal power supply scheme, and the updated execution process is re-rendered into the 3D simulation model, forming a closed-loop adaptive adjustment mechanism of "user interaction—parameter adjustment—scheme re-optimization—visual feedback".

[0062] Through the aforementioned closed-loop mechanism, every interactive operation by the user in the 3D visualization scene can be transmitted in a structured manner to the parameter layer of power supply constraint rules and load allocation strategies, realizing real-time human-machine collaborative optimization of emergency power supply simulation schemes and improving the flexibility and practicality of emergency power supply decision-making.

[0063] like Figures 2-7 As shown, the method further includes: The node state tensor of potential risk nodes and the propagation strength tensor of risk propagation paths are coupled and superimposed into the topology matrix of the power supply network topology corresponding to each candidate scheme, forming a risk-incentivized topology carrying risk information. The node state tensor of a potential risk node contains multi-dimensional state parameters such as the node's current voltage deviation, load factor, connectivity, and historical fault frequency. Voltage deviation is obtained by dividing the difference between the current measured voltage and the rated voltage by the rated voltage, ranging from -0.3 to +0.3 per unit. A node exceeding ±0.1 per unit is marked as medium-risk, and exceeding ±0.2 per unit is marked as high-risk. The load factor is the ratio of the node's current carrying power to its rated power, ranging from 0 to 1.2; exceeding 0.9 triggers an overload warning. Connectivity is the number of other nodes directly connected to the node, ranging from 1 to 20; nodes with a connectivity greater than 5 play a critical hub role in the topology. Historical fault frequency statistics show the cumulative number of abnormal events such as equipment tripping and protection actions that have occurred at this node in the past 30 days, ranging from 0 to 50 times. A node is assigned a higher risk weight if the number of such events exceeds 5. The node state tensor is represented as a four-dimensional vector, with each dimension corresponding to a state parameter. The vector elements are normalized and range from 0 to 1.

[0064] The propagation strength tensor of a risk propagation path describes the ability of a risk to spread along network connections, including parameters such as path impedance, transmission capacity margin, protection configuration completeness, and path redundancy. Path impedance is calculated by summing line resistance, reactance, and transformer equivalent impedance, in ohms, with a typical range of 0.01 to 5 ohms. Transmission capacity margin is the difference between the path's rated transmission power and the current transmission power, divided by the rated transmission power, ranging from 0 to 1. A margin below 0.2 indicates the path is close to full load. Protection configuration completeness assesses the completeness and timing coordination of protection devices along the path, a normalized score ranging from 0 to 1. Path redundancy calculates the number of backup paths that can bypass the current path and connect the same node pairs, with a value ranging from 0 to 5. The propagation strength tensor is represented as a four-dimensional vector, and the weighted combination of vector elements maps to a comprehensive propagation strength coefficient, with a value ranging from 0.1 to 1.0.

[0065] The power supply network topology corresponding to each candidate scheme is stored in the form of a topology matrix. The matrix dimension is a square matrix consisting of the number of nodes multiplied by the number of nodes. Matrix elements take values ​​of 0 or 1 to indicate whether there is a direct connection between nodes. For node pairs with connections, the matrix elements are replaced with the path propagation strength coefficient of that connection. The coupling and superposition operation superimposes the risk weight values ​​in the node state tensor onto the diagonal elements of the topology matrix. The original diagonal elements are initially 0, and after superposition, the diagonal elements are equal to the node's own risk weight, with values ​​ranging from 0 to 1. At the same time, the propagation strength coefficient of the risk propagation path is updated to the off-diagonal elements of the topology matrix. After coupling and superposition, the topology matrix is ​​transformed into a risk-incentivized topology matrix, which simultaneously encodes network connectivity, node risk states, and path propagation capabilities.

[0066] A hierarchical decomposition of the inter-node coupling relationships in the risk-incentivized topology yields direct-coupled and indirect-coupled subgraphs. The direct-coupled subgraph extracts node connections corresponding to all non-zero, off-diagonal elements in the risk-incentivized topology matrix, preserving the original connection weights to form a subgraph matrix containing only first-order neighbor relationships. The indirect-coupled subgraph is obtained by exponentiation of the direct-coupled subgraph matrix. Calculating the second power of the direct-coupled subgraph matrix yields a second-order indirect-coupled subgraph, and calculating the third power yields a third-order indirect-coupled subgraph. In the second-order indirect-coupled subgraph matrix, non-zero elements represent node pairs connected by a single intermediate node, with element values ​​being the product and sum of the weights of the two path segments. In practical applications, third- or fourth-order indirect-coupled subgraphs are typically calculated. For medium-sized power grids with 100 to 500 nodes, calculating to the fourth order can cover over 95% of the effective propagation paths.

[0067] Based on the topological transitivity between directly coupled and indirectly coupled subgraphs, a multi-order coupling recombination tensor is constructed. This tensor stacks coupled subgraphs of various orders hierarchically into a three-dimensional tensor structure. The first and second dimensions correspond to node indices, and the third dimension corresponds to the coupling order. The zeroth layer of the third dimension stores the matrix of directly coupled subgraphs, the first layer stores the matrix of second-order indirectly coupled subgraphs, and the second layer stores the matrix of third-order indirectly coupled subgraphs. A weighted sum is applied to the third dimension of the tensor, with the weight coefficients decreasing exponentially with increasing order. The weights are set to 1.0 for first-order, 0.5 for second-order, 0.25 for third-order, and 0.125 for fourth-order. The total weights are normalized to 2.0 to maintain numerical stability. The weighted sum yields the comprehensive coupling matrix, whose element values ​​reflect the comprehensive risk transmission strength between nodes after considering multi-hop propagation paths.

[0068] The risk states in the risk-incentivized topology are propagated iteratively over time using a multi-order coupled recombined tensor to obtain a dynamic evolution sequence of risk radiating outward along the topological path. The iterative propagation employs a discrete-time stepping method, with each time step ranging from 0.1 to 1 second, adaptively adjusted based on the network's transient response characteristics. The initial state of the iterative propagation is represented by the diagonal elements of the risk-incentivized topology matrix, indicating the initial risk level of each node. The initial risk vector dimension is the number of nodes, with elements ranging from 0 to 1. Each time step's iterative calculation multiplies the risk state vector of each node at the current moment with the comprehensive coupling matrix of the multi-order coupled recombined tensor to obtain the incoming risk quantity received by each node at the next moment. The incoming risk quantity is superimposed on the node's current risk state, using a saturated addition rule, meaning the sum does not exceed 1.0; if the result exceeds 1.0, it is truncated to 1.0. The iteration terminates when the maximum value of the change in risk state of all nodes between two consecutive time steps is less than 0.001, or when the number of iterations reaches a preset upper limit, such as 1000 steps. During the iteration process, a snapshot of the risk state of all nodes at each time step is recorded, forming a two-dimensional dynamic evolution sequence matrix with the time dimension being the number of iteration steps and the spatial dimension being the number of nodes.

[0069] The spatial diffusion pattern of risk is extracted from the dynamic evolution sequence. For each column of the dynamic evolution sequence matrix, the time when the risk reaches a preset threshold is calculated. The threshold is set to 0.5, indicating that the risk level of a node exceeds the moderate intensity. The times when the risk of all nodes first reaches the threshold are statistically analyzed, and the time values ​​are mapped to color gradients and rendered as a heatmap in a 3D scene. Quantitative indicators of the diffusion pattern include the risk propagation rate, the radius of influence, and the main axis of the diffusion direction. The risk propagation rate is calculated as the ratio of the risk arrival time to the topological hops from the node to the risk source node, in hops per second, with a typical value range of 0.5 to 5 hops per second. The radius of influence is defined as the geographic outer radius of the set of nodes whose risk level exceeds the threshold, in kilometers, with a value range of 1 to 50 kilometers.

[0070] Based on the spatial diffusion pattern, the node links where risk propagation occurs are traced back to their origins. This tracing begins with a dynamic evolution sequence matrix. For each node exceeding the risk threshold, the source node of that risk at the previous time step is queried. A source node is defined as a node in the comprehensive coupling matrix that has a non-zero connection to the current node and whose risk state at the previous time step was higher than 0.1. If there are multiple source nodes, the node with the largest comprehensive coupling weight is selected as the primary source, and the directed connection between the current node and the source node is recorded. The tracing query is recursively executed until the initial risk source node is reached, and the entire tracing path forms a node link. Node links are stored as a list of directed edges, with each edge containing attributes such as the starting node identifier, ending node identifier, propagation weight, and propagation time.

