Multi-type mobile emergency resource optimal scheduling based on data-model hybrid driving method and power distribution network resilience improvement strategy

Abstract

The application provides a multi-type mobile emergency resource optimal scheduling and power distribution network resilience improvement strategy based on a data-model hybrid driving method, which starts from the dynamic interaction relationship between mobile energy storage vehicles and emergency repair teams, coordinates the mobile energy storage vehicles and the emergency repair teams to carry out load recovery and road repair, adjusts the emergency repair teams to repair the damaged roads of the traffic network under the consideration of the uncertain duration required for repairing the roads, and adjusts the mobile energy storage vehicles to recover the load of the power distribution network nodes under the consideration of the dynamic change of the road state of the traffic network; a graph neural network multi-agent reinforcement learning algorithm GATD3QN based on a graph attention mechanism is used to support the data-driven part, so as to solve the optimal multi-type mobile emergency resource routing behavior strategy; the scheduling model of the mobile energy storage vehicles and the emergency repair teams, the power distribution network reconstruction model and the power distribution network optimal power flow model are constructed into an MISOCP model, and the optimal multi-type mobile emergency resource scheduling behavior strategy is solved.

Description

Technical Field

[0001] This invention belongs to the field of distribution network resilience enhancement technology, specifically involving a strategy for optimizing the scheduling of multiple types of mobile emergency resources and enhancing distribution network resilience based on a data-model hybrid driving method. Background Technology

[0002] As a critical infrastructure, the power distribution system is the foundation for the normal operation of other infrastructures. Its safe and stable operation plays an extremely important role in national economic development and social stability. However, in recent years, extreme events such as natural disasters and man-made attacks have increasingly threatened the safety of power distribution systems. The impact of extreme events on the power distribution network is extremely wide-ranging. In order to minimize the impact of extreme events, it is necessary to further improve the safe and stable operation capability of the power distribution network. Research on strategies to enhance the resilience of the power distribution network has become an inevitable trend in the new era.

[0003] With the rapid development of urban electrification, the number of charging stations in transportation networks has surged, the energy exchange capacity between transportation and distribution networks has continuously improved, and the mobile energy storage flowing through transportation networks has increased significantly, forming a large-scale, diverse, and topologically flexible discrete energy network. This provides a material and energy foundation for enhancing the resilience of distribution networks. Mobile energy storage vehicles (MEVs) are vehicles capable of storing and flexibly delivering electrical energy to different locations. In the event of natural disasters or power outages, MEVs can quickly connect to charging station nodes to provide temporary power to critical facilities and equipment, making them the most widely used emergency power supply resource for power grids. However, extreme events can also directly or indirectly affect transportation networks, causing road damage and traffic paralysis. This significantly limits the mobility of MEVs within transportation networks, impacting the resilience of distribution networks. Therefore, it is necessary to dispatch emergency repair teams to repair damaged roads. Most existing research on distribution network resilience enhancement assumes a fixed road fault repair time, or even that the fault persists, which is inconsistent with reality. Furthermore, existing research neglects the impact of the dynamic interaction of mobile emergency resources on optimal dispatch decisions. Some studies have chosen pure model-based methods to solve mathematical models for the stochastic optimal scheduling of mobile emergency resources. While this method offers advantages such as strong interpretability and relatively accurate results, it suffers from poor generalization ability and is susceptible to the influence of numerous integer variables and nonlinear constraints, resulting in poor real-time performance and failing to meet the real-time decision-making requirements of practical mobile emergency resources. Other techniques apply reinforcement learning methods (pure data-based methods) to the optimal scheduling of mobile emergency resources, enabling agents to make real-time decisions. However, these methods suffer from drawbacks such as long training times and the potential to find infeasible strategies. Summary of the Invention

[0004] Therefore, in view of the defects and shortcomings of the existing technologies, it is necessary to find an efficient and accurate method to solve the multi-type mobile emergency resource scheduling strategy that takes into account the dynamic changes of road conditions and adjacency relationships in the transportation network, so as to improve the resilience of the distribution network after extreme events.

[0005] In view of this, the purpose of this invention is to provide a multi-type mobile emergency resource scheduling and distribution network resilience enhancement strategy based on a data-model hybrid driving method. This strategy starts from the dynamic interaction between mobile energy storage vehicles and emergency repair teams, and coordinates both mobile energy storage vehicles and emergency repair teams for load restoration and road maintenance. This strategy can consider scheduling emergency repair teams to repair damaged roads in the transportation network when the time required for road repair is uncertain, and can also consider scheduling mobile energy storage vehicles to restore loads to distribution network nodes when the road conditions of the transportation network change dynamically. Furthermore, this strategy innovatively combines data methods and model methods to solve the aforementioned complex, high-dimensional, nonlinear, multi-type mobile emergency resource scheduling problem. The data-driven part is supported by the GATD3QN graph neural network multi-agent reinforcement learning algorithm based on graph attention mechanisms. This algorithm combines the advantages of attention-based graph neural networks, Double DQN, and Dueling DQN algorithms to optimize the feature extraction capability and Q-value calculation of traditional DQN algorithms. It also employs a priority experience replay strategy to improve the algorithm's sampling efficiency and training effect. This algorithm can deeply mine the characteristic information in the power-transportation coupled network graph data and the adjacency relationships of mobile emergency resources to solve for the optimal routing behavior strategy for multiple types of mobile emergency resources. The model-driven part constructs the scheduling model of mobile energy storage vehicles and emergency repair teams, the distribution network reconfiguration model, and the optimal power flow model of the distribution network into the MISOCP model and solves for the optimal scheduling behavior strategy for multiple types of mobile emergency resources, including power scheduling decisions for mobile energy storage vehicles and road repair or return-to-warehouse decisions for emergency repair teams, to improve the resilience of the distribution network.

[0006] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0007] A multi-type mobile emergency resource optimization scheduling and distribution network resilience enhancement strategy based on a data-model hybrid approach is proposed. Starting from the dynamic interaction between mobile energy storage vehicles and emergency repair teams, it coordinates both vehicles for load restoration and road repair. It considers scheduling emergency repair teams to repair damaged roads in the transportation network under uncertain road repair times, and scheduling mobile energy storage vehicles to restore loads to distribution network nodes under dynamic changes in road conditions. The approach combines data and model methods to solve the above multi-type mobile emergency resource scheduling problems. The data-driven part is supported by the GATD3QN multi-agent reinforcement learning algorithm based on graph attention mechanisms. The GATD3QN algorithm combines attention-based graph neural networks, Double DQN algorithms, and Dueling... The DQN algorithm optimizes the feature extraction capability and Q-value calculation of the traditional DQN algorithm, and adopts a priority experience replay strategy to improve the sampling efficiency and training effect of the algorithm, so as to solve the optimal routing behavior strategy for multiple types of mobile emergency resources. The model-driven part constructs the scheduling model of mobile energy storage vehicles and emergency repair teams, the distribution network reconfiguration model and the optimal power flow model of distribution network into the MISOCP model, and solves the optimal scheduling behavior strategy for multiple types of mobile emergency resources, including the power scheduling decision of mobile energy storage vehicles, the road repair decision or return to the warehouse decision of emergency repair teams, so as to improve the resilience of distribution network.

[0008] Furthermore, the strategy includes the following steps when executed:

[0009] Step S1: Construct and initialize the power-transportation coupled network model, the mobile energy storage vehicle graph reinforcement learning model, and the emergency repair team graph reinforcement learning model;

[0010] Step S2: Simulate a power-transportation coupled network failure scenario after an extreme event occurs, and set up line disconnection in the distribution network and road damage in the transportation network;

[0011] Step S3: Initialize the environment for the graph reinforcement learning algorithm and the fault scenario for the power-transportation coupled network;

[0012] Step S4: Based on the power-transportation coupled network information and the mobile emergency resource's own state information, construct a graph neural network reinforcement learning algorithm to determine the state of the mobile energy storage vehicle. and the status of the emergency repair team Obtain the state matrix o t Construct an adjacency matrix A based on the adjacency relationships of mobile emergency resources. t ;

[0013] Step S5: Calculate the mobile energy storage vehicle routing behavior strategy based on the ε-Greedy policy and the graph neural network reinforcement learning algorithm as the data-driven part. Routing behavior strategy of emergency repair teams

[0014] Step S6: Execute the mobile energy storage vehicle routing behavior policy respectively. Routing behavior strategy of emergency repair teams It also assesses and updates the status of mobile energy storage vehicles and emergency repair teams;

[0015] Step S7: The model-driven part calculates the power scheduling strategy and reward function r of the mobile energy storage vehicle based on the mobile energy storage vehicle scheduling model combined with the distribution network reconfiguration model and the distribution network optimal power flow model. t M Based on the emergency repair team scheduling model, the maintenance scheduling strategy and reward function r of the emergency repair team are calculated. t R ;

[0016] Step S8: Update the mobile energy storage vehicle state using the graph neural network reinforcement learning algorithm. and the status of the emergency repair team Obtain the next state matrix o t+1 Update the adjacency matrix A t+1 ;

[0017] Step S9: Transfer the information from the current step and initial priority Storing in the mobile energy storage vehicle image reinforcement learning memory unit D M ,Will and initial priority Stored in the emergency repair team's image reinforcement learning memory unit D R ;

[0018] Step S10: The importance of samples is sampled for the memory cells using a priority experience replay strategy, and the weights of the graph neural networks for the mobile energy storage vehicle and the emergency repair team are updated based on the stochastic gradient descent method.

[0019] Step S11: Determine if the end time T has been reached. End If not, proceed to steps S3 to S10;

[0020] Step S12: Determine whether the training end sequence number E has been reached. End If not, proceed to steps S3 to S11; if yes, output the graph reinforcement learning algorithm parameters, routing, and scheduling strategies for the mobile energy storage vehicle and the emergency repair team.

[0021] Further, in step S1, the power-transportation coupled network model, the mobile energy storage vehicle graph reinforcement learning model, and the emergency repair team graph reinforcement learning model are initialized, and the number of training sessions is initialized.

[0022] The nodes and connecting lines in the distribution network are considered as points and edges, forming the basic distribution network diagram; charging station nodes are added to this diagram to create a distribution network model. Where N PDS L represents the set of nodes in the distribution network. PDS Represents a set of distribution network lines. Represents the set of charging station nodes;

[0023] The basic transportation network graph is constructed by treating the nodes and roads connecting them as points and edges; charging station nodes and emergency repair team warehouses are added to complete the transportation network graph model. Where N TN L represents the set of nodes in the transportation network. TN Represents the set of roads in a transportation network; This represents the set of charging station nodes in the transportation network; This represents the set of emergency repair team warehouse nodes in the transportation network;

[0024] By treating all mobile energy storage vehicles and emergency repair teams as agents, and each agent as a node, while the connections between them are considered as edges, a dynamic network model of mobile emergency resources is constructed. Where M represents a collection of mobile energy storage vehicles, and W represents a collection of emergency repair teams. For mobile emergency resource adjacency relationships The adjacency matrix formed.

[0025] Furthermore, in step S2, some lines in the distribution network are randomly interrupted to construct a set L of damaged lines. PDS,off This is used to simulate the outage of weak lines in the power distribution network after extreme events; and a set of damaged roads is constructed by randomly setting up some damaged roads in the transportation network. This simulates situations where parts of the transportation network are blocked due to obstacles caused by extreme events.

[0026] Furthermore, step S3 specifically includes: initializing various variables in the graph reinforcement learning algorithm environment and initializing the power-transportation coupled network fault scenario, and initializing the training time.

[0027] Further, in step S4:

[0028] Based on the power-transportation coupled network information and the state information of mobile emergency resources, a state graph neural network reinforcement learning algorithm for mobile energy storage vehicles is constructed. Including the mobile energy storage vehicle's own status information Adjacent mobile emergency resource status information Traffic network road information for the location Distribution network node load data P t EDNew energy output P t WT ;

[0029] Based on the power-transportation coupled network information and the state information of mobile emergency resources, a state graph neural network reinforcement learning algorithm for emergency repair teams is constructed. Including the mobile energy storage vehicle's own status information Adjacent mobile emergency resource status information Traffic network road information for the location Traffic network road fault information and the time required to repair damaged roads

[0030] Then, construct an adjacency matrix A based on the adjacency relationships of mobile emergency resources. t Adjacency matrix A t It is used as input into graph reinforcement learning algorithms to participate in the calculation of state-action values.

[0031] Furthermore, in step S5, firstly, the random number p at time t is calculated. t Its value range is [0,1), and the ε-Greedy strategy is used to compare the current ε. t Value and random number p t If there is p t <ε t Then, a random method is used to generate the routing behavior strategy for mobile energy storage vehicles. Routing behavior strategy of emergency repair teams

[0032] If there is p t ≥ε t Then, a routing behavior strategy for the mobile energy storage vehicle is generated using a graph neural network multi-agent reinforcement learning algorithm. Routing behavior strategy of emergency repair teams

[0033] The graph reinforcement learning algorithm for mobile energy storage vehicles and the graph reinforcement learning algorithm for emergency repair teams use the same neural network structure: one neural network controls multiple mobile energy storage vehicles, and another neural network controls multiple emergency repair teams; the neural network structure is as follows: first, a fully connected input layer, which takes the input mobile emergency resource state set o as input. t Perform preliminary feature extraction and dimension transformation, and output. This is followed by two graph attention network layers, which apply an attention mechanism to the state matrix. Mobile emergency resource adjacency matrix A t Attention coefficients are calculated, and features are extracted based on the attention matrix. The output is based on adjacency relationships. Features The final layer employs the Dueling DQN algorithm strategy, splitting the feature information into two fully connected branches: the first branch outputs the scalar value V(o) of the state function. t A t The output status of the second branch. The action advantage value function vector

[0034] Furthermore, in step S7, a mobile energy storage vehicle scheduling model, a distribution network reconfiguration model, and a distribution network optimal power flow model were constructed.

[0035] An emergency repair team dispatch model was constructed;

[0036] Construct an objective function based on the tasks of load restoration at distribution network nodes and road repair for damaged transportation networks;

[0037] During the calculation, constraints (34) and (35) were linearized using the Big M method, and constraint (39) was linearized using the second-order cone relaxation method; constraints (46) and (47) of the emergency repair team scheduling problem were linearized using the McCormick envelope method.

[0038] The model-driven part uses formulas (51)-(53) as the objective function and formulas (20)-(50) as constraints to construct a mixed integer second-order cone programming model (MISOCP), which is solved using the Gurobi solver. The solution yields the load of the distribution network nodes restored by the mobile energy storage vehicle, and the reward function of the mobile energy storage vehicle is calculated using formula (54).

[0039]

[0040] In the formula, Let represent the set of distribution network nodes whose load is supported by the m-th mobile energy storage vehicle at time t; Let t be the distribution network node The amount of load recovery; For distribution network nodes The load level weighting coefficient; the reward value for the mobile energy storage vehicle during its movement between charging stations in the transportation network is 0;

[0041] By solving the MISOCP, the road maintenance strategy or dispatch strategy of the emergency repair team can be obtained, and the reward function of the emergency repair team can be calculated.

[0042] Furthermore, in step S10, in the mobile energy storage vehicle image reinforcement learning memory unit D... M Samples N were collected using an importance sampling method. s And the importance weight of each sample is calculated using formula (62):

[0043]

[0044] In the formula, N D Represents memory unit D M The total number of samples in the sample; This indicates the priority of the sample; β is a hyperparameter.

[0045] Combining the strategy of the Double DQN algorithm, the loss function is calculated using formulas (63) and (64):

[0046]

[0047] In the formula, and These are the current network parameters and the target network parameters of the mobile energy storage vehicle, respectively; γ is a discount factor reflecting the impact of the future Q value on the current action, with a value of [0,1].

[0048] The neural network parameters of the mobile energy storage vehicle graph reinforcement learning algorithm are updated based on the stochastic gradient descent method and sample importance weights, as specifically expressed as follows:

[0049]

[0050] In the formula, α M The learning rate of the gradient descent algorithm for the current network of the mobile energy storage vehicle; and after a fixed number of steps N. up Then, the parameters of the target network for the mobile energy storage vehicle are configured using the following method. Update:

[0051]

[0052] The TD-error value is calculated according to formula (63). Calculate the new priority for each sample Where ∈ is a constant used to prevent the priority from being zero; in the mobile energy storage vehicle graph reinforcement learning memory unit D M The priority of each sample is updated in the process;

[0053] The parameters of the graph reinforcement learning algorithm neural network for emergency repair teams are updated.

[0054] And an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the steps of the multi-type mobile emergency resource optimization scheduling and distribution network resilience enhancement strategy based on the data-model hybrid driving method as described above.

[0055] Compared with existing technologies, this invention and its preferred solutions can solve the complex, high-dimensional, nonlinear, stochastic optimization scheduling problem of multiple types of mobile emergency resources, and its beneficial effects include at least the following:

[0056] First, this invention dispatches emergency repair teams to repair faulty roads in the transportation network, taking into account the impact of dynamic changes in the condition of the transportation network on the resilience of the power distribution network. This avoids the problems of existing technologies that ignore road damage or assume that road repair takes a fixed time. As a result, this invention can be more effectively applied to the emergency repair of power-transportation coupled networks after disasters, and achieves better results in improving the resilience of the power distribution network.

[0057] Secondly, this invention delves into the impact of dynamic interaction relationships among various types of mobile emergency resources on their collaborative scheduling. All mobile emergency resources (i.e., mobile energy storage vehicles and emergency repair teams) are treated as agents, and the dynamic adjacency relationships of all agents are abstracted as edges. This transforms the collaborative and cooperative relationships of all mobile energy storage vehicle and emergency repair team agents into a dynamic graph structure of mobile emergency resources. Simultaneously, effective feature information is extracted from the traffic flow changes generated by the movement of mobile emergency resources in the transportation network and the changes in adjacency relationships within the dynamic graph structure of mobile emergency resources to solve for the optimal routing and scheduling strategies.

[0058] Finally, this invention combines the efficiency of data-driven methods with the accuracy of model-driven methods to propose a novel data-model hybrid driving approach. This method can more effectively solve mathematical models for optimizing the scheduling of various types of mobile emergency resources, considering uncertainties such as road repair time, dynamic changes in adjacency relationships, traffic flow variations, and the time required for road repair. In the data-driven part, to study the dynamic interaction relationships of mobile emergency resources, this method constructs a novel graph attention neural network multi-agent reinforcement learning algorithm, GATD3QN. Compared to existing technologies, this algorithm abandons the use of traditional fully connected neural networks and instead employs an attention-based graph neural network to extract features from complex graph information and dynamic adjacency relationships. Compared to traditional fully connected neural networks, the graph attention neural network, by introducing node-level attention, can focus on the most relevant adjacent nodes in the dynamic graph and effectively aggregate information, improving the model's ability to capture important local and global patterns, thereby enhancing the ability of mobile emergency resources to mine effective information from adjacent agents. Furthermore, addressing the shortcomings of traditional DQN algorithms such as Q-value overestimation, training instability, and low sample efficiency, the GATD3QN algorithm combines the advantages of Double DQN and Dueling DQN algorithms to optimize the feature extraction capability and Q-value calculation of traditional DQN algorithms. It also employs a Prioritized Experience Replay (PER) strategy to improve the algorithm's sampling efficiency and training effect (corresponding to the design of the GATD3QN algorithm, which samples from the sample experience pool and calculates TD-error for back-updating neural network parameters). The data-driven part employs the GATD3QN algorithm, which combines the excellent non-Euclidean data information extraction capability of graph attention networks with the powerful sequential stochastic optimization decision-making capability of reinforcement learning algorithms. This effectively extracts key information from the dynamic adjacency relationships and power-transportation coupled network graph data of mobile energy storage vehicles and emergency repair teams to formulate routing behavior strategies for these vehicles. In the model-driven part, the method constructs the scheduling model of mobile energy storage vehicles and emergency repair teams, the distribution network reconfiguration model, and the optimal power flow model of the distribution network into the MISOCP model and solves for the optimal multi-type mobile emergency resource scheduling behavior strategies, including power scheduling decisions for mobile energy storage vehicles and road repair or return-to-warehouse decisions for emergency repair teams.

[0059] In summary, this invention can simultaneously consider the dynamic changes in road conditions of the transportation network and the dynamic changes in the relationships between adjacent mobile emergency resources. By solving the routing and scheduling strategies of mobile energy storage vehicles and emergency repair teams through a data-model hybrid driving method, it restores the load of distribution network nodes and repairs damaged roads in the transportation network, thereby improving the resilience of the distribution network. Attached Figure Description

[0060] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0061] Figure 1 This is a flowchart illustrating the strategy implementation of an embodiment of the present invention. Detailed Implementation

[0062] To make the features and advantages of this patent more apparent and understandable, specific embodiments are provided below for detailed explanation:

[0063] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0064] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0065] like Figure 1 The diagram illustrates a strategy for optimizing and scheduling multiple types of mobile emergency resources and enhancing the resilience of the distribution network based on a data-model hybrid driving method, as provided in this embodiment of the invention. The implementation process includes the following steps:

[0066] S1: Initialize the power-transportation coupled network model, the mobile energy storage vehicle graph reinforcement learning model, and the emergency repair team graph reinforcement learning model;

[0067] S2: Simulates a power-transportation coupled network fault scenario after an extreme event: a line disconnection is set in the distribution network, and road damage is set in the transportation network;

[0068] S3: Initialization of the graph reinforcement learning algorithm environment and initialization of the power-transportation coupled network fault scenario.

[0069] S4: Based on the power-transportation coupled network information and the state information of the mobile emergency resources themselves, construct a graph neural network reinforcement learning algorithm to determine the state of the mobile energy storage vehicle. and the status of the emergency repair team Obtain the state matrix o t Construct an adjacency matrix A based on the adjacency relationships of mobile emergency resources. t ;

[0070] S5: Calculate the routing behavior strategy of the mobile energy storage vehicle based on the ε-Greedy policy and the graph neural network reinforcement learning algorithm (data-driven method). Routing behavior strategy of emergency repair teams

[0071] S6: Execute the mobile energy storage vehicle routing behavior policy respectively. Routing behavior strategy of emergency repair teams It also assesses and updates the status of mobile energy storage vehicles and emergency repair teams;

[0072] S7: Calculate the power scheduling strategy and reward function r of the mobile energy storage vehicle based on the mobile energy storage vehicle scheduling model combined with the distribution network reconfiguration model and the distribution network optimal power flow model (model-driven method). t M Based on the emergency repair team scheduling model (model-driven method), the maintenance scheduling strategy and reward function r of the emergency repair team are calculated. t R ;

[0073] S8: Update the mobile energy storage vehicle state using the graph neural network reinforcement learning algorithm. and the status of the emergency repair team Obtain the next state matrix o t+1 Update the adjacency matrix A t+1 ;

[0074] S9: Transfer the information from the current step and initial priority Storing in the mobile energy storage vehicle image reinforcement learning memory unit D M ,Will and initial priority Stored in the emergency repair team's image reinforcement learning memory unit D R ;

[0075] S10: The importance of samples is sampled by prioritizing experience replay strategy for memory units, and the weights of the graph neural networks of mobile energy storage vehicles and emergency repair teams are updated based on the stochastic gradient descent method.

[0076] S11: Determine if the end time T has been reached. End If not, proceed to steps S3 to S10;

[0077] S12: Determine if the training end time E has been reached. End If not, proceed to steps S3 to S11; if yes, output the graph reinforcement learning algorithm parameters, routing, and scheduling strategies for the mobile energy storage vehicle and the emergency repair team.

[0078] The following is a detailed description of the solution in this embodiment:

[0079] I. Initialize the power-transportation coupled network model, the mobile energy storage vehicle graph reinforcement learning model, and the emergency repair team graph reinforcement learning model.

[0080] (1) Initialize the number of training sessions e = 0;

[0081] (2) Initialization of the power-transportation coupled network model, including: voltage limits of distribution network nodes, parameter settings of lines and transformers, upper and lower constraints of voltage and output of distributed generators, output of renewable energy, load factor of distribution network, access location of charging stations in distribution network; traffic network nodes, road length, road capacity and free driving speed, location of charging stations in traffic network, and location of emergency repair team warehouses.

[0082] (3) The nodes and connecting lines in the distribution network are considered as points and edges forming the basic distribution network diagram; charging station nodes are added to form a distribution network diagram model. Where N PDS L represents the set of nodes in the distribution network. PDS Represents a set of distribution network lines. This represents the set of charging station nodes. Initialize the load factor of the distribution network nodes;

[0083] (4) The nodes in the transportation network and the roads connecting the nodes are regarded as points and edges to form the basic transportation network graph; charging station nodes and emergency repair team warehouses are added to form the transportation network graph model. Where N TN L represents the set of nodes in the transportation network. TN Represents the set of roads in a transportation network; This represents the set of charging station nodes in the transportation network; This represents the set of emergency repair team warehouse nodes in the transportation network;

[0084] (5) Treat all mobile energy storage vehicles and emergency repair teams as agents, and each agent as a node, while the connections between them are treated as edges, to construct a dynamic network model of mobile emergency resources. Where M represents a collection of mobile energy storage vehicles, and W represents a collection of emergency repair teams. For mobile emergency resource adjacency relationships The adjacency matrix formed;

[0085] (6) Initialization of the graph reinforcement learning model for the mobile energy storage vehicle, including: the number of graph neural network layers and neuron settings, weights W M and bias b M Initialization; learning rate α M Sampling batch size N s and memory unit D M Initialize hyperparameters such as capacity;

[0086] (7) Initialization of the graph reinforcement learning model for the emergency repair team, including: the number of graph neural network layers and neuron settings, weights W R and bias b RInitialization; learning rate α R Sampling batch size N s and memory unit D R Initialize hyperparameters such as capacity.

[0087] II. Simulate power-transportation coupled network failure scenarios after extreme events occur.

[0088] To study the impact of extreme events on the power-transportation coupled network, some lines in the distribution network were randomly interrupted, constructing a set L of damaged lines. PDS,off This study simulates the disruption of weak power distribution lines in real-world power grids after extreme events; and randomly introduces damaged sections of roads within a transportation network to construct a set of damaged roads. This simulates real-life traffic network scenarios where obstacles cause traffic congestion on certain roads due to extreme events.

[0089] III. Initialization Graph Reinforcement Learning Algorithm Environment and Power-Transportation Coupled Network Fault Scenarios

[0090] (1) Initialize the training time t = 0;

[0091] (2) Initialize the variables in the graph reinforcement learning algorithm environment, including: the initial position of the mobile energy storage vehicle (charging station node), the initial SOC value of the mobile energy storage vehicle, the initial speed of the mobile energy storage vehicle, the initial position of the emergency repair team (each repair team warehouse), the repair materials of the emergency repair team, the initial speed of the emergency repair team, and the traffic flow and speed of each road in the transportation network.

[0092] (3) Initialize the fault scenario of the power-transportation coupled network, that is, the set of damaged lines L constructed in the initialization step S2. PDS,off Collection of damaged roads

[0093] IV. Based on the power-transportation coupled network information and the state information of mobile emergency resources themselves, a graph neural network reinforcement learning algorithm is constructed to determine the state of mobile energy storage vehicles. and the status of the emergency repair team Obtain the state matrix o t Construct an adjacency matrix A based on the adjacency relationships of mobile emergency resources. t .

[0094] (1) Information is collected from the power-transportation coupled network graph model and the mobile emergency resource dynamic graph model to construct the state of the mobile energy storage vehicle graph neural network reinforcement learning algorithm. Including the mobile energy storage vehicle's own status information Adjacent mobile emergency resource status information Traffic network road information for the location Distribution network node load data P t ED New energy output P t WT Specifically, it is expressed as:

[0095]

[0096] In the formula, M is the set of mobile energy storage vehicles; T is the time set; and m is the number of the mobile energy storage vehicle. This indicates the location information of the mobile energy storage vehicle in the transportation network at time t; S represents the speed at which a mobile energy storage vehicle moves within a transportation network. m,t This indicates the remaining SOC value of the mobile energy storage vehicle; and These represent the information of the mobile energy storage vehicles or emergency repair teams adjacent to the m-th mobile energy storage vehicle at time t; N r J represents the number of adjacent emergency resources for the mobile energy storage vehicle; J represents the number of adjacent roads for the mobile energy storage vehicle. This indicates the number of the j-th road adjacent to the m-th mobile energy storage vehicle at time t. and Indicates road The first node number and the last node number, and Representing roads The status, traffic speed, and number of vehicles on the road; Indicates road Length; Indicates the current location number of the transportation network node. Indicates the selected next node position number, Indicates the position number of the previous node. Indicates the road number to be entered and RT indicates the previous road number. w,t Indicates the amount of repair material remaining in the RC;

[0097] (2) Information is collected from the power-transportation coupled network graph model and the mobile emergency resource dynamic graph model to construct the state of the emergency repair team graph neural network reinforcement learning algorithm. Including the mobile energy storage vehicle's own status information Adjacent mobile emergency resource status information Traffic network road information for the location Traffic network road fault information and the time required to repair damaged roads Specifically, it is expressed as follows:

[0098]

[0099]

[0100] (3) Constructing the state matrix

[0101] (4) Collect information from the dynamic graph model of mobile emergency resources to construct the adjacency matrix A t Specifically, it is expressed as:

[0102]

[0103] In the formula, Let N be the adjacency relationship of all mobile emergency resources at time t; v The quantity of mobile emergency resources can be expressed as N. v =Calculate |M|+|W|; This is a binary variable representing the adjacency relationship between mobile emergency resources i and j. It is 1 if they are adjacent and 0 otherwise. The adjacency matrix A t It will be input into the graph reinforcement learning algorithm to participate in the calculation of state-action values.

[0104] V. Calculate the routing behavior strategy of the mobile energy storage vehicle based on the ε-Greedy policy and the graph neural network reinforcement learning algorithm (data-driven method). Routing behavior strategy of emergency repair teams

[0105] (1) Using the ε-Greedy strategy: First, calculate the random number p at time t. t Compare the current ε t Value and random number p t If there is p t <ε t Then, a random method is used to generate the routing behavior strategy for mobile energy storage vehicles. Routing behavior strategy of emergency repair teams Specifically, as shown in formula (150):

[0106]

[0107] In the formula, This represents the set of routing actions for mobile energy storage vehicles and emergency repair teams; when action a is selected... i,t =0 indicates stopping in place, and also indicates that the mobile emergency resource has been routed to its destination (mobile energy storage vehicle is a charging station, emergency repair team is a damaged road or warehouse node), ready to receive subsequent model scheduling decisions; when action a is selected... i,t ≠0 indicates that mobile energy storage vehicles and emergency repair teams choose to enter the next road according to the routing strategy to reach their destination;

[0108] (2) If there is p t ≥ε t Then, a routing behavior strategy for the mobile energy storage vehicle is generated using a graph neural network multi-agent reinforcement learning algorithm. Routing behavior strategy of emergency repair teams

[0109]

[0110] In the formula, argmax(·) represents the parameter corresponding to the maximum value; For mobile energy storage vehicles, the graph reinforcement learning algorithm at time t provides parameters of the neural network. Based on the mobile emergency resource status set o t Routing actions Mobile emergency resource adjacency matrix A t The calculated action value function; For the emergency repair team, the graph reinforcement learning algorithm at time t provides parameters of the neural network. Based on the mobile emergency resource status set o t Routing actions Mobile emergency resource adjacency matrix A t The calculated action value function;

[0111] The graph reinforcement learning algorithm for mobile energy storage vehicles and the graph reinforcement learning algorithm for emergency repair teams mentioned in this example use the same neural network structure. That is, the same neural network controls multiple mobile energy storage vehicles, and another neural network controls multiple emergency repair teams. The specific neural network structure is as follows: First, there is a fully connected input layer, which takes the input mobile emergency resource state set o as input. t Perform preliminary feature extraction and dimension transformation, and output. This is followed by two graph attention network layers, which apply an attention mechanism to the state matrix. Mobile emergency resource adjacency matrix A t Attention coefficients are calculated, and features are extracted based on the attention matrix. The output is based on adjacency relationships. Features The final layer employs the Dueling DQN algorithm strategy, splitting the feature information into two fully connected branches: the first branch outputs the scalar value V(o) of the state function. t A t The output status of the second branch. The action advantage value function vector Therefore, the calculation of the Q value in formulas (151) and (152) can be expressed as:

[0112]

[0113] In the formula, ψ t V(o) t A t The parameters of the fully connected neural network in the branch containing the given information; for The parameters of the fully connected neural network for the branch. The mobile energy storage vehicle and the emergency repair team select routing actions based on the action value functions calculated by their respective neural networks. and

[0114] VI. Implement the routing behavior strategy for mobile energy storage vehicles separately. Routing behavior strategy of emergency repair teams It also assesses and updates the status of mobile energy storage vehicles and emergency repair teams.

[0115] VII. Calculate the power scheduling strategy and reward function r of the mobile energy storage vehicle based on the mobile energy storage vehicle scheduling model, emergency repair team scheduling model, distribution network reconfiguration model, and distribution network optimal power flow model (model-driven method). t M Maintenance scheduling strategy and reward function r of emergency repair teams t R .

[0116] (1) Construct a mobile energy storage vehicle scheduling model, a distribution network reconfiguration model, and a distribution network optimal power flow model, as shown below:

[0117]

[0118]

[0119] In the formula, formulas (154)-(159) are the mobile energy storage vehicle scheduling model; This represents the set of charging station nodes in the transportation network; This is a binary variable representing whether the mobile energy storage vehicle is located at node n in the transportation network. tn If it is at this node, it is 1; otherwise, it is 0. and These are binary variables, representing the charging and discharging states of the m-th mobile energy storage vehicle at time t; and Let represent the charging active power, discharging active power, and reactive power at time t, respectively. and These represent the upper limits of active power and reactive power, respectively; S m,t This represents the state of charge (SOC) of the battery at time t. and These are the charging efficiency coefficient and the discharging efficiency coefficient, respectively.S m and These are the upper and lower limits of the SOC value, respectively; Formulas (154)-(156) restrict the charging and discharging power and reactive power of the mobile energy storage vehicle; Formula (157) restricts the charging and discharging behavior of the mobile energy storage vehicle, that is, the mobile energy storage vehicle can only be in a charging state or a discharging state, and if the current location of the mobile energy storage vehicle is not located at a charging station node of the transportation network, it cannot charge or discharge; Formulas (158)-(159) constrain the SOC value of the mobile energy storage vehicle;

[0120] In the formulas, equations (160)-(165) represent the distribution network reconfiguration model; L PDS L PDS,off L PDS,switch and Represent the set of distribution network lines, the set of damaged lines, the set of lines equipped with tie switches, and the set of distribution network source nodes, respectively; ρ(·) and φ(·) represent the set of parent nodes and the set of child nodes, respectively; α ij,t This represents the connection status of the distribution network line (i,j) at time t; 1 indicates connection and 0 indicates disconnection. PDS Indicates the number of nodes in the distribution network; If it is a binary variable, then at time t, the distribution network node n pds If it is the source node, it is 1; otherwise, it is 0. ij,t represents the virtual power flow into distribution network node j at time t; M is a large constant. Constraint (160) is a constraint on the line state based on the distribution network line fault setting in Example 3; Constraint (161) is a distribution network radiation constraint, which describes the relationship between the distribution network line state (normal line, damaged line and line equipped with remote switch) and the distribution network source node; Constraints (162)-(165) describe the virtual power flow balance constraints of the distribution network;

[0121] In the formulas, equations (166)-(176) represent the optimal power flow models for the distribution network; and Let represent the active power and reactive power flowing through the distribution network line (i,j) at time t, respectively. and These represent the active and reactive power outputs of the generator at node j in the distribution network, respectively. and These represent the active and reactive power outputs of the wind turbine WT, respectively. and Let represent the active power exchange and reactive power exchange between the mobile energy storage vehicle and node j, respectively. and Both represent the load loss at node j; and These represent the load demand of node j; and These represent the maximum active power and maximum reactive power of the distribution network line (i,j), respectively. This represents the upper limit of the capacity of the distribution network line (i,j); The voltage square value at node j in the distribution network at time t; r ij and x ij These represent the line resistance and reactance values ​​of the distribution network line (i,j), respectively. and Let represent the upper and lower limits of the squared voltage of node j in the distribution network at time t. Constraints (166) and (167) describe the active power balance and reactive power balance of the distribution network lines; constraints (168) and (169) describe the active and reactive power of the mobile energy storage vehicle charging (or discharging) at the charging station node at time t; constraint (170) restricts the active and reactive power of the mobile energy storage vehicle to 0 when it is not at the charging station node; constraints (171) and (172) limit the active and reactive power flowing through the distribution network lines at time t; constraint (173) limits the heat capacity of line (i,j); constraints (174)-(176) describe the voltage relationship between adjacent nodes in the distribution network.

[0122] (2) Construct an emergency repair team dispatch model, as shown below:

[0123]

[0124]

[0125] In the formula, W represents the assembly of emergency repair teams; and All are binary variables. If at time t the w-th emergency repair team is located at node n of the transportation network... tn Or road tn The value is 1 if it is not 1, and 0 otherwise. This represents a collection of roads in the transportation network that are damaged. This is a binary variable, representing the damage to the road caused by the w-th emergency repair team at time t. The repair decision is set to 1 if the item is repaired, and 0 otherwise. It is a binary variable representing the road at time t. The state; To repair the damaged road Required time; β w,t This is a binary variable used to determine whether the emergency repair team has reached the warehouse node; β w,tA binary variable is used to determine whether the emergency repair team that has reached the warehouse location needs to be dispatched further. Constraints (177)-(182) represent the emergency repair team's decision on road repair, while constraints (183)-(184) refer to this formula to decide whether the emergency repair team should stay at the warehouse. Constraint (177) indicates that the emergency repair team has not reached the damaged road in the transportation network. When the road repair decision variable is set to 0, constraint (178) describes the relationship between the emergency repair team's repair decision variable and the state of damaged roads in the traffic network. If the road network at time t is damaged, the road network at time t will be damaged. If the road is still damaged, the emergency repair team can repair it; constraint (179) constrains the road status of the traffic network based on the road fault settings in this example three; constraint (180) indicates that each damaged road in the traffic network Only one emergency repair team can repair the damaged road; constraint (181) indicates that the emergency repair team can only repair the damaged road. Repaired After that time, the road Traffic can be restored at time t; constraint (182) assumes that the damaged roads in the transportation network will not be damaged again after they are repaired; constraint (183) calculates whether the emergency repair team has arrived at the warehouse node, while constraint (184) constrains the subsequent dispatch of the emergency repair team by the state of all damaged roads and the location of the emergency repair team at time t, that is, calculates the dispatch status of the emergency repair team.

[0126] (3) Construct an objective function based on the tasks of load restoration at distribution network nodes and road repair for damaged transportation networks, as shown below:

[0127]

[0128] In the formula, c pds and c tn Let n be the distribution network node. pds Unit load loss cost and transportation network roads Damage cost per unit time; Represents node n pds Importance coefficient; Objective function (185) represents minimizing the power-transportation coupled network loss cost; Formula (186) calculates the distribution network node load loss cost; Formula (187) calculates the transportation network road damage cost and emergency repair team dispatch cost;

[0129] (4) Constraints (168) and (169) are linearized using the Big M method, and constraint (173) is linearized using the second-order cone relaxation method; while constraints (180) and (181) of the emergency repair team scheduling problem are linearized using the McCormick envelope method.

[0130] (5) (Model-Driven Part) A mixed-integer second-order cone programming model, MISOCP, is constructed using formulas (185)-(187) as the objective function and formulas (154)-(176) as constraints. The model is solved using the Gurobi commercial solver. The load on the distribution network nodes restored by the mobile energy storage vehicle is obtained, and the reward function for the mobile energy storage vehicle is calculated using formula (188).

[0131]

[0132] In the formula, Let represent the set of distribution network nodes whose load is supported by the m-th mobile energy storage vehicle at time t; Let t be the distribution network node The amount of load recovery; For distribution network nodes The load level weighting coefficient. It's important to note here that the mobile energy storage vehicle's dispatch cost is not considered; therefore, the reward value for the mobile energy storage vehicle during its movement between charging stations on the transportation network is 0.

[0133] Meanwhile, the road maintenance strategy or dispatch strategy of the emergency repair team is obtained by solving MISOCP, and the reward function of the emergency repair team is calculated by formulas (189)-(195).

[0134]

[0135]

[0136] In the formula, This represents the travel time from the starting node b to the destination node p in the transportation network. For road l tn Free passage time; α tn and These are all road delay coefficients, typically with values ​​of 0.15 and 4; The reward value obtained by the emergency repair team upon completing the task at time t; The road repaired by the emergency repair team Contribution reward value; This is the penalty value; and All are weighting coefficients; The emergency repair team completed the repairs on the damaged roads in the transportation network at the designated time. The reward obtained from the repair is set to a constant. The reward value for the emergency repair team to return to the initial warehouse node at time t after all the damaged roads have been repaired is set to a constant. c1 and c2 are binary variables, representing a value of 1 if the road chosen at time t is the same as that chosen at time t-1, and 0 otherwise; c1 and c2 are constants. Formula (193) calculates the number of damaged roads in the transportation network repaired by the w-th emergency repair team at time t. The proportion of the number of vehicles to the total number of vehicles; the three penalty terms in formula (194) represent: the penalty when the emergency repair team chooses the road passed in the previous moment as the next road to enter, the penalty when the emergency repair team repairs the road but fails to return to its repair station node at the last moment, and the penalty when the road is not repaired and the warehouse is not returned at the last moment.

[0137] 8. Update the state of the mobile energy storage vehicle using the graph neural network reinforcement learning algorithm. and the status of the emergency repair team Obtain the next state matrix o t+1 Update the adjacency matrix A t+1 ;

[0138] 9. Transfer the information from the current step and initial priority Storing in the mobile energy storage vehicle image reinforcement learning memory unit D M ,Will and initial priority Stored in the emergency repair team's image reinforcement learning memory unit D R .

[0139] 10. The importance of samples is sampled by prioritizing experience replay strategy for memory units, and the weights of the graph neural networks of mobile energy storage vehicles and emergency repair teams are updated based on the stochastic gradient descent method.

[0140] (1) In the mobile energy storage vehicle, the learning and memory unit D is reinforced. M Samples N were collected using an importance sampling method. s And the importance weight of each sample is calculated using formula (62):

[0141]

[0142] In the formula, N D Represents memory unit D M The total number of samples in the sample; This indicates the priority of the sample; β is a hyperparameter.

[0143] (2) Combining the strategy of the Double DQN algorithm, the loss function is calculated using formulas (197) and (198):

[0144]

[0145] In the formula, and These are the current network parameters and the target network parameters of the mobile energy storage vehicle, respectively; γ is a discount factor reflecting the impact of the future Q value on the current action, with a value of [0,1].

[0146] (3) The neural network parameters of the mobile energy storage vehicle graph reinforcement learning algorithm are updated based on the stochastic gradient descent method and sample importance weights, as specifically expressed as:

[0147]

[0148] In the formula, α M Let N be the learning rate of the gradient descent algorithm in the current network of the mobile energy storage vehicle. And after a fixed number of steps N... up Then, the parameters of the target network for the mobile energy storage vehicle are configured using the following method. Update:

[0149]

[0150] (4) The TD-error value calculated according to formula (63) Calculate the new priority for each sample Here, ∈ is a small constant to prevent the priority from being zero. In the graph reinforcement learning memory unit D of the mobile energy storage vehicle... M The priority of each sample is updated in the process;

[0151] (5) Update the parameters of the neural network of the graph reinforcement learning algorithm for the emergency repair team according to (1)-(4).

[0152] 11. Determine if the end time T has been reached. End If not, then proceed to steps three through ten.

[0153] 12. Determine if the training session has reached its end date (E). End If not, then execute steps 3 through 11; if yes, then output the graph reinforcement learning algorithm parameters, routing, and scheduling strategies for the mobile energy storage vehicle and the emergency repair team.

[0154] This invention provides a strategy for optimizing the scheduling of multiple types of mobile emergency resources and enhancing the resilience of distribution networks based on a data-model hybrid driving method. The method first treats mobile emergency resources (i.e., mobile energy storage vehicles and emergency repair teams) as agents, abstracting the dynamic adjacency relationships of all agents as edges, and then transforms the collaborative relationships of all mobile energy storage vehicle and emergency repair team agents into a dynamic graph structure. In the data-driven part, the method employs a graph neural network reinforcement learning algorithm based on an attention mechanism. This combines the excellent non-Euclidean data information extraction capability of graph attention networks with the powerful sequential stochastic optimization decision-making capability of reinforcement learning algorithms, effectively extracting key information from the dynamic adjacency relationships and power-transportation coupled network graph data of mobile energy storage vehicles and emergency repair teams to formulate routing behavior strategies for these vehicles. In the model-driven part, the scheduling model of mobile energy storage vehicles and emergency repair teams, the distribution network reconfiguration model, and the optimal power flow model of the distribution network are constructed into a MISOCP model, and the optimal scheduling behavior strategy for multiple types of mobile emergency resources is solved to achieve distribution network resilience enhancement. In summary, this invention can simultaneously consider the dynamic changes in road conditions of the transportation network and the dynamic changes in the relationships between adjacent mobile emergency resources. By solving the routing and scheduling strategies of mobile energy storage vehicles and emergency repair teams through a data-model hybrid driving method, it restores the load of distribution network nodes and repairs damaged roads in the transportation network, thereby improving the resilience of the distribution network.

[0155] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0156] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0157] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0158] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0159] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0160] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0161] This patent is not limited to the above-described preferred implementation method. Anyone can derive other forms of multi-type mobile emergency resource optimization scheduling and distribution network resilience enhancement strategies based on a data-model hybrid driving method under the guidance of this patent. All equivalent changes and modifications made within the scope of this patent application shall fall within the scope of this patent.

Claims

1. A strategy for optimizing the scheduling of multiple types of mobile emergency resources and improving the resilience of distribution networks based on a data-model hybrid driving method, characterized in that: Starting from the dynamic interaction between mobile energy storage vehicles and emergency repair teams, this paper coordinates both for load restoration and road repair. It considers the uncertainty of road repair time when dispatching emergency repair teams to repair damaged roads in the transportation network, and the dynamic changes in the road conditions of the transportation network when dispatching mobile energy storage vehicles to restore loads to distribution network nodes. Data methods and modeling methods are combined to solve these various types of mobile emergency resource scheduling problems. The data-driven part is supported by the GATD3QN graph neural network multi-agent reinforcement learning algorithm based on graph attention mechanism. The GATD3QN algorithm combines a graph neural network based on attention mechanism and Double... The DQN and DuelingDQN algorithms optimize the feature extraction capability and Q-value calculation of the DQN algorithm, and adopt a priority experience replay strategy to improve the sampling efficiency and training effect of the algorithm, so as to solve the optimal multi-type mobile emergency resource routing behavior strategy. The model-driven part constructs the scheduling model of mobile energy storage vehicles and emergency repair teams, the distribution network reconfiguration model and the optimal power flow model of distribution network into the MISOCP model and solves the optimal multi-type mobile emergency resource scheduling behavior strategy, including the power scheduling decision of mobile energy storage vehicles, the road repair decision or return to the warehouse decision of emergency repair teams, so as to improve the resilience of distribution network. Includes the following steps: Step S1: Construct and initialize the power-transportation coupled network model, the mobile energy storage vehicle graph reinforcement learning model, and the emergency repair team graph reinforcement learning model; Step S2: Simulate a power-transportation coupled network failure scenario after an extreme event occurs, and set up line disconnection in the distribution network and road damage in the transportation network; Step S3: Initialize the environment for the graph reinforcement learning algorithm and the fault scenario for the power-transportation coupled network; Step S4: Based on the power-transportation coupled network information and the mobile emergency resource's own state information, construct a graph neural network reinforcement learning algorithm to determine the state of the mobile energy storage vehicle. and the status of the emergency repair team Obtain the state matrix o t Construct an adjacency matrix A based on the adjacency relationships of mobile emergency resources. t ; Step S5: Calculate the mobile energy storage vehicle routing behavior strategy based on the ε-Greedy policy and the graph neural network reinforcement learning algorithm as the data-driven part. Routing behavior strategy of emergency repair teams Step S6: Execute the mobile energy storage vehicle routing behavior policy respectively. Routing behavior strategy of emergency repair teams It also assesses and updates the status of mobile energy storage vehicles and emergency repair teams; Step S7: The model-driven part calculates the power scheduling strategy and reward function r of the mobile energy storage vehicle based on the mobile energy storage vehicle scheduling model combined with the distribution network reconfiguration model and the distribution network optimal power flow model. t M Based on the emergency repair team scheduling model, the maintenance scheduling strategy and reward function r of the emergency repair team are calculated. t R ; Step S8: Update the mobile energy storage vehicle state using the graph neural network reinforcement learning algorithm. and the status of the emergency repair team Obtain the next state matrix o t+1 Update the adjacency matrix A t+1 ; Step S9: Transfer the information from the current step and initial priority Storing in the mobile energy storage vehicle image reinforcement learning memory unit D M ,Will and initial priority Stored in the emergency repair team's image reinforcement learning memory unit D R ; Step S10: The importance of samples is sampled for the memory cells using a priority experience replay strategy, and the weights of the graph neural networks for the mobile energy storage vehicle and the emergency repair team are updated based on the stochastic gradient descent method. Step S11: Determine if the end time T has been reached. End If not, proceed to steps S3 to S10; Step S12: Determine whether the training end sequence number E has been reached. End If not, proceed to steps S3 to S11; if yes, output the graph reinforcement learning algorithm parameters, routing, and scheduling strategies for the mobile energy storage vehicle and the emergency repair team.

2. The strategy for optimizing scheduling of multiple types of mobile emergency resources and improving the resilience of distribution networks based on a data-model hybrid driving method as described in claim 1, characterized in that: In step S1, the power-transportation coupled network model, the mobile energy storage vehicle graph reinforcement learning model, and the emergency repair team graph reinforcement learning model are initialized, and the number of training sessions is initialized. The nodes and connecting lines in the distribution network are considered as points and edges, forming the basic distribution network diagram; charging station nodes are added to this diagram to create a distribution network model. Where N PDS L represents the set of nodes in the distribution network. PDS Represents a set of distribution network lines. Represents the set of charging station nodes; The basic transportation network graph is constructed by treating the nodes and roads connecting them as points and edges; charging station nodes and emergency repair team warehouses are added to complete the transportation network graph model. Where N TN L represents the set of nodes in the transportation network. TN Represents the set of roads in a transportation network; This represents the set of charging station nodes in the transportation network; This represents the set of emergency repair team warehouse nodes in the transportation network; By treating all mobile energy storage vehicles and emergency repair teams as agents, and each agent as a node, while the connections between them are considered as edges, a dynamic network model of mobile emergency resources is constructed. Where M represents a collection of mobile energy storage vehicles, and W represents a collection of emergency repair teams. For mobile emergency resource adjacency relationships The adjacency matrix formed.

3. The strategy for optimizing scheduling of multiple types of mobile emergency resources and improving the resilience of distribution networks based on a data-model hybrid driving method as described in claim 2, characterized in that: In step S2, some lines in the distribution network are randomly interrupted to construct a set L of damaged lines. PDS,off This is used to simulate the outage of weak lines in the power distribution network after extreme events; and a set of damaged roads is constructed by randomly setting up some damaged roads in the transportation network. This simulates situations where parts of the transportation network are blocked due to obstacles caused by extreme events.

4. The strategy for optimizing and scheduling multi-type mobile emergency resources and improving distribution network resilience based on a data-model hybrid driving method as described in claim 1, characterized in that: Step S3 specifically includes: initializing various variables in the graph reinforcement learning algorithm environment and initializing the power-transportation coupled network fault scenario, and initializing the training time.

5. The strategy for optimizing and scheduling multi-type mobile emergency resources and improving distribution network resilience based on a data-model hybrid driving method as described in claim 3, characterized in that: In step S4: Based on the power-transportation coupled network information and the state information of mobile emergency resources, a state graph neural network reinforcement learning algorithm for mobile energy storage vehicles is constructed. Including the mobile energy storage vehicle's own status information Adjacent mobile emergency resource status information Traffic network road information for the location Distribution network node load data P t ED New energy output P t WT Specifically, it is expressed as: In the formula, M is the set of mobile energy storage vehicles; T is the time set; and m is the number of the mobile energy storage vehicle. This indicates the location information of the mobile energy storage vehicle in the transportation network at time t; S represents the speed at which a mobile energy storage vehicle moves within a transportation network. m,t This indicates the remaining SOC value of the mobile energy storage vehicle; and These represent the information of the mobile energy storage vehicles or emergency repair teams adjacent to the m-th mobile energy storage vehicle at time t; N r J represents the number of adjacent emergency resources for the mobile energy storage vehicle; J represents the number of adjacent roads for the mobile energy storage vehicle. This indicates the number of the j-th road adjacent to the m-th mobile energy storage vehicle at time t. and Indicates road The first node number and the last node number, and Representing roads The status, traffic speed, and number of vehicles on the road; Indicates road Length; Indicates the current location number of the transportation network node. Indicates the selected next node position number, Indicates the position number of the previous node. Indicates the road number to be entered and RT indicates the previous road number. w,t Indicates the amount of repair material remaining in the RC; Based on the power-transportation coupled network information and the state information of mobile emergency resources, a state graph neural network reinforcement learning algorithm for emergency repair teams is constructed. Including the mobile energy storage vehicle's own status information Adjacent mobile emergency resource status information Traffic network road information for the location Traffic network road fault information and the time required to repair damaged roads Specifically, it is expressed as follows: In the formula, W represents the assembly of emergency repair teams; Thus, the state matrix is ​​constructed. Then, construct an adjacency matrix A based on the adjacency relationships of mobile emergency resources. t Specifically, it is expressed as: In the formula, Let N be the adjacency relationship of all mobile emergency resources at time t; v The number of mobile emergency resources is determined by N. v =Calculate |M|+|W|; This is a binary variable representing the adjacency relationship between mobile emergency resources i and j. It is 1 if they are adjacent and 0 otherwise. Adjacency matrix A t It is used as input into graph reinforcement learning algorithms to participate in the calculation of state-action values.

6. The strategy for optimizing and scheduling multi-type mobile emergency resources and improving distribution network resilience based on a data-model hybrid driving method as described in claim 5, characterized in that: In step S5, firstly, the random number p at time t is calculated. t Its value range is [0,1), and the ε-Greedy strategy is used to compare the current ε. t Value and random number p t If there is p t <ε t Then, a random method is used to generate the routing behavior strategy for mobile energy storage vehicles. Routing behavior strategy of emergency repair teams Specifically, as shown in formula (16): In the formula, This represents the set of routing actions for mobile energy storage vehicles and emergency repair teams; when action a is selected... i,t =0 indicates stopping in place, corresponding to the mobile emergency resource routing to the destination location, ready to receive subsequent model scheduling decisions; when action a is selected... i,t ≠0 indicates that mobile energy storage vehicles and emergency repair teams choose to enter the next road according to the routing strategy to reach their destination; If there is p t ≥ε t Then, a routing behavior strategy for the mobile energy storage vehicle is generated using a graph neural network multi-agent reinforcement learning algorithm. Routing behavior strategy of emergency repair teams Specifically, as shown in formulas (17) and (18): In the formula, argmax(·) represents the parameter corresponding to the maximum value; For mobile energy storage vehicles, the graph reinforcement learning algorithm at time t provides parameters of the neural network. The following is based on the mobile emergency resource status set o t Routing actions Mobile emergency resource adjacency matrix A t The calculated action value function; For the emergency repair team, the graph reinforcement learning algorithm at time t provides parameters of the neural network. The following is based on the mobile emergency resource status set o t Routing actions Mobile emergency resource adjacency matrix A t The calculated action value function; The graph reinforcement learning algorithm for mobile energy storage vehicles and the graph reinforcement learning algorithm for emergency repair teams use the same neural network structure: one neural network controls multiple mobile energy storage vehicles, and another neural network controls multiple emergency repair teams; the neural network structure is as follows: first, a fully connected input layer, which takes the input mobile emergency resource state set o as input. t Perform preliminary feature extraction and dimension transformation, and output. This is followed by two graph attention network layers, which apply an attention mechanism to the state matrix. