A Power Grid Construction Risk Event Evolution Prediction Method Based on the Global Supply Chain Structure

Abstract

This invention relates to the field of risk management technology for power grid construction projects, specifically disclosing a method for predicting the evolution of risk events in power grid construction based on a global supply chain structure. The method includes: constructing a risk-aware intelligent agent model based on multi-source heterogeneous data, outputting the probability distribution of node risk states; constructing a risk evolution prediction model based on Markov chains and Bellman equations, generating long-term cumulative risk loss and a comprehensive evolutionary risk index; and constructing a key resource allocation model based on biblical optimization, achieving the optimal trade-off between resource input and remedial costs under Wasserstein fuzzy sets. This invention achieves accurate perception, nonlinear evolution prediction, and proactive robust decision-making for risks in the power grid construction supply chain, significantly improving supply chain resilience and emergency response capabilities.

Description

Technical Field

[0001] This invention relates to the field of risk management technology for power grid construction projects, and more specifically to a method for predicting the evolution of risk events in power grid construction based on the global supply chain structure. Background Technology

[0002] Currently, existing power grid construction risk management technologies have the following shortcomings: First, they rely on static thresholds and passive response mechanisms at the risk perception level, making it difficult to achieve real-time fusion and dynamic perception of multi-source heterogeneous data; second, they rely too much on linear assumptions at the risk evolution modeling level, failing to characterize the nonlinear coupling and sudden transition phenomena of risks; third, they adopt static optimization methods at the resource allocation level, which are out of touch with environmental uncertainties and lack proactive decision-making capabilities; and fourth, multi-source data are isolated, resulting in insufficient model generalization ability.

[0003] Therefore, how to achieve accurate and dynamic risk perception and early warning, grasp the nonlinear evolution law and propagation path of risk, and improve the resilience and rapid recovery capability of the supply chain system are problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a method for predicting the evolution of power grid construction risk events based on the global structure of the supply chain, which overcomes or at least partially solves the above problems.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for predicting the evolution of power grid construction risk events based on the global supply chain structure includes the following steps: S1. Construct a risk-aware intelligent agent model. Each intelligent agent corresponds to a node in the power grid construction supply chain network. It is used to receive and process multi-source heterogeneous data and output the probability distribution of the risk status of each node at the current moment.

[0007] S2. Construct a risk evolution prediction model; The risk evolution prediction model is based on the probability distribution of risk states, uses Markov chains to simulate the dynamic transition process of node states over time, and combines the Bellman equation to calculate the long-term cumulative risk loss under each state, generating a comprehensive evolution risk index and a prediction of future evolution risk distribution.

[0008] S3. Based on the comprehensive evolutionary risk index and the prediction of future evolutionary risk distribution, a key resource allocation model is constructed. The dual reconstruction method is used to solve the key resource allocation model and output the optimal resource allocation scheme.

[0009] Preferably, in S1, the multi-source heterogeneous data includes internal attribute data, external environment data, and relationship interaction data.

[0010] Normalize multi-source heterogeneous data to obtain standardized feature vectors.

[0011] A weighted fusion function is constructed, and the weights of multi-source heterogeneous data are dynamically allocated through an attention mechanism to generate comprehensive risk feature values.

[0012] The comprehensive risk feature value is mapped to the probability distribution of the node in the preset risk state space through the Softmax function.

[0013] The deterministic risk state of a node at the current moment is determined based on the maximum a posteriori probability criterion.

[0014] Preferably, the internal attribute data includes material inventory level, task progress percentage, historical failure frequency, and material sampling pass rate; External environmental data includes raw material market price volatility index, weather suitability index, and policy production restriction indicators; The relationship interaction data includes the average latency of preceding nodes and the criticality index of nodes in the network.

[0015] Preferably, S2 includes: The risk state probability distribution output by the risk-aware agent model is used as the initial state vector of the Markov chain.

[0016] Based on the current environmental characteristics and dynamic transition process, a dynamic state transition function based on multinomial logistic regression is constructed, and the state transition probability is calculated.

[0017] Based on state transition probabilities and immediate risk losses, a Bellman equation is constructed, and a value iteration algorithm is used to solve for the long-term cumulative risk losses under each state.

[0018] The Bellman equation is expressed as:

[0019] in, Indicates the current state The immediate risk of loss, This represents the discount factor, used to reflect the degree to which risk diminishes over time. This is represented as the state transition probability. Represents the standardized feature vector. It represents the risk value of the possible state at the next moment.

[0020] Based on the initial state vector and dynamic state transition function, the system outputs a prediction of the future evolutionary risk distribution and calculates the comprehensive evolutionary risk index of the nodes.

[0021] Preferably, the comprehensive evolutionary risk index is obtained by weighted summation of the probability distribution of risk states and long-term cumulative risk losses.

[0022] Preferably, the key resource allocation model is a bibliometric optimization model based on a two-stage architecture, whose decision variables include resource input decision variables in the first stage and emergency relief decision variables in the second stage.

[0023] The critical resource allocation model aims to minimize the sum of resource input costs, key monitoring costs, and expected remediation costs under the worst-case probability distribution.

[0024] The worst-case probability distribution is selected from a fuzzy set centered on the predicted distribution output by the risk evolution prediction model and with the Wasserstein distance as the radius.

[0025] Preferably, the decision variables for resource allocation in the first stage include the initial amount of key resources allocated to each node, and the decision variable for whether to list the node as a key monitoring node.

[0026] The decision variables for the second stage of emergency response include the actual start time and the proportion of expedited work under random scenarios.

[0027] Preferably, the sub-Bruker optimization model includes risk exposure control constraints and dynamic resource capacity constraints.

[0028] Risk exposure control constraints are based on a comprehensive evolutionary risk index, which mandates that the sum of the risk indices of unmonitored nodes shall not exceed a preset safety threshold.

[0029] When a node with dynamic resource capacity constraints is listed as a key monitoring node, strategic reserve resources can be invoked.

[0030] Preferably, in S3, the process of solving the bibloc bar optimization model includes: The problem is transformed into a mixed-integer second-order cone programming problem using strong duality theory, and then solved iteratively using a column and constraint generation algorithm.

[0031] The main problem is to find the optimal resource allocation under a finite subset of scenarios, and the subproblem is the worst scenario with the greatest degree of constraint violation under the main problem allocation.

[0032] The column and constraint generation algorithm identifies the worst random scenario and adds it to the main problem through alternating iterations of the main problem and subproblems, and repeats the operation until convergence, outputting the optimal resource allocation scheme.

[0033] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method for predicting the evolution of power grid construction risk events based on the global structure of the supply chain. The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least the following: 1. It realizes dynamic perception and nonlinear evolution prediction of power grid construction supply chain risks, overcoming the shortcomings of traditional methods in terms of risk identification lag, linear evolution modeling, and static resource allocation; 2. The robustness and recovery capability of the system in uncertain environments are improved by using the sub-bar optimization method. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0035] Figure 1 This is a technical framework diagram provided in the embodiments of the present invention. Detailed Implementation

[0036] 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.

[0037] This invention discloses a method for predicting the evolution of power grid construction risk events based on the global supply chain structure, specifically including the following steps: S1. Construct a risk-aware intelligent agent model. Each intelligent agent corresponds to a node in the power grid construction supply chain network. It is used to receive and process multi-source heterogeneous data and output the probability distribution of the risk status of each node at the current moment.

[0038] In S1, multi-source heterogeneous data includes internal attribute data, external environment data, and relationship interaction data.

[0039] Normalize multi-source heterogeneous data to obtain standardized feature vectors.

[0040] A weighted fusion function is constructed, and the weights of multi-source heterogeneous data are dynamically allocated through an attention mechanism to generate comprehensive risk feature values.

[0041] The comprehensive risk feature value is mapped to the probability distribution of the node in the preset risk state space through the Softmax function.

[0042] The deterministic risk state of a node at the current moment is determined based on the maximum a posteriori probability criterion.

[0043] Preferably, the internal attribute data includes material inventory level, task progress percentage, historical failure frequency, and material sampling pass rate; External environmental data includes raw material market price volatility index, weather suitability index, and policy production restriction indicators; The relationship interaction data includes the average latency of preceding nodes and the criticality index of nodes in the network.

[0044] S2. Construct a risk evolution prediction model; The risk evolution prediction model is based on the probability distribution of risk states, uses Markov chains to simulate the dynamic transition process of node states over time, and combines the Bellman equation to calculate the long-term cumulative risk loss under each state, generating a comprehensive evolution risk index and a prediction of future evolution risk distribution.

[0045] S2 includes: The risk state probability distribution output by the risk-aware agent model is used as the initial state vector of the Markov chain.

[0046] Based on the current environmental characteristics and dynamic transition process, a dynamic state transition function based on multinomial logistic regression is constructed, and the state transition probability is calculated.

[0047] Based on state transition probabilities and immediate risk losses, a Bellman equation is constructed, and a value iteration algorithm is used to solve for the long-term cumulative risk losses under each state.

[0048] The Bellman equation is expressed as:

[0049] in, Indicates the current state The immediate risk of loss, This represents the discount factor, used to reflect the degree to which risk diminishes over time. This is represented as the state transition probability. Represents the standardized feature vector. It represents the risk value of the possible state at the next moment.

[0050] Based on the initial state vector and dynamic state transition function, the system outputs a prediction of the future evolutionary risk distribution and calculates the comprehensive evolutionary risk index of the nodes.

[0051] S3. Based on the comprehensive evolutionary risk index and the prediction of future evolutionary risk distribution, a key resource allocation model is constructed. The dual reconstruction method is used to solve the key resource allocation model and output the optimal resource allocation scheme.

[0052] The comprehensive evolutionary risk index is obtained by weighted summation of the probability distribution of risk states and long-term cumulative risk losses.

[0053] The critical resource allocation model is a bibliometric optimization model based on a two-stage architecture, whose decision variables include resource input decision variables in the first stage and emergency relief decision variables in the second stage.

[0054] The decision variables for resource allocation in the first phase include the initial amount of critical resources allocated to each node, and whether to list that node as a key monitoring node.

[0055] The decision variables for the second stage of emergency response include the actual start time and the proportion of expedited work under random scenarios.

[0056] Preferably, the sub-Bruker optimization model includes risk exposure control constraints and dynamic resource capacity constraints.

[0057] Risk exposure control constraints are based on a comprehensive evolutionary risk index, which mandates that the sum of the risk indices of unmonitored nodes shall not exceed a preset safety threshold.

[0058] When a node with dynamic resource capacity constraints is listed as a key monitoring node, strategic reserve resources can be invoked.

[0059] The critical resource allocation model aims to minimize the sum of resource input costs, key monitoring costs, and expected remediation costs under the worst-case probability distribution.

[0060] The worst-case probability distribution is selected from a fuzzy set centered on the predicted distribution output by the risk evolution prediction model and with the Wasserstein distance as the radius.

[0061] In S3, the process of solving the bipartite bar optimization model includes: The problem is transformed into a mixed-integer second-order cone programming problem using strong duality theory, and then solved iteratively using a column and constraint generation algorithm.

[0062] The main problem is to find the optimal resource allocation under a finite subset of scenarios, and the subproblem is the worst scenario with the greatest degree of constraint violation under the main problem allocation.

[0063] The column and constraint generation algorithm identifies the worst random scenario and adds it to the main problem through alternating iterations of the main problem and subproblems, and repeats the operation until convergence, outputting the optimal resource allocation scheme.

[0064] like Figure 1 As shown, a specific embodiment will be used to explain the current prediction method in detail.

[0065] S1. Construct a risk-aware intelligent agent model. The agent perceives risks from multi-source heterogeneous data. By inputting external environmental data, internal attribute data, and relational interaction data, the agent's attention mechanism is used to perceive risks. This lays the groundwork for a risk evolution prediction model based on Markov chains and Bellman equations. The results serve as the initial input for the risk evolution prediction model.

[0066] To achieve real-time risk perception across multiple elements including "people, machines, materials, and environment," this model defines a set of risk perception agents. Each intelligent agent corresponds to a node in the supply chain network, that is, to each node in the directed acyclic graph constructed based on engineering construction.

[0067] intelligent agent At any moment The specific data input structure is defined as the feature vector. The vector consists of three parts:

[0068] in Represents the eigenvector. Represents internal attribute data. Represents external environment data. This represents data exchanged based on relationships.

[0069] Internal attribute data Including current inventory levels Current task progress percentage Frequency of historical faults / defaults and the recent pass rate of material spot checks .

[0070] External environment data Including raw material market price volatility index Meteorological suitability index at the construction site And policy / environmental protection production and employment restrictions quotas .

[0071] Relationship Interaction Data Including the average delay time of the preceding node And the key index of this node in the CPM network .

[0072] Optionally, normalization processing is performed on multi-source heterogeneous data, since the input data , , Having different dimensions and physical units, the Min-Max normalization method is first used to map all features to... Interval, eliminating the influence of dimensions. Let the eigenvector be... The first in Each feature component is The standardized value is :

[0073] Thus, the standardized feature vector is obtained. .

[0074] Next, the normalized multi-source heterogeneous data is fused. Specifically, the agent uses an attention mechanism to dynamically allocate weights from different data sources, establishes a weighted fusion function, and calculates the comprehensive risk score of the nodes. Combined with the standardized feature vector of the input This process is achieved through a single-layer neural network:

[0075] in, The feature weight matrix, obtained through prior data training, reflects the contribution of each element to the risk. The bias vector represents the inherent basic risk level of a node. This represents the activation function, used to introduce nonlinear factors and simulate the nonlinear characteristics of a surge in risk; This represents the overall risk characteristic value of the output.

[0076] The agent outputs the current risk state (normal) of each risk state. ,fluctuation Interruption The probability distribution of the comprehensive risk characteristics is obtained by using the Softmax function. Mapped to state space probability distribution vector on :

[0077] in, Represents a node At any moment In the first The probability of a certain risk state; Represents the class weight vector of the classifier; output result .

[0078] The agent needs to output a unique current state label. The maximum possible unique state is obtained by using the maximum a posteriori probability criterion:

[0079] The agent, based on the feature vector Monitor and output the initial risk status of the current node. .

[0080] S2. A risk evolution probability prediction model based on Markov chains and Bellman equations is used. Markov chains describe the dynamic evolution of risk events over time, and Bellman equations are used to calculate the long-term cumulative effect of risk. By mapping the real-time state probability distribution output by the risk-aware agent model to the initial state vector of the Markov chain and Bellman equation risk evolution prediction model, coupling from static perception to dynamic evolution is achieved. Specific steps include: Let intelligent agents At any moment The output risk state probability distribution is Define the initial state vector of the Markov prediction model. The value is directly taken from this probability distribution:

[0081] Risk evolution is not a static process; the state transition probability is influenced by the standardized eigenvector. The real-time impact (such as deteriorating weather or large-scale upstream delays) is considered. This model uses multinomial logistic regression to construct a dynamic transfer function, with nodes... From the current state Transition to the next state The probability is The calculation formula is as follows:

[0082] in, For state Transferred to The feature weight vector, This is the state space.

[0083] The parameters are learned using maximum likelihood estimation in order to determine the weight vector. Constructing a log-likelihood function using prior datasets The stochastic gradient ascent algorithm is used to iteratively solve the problem until convergence.

[0084]

[0085] in, For learning rate, .

[0086] This process enables the model to learn from prior data the driving mechanisms of different environmental factors on risk evolution.

[0087] Next, risk loss assessment is conducted based on the Bellman equation. To predict the long-term impact of risk evolution, a risk value function is introduced. This function indicates the state of the system. At that time, the expected cumulative risk loss in the future, according to the recursive definition of the Bellman equation:

[0088] in, Current state The immediate risk of loss. For example, when... In this case, the immediate loss is the cost of work stoppage for that day.

[0089] Discount factor ( This reflects the degree to which risk diminishes over time.

[0090] : State transition probability.

[0091] Value at risk of the possible state at the next moment.

[0092] The risk loss is calculated using a value iteration algorithm: Initialization: Let Set a convergence threshold .

[0093] Iterative Update: Entering the... In the next iteration, calculate:

[0094] Convergence criterion: If If the iteration stops, output the final cumulative risk loss. .

[0095] Based on the above calculations, two types of key parameters are generated and input into the key resource allocation model. a) Future evolutionary distribution prediction: based on the initial vector. and transition matrix Predicting the future State distribution at time: a) Used for Monte Carlo sampling to generate a set of scenarios with random durations and costs; b) Comprehensive evolutionary risk index: Combining current probability and long-term potential, calculate the comprehensive risk score of nodes, which serves as a weighting coefficient in the optimization objective function: The higher the index, the higher the robustness weight that needs to be assigned to that node in the optimization model.

[0096] S3. Based on the comprehensive evolutionary risk index and future state distribution scenarios, a key resource allocation model is constructed. This key resource allocation model does not rely on a single exact distribution, but instead seeks the optimal solution under the worst distribution in a fuzzy set centered on the predicted distribution, thereby ensuring that the decision-making scheme remains robust under extreme risks.

[0097] The optimization model based on Wasserstein distance aims to minimize the sum of the initial resource cost and the expected remedial cost under the worst distribution, while satisfying all engineering constraints.

[0098] Variables for resource input decisions in the first stage: Assigned to task resources The initial quantity.

[0099] List of key monitoring targets.

[0100] : Should the task be... Listed as a key monitoring node (triggered by the ERI index).

[0101] Second-stage emergency response decision variables: In random scenarios Next, the task The actual start time.

[0102] In random scenarios Next, regarding the task The proportion of work done to expedite the process.

[0103] The model-based solution algorithm utilizes strong duality theory to solve the inner-layer maximization problem. This is transformed into a finite-dimensional minimization problem. The degenerate bar optimization model utilizes the strong duality theorem to equivalently transform the original DRO problem into a mixed-integer second-order cone programming problem:

[0104]

[0105]

[0106] : The support set of random parameters, including random vectors such as project schedule fluctuations and cost changes.

[0107] The objective function is:

[0108] This indicates the cost of purchasing and allocating basic resources.

[0109] This indicates the additional overhead of configuring monitoring methods for high-risk nodes.

[0110] This represents the predicted distribution output by the risk evolution prediction model. Centered on Wasserstein fuzzy set with radius middle Find the worst distribution And minimize the expected two-stage loss under this distribution. .

[0111] The key parameters include: : Predicted distribution scene set generated by Markov model This constitutes an empirical distribution.

[0112] : For nodes Fixed costs of implementing key monitoring.

[0113] : Robust radius, representing the degree of distrust in the prediction accuracy of the risk evolution prediction model.

[0114] The maximum residual risk exposure threshold allowed by the system.

[0115] The constraints include: Constraint 1: Risk Exposure Control Constraint Based on ERI. To force the system to focus on high-risk nodes, it is stipulated that the sum of the risk indices of unmonitored nodes must not exceed a safety threshold. This will force the model to prioritize high-risk nodes. The node is set as .

[0116]

[0117] The set of nodes in a power grid engineering project network. .

[0118] The comprehensive evolution risk index reflects the nodes Potential risk potential.

[0119] Constraint 2: Dynamic resource capacity constraints triggered by monitoring. Nodes listed as key monitoring points ( This will grant access to strategic reserve resources, thereby increasing resource availability. To mitigate construction delays.

[0120]

[0121] : Standard resource library capacity.

[0122] The standby resource quota released by single-node monitoring is used to unlock dynamic resource capacity. .

[0123] : A collection of key resource types .

[0124] Constraint 3: Second-stage ex post facto remedial variable function definition:

[0125] st

[0126] in, This is the penalty coefficient for overtime, and the construction period. Resource input The function, and subject to random vectors Disturbance.

[0127] Furthermore, a column and constraint generation algorithm is used to iteratively solve the transformed main problem and subproblems: The main problem is to find the optimal resource allocation within a finite subset of scenarios. .

[0128] Sub-problem: Find the worst-case scenario that maximizes constraint violation under the current configuration. .

[0129] Iteration: Add the identified adverse scenarios to the main problem, repeat until convergence, and output the optimal resource allocation scheme.

[0130] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0131] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the evolution of power grid construction risk events based on the global supply chain structure, characterized in that, Includes the following steps: S1. Construct a risk-aware intelligent agent model. Each intelligent agent corresponds to a node in the power grid construction supply chain network. It is used to receive and process multi-source heterogeneous data and output the probability distribution of the risk status of each node at the current moment. S2. Construct a risk evolution prediction model; the risk evolution prediction model is based on the probability distribution of the risk state, uses Markov chains to simulate the dynamic transition process of node states over time, and combines the Bellman equation to calculate the long-term cumulative risk loss under each state, generating a comprehensive evolution risk index and a prediction of future evolution risk distribution. S3. Based on the comprehensive evolutionary risk index and the prediction of future evolutionary risk distribution, construct a key resource allocation model, solve the key resource allocation model using the dual reconstruction method, and output the optimal resource allocation scheme.

2. The method for predicting the evolution of power grid construction risk events based on the global supply chain structure according to claim 1, characterized in that, In S1, the multi-source heterogeneous data includes internal attribute data, external environment data, and relationship interaction data; The multi-source heterogeneous data is normalized to obtain a standardized feature vector; A weighted fusion function is constructed, and the weights of the multi-source heterogeneous data are dynamically allocated through an attention mechanism to generate a comprehensive risk feature value. The comprehensive risk feature value is mapped to the probability distribution of the node in the preset risk state space through the Softmax function; The deterministic risk state of a node at the current moment is determined based on the maximum a posteriori probability criterion.

3. The method for predicting the evolution of power grid construction risk events based on the global supply chain structure according to claim 2, characterized in that, The internal attribute data includes material inventory levels, task progress percentage, historical failure frequency, and material sampling pass rate. The external environment data includes raw material market price fluctuation index, weather suitability index and policy production restriction indicators; The relationship interaction data includes the average latency of the preceding nodes and the criticality index of the nodes in the network.

4. The power grid construction risk event evolution prediction method based on the global supply chain structure according to claim 2, characterized in that, S2 include: The risk state probability distribution output by the risk-aware agent model is used as the initial state vector of the Markov chain. Based on the current environmental characteristics and the dynamic transition process, a dynamic state transition function based on multinomial logistic regression is constructed, and the state transition probability is calculated. Based on the state transition probabilities and immediate risk losses, the Bellman equation is constructed, and the long-term cumulative risk losses under each state are solved using the value iteration algorithm. The Bellman equation is expressed as: in, Indicates the current state The immediate risk of loss, This represents the discount factor, used to reflect the degree to which risk diminishes over time. This is represented as the state transition probability. Represents the standardized feature vector. Value at risk represents the possible state at the next moment; Based on the initial state vector and the dynamic state transition function, the future evolution risk distribution prediction is output, and the comprehensive evolution risk index of the node is calculated.

5. The method for predicting the evolution of power grid construction risk events based on the global supply chain structure according to claim 4, characterized in that, The comprehensive evolutionary risk index is obtained by weighted summation of the probability distribution of the risk state and the long-term cumulative risk loss.

6. The method for predicting the evolution of power grid construction risk events based on the global supply chain structure according to claim 1, characterized in that, The key resource allocation model is a two-stage architecture-based sub-Brussels bar optimization model, whose decision variables include the first-stage resource input decision variables and the second-stage emergency relief decision variables. The key resource allocation model aims to minimize the sum of resource input costs, key monitoring costs, and expected remediation costs under the worst-case probability distribution. The worst-case probability distribution is selected from a fuzzy set centered on the predicted distribution output by the risk evolution prediction model and with the Wasserstein distance as the radius.

7. The method for predicting the evolution of power grid construction risk events based on the global supply chain structure according to claim 6, characterized in that, The resource allocation decision variables in the first phase include the initial amount of critical resources allocated to each node, and the decision variable of whether to list the node as a key monitoring node. The decision variables for the second stage of emergency remediation include the actual start time and the proportion of expedited work under random scenarios.

8. The method for predicting the evolution of power grid construction risk events based on the global supply chain structure according to claim 7, characterized in that, The sub-Bruker optimization model includes risk exposure control constraints and dynamic resource capacity constraints; The risk exposure control constraint is based on the comprehensive evolution risk index, and forces the sum of the risk indices of unmonitored nodes not to exceed a preset safety threshold; When the dynamic resource capacity constraint is listed as a key monitored node, strategic reserve resources can be invoked.

9. The method for predicting the evolution of power grid construction risk events based on the global supply chain structure according to claim 6, characterized in that, In S3, the process of solving the sub-Bruker optimization model includes: The problem is transformed into a mixed-integer second-order cone programming problem using strong duality theory, and then solved iteratively using a column and constraint generation algorithm. The main problem is to find the optimal resource allocation under a finite subset of scenarios, and the sub-problem is the worst scenario with the greatest degree of constraint violation under the configuration of the main problem. The column and constraint generation algorithm identifies the worst-case random scenario and adds it to the main problem through alternating iterations of the main problem and the sub-problems, repeating the operation until convergence, and outputs the optimal resource allocation scheme.

Images