[0071] By performing topological closure operations on node links, the set of network nodes affected by the cascading effect of the risk is delineated, thus obtaining the cascading impact range of the risk. The topological closure operation extracts all nodes that have appeared in the node link set, constructing an initial node set. For each node in the initial node set, all neighbor nodes with non-zero connections and connection weights greater than 0.1 in the comprehensive coupling matrix are queried, and these neighbor nodes are added to the node set. The neighbor expansion operation is recursively performed until two consecutive expansions do not add any new nodes, or the node set size exceeds 80% of the total number of nodes in the network. The final node set represents the cascading impact range of the risk. The ratio of the number of nodes in the set to the total number of nodes in the network is defined as the cascading impact coverage rate, ranging from 0.01 to 1.0. The cascading impact range is output as a list of node identifiers, with the list length equal to the total number of affected nodes, typically ranging from 10 to 300. Affected nodes are sorted in ascending order by the time the risk arrived, providing a quantitative basis for subsequent emergency decision-making.

[0072] A second aspect of this invention provides an artificial intelligence-based three-dimensional simulation system for emergency power supply, comprising: The modeling unit is used to acquire topology data, equipment operating status data, and spatial environment data of the emergency power supply system, and to construct a three-dimensional simulation model of the emergency power supply system based on the topology data, equipment operating status data, and spatial environment data. An evolution unit is used to extract time-series features and perform evolutionary reasoning on the equipment operation status data, generate state evolution data of each equipment node in the future time period, map the state evolution data to the corresponding equipment node in the three-dimensional simulation model, and dynamically update the three-dimensional simulation model. The simulation unit is used to perform simulation calculations on the power supply path in the three-dimensional simulation model based on power supply constraint rules and load distribution strategies, and to obtain a set of candidate power supply schemes; The evaluation unit is used to establish a knowledge structure containing the correlation between the state evolution data and historical fault events, identify potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set through correlation reasoning in the knowledge structure, and perform risk assessment on the power supply scheme candidate set based on the potential risk nodes and the risk propagation paths to generate the optimal power supply scheme. The adjustment unit is used to visualize and render the execution process of the optimal power supply scheme in the three-dimensional simulation model, collect user interaction operation commands for the execution process, and adaptively adjust the power supply constraint rules and the load allocation strategy using the interaction operation commands.

[0073] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0074] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0075] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence-based three-dimensional simulation method for emergency power supply, characterized in that, include: Acquire topology data, equipment operating status data, and spatial environment data of the emergency power supply system, and construct a three-dimensional simulation model of the emergency power supply system based on the topology data, equipment operating status data, and spatial environment data; The device operation status data is subjected to time-series feature extraction and evolutionary reasoning to generate state evolution data of each device node in the future time period, and the state evolution data is mapped to the corresponding device node in the three-dimensional simulation model, and the three-dimensional simulation model is dynamically updated. Based on power supply constraint rules and load allocation strategies, the power supply paths in the three-dimensional simulation model are simulated and calculated to obtain a set of candidate power supply schemes; A knowledge structure containing the correlation between the state evolution data and historical fault events is established. Through correlation reasoning, potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set are identified in the knowledge structure. Based on the potential risk nodes and the risk propagation paths, the power supply scheme candidate set is risk-assessed to generate the optimal power supply scheme. The execution process of the optimal power supply scheme is visualized and rendered in the three-dimensional simulation model, and the user's interactive operation commands for the execution process are collected. The power supply constraint rules and the load allocation strategy are adaptively adjusted using the interactive operation commands.

2. The method according to claim 1, characterized in that, The process involves extracting temporal features and performing evolutionary reasoning on the equipment operating status data to generate state evolution data for each equipment node within a future time period. This state evolution data is then mapped to the corresponding equipment nodes in the 3D simulation model. Dynamic updates to the 3D simulation model include: The equipment operation status data is decomposed into multi-scale time series, and the long-term trend component, periodic component and random disturbance component are analyzed and recombined. The baseline evolution direction of each equipment node is projected and mapped according to the long-term trend component. Based on the periodic component identification device's cyclic mode, and by coupling the cyclic mode with the reference evolution direction, a steady-state evolution trajectory of the device node is generated. A disturbance propagation network is constructed using the random disturbance components. The transmission relationship and influence weight of the random disturbance components between device nodes are decoupled from their source. The steady-state evolution trajectory is adaptively fused based on the transmission relationship and influence weight to obtain a dynamic evolution trajectory that considers the coupling effect between nodes. The state parameters of the device nodes at each moment in the dynamic evolution trajectory are aggregated to form state evolution data, and a spatial mapping relationship is established between the topological location information of the device nodes and the corresponding device nodes in the three-dimensional simulation model. Based on the spatial mapping relationship, the state parameters and visual representations of the corresponding device nodes are updated in chronological order in the three-dimensional simulation model to achieve dynamic updating of the three-dimensional simulation model.

3. The method according to claim 1, characterized in that, Based on power supply constraint rules and load allocation strategies, the power supply paths in the three-dimensional simulation model are simulated and calculated to obtain a candidate set of power supply schemes, including: The power supply constraint rules are transformed into a multi-dimensional constraint space, and a feasible domain boundary for the power supply path is established in the multi-dimensional constraint space. The topology of the connection path between the power supply node and the load node in the three-dimensional simulation model is filtered through the feasible domain boundary to generate an initial set of feasible paths. For each path in the initial feasible path set, a load-path mapping matrix is ​​constructed based on the load allocation strategy, and the power flow distribution of each path under different load allocation modes is calculated using the mapping matrix. The power flow distribution is substituted into the multidimensional constraint space for constraint verification, the path state points that trigger the constraint boundary are identified, and the path state points are adaptively adjusted so that the power flow distribution returns to the interior of the feasible domain boundary. Based on the regulated power flow distribution, the resource utilization efficiency index and safety margin index of each power supply path under the load allocation strategy are calculated, and the resource utilization efficiency index and safety margin index are weighted and fused to obtain the comprehensive evaluation value of each power supply path. Based on the comprehensive evaluation value, the initial feasible path set is optimized and sorted, and power supply paths that meet the evaluation threshold and their corresponding load allocation schemes are selected and combined to form a power supply scheme candidate set.

4. The method according to claim 3, characterized in that, For each path in the initial feasible path set, a load-path mapping matrix is ​​constructed based on the load allocation strategy, and the power flow distribution of each path under different load allocation modes is calculated using the mapping matrix, including: The load demand vector and path capacity vector are projected backward from the load allocation strategy, and a mapping relationship matrix representing the allocation weight between load nodes and power supply paths is constructed based on the coupling relationship between the load demand vector and the path capacity vector. The mapping relationship matrix is ​​decoupled by feature to obtain the load space basis vector and the path space basis vector. Multiple load allocation patterns are generated by multi-order coupling and recombination of the load space basis vector and the path space basis vector. For each load distribution mode, the distribution weights in the mapping matrix are calculated and transformed into the actual load carrying capacity of each power supply path. The actual load carrying capacity is then integrated and excited into the electrical topology network corresponding to the initial feasible path set. The power flow distribution on each path is obtained by solving the power flow equation. Cross-validate the power flow distribution with the allocation weights in the mapping matrix, invert and lock out the path nodes where the power distribution deviates, and dynamically coordinate their allocation weights until the power flow distribution and the allocation weights reach a convergence state, thus obtaining the power flow distribution of each path under different load allocation modes.

5. The method according to claim 1, characterized in that, By using associative reasoning to identify potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set within the knowledge structure, and by performing a risk assessment on the power supply scheme candidate set based on the potential risk nodes and risk propagation paths, the optimal power supply scheme is generated, including: The network topology feature vectors of each candidate scheme in the power supply scheme candidate set are input into the knowledge structure, and a risk reasoning graph is constructed in the knowledge structure. By performing correlation reasoning in the risk reasoning graph, potential risk nodes in each candidate scheme that match historical failure modes are identified. For the potential risk node, the risk causal chain associated with it is traced in the risk reasoning graph, and the node propagation relationship in the risk causal chain is converted into a risk propagation intensity matrix. By decoupling the risk propagation intensity matrix by features, the risk propagation path starting from the potential risk node is obtained. The potential risk nodes and the risk propagation paths are integrated and incentivized into the power supply network topology corresponding to each candidate scheme. By performing multi-level coupling and reorganization in the power supply network topology, the diffusion pattern and cascading impact range of the risk in the network are deduced. Based on the aforementioned diffusion trend and the cascading impact range, a comprehensive risk assessment function is constructed that includes risk coverage and risk propagation rate. The comprehensive risk assessment function is then used to quantitatively evaluate each candidate scheme in the power supply scheme candidate set to obtain the risk assessment value of each candidate scheme. The candidate schemes in the power supply scheme candidate set are ranked and selected according to the risk assessment value, and the candidate scheme with the lowest risk assessment value is determined as the optimal power supply scheme.

6. The method according to claim 5, characterized in that, The potential risk nodes and the risk propagation paths are fused and incentivized into the power supply network topology corresponding to each candidate scheme. Through multi-level coupling and reorganization within the power supply network topology, the diffusion pattern and cascading impact range of the risk in the network are deduced, including: The node state tensor of potential risk nodes and the propagation strength tensor of risk propagation paths are coupled and superimposed into the topology matrix of the power supply network topology corresponding to each candidate scheme, forming a risk-incentivized topology that carries risk information. The coupling relationships between nodes in the risk incentive topology are decomposed hierarchically to obtain direct coupling subgraphs and indirect coupling subgraphs. Based on the topological transitivity between the direct coupling subgraphs and indirect coupling subgraphs, a multi-order coupling recombination tensor is constructed. The risk state in the risk-incentive topology is propagated in a temporal iterative manner by the multi-order coupled recombination tensor to obtain a dynamic evolution sequence of risk radiating outward along the topological path, and the spatial diffusion pattern of risk is extracted from the dynamic evolution sequence. Based on the aforementioned spatial diffusion trend, the node links where risk transmission occurs are traced back and located. By performing topological closure operations on the node links, the set of network nodes affected by the risk through cascading effects is delineated, thus obtaining the cascading impact range of the risk.

7. The method according to claim 1, characterized in that, The execution process of the optimal power supply scheme is visualized and rendered in the three-dimensional simulation model, and user interaction commands for the execution process are collected. The power supply constraint rules and load allocation strategy are adaptively adjusted using these interaction commands, including: The power supply path topology and time-series power flow data in the optimal power supply scheme are mapped to the rendering space of the three-dimensional simulation model, and a dynamic visualization scene reflecting the execution process of the optimal power supply scheme is generated through multi-level visual encoding. An interactive perception layer is deployed in the dynamic visualization scenario. The interactive perception layer captures the user's interactive operation behavior for power supply nodes, power supply paths and load distribution status in real time. The interactive operation behavior is subjected to intent recognition and semantic parsing to generate interactive operation instructions containing adjustment targets and adjustment parameters. The user's intention to modify the power supply constraints and load allocation weights is projected backward from the interactive operation instructions, and the modification intention is converted into constraint rule adjustment vector and strategy weight adjustment vector. The constraint rule adjustment vector is fused and excited into the constraint space boundary of the power supply constraint rule, and the strategy weight adjustment vector is fused and excited into the weight allocation matrix of the load allocation strategy. The power supply constraint rule and the load allocation strategy are adaptively adjusted through dynamic coordination.

8. An artificial intelligence-based emergency power supply three-dimensional simulation system, used to implement the method as described in any one of claims 1-7, characterized in that, include: The modeling unit is used to acquire topology data, equipment operating status data, and spatial environment data of the emergency power supply system, and to construct a three-dimensional simulation model of the emergency power supply system based on the topology data, equipment operating status data, and spatial environment data. An evolution unit is used to extract time-series features and perform evolutionary reasoning on the equipment operation status data, generate state evolution data of each equipment node in the future time period, map the state evolution data to the corresponding equipment node in the three-dimensional simulation model, and dynamically update the three-dimensional simulation model. The simulation unit is used to perform simulation calculations on the power supply path in the three-dimensional simulation model based on power supply constraint rules and load distribution strategies, and to obtain a set of candidate power supply schemes; The evaluation unit is used to establish a knowledge structure containing the correlation between the state evolution data and historical fault events, identify potential risk nodes and risk propagation paths corresponding to each candidate scheme in the power supply scheme candidate set through correlation reasoning in the knowledge structure, and perform risk assessment on the power supply scheme candidate set based on the potential risk nodes and the risk propagation paths to generate the optimal power supply scheme. The adjustment unit is used to visualize and render the execution process of the optimal power supply scheme in the three-dimensional simulation model, collect user interaction operation commands for the execution process, and adaptively adjust the power supply constraint rules and the load allocation strategy using the interaction operation commands.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images