A power-traffic coupled network scenario generation method based on topology-embedded dynamics

Abstract

The application discloses a power-traffic coupling network scene generation method based on topology-embedded dynamics, comprising the following steps: defining the economic target of power-traffic coupling network cooperative operation and the resilience target of emergency recovery; constructing a physical-behavior fusion "car-station-road-network" multidimensional mechanism model, including an extreme scene mechanism model, a traffic network-charging network-distribution network coupling topology model and a user behavior decision preference model; constructing a unified representation model of topology-embedded dynamics, which is used for generating unified feature representation of the coupling network, and the unified representation model comprises a network topology representation module and a flow dynamics representation module; based on the condition of the unified feature representation, generating an extreme operation scene by using a conditional diffusion model based on a fusion decoupling generation architecture and a physical information regularization training mechanism. The technical scheme of the application provides high-fidelity and interpretable scene support for power-traffic coupling system operation optimization and emergency decision.

Description

Technical Field

[0001] This invention relates to the field of power grid and transportation network coupled operation technology, and specifically to a method for generating power-transportation coupled network scenarios based on topology embedded dynamics. Background Technology

[0002] In recent years, with the popularization of electric vehicles and the deep integration of transportation and energy systems, the coupled operation and collaborative optimization of power grids and transportation networks have become a research hotspot in the fields of smart grids and intelligent transportation. Existing research mainly focuses on data-driven methods such as graph neural networks and reinforcement learning to predict and schedule power flow and traffic volume, or to use heuristic strategies for emergency recovery in specific extreme scenarios. Preliminary collaborative decision support between the two networks at time scales such as day-ahead and intraday has been achieved.

[0003] However, existing methods still face significant challenges: On the one hand, the multidimensional network topology and dynamic behavior of power-transportation-charging facilities are usually modeled separately, lacking a unified representation framework integrating "topology-flow-behavior," making it difficult to accurately depict the cross-network propagation and interaction effects of traffic flow and power flow under abnormal conditions such as faults and congestion. On the other hand, existing scenario generation methods mostly rely on historical data or simplified mechanisms, making it difficult to integrate extreme physical constraints and user behavior uncertainties. This results in poor physical consistency and weak interpretability of the generated training scenarios, limiting the generalization ability and reliability of collaborative strategies in real and complex environments.

[0004] Therefore, there is an urgent need to construct a method for generating power-transportation coupled network scenarios that can deeply integrate multi-source mechanisms and data, support unified representation of topological embedded dynamics, and have high-fidelity scenario generation capabilities. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings and deficiencies of the prior art and propose a method for generating power-transportation coupled network scenarios based on topology embedded dynamics, providing high-fidelity and interpretable scenario support for the operation optimization and emergency decision-making of power-transportation coupled systems.

[0006] To achieve the above objectives, this invention provides a method for generating power-transportation coupled network scenarios based on topology-embedded dynamics, comprising: Define the economic objectives and emergency recovery resilience objectives for the coordinated operation of the power-transportation coupled network; Construct a physical-behavioral integrated "vehicle-station-road-network" multi-dimensional mechanism model, including an extreme scenario mechanism model, a coupled topology model of transportation network-charging network-distribution network, and a user behavior decision preference model; A unified representation model of topology-embedded dynamics is constructed to generate a unified feature representation of coupled networks. The unified representation model includes a network topology representation module for representing the interaction relationship of multiple elements of "vehicle-station-road-network" and a flow dynamics representation module for characterizing the dynamic propagation process of traffic. Using the unified feature representation as a condition, an extreme operating scenario is generated based on a conditional diffusion model that integrates a decoupled generation architecture and a physical information regularization training mechanism.

[0007] In some embodiments, the economic objective of the coordinated operation is to minimize the overall economic cost of the power and transportation systems, including the power generation / purchase cost of the power system and the user travel time cost of the transportation system; the corresponding constraints include distribution network power flow balance constraints, distribution network safe operation constraints, transportation network flow conservation constraints, and electric vehicle state of charge constraints. The resilience objective of the emergency recovery is to minimize the power loss of critical loads and the emergency discharge cost of electric vehicles. The corresponding constraints include the number of distribution network reconfiguration constraints, islanded operation power balance constraints, voltage safety constraints under emergency conditions, and electric vehicle emergency discharge capability constraints.

[0008] In some embodiments, constructing the extreme scenario mechanism model includes: Fault chain system defined based on complex network theory S ( n ) ={S G (n),R,E} ,in S G A set of fault events, R For the relationship between events, E Assuming environmental factors are considered, and a stock-flow diagram of system dynamics is introduced, the physical damage process is divided into two categories: transient impact and steady-state accumulation. Define the node disaster loss rate, which is determined by the cascading effects of the parent node, the node's own vulnerability, and the environmental amplification factor. The node vulnerability is composed of physical exposure, disaster vulnerability, and emergency response capability. The cumulative disaster loss degree is defined, which is obtained by integrating the disaster loss rate of the node over time, and is used to characterize the cumulative damage state of the node. For wind disasters and floods / debris flow disasters, specific fault triggering and propagation models are established respectively. Among them, the wind disaster model defines the line galloping rate and cumulative galloping degree. When the cumulative galloping degree exceeds the threshold, the line fault is triggered. The flood / debris flow model constructs a nonlinear relationship between tree lodging rate and rainfall factor.

[0009] In some embodiments, constructing the coupled topology model of the transportation network-charging network-distribution network includes: The distribution network is abstracted as a directed weighted graph. G D =( V D ,E D ), where the set of nodes V D Includes substation nodes, load nodes, and charging station nodes, and branch set. E D Characterizing power distribution lines and tie switches; nodes i The grid-connected power balance satisfies: in, P L,i，t Represents a node i In t The basic active power load requirement at any given time; P G,i，t Represents a node i In t Active power output of distributed power sources or units at all times. Oh S,i For access nodes i A collection of charging stations P CS,s,t For charging stations s exist t Aggregate power consumption at any given moment; Abstracting the urban transportation network into a directed graph G T =( V T ,E T ), where the set of nodes V T Includes traffic locations such as road intersections, vehicle origin / destination, and charging station locations; road segment collection. E T Urban road segments are characterized; and road resistance functions are used to characterize the travel time of road segments. t a With traffic f a Nonlinear relationship: in, To represent the growth factor of road segment impedance as traffic flow increases, To represent the congestion sensitivity of road segment travel time to traffic saturation, the exponential function is used. Traffic section a Reference travel time in free-flow conditions; Define a set of charging stations S={s 1 ,s 2 ,…,s K}, and establish a mapping function. f map :S→(V D ×V T ) This enables each charging station s Physically, this corresponds to one power node and one transportation node; the operation of the charging station is limited by the transformer capacity. P cap,s and the number of charging stations M s And queuing space.

[0010] In some embodiments, the user behavior decision preference model is used to characterize the spatial flexibility and decision bias of electric vehicle users in fault or congestion scenarios, including: The user space transfer decision model based on price elasticity describes the dynamic migration pattern of charging load between charging stations by introducing the price elasticity coefficient and response rate. A fault scenario user behavior decision model is constructed based on prospect theory to build a bounded rationality decision model, describing the changes in charging / discharging behavior of electric vehicle users in extreme scenarios; The user preference boundary characterization model uses a polyhedral uncertainty set to define the physical boundary of load deviation under fault scenarios.

[0011] In some embodiments, the price elasticity-based spatial transfer decision model uses a linear demand response equation to describe the user's charging decision power at charging station s at time t, including: in, For users t At the charging station s The charging decision power, This serves as the baseline load prior to participation in the response; For charging stations s Price deviation before and after the response; This is the self-response coefficient, reflecting the load reduction caused by the price increase at this station; The cross-response coefficient reflects the impact of price changes at other charging stations on this station. To indicate adjacent charging stations r exist t Real-time response to price discrepancies before and after, Let represent the set of all charging stations participating in the collaborative operation decision-making within the system.

[0012] In some embodiments, the fault scenario user behavior decision model includes: The perceived value function describes a user's perception of travel time. T and battery anxiety E Perceived value includes: in, x k For actual road conditions or electricity prices, Indicates the user's opinion on the first k The importance weight coefficients of various influencing factors (such as time, cost, and anxiety) This represents a value perception function based on user psychological expectations. ref k This represents the normal value expected by users. This refers to the user's marginal sensitivity parameter in the revenue region. For the user's marginal sensitivity parameter in the loss region, The loss aversion coefficient, l >1 reflects the user's avoidance of losses caused by faults; A probabilistic model for route and charging station selection, based on cumulative foreground values, guides users to choose specific routes. r or charging station s probability P choice Follows the Logit distribution: in, Parameters representing the perceived rationality of users in their route or charging station selection decisions. Indicates a specific path r or charging station s User preference weighting factors This represents the sum of all possible path and site preference weights, used as a probability normalization factor.

[0013] In some embodiments, the user preference boundary characterization model includes: in, For charging stations s The charging power load deviation, and These are the upper bounds of the uncertainty of the load self-response and cross-response under fault conditions, respectively. and This represents the irrational price sensitivity coefficient influenced by the psychological impact of disasters in fault scenarios. and This is the benchmark price sensitivity coefficient under normal operating conditions. For charging stations s exist t Real-time response to price discrepancies before and after, Other charging stations r exist t Real-time response to price discrepancies before and after, For charging stations r exist t Always participate in the baseline charging load before response. This is the set of all charging stations in the system.

[0014] In some embodiments, the network topology representation module is used to generate a unified spatiotemporal feature representation, including: The heterogeneous graph representation of the coupled network is defined as follows: G= ( V,E,A ); where, node set V=V P ∪V T ∪V C , V P For power nodes in the distribution network, V T As a node in the transportation network, V C For charging / discharging station coupled nodes; edge set E Includes physical connection edges and logical coupling edges representing the interaction between energy replenishment and power; adjacency matrix A By predefined physical topology A pre Adaptive topology for data learning A ada Together constitute; Constructing the feature matrix X Its row vector x i At least include nodes i generator power P G,i Basic load P L,i Charging load P EV,i Voltage amplitude V i Traffic flow on road sections f road and the state of charge of electric vehicles SOC k Based on the system's operating status, distinguish the feature matrix of typical scenarios. X norm With loads that have lost powerP load,i and emergency discharge power P dis,k Fault scenario feature matrix X fault ; Through the mask matrix F Logical reconstruction of the topology under fault scenarios, resulting in a fault topology matrix. A′ Represented as Where ⊙ represents the Hadamard product, if a node or branch fails... F The corresponding element is 0 if it is in the middle, otherwise it is 1; Spatiotemporal aggregation is performed using an adaptive temporal topological graph convolutional network with node attention, and its node attention matrix... S ij ′ It is calculated by querying Q, key K, and value V; Based on formula Obtaining a unified feature representation Z Where σ is the activation function, A ij For adaptive adjacency matrix, X temp Θ is the feature matrix extracted by temporal dilated convolution, and Θ is the learnable filter weight matrix; The flow dynamics characterization module is used to derive the fault propagation path, including: Construct a fault propagation scenario tree with the initial fault state as the root node; A strategy combining the A* heuristic algorithm with breadth-first search is employed, by evaluating the function... f ( n ) =g ( n ) +h ( n Search for the evolutionary path with the highest risk, among which g ( n This represents the cumulative failure probability or cost loss from the initial failure to the current state. h ( n () is a heuristic estimation function that predicts the current state to the system's collapse boundary; It automatically identifies traffic flow redistribution paths caused by power outages in distribution network lines and charging stations, forming a complete scenario chain covering "fault source - propagation path - final effect".

[0015] In some embodiments, the generation of extreme operating scenarios based on a conditional diffusion model that integrates a decoupled generation architecture and a physical information regularization training mechanism, using the output unified feature representation as a condition, includes: A decoupled generation architecture is constructed, which processes the power node status, traffic segment flow, and charging station coupling information into mutually independent but logically related feature tags, and configures an independent noise state for each feature tag to simulate the asynchronous evolution of grid-side load transfer and traffic-side flow rerouting during the generation process. Using the Transformer architecture as the core of denoising, a conditional diffusion model is constructed; in the inverse denoising process, the conditional diffusion model represents the unified feature. Z Mask topology matrix F Heterogeneous features are fused from high-precision road network map features and power operation criteria, and the cascading flow transfer effect between long-distance road segments is characterized by a self-attention mechanism; its denoising evolution process is represented as follows: ,in for Noisy scene data at any given moment Resilient guiding objectives to minimize power loss losses; In the training of the conditional diffusion model, a physical information loss function is constructed. Loss phy The physical information loss function includes power flow balance loss. Traffic flow conservation loss Voltage deviation penalty and security status penalties At least one of the following, and the physical information loss function is used as part of the model training objective to force the generated data to meet physical constraints; In the training objective function of the conditional diffusion model, a Tikhonov regularization term is introduced. This term constrains the norm of the model's score function to prevent the model from collapsing to the center of the historical data distribution and to guide the model in exploring boundary conditions. The training objective function is expressed as follows: in, Indicates parameters The optimization objective is to minimize the expected loss of the variable. It is Gaussian noise. This is a diffusion model with a conditional diffusion converter. They are respectively t The latent space state at time step, the diffusion time step, and the guiding condition vector. l phy The weighting coefficients for the physical information loss term. The regularization coefficient is . For regularized fractional functions, This is the time-dependent regularization matrix.

[0016] Compared with the prior art, the technical solution of the present invention has at least the following beneficial effects: This invention proposes a full-process scenario generation architecture, from multi-source physical mechanism modeling to topology-embedded dynamic representation, and then to physical information fusion and diffusion generation. This architecture deeply couples the power flow distribution of the power grid with the traffic evolution of the transportation network, providing a scientific, closed-loop, and highly interpretable technical solution for resilience assessment of power-transportation coupled networks under extreme disasters, from underlying physical logic to high-level scenario output.

[0017] This invention establishes a multi-dimensional physical-behavioral fusion model of "vehicle-station-road-network," utilizing an adaptive temporal topological graph convolutional network and a mask matrix mechanism to achieve a unified representation of heterogeneous network characteristics under both normal operation and extreme failure scenarios. This model can accurately characterize the spatiotemporal interaction features and cascading failure evolution process among cross-network coupled elements, effectively solving the challenges of collaborative representation of multi-source heterogeneous data and quantitative description of bounded rationality decisions in complex coupled systems.

[0018] This invention constructs a diffusion model generation paradigm that embeds power-traffic physics criteria and the Tikhonov regularization mechanism, integrating hard constraints such as power grid power flow balance and traffic flow conservation into the latent space denoising generation process. Through a decoupled diffusion architecture and memory enhancement mechanism, this algorithm can automatically "mine out" and reconstruct rare but highly destructive extreme boundary conditions from historical data, while ensuring that the generated scenarios strictly adhere to physical mechanisms. This significantly improves the fidelity, diversity, and reliability of the coupled system stress test scenarios. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a method for generating a power-transportation coupled network scenario based on topology embedded dynamics, provided by an embodiment of the present invention; Figure 2 This is an architecture diagram of a method for generating power-transportation coupled network scenarios based on topology embedded dynamics, provided in an embodiment of the present invention. Figure 3 This is a flowchart of a method for generating a power-transportation coupled network scenario based on topology embedded dynamics, provided as an embodiment of the present invention. Detailed Implementation

[0020] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a further detailed explanation of the method for generating power-transportation coupled network scenarios based on topological embedded dynamics proposed in this invention. The advantages and features of this invention will become clearer from the following description. It should be noted that the accompanying drawings are in a very simplified form and use non-precise scales, used only to facilitate and clearly illustrate the embodiments of this invention. Please refer to the accompanying drawings to make the objectives, features, and advantages of this invention more apparent and understandable. It should be understood that the structures, scales, sizes, etc., depicted in the accompanying drawings are only for illustrative purposes to aid those skilled in the art and are not intended to limit the implementation conditions of this invention. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and objectives achieved by this invention, should still fall within the scope of the technical content disclosed in this invention.

[0021] This embodiment provides a method for generating power-transportation coupled network scenarios based on topology-embedded dynamics. It covers the entire process from multi-source data input and physical mechanism modeling to unified feature representation of topology-embedded dynamics, ultimately achieving the generation of extreme scenarios from mechanistic data fusion. The multi-source data input refers to power data, traffic data, charging and user data, and meteorological environmental data, which provide the data foundation for subsequent topology representation, mechanism analysis, and extreme scenario generation. For details, please refer to... Figure 1 The method for generating power-transportation coupled network scenarios based on topology embedded dynamics includes: Step S1: Define the economic objectives and emergency recovery resilience objectives for the coordinated operation of the power-transportation coupled network. Specifically, step S1 includes: S11: The economic objective of coordinated operation is defined as minimizing the combined economic cost of the power and transportation systems, which includes the generation / purchase cost of the power system and the user travel time cost of the transportation system. The objective function is expressed as: in, The overall economic cost of the power and transportation systems, The cost of purchasing / generating electricity for the power system. The cost of travel time for users of the transportation system.

[0022] Specifically, in, Indicates the power generation cost coefficient; Represents a node i The active power generated by the generator at that location; This indicates the power purchased from the main grid. This indicates the unit price of electricity. This represents the unit time cost for travel users; E R This represents a set of regular road sections; x a Indicates regular road section a Traffic flow on the road; t a Indicates the section of road the vehicle passed through. a Required travel time; E c This represents a set of charging sections. x v This indicates the flow of electric vehicles (EVs) heading to charging stations; t travel +t charge This indicates the total time spent during the charging process (including the time spent driving to the charging station). t travel and charging dwell time t charge ).

[0023] Specifically, the constraints corresponding to the economic objectives of coordinated operation include: Distribution network power flow balance constraints are used to ensure that the active and reactive power of each node in the power network satisfy Kirchhoff's laws, specifically: In the formula: Represents a node i Aggregated charging load for electric vehicles; P L,i Represents a node i The basic active power load demand of the facility; P G,i Represents a node i Active power output of distributed power sources or generating units; P i,net (V,θ) Represents a node i The net active power injected into the grid is determined by the network voltage magnitude vector. V and phase angle vector i Decide; Q G,i and Q L,i Representing nodes respectively i The reactive power output and the reactive load of the foundation. Q i,net ( V,θ ) represents a nodei Net reactive power injected into the power grid.

[0024] Distribution network safety operation constraints are used to ensure system safety. The voltage amplitude at each node must be kept within the allowable deviation range, and the power flowing through each line must not exceed the line's thermal stability limit. Specifically: In the formula: Vi Indicates distribution network node i The real-time voltage amplitude; V min , V max These represent the lower and upper limits of the system's permissible safe operating voltage range, respectively. S ij Indicates the connection node i With nodes j The apparent power flowing through the branch; S ij max This indicates the maximum apparent power allowed to pass through the branch, which is the thermal stability limit capacity of the line.

[0025] Traffic flow conservation constraints are used to ensure that the generated traffic flow scenario conforms to physical continuity. Specifically, for any node in the traffic network, the number of inflowing vehicles per unit time should be equal to the sum of the changes in the number of outflowing vehicles and the number of vehicles stranded at that node, expressed as: In the formula: q in This represents the flow of vehicles entering a traffic network node per unit of time. q out This represents the flow of vehicles leaving a traffic network node per unit of time. ΔN node This indicates the change in the number of vehicles remaining within the node area during the observation period.

[0026] Electric vehicle state-of-charge constraints are used to limit the remaining charge of the coupled unit electric vehicle to prevent overcharging and over-discharging. Specifically: In the formula: For the first k The car is t State of charge at time t, and They represent the first k The car is t The lower and upper limits of the state of charge at time t.

[0027] Step S12: Define the resilience objective of emergency recovery as minimizing the power loss losses of critical loads and the emergency discharge costs of electric vehicles; After a fault occurs, the objective function is to maximize the system's power supply recovery capability and minimize the recovery cost through distribution network reconfiguration and emergency dispatch of electric vehicles. In the formula: The costs of power loss for critical loads and the emergency discharge costs of electric vehicles. T This represents the total number of time periods in the fault recovery cycle. Δt The duration of a single time period; For the set of load nodes in the distribution network; For nodes i exist t The amount of load shedding at any given moment (i.e., the power that could not be restored). For nodes i The importance weight of the load; A collection of electric vehicles to participate in emergency support; For the first k electric vehicles t Discharge power at any given time For the first k The unit compensation cost for each electric vehicle participating in emergency discharge.

[0028] Specifically, the constraints corresponding to the resilience target for emergency recovery include: Distribution network reconfiguration frequency constraints are used to limit the number of network topology operations to prevent frequent switch operations during emergencies from causing equipment damage or excessive control complexity. Specifically: In the formula: T: represents the total number of time periods in the fault recovery cycle; Oh line It represents the set of all operable branches (including tie switch lines) in the distribution network. For the line ij exist t The switch state at any given moment (0 is open, 1 is closed). This represents the maximum number of actions allowed.

[0029] Islanding power balance constraints characterize the requirement that the total power of distributed generation and EV discharge must be balanced with the restored load in real time within the islanded area caused by a fault. Specifically: In the formula, P DG,t express t The total active power output of distributed power sources within the islanded area at any given moment;P dis,k,t express t Time of the first k The discharge power of electric vehicles participating in emergency response; P load,i,t Represents a node i exist t The amount of power loss that could not be restored immediately; For nodes i exist t The amount of load shedding at any given moment (i.e., the power that could not be restored). P loss,t This indicates the power loss of the distribution network during the fault recovery process.

[0030] Emergency voltage safety constraints, taking into account the special characteristics of fault conditions, allow node voltages to fluctuate within emergency limits. Specifically: In the formula, V i,t Indicates distribution network node i exist t Voltage amplitude at any given moment; V min emerg , V max emerg These represent the lower and upper limits of voltage fluctuations allowed by the system under extreme emergency conditions, respectively. This range is usually slightly wider than the normal operating limits.

[0031] EV emergency discharge capacity constraints are used to ensure that when an electric vehicle participates in reverse power supply, its remaining charge is not less than the minimum charge required for emergency evacuation. Specifically: In the formula, For the first k The car is t State of charge at time t, The minimum state of charge set for users to meet emergency travel needs.

[0032] Step S2: Construct a physical-behavioral integrated multi-dimensional mechanism model of "vehicle-station-road-network", including an extreme scenario mechanism model, a coupled topology model of transportation network-charging network-distribution network, and a user behavior decision preference model; for details, please refer to Figure 2 Step S2 includes: Step S21: Construct an extreme scenario mechanism model; based on an extreme scenario evolution mechanism model with dual coupling of "disaster-environment-system", characterize the physical damage patterns of disasters to facilities, and quantify the damage processes of typhoons, rainstorms, and secondary disasters to power facilities and transportation networks in time and space through system dynamics methods. Specifically, this includes: Dynamic coupling equation between disaster chain and physical damage: Based on complex network theory, defining the fault chain system under extreme disasters. S ( n ) ={S G (n),R,E} ,in S G A set of fault events, R For the relationship between events, E Environmental factors are considered. A stock-flow diagram from system dynamics is introduced to classify the physical damage process into two categories: "transient impact" and "steady-state accumulation."

[0033] Define the node disaster loss rate to characterize a specific moment. t The instantaneous rate of damage to power or transportation nodes caused by external disasters. n i ( t ), affected by the cascading effect of the parent node. q i ( t ), the inherent fragility of nodes V i and environmental instability E i ( t A joint decision was made, specifically: In the formula, This indicates node vulnerability, determined by physical exposure. V exp Disaster vulnerability V sus and emergency response capabilities V res constitute. E i ( t This is the environmental amplification factor; when environmental indicators exceed a threshold... E 0 This significantly amplifies the consequences of disasters.

[0034] Define cumulative disaster loss degree to characterize the cumulative damage state of a node as the disaster continues. N ( t (i.e., the probability integral of failure occurrence), obtained by integrating the node disaster loss rate over time: Cascading fault propagation mechanisms induced by multiple disasters: Specific fault triggering and propagation models are established for different types of extreme scenarios, including: Wind-induced fault propagation: Modeling tower collapse and line breakage caused by wind loads exceeding tower wind resistance design. Defining line galloping rate. SRL With cumulative dance degree SDL When the cumulative sway exceeds a threshold, a line fault is triggered, including: In the formula, WF Wind force factor; C wind-line The coupling coefficient between wind and the transmission line; S line Line sensitivity; I parent The cascading effects of upstream tower failures This represents the instantaneous fault triggering probability of overhead lines under wind load. This represents the structural reliability configuration factor of power facilities (such as poles and towers), used to reflect the differences in disaster resistance capabilities of different design standards.

[0035] Flood / Debris Flow-Induced Fault Propagation: Modeling how heavy rain causes soil loosening, leading to fallen trees and debris flows, which in turn cause road network disruptions and power grid pole collapses. Constructing tree fall rate data. R tree The nonlinear relationship with the rainfall factor RF includes: in, R tree Tree lodging rate, RF is the rainfall factor. The baseline lodging coefficient is affected by local stand density and soil quality. This is an index representing the sensitivity of tree fall rate to the intensity of rainfall factors.

[0036] Step S22: Construct a coupled topology model of the transportation network, charging network, and distribution network.

[0037] Specifically, please refer to Figure 2 Step S22 involves constructing a multi-dimensional topology model of "vehicle-station-road-network" with deep physical spatial coupling, defining the physical connection parameters between multiple networks, and using graph theory methods to uniformly describe the power distribution system, urban road system, and intermediate charging facility layer. Specifically, this includes: Distribution network topology modeling: abstracting the distribution network as a directed weighted graph. G D=( V D ,E D This is used to characterize the distribution and flow of electricity. The set of nodes... V D Includes substation nodes, transformer nodes, load nodes, and charging station nodes that serve as distributed power sources; branch set E D Characterizing power distribution lines and interconnecting switches, each branch has resistance. r ij Reactance x ij and transmission capacity limit Physical properties, etc. Nodes i The grid-connected power balance satisfies: ;in: P L,i，t Represents a node i In t The basic active power load requirement at any given time; P G,i，t Represents a node i In t Active power output of distributed power sources or units at all times. Oh S,i For access nodes i A collection of charging stations P CS,s,t For charging stations s exist t Aggregate power consumption at any given moment.

[0038] Transportation network topology modeling: abstracting the urban transportation network into a directed graph. G T =( V T ,E T This is used to characterize the vehicle's travel path and spatiotemporal distribution. The node set... V T Includes road intersections, vehicle origin / destination (OD) points, and the traffic location of charging stations. Road segment collection. E T Used to characterize urban road segments, each segment a Having length l a Free flow velocity v f,a and traffic capacity Ca The BPR function is introduced to characterize the travel time of a road segment. t a With traffic f aNonlinear relationship: in, To represent the growth factor of road segment impedance as traffic flow increases, To represent the congestion sensitivity of road segment travel time to traffic saturation, the exponential function is used. Traffic section a Baseline travel time in free-flow (zero flow) conditions.

[0039] Charging network layer coupling mapping modeling: As the "interface" between the distribution network and the transportation network, the topology model of the charging network consists of a set of coupling mapping relationships. Define the set of charging stations. S={s 1 ,s 2 ,…,s K Establish a mapping function. f map :S→(V D × V T ) That is, each charging station s corresponds to a power node in physical location. i∈V D and a traffic node j∈V T .

[0040] Power-side constraints: Charging stations exhibit time-varying loads on the power grid, and their power output is limited by the capacity of the distribution transformers within the station. P cap,s : in, For the first k electric vehicles t Real-time charging / discharging power is constantly connected to the system via charging piles.

[0041] Traffic-side constraints: Charging stations act as a "convergence" or "source" of vehicles on the transportation network, and their service capacity is limited by the number of charging piles. M s and queuing space : in, N EV,s,t for t The total number of vehicles charging and queuing at the station.

[0042] Step S23: Construct a behavioral decision-making preference model. Specifically, construct a user behavioral decision-making model for fast-charging electric vehicles that integrates external environmental factors and users' internal psychological expectations. This model aims to accurately characterize users' spatial flexibility and decision-making biases in fault or congestion scenarios, including: A price elasticity-based user spatial migration decision model describes the spatial response behavior of users to price adjustment signals triggered by power system failures or traffic congestion. By introducing a response coefficient, it characterizes the dynamic migration pattern of charging load between charging stations. Specifically, users in... t At the charging station s Charging decision power The following linear demand response equation must be satisfied: In the formula, For users t At the charging station s The charging decision power, This serves as the baseline load prior to participation in the response; For charging stations s Price deviation before and after the response; This is the self-response coefficient, reflecting the load reduction caused by the price increase at this station; The cross-response coefficient reflects the impact of price changes at other charging stations on this station. To indicate adjacent charging stations r exist t Real-time response to price discrepancies before and after, Let represent the set of all charging stations participating in the collaborative operation decision-making within the system.

[0043] User behavior decision-making model under fault scenarios: This model introduces a bounded rationality decision-making model based on prospect theory to describe changes in the charging / discharging behavior of electric vehicle users after extreme scenarios lead to changes in the physical network topology. Specifically, it includes: Perceived value function: used to characterize a user's perception of travel time in fault scenarios. T and battery anxiety E Perceived value V User It is no longer linear, but a nonlinear deviation based on a reference point (normal state): In the formula, x k For actual road conditions or electricity prices, Indicates the user's opinion on the first k The importance weight coefficients of various influencing factors (such as time, cost, and anxiety) This represents a value perception function based on user psychological expectations. ref k This represents the normal value expected by users. This refers to the user's marginal sensitivity parameter in the revenue region. For the user's marginal sensitivity parameter in the loss region, The loss aversion coefficient, l >1 reflects the user's tendency to avoid losses caused by faults.

[0044] Path and charging station selection probabilities: used to generate a dynamic response for traffic flow redistribution after a fault and the willingness of electric vehicles to perform emergency discharge. Based on accumulated foreground values, users select specific paths. r or charging station s probability P choice Follows the Logit distribution: in, Parameters representing the perceived rationality of users in their route or charging station selection decisions. Indicates a specific path r or charging station s User preference weighting factors This represents the sum of all possible path and site preference weights, used as a probability normalization factor.

[0045] User preference boundary characterization model: The physical boundary of load deviation under fault scenarios is defined using a polyhedral uncertainty set, specifically: in, For charging stations s The charging power load deviation, and These are the upper bounds of the uncertainty of the load self-response and cross-response under fault conditions, respectively. and This represents the irrational price sensitivity coefficient influenced by the psychological impact of disasters in fault scenarios. and This is the benchmark price sensitivity coefficient under normal operating conditions. For charging stations s exist t Real-time response to price discrepancies before and after, For other charging stations r exist t Real-time response to price discrepancies before and after, For charging stations r exist t Always participate in the baseline charging load before response. This is the set of all charging stations in the system.

[0046] The model features of the user preference boundary characterization model include: Psychologically sensitive zone: In the "linear zone" where the charging price difference at charging stations is small, user decisions are dominated by psychological factors and are more biased.

[0047] Rational saturation zone: When the price incentive caused by the fault reaches a threshold, user behavior enters the "saturation zone", load transfer tends to stabilize, and uncertainty reaches a minimum.

[0048] Fault redundancy: Introducing an uncertainty budget limits the total deviation of all charging stations, preventing the generation of overly conservative scenario data.

[0049] Step S3: Construct a unified representation model of topology-embedded dynamics to generate a unified feature representation of the coupled network. This unified representation model includes a network topology representation module for representing the interaction relationships of multiple elements within the "vehicle-station-road-network" system, and a flow dynamics representation module for characterizing the dynamic propagation process of traffic. Please refer to [link to relevant documentation]. Figure 3 Specifically, it includes: Step S31: Construct a network topology representation module. Use an adaptive adjacency matrix to represent interaction relationships. When a fault occurs, use a mask matrix... F Dynamically adjusting the topology to form a topology condition vector, specifically including: Construction of Heterogeneous Graphs for Coupled Networks: Defining the global heterogeneous graph of a power-transportation coupled system as follows: G= ( V,E,A ). Node set V : V=V P ∪V T ∪V C .in V P For power nodes in the distribution network, V T For road segments / intersections in the traffic network, V C This represents the coupled nodes of the charging / discharging station. (Edge set) E It includes physical connections between power systems and transportation networks, as well as logically coupled edges representing the interaction between vehicle-to-station energy replenishment and the power interaction between station and network. Adaptive adjacency matrix. A It is no longer determined solely by physical connections, but by a predefined physical topology. A pre and adaptive topology obtained through data learning A ada Together, they constitute a characterization of the implicit correlation between electricity load flow and traffic flow transfer.

[0050] Topological differentiation representation between normal and fault scenarios: by constructing an adjacency matrix A With characteristic matrix X This transforms the multidimensional heterogeneous information of the coupled "vehicle-station-road-network" system into tensor form. Based on the system's operating status, it is divided into two modes: normal operation and fault operation. Feature matrix X Construction: Define the characteristic matrix X =[ x 1, x 2,…, x N ] T ∈R N×F ,in N The total number of nodes in the coupled network. F For feature dimensions. Each node i eigenvectors x i Specifically defined as: In the formula P G,i , P L,i , P EV,i These are the nodes defined in step S11. i Generator power, base load, and charging load; V i The node voltage amplitude is defined in step S11. x a and SOC k These are the traffic flow of the road segment and the state of charge of the electric vehicle, as defined in step S11.

[0051] In typical scenarios, the feature matrix X norm Record the system's operating parameters under steady-state conditions; under fault scenarios, the feature matrix... X fault Includes the power loss load defined in step S12 P load,i With emergency discharge power P dis,k Disturbance data, etc.

[0052] Adjacency matrix A Construction: Adjacency matrix in conventional scenarios A Representing the initial physical connection relationship of "vehicle-station-road-network", if nodes i With nodes jIf there is a physical or logical connection (such as a power line, road section, or charging pile access point), then A ij >0, otherwise 0. A mask matrix is ​​introduced in fault scenarios. F Logical reconstruction of the topology damaged by the disaster. Topology matrix under the fault scenario. A′ Represented as: In the formula, ⊙ represents the Hadamard product. If the node / branch i If a malfunction occurs, F Corresponding element f i = 0, otherwise f i =1. The complex physical evolution of faults is transformed into sparse tensors that the model can directly process.

[0053] Topological representation based on adaptive attention graph convolution: An adaptive temporal topological graph convolutional network with node attention is used to perform spatiotemporal aggregation on the above matrix to generate a unified representation of the coupled network. Z ,include: Node attention mechanism: To identify the differentiated contributions of different sensor data to fault diagnosis and scene generation, an attention matrix based on query (Q), key (K), and value (V) is constructed. S ij ′ : in, For nodes i With nodes j Attention rating coefficient between them For nodes i With nodes k Attention rating coefficient between them The node attention score matrix, Let be the learnable weight vector for spatial attention. It is a non-linear activation function. For querying the matrix, This is the transpose of the key matrix. This is a learnable bias term for spatial attention.

[0054] The sensor data includes, but is not limited to, voltage / power data collected by the power distribution network phasor measurement unit, road flow data collected by the traffic flow detector, and battery status data obtained by the charging station management system.

[0055] The node attention mechanism can dynamically adjust the correlation weights between different sensor nodes, enabling the model to focus on core features that are highly correlated with extreme faults (such as voltage over-limit points or flow fluctuation areas).

[0056] Adaptive graph convolutional aggregation: This approach updates topological weights using an adaptive adjacency matrix learning module, making it universally applicable to heterogeneous robot platforms or different topologies. It also incorporates features extracted from temporally dilated convolutions. X temp The adjacency matrix is ​​fused with attention-weighted adjacency matrix based on the formula. A unified representation of coupled networks in the latent space is obtained. Z In the formula s Indicates the activation function; A ij This represents the adaptive adjacency matrix obtained by the graph learning module; S ij ′ Represents the node attention matrix; X temp Θ represents the temporal feature matrix extracted after temporal dilated convolution (TDN); Θ represents the learnable filter weight matrix. Z This represents the unified feature representation of the coupled network in the final output.

[0057] Step S32: Construct a flow dynamics characterization module. This module is used to systematically characterize the dynamic propagation characteristics of power flow and traffic flow in a coupled network. Through graph search algorithms and knowledge reasoning techniques, it identifies key evolution paths and their impact range in extreme scenarios. It also derives fault mechanism propagation paths based on heuristic graph search, specifically including: Constructing a fault propagation scenario tree: A strategy combining the A* heuristic algorithm and breadth-first search is employed. The initial state where the fault occurs is taken as the root node, and the tree is constructed using an evaluation function. f ( n ) =g ( n ) +h ( n The most risky evolutionary path for search systems. g ( n ) represents the cumulative failure probability or cost loss from the initial failure to the current state; h ( n ) is a heuristic estimation function used to predict the potential risk from the current state to the system's collapse boundary.

[0058] Path generation: Automatically identify traffic flow redistribution paths caused by power grid line tripping and large-scale power outages at charging stations, forming a complete scenario chain covering "fault source - propagation path - final effect".

[0059] Step S4: Using the unified feature representation as a condition, and based on the conditional diffusion model of the fusion-decoupled generation architecture and the physical information regularization training mechanism, generate extreme operating scenarios. Please refer to [link to relevant documentation]. Figure 3 Specifically, it includes: Step S41: Generation of a multi-dimensional coupled scenario based on a decoupled diffusion transformer. An adaptive decoupled diffusion architecture incorporating topological features is employed to achieve high-fidelity simulation of the dynamic process of traffic transfer and load shifting between the power and transportation networks. Specifically, this includes: The topology-embedded decoupled generation architecture: Drawing on the idea of ​​decoupled diffusion, the coupled information of power node states, traffic flow segments, and charging stations is processed into refined feature labels that are independent but logically related. Each label has an independent noise state, allowing the model to dynamically simulate the asynchronous evolution of "load transfer" on the grid side and "traffic rerouting" on the traffic side during the generation process.

[0060] Feature fusion based on WcDT: Using the Transformer architecture as the denoising core, the mask topology matrix extracted in step S3 is fused together. F Heterogeneous features are fused from high-precision road network map features and power operation criteria. A self-attention mechanism is used to characterize the cascading traffic transfer effect caused by charging station failures across long-distance road segments. The inverse denoising evolution operator is expressed as: In the formula, where for Noisy scene data at any given moment Z The unified feature representation of "vehicle-station-road-network" output in step S31; Resilience-guided objectives such as minimizing power loss losses.

[0061] Extreme scenario dynamics characterization: By performing physical consistency checks on each frame of generated scenario data, the system accurately describes the process of traffic flow shifting from congested road sections to surrounding branch roads after a fault occurs, as well as the power flow redistribution achieved by the distribution network through tie switch reconfiguration. The resulting typical extreme scenario dataset can directly support the resilience assessment and decision optimization of coupled systems under extreme disasters.

[0062] Step S42: Transformation of training mechanisms based on physical information regularization and memory enhancement.

[0063] Step S42 transforms the established extreme scenario mechanism, user decision preference model, and physical constraint boundaries into the inherent constraint mechanism of the generative adversarial diffusion model, ensuring that the generated scenario possesses both extreme stress characteristics and strictly adheres to physical laws. Specifically, this includes: Physical information loss function construction: Mapping hard constraints such as power flow balance and voltage security in step S11 and islanded power balance in step S12 into penalty terms. Loss phy By calculating the residuals of the generated data in the physical mechanism equations and using them as an important component of the training gradient, the model is forced to avoid infeasible regions in the latent space that violate physical laws.

[0064] In the formula, The penalties include power flow balance loss, traffic flow conservation loss, voltage deviation penalty, and safety state penalty. These are the weighting coefficients.

[0065] For the set of power grid nodes, Indicates distribution network node i exist t Voltage amplitude at time 10:00 Represents a node i The generator is located at The active power emitted at all times, for t Nodes are constantly purchased (or injected) from the mainnet. i power, Represents a node i Electric vehicles Aggregate charging load at all times; P L,i,t Represents a node i In The basic active power load requirement at any given time; Adjacent nodes j exist t Voltage amplitude at time 10:00 branch road ij electrical conductivity, For the festival i and j The cosine value of the voltage phase angle difference between them For nodes i The equivalent susceptance, For nodes i The set of directly connected neighbor nodes. Indicates the index of the connected node.

[0066] in, For time period, For a set of traffic nodes, Inflow or outflow node j All road sections a The set, For road section a existt Real-time traffic flow at any given moment This represents the change in the number of vehicles stranded at the node.

[0067] in, For time period, For the set of power grid nodes, This is the upper limit of the system's permissible safe voltage range. This is the lower limit of the system's permissible safe voltage range. Indicates distribution network node i exist t Voltage amplitude at a given time.

[0068] in, This serves as an index for electric vehicles; EV represents the collection of electric vehicles participating in emergency support. For the first k The car is t State of charge at time t, This represents the upper limit of the state of charge (SOC) of an electric vehicle. The minimum state of charge set for users to meet emergency travel needs.

[0069] Anti-memory mechanism based on Tikhonov regularization: To avoid the generative model simply reproducing historical failure data, an improved Tikhonov regularization term is introduced. This mechanism prevents the model from collapsing to the center of the historical experience distribution by constraining the norm of the score function, thereby guiding the model to explore more challenging boundary conditions. Its optimization objective function is expressed as: in, Indicates parameters The optimization objective is to minimize the expected loss of the variable. It is Gaussian noise. This is a diffusion model with a conditional diffusion converter. They are respectively t The latent space state at time step, the diffusion time step, and the guiding condition vector. l phy The weighting coefficients for the physical information loss term. The regularization coefficient is . For regularized fractional functions, This is the time-dependent regularization matrix.

[0070] in, This represents the noise intensity scheduling function during the diffusion process. This represents the cumulative noise control coefficient in the diffusion model.

[0071] in, Represents the identity matrix. Represents the latent variable z gradient operator, p This represents conditional probability.

[0072] in, s Indicates the step index in the diffusion process. express s The step size of the noise variance injected at each time step. in, This represents the minimum noise intensity. This represents the maximum noise intensity. This is the standard deviation adjustment factor corresponding to the step size index in the noise scheduling scheme.

[0073] in, For adaptive regularization weights used to balance physical consistency during the diffusion process, This represents the conditional variance of forward diffusion.

[0074] In summary, the power-transportation coupled network scenario generation method based on topology-embedded dynamics provided in this embodiment proposes a full-process scenario generation architecture, from multi-source physical mechanism modeling to topology-embedded dynamics representation, and then to physical information fusion and diffusion generation. By deeply coupling the power flow distribution of the power grid with the traffic flow evolution of the transportation network, it provides a scientific, closed-loop, and highly interpretable technical solution for the resilience assessment of power-transportation coupled networks under extreme disasters, from the underlying physical logic to the high-level scenario output.

[0075] This invention establishes a multi-dimensional physical-behavioral fusion model of "vehicle-station-road-network," utilizing an adaptive temporal topological graph convolutional network and a masking matrix mechanism to achieve a unified representation of heterogeneous network characteristics under both normal operation and extreme failure scenarios. This model can accurately characterize the spatiotemporal interaction features and cascading failure evolution process among cross-network coupled elements, effectively solving the challenges of collaborative representation of multi-source heterogeneous data and quantitative description of bounded rationality decisions in complex coupled systems.

[0076] This invention constructs a diffusion model generation paradigm that embeds power-traffic physics criteria and Tikhonov regularization mechanism. It integrates hard constraints such as power grid power flow balance and traffic flow conservation into the denoising generation process of the latent space. Through decoupling the diffusion architecture and memory enhancement mechanism, it can automatically "mine out" and restore rare but highly destructive extreme boundary conditions in historical data while ensuring that the generated scenario strictly follows the physical mechanism. This significantly improves the fidelity, diversity and reliability of the stress test scenario of the coupled system.

[0077] Based on the same inventive concept, this embodiment also provides a power-transportation coupled network scene generation system based on topology embedded dynamics, including: The multi-source data integration and real-time monitoring module is responsible for real-time access to data on power distribution network operation status, urban traffic flow density, charging station occupancy rate, and meteorological environment, and performs refined preprocessing. Using the feature matrix X construction method defined in step S31, heterogeneous data is transformed into unified tensor features, enabling non-Gaussian denoising of sensor noise, ensuring the accuracy of the input data, and providing a high-quality data foundation for subsequent mechanism analysis.

[0078] The multidimensional mechanism analysis and topology representation module integrates the extreme disaster evolution mechanism and user decision preference model from step S2 to perceive the physical vulnerability of the system online. By running an adaptive temporal topology graph convolutional network, the module can calculate the dynamic interaction weights of "vehicle-station-road-network" in real time and generate a topology representation Z with an embedded fault mask matrix F. The core function of this module is to reveal the cross-network cascading reaction path between power grid faults and traffic congestion.

[0079] The physics-guided extreme scenario generation module, based on the core algorithm of step S4, utilizes WcDT and a decoupled generation strategy to automatically "mine" potential extreme stress scenarios for the current operating conditions. This module transforms the economic and resilience constraints defined in step S1 into a physical loss function, ensuring that the generated scenarios not only conform to physical laws such as power flow conservation but also reach the critical boundaries of system operation. The generated scenario set provides high-fidelity simulation data for system stress testing and emergency plan development.

[0080] The system resilience assessment and strategy optimization module quantifies the power outage risk and congestion level of the system under different disasters by running collaborative scheduling and emergency recovery algorithms on a generated set of extreme scenarios. Using the resilience objective function established in step S12, this module can automatically optimize the reconfiguration scheme of the distribution network interconnection switches and the orderly discharge compensation strategy for electric vehicles. This module aims to maximize the system's power supply recovery capability and minimize recovery costs through cross-network collaborative decision-making.

[0081] The interactive decision visualization and early warning command module constructs a digital twin interface based on a Geographic Information System (GIS) to intuitively display the spatiotemporal evolution of the "vehicle-station-road-network" under extreme scenarios. This module can mark voltage over-limit points, traffic congestion areas, and affected charging hubs in real time, and provide operators with logical rule suggestions based on knowledge graph reasoning. Through the human-computer interaction interface, command personnel can issue emergency dispatch commands with a single click, achieving rapid response and intelligent closed-loop control in the face of extreme disasters.

[0082] It should be noted that the apparatus and methods disclosed in the embodiments herein can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments herein. In this regard, each block in a flowchart or block diagram may represent a module, program, or part of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system to perform the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0083] In addition, the functional modules in the various embodiments of this article can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0084] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A method for generating power-transportation coupled network scenarios based on topological embedded dynamics, characterized in that, include: Define the economic objectives and emergency recovery resilience objectives for the coordinated operation of the power-transportation coupled network; A multi-dimensional mechanism model integrating physical and behavioral aspects, namely "vehicle-station-road-network," is constructed, including an extreme scenario mechanism model, a coupled topology model of the transportation network-charging network-distribution network, and a user behavior decision-making preference model. Constructing the mechanism model for the extreme scenario includes: Based on complex network theory, a fault chain system S(n) = {SG(n), R, E} is defined, where SG is the set of fault events, R is the relationship between events, and E is the environmental factor. The stock-flow diagram of system dynamics is introduced to divide the physical damage process into two types of effects: transient impact and steady-state accumulation. Define the node disaster loss rate, which is determined by the cascading effects of the parent node, the node's own vulnerability, and the environmental amplification factor. The node vulnerability is composed of physical exposure, disaster vulnerability, and emergency response capability. The cumulative disaster loss rate is defined, which is obtained by integrating the disaster loss rate of the node over time, and is used to characterize the cumulative damage state of the node. For wind disasters and floods / debris flow disasters, specific fault triggering and propagation models are established respectively. Among them, the wind disaster model defines the line galloping rate and cumulative galloping degree. When the cumulative galloping degree exceeds the threshold, the line fault is triggered. The flood / debris flow model constructs a nonlinear relationship between tree lodging rate and rainfall factor. A unified representation model of topology-embedded dynamics is constructed to generate a unified feature representation of coupled networks. The unified representation model includes a network topology representation module for representing the interaction relationship of multiple elements of "vehicle-station-road-network" and a flow dynamics representation module for characterizing the dynamic propagation process of traffic. Using the unified feature representation as a condition, an extreme operating scenario is generated based on a conditional diffusion model that integrates a decoupled generation architecture and a physical information regularization training mechanism.

2. The scene generation method as described in claim 1, characterized in that, The economic objective of the coordinated operation is to minimize the comprehensive economic cost of the power and transportation systems, including the power generation / purchase cost of the power system and the user travel time cost of the transportation system; the corresponding constraints include power flow balance constraints of the distribution network, safe operation constraints of the distribution network, flow conservation constraints of the transportation network, and state of charge constraints of electric vehicles. The resilience objective of the emergency recovery is to minimize the power loss of critical loads and the emergency discharge cost of electric vehicles. The corresponding constraints include the number of distribution network reconfiguration constraints, islanded operation power balance constraints, voltage safety constraints under emergency conditions, and electric vehicle emergency discharge capability constraints.

3. The scene generation method as described in claim 1, characterized in that, Constructing the coupled topology model of the transportation network-charging network-distribution network includes: The distribution network is abstracted as a directed weighted graph. G D =( V D ,E D ), where the set of nodes V D Includes substation nodes, load nodes, and charging station nodes, and branch set. E D Characterizing power distribution lines and tie switches; nodes i The grid-connected power balance satisfies: in, P L,i，t Represents a node i In t The basic active power load requirement at any given time; P G,i，t Represents a node i In t Active power output of distributed power sources or units at all times. Ω S,i For access nodes i A collection of charging stations P CS,s,t For charging stations s exist t Aggregate power consumption at any given moment; Abstracting the urban transportation network as a directed graph G T =( V T ,E T ), where the set of nodes V T Includes traffic locations such as road intersections, vehicle origin / destination, and charging station locations, and a collection of road segments. E T Urban road segments are characterized; and road resistance functions are used to characterize the travel time of road segments. t a With traffic f a Nonlinear relationship: in, To represent the growth factor of road segment impedance as traffic flow increases, To represent the congestion sensitivity of road segment travel time to traffic saturation, the exponential function is used. Traffic section a Reference travel time in free-flow conditions; Traffic section a Traffic capacity; Define a set of charging stations S={s 1 ,s 2 ,…,s K }, and establish a mapping function. f map :S→(V D ×V T ) This enables each charging station s Physically, this corresponds to one power node and one transportation node; the operation of the charging station is limited by the transformer capacity. P cap,s and the number of charging stations M s And queuing space.

4. The scene generation method as described in claim 1, characterized in that, The user behavior decision preference model is used to characterize the spatial flexibility and decision bias of electric vehicle users in fault or congestion scenarios, including: The user space transfer decision model based on price elasticity describes the dynamic migration pattern of charging load between charging stations by introducing the price elasticity coefficient and response rate. A fault scenario user behavior decision model is constructed based on prospect theory to build a bounded rationality decision model, describing the changes in charging / discharging behavior of electric vehicle users in extreme scenarios; The user preference boundary characterization model uses a polyhedral uncertainty set to define the physical boundary of load deviation under fault scenarios.

5. The scene generation method as described in claim 4, characterized in that, The price elasticity-based spatial transfer decision model uses a linear demand response equation to describe a user's charging decision power at charging station s at time t, including: in, For users t At the charging station s The charging decision power, This serves as the baseline load prior to participation in the response; For charging stations s Price deviation before and after the response; This is the self-response coefficient, reflecting the load reduction caused by the price increase at this station; The cross-response coefficient reflects the impact of price changes at other charging stations on this station. To indicate adjacent charging stations r exist t Real-time response to price discrepancies before and after, Let represent the set of all charging stations participating in the collaborative operation decision-making within the system.

6. The scene generation method as described in claim 4, characterized in that, The user behavior decision-making model for the fault scenario includes: The perceived value function describes a user's perception of travel time. T and battery anxiety E Perceived value includes: in, x k For actual road conditions or electricity prices, Indicates the user's opinion on the first k The importance weight coefficient of each type of influencing factor This represents a value perception function based on user psychological expectations. ref k This represents the normal value expected by users. This refers to the user's marginal sensitivity parameter in the revenue region. For the user's marginal sensitivity parameter in the loss region, The loss aversion coefficient, λ >1 reflects the user's avoidance of losses caused by faults; A probabilistic model for route and charging station selection, based on cumulative foreground values, guides users to choose specific routes. r or charging station s probability P choice Follows the Logit distribution: in, Parameters representing the perceived rationality of users in their route or charging station selection decisions. Indicates a specific path r or charging station s User preference weighting factors This represents the sum of all possible path and site preference weights, used as a probability normalization factor.

7. The scene generation method as described in claim 4, characterized in that, The user preference boundary characterization model includes: in, For charging stations s The charging power load deviation, and These are the upper bounds of the uncertainty of the load self-response and cross-response under fault conditions, respectively. and This represents the irrational price sensitivity coefficient influenced by the psychological impact of disasters in fault scenarios. and This is the benchmark price sensitivity coefficient under normal operating conditions. For charging stations s exist t Real-time response to price discrepancies before and after, Other charging stations r exist t Real-time response to price discrepancies before and after, For charging stations r exist t Always participate in the baseline charging load before response. This is the set of all charging stations in the system.

8. The scene generation method as described in claim 1, characterized in that, The network topology representation module is used to generate a unified spatiotemporal feature representation, including: The heterogeneous graph representation of the coupled network is defined as follows: G= ( V,E,A ); where, node set V=V P ∪V T ∪V C , V P For power nodes in the distribution network, V T As a node in the transportation network, V C For charging / discharging station coupled nodes; edge set E Includes physical connection edges and logical coupling edges representing the interaction between energy replenishment and power; adjacency matrix A By predefined physical topology A pre Adaptive topology for data learning A ada Together constitute; Constructing the feature matrix X Its row vector x i At least include nodes i generator power P G,i Basic load P L,i Charging load P EV,i Voltage amplitude V i Traffic flow on road sections f road and the state of charge of electric vehicles SOC k Based on the system's operating status, distinguish the feature matrix of typical scenarios. X norm With loads that have lost power P load,i and emergency discharge power P dis,k Fault scenario feature matrix X fault ; Through the mask matrix F Logical reconstruction of the topology under fault scenarios, resulting in a fault topology matrix. A′ Represented as Where ⊙ represents the Hadamard product, if a node or branch fails... F The corresponding element is 0 if it is in the middle, otherwise it is 1; Spatiotemporal aggregation is performed using an adaptive temporal topological graph convolutional network with node attention, and its node attention matrix... S ij ′ It is calculated by querying Q, key K, and value V; Based on formula Obtaining a unified feature representation Z Where σ is the activation function, A ij For adaptive adjacency matrix, X temp Θ is the feature matrix extracted by temporal dilated convolution, and Θ is the learnable filter weight matrix; The flow dynamics characterization module is used to derive the fault propagation path, including: Construct a fault propagation scenario tree with the initial fault state as the root node; A strategy combining the A* heuristic algorithm with breadth-first search is employed, through the evaluation function... f ( n ) =g ( n ) +h ( n Search for the evolutionary path with the highest risk, among which g ( n This represents the cumulative failure probability or cost loss from the initial failure to the current state. h ( n () is a heuristic estimation function that predicts the current state to the system's collapse boundary; It automatically identifies traffic flow redistribution paths caused by power outages in distribution network lines and charging stations, forming a complete scenario chain covering "fault source - propagation path - final effect".

9. The scene generation method as described in claim 1, characterized in that, The conditional diffusion model, based on the unified feature representation and the fusion decoupled generation architecture and physical information regularization training mechanism, generates extreme operating scenarios, including: A decoupled generation architecture is constructed, which processes the power node status, traffic segment flow, and charging station coupling information into mutually independent but logically related feature tags, and configures an independent noise state for each feature tag to simulate the asynchronous evolution of grid-side load transfer and traffic-side flow rerouting during the generation process. Using the Transformer architecture as the core of denoising, a conditional diffusion model is constructed; in the inverse denoising process, the conditional diffusion model represents the unified feature. Z Mask topology matrix F Heterogeneous features are fused from high-precision road network map features and power operation criteria, and the cascading flow transfer effect between long-distance road segments is characterized by a self-attention mechanism; its denoising evolution process is represented as follows: ,in for Noisy scene data at any given moment Resilient guiding objectives to minimize power loss losses; In the training of the conditional diffusion model, a physical information loss function is constructed. Loss phy The physical information loss function includes power flow balance loss. Traffic flow conservation loss Voltage deviation penalty and security status penalties At least one of the following, and the physical information loss function is used as part of the model training objective to force the generated data to meet physical constraints; In the training objective function of the conditional diffusion model, a Tikhonov regularization term is introduced. This term constrains the norm of the model's score function to prevent the model from collapsing to the center of the historical data distribution and to guide the model in exploring boundary conditions. The training objective function is expressed as follows: in, Indicates parameters The optimization objective is to minimize the expected loss of the variable. It is Gaussian noise. For the diffusion model with a conditional diffusion converter, They are respectively t The latent space state at time step, the diffusion time step, and the guiding condition vector. λ phy The weighting coefficients for the physical information loss term. The regularization coefficient is . For regularized fractional functions, This is the time-dependent regularization matrix.

Images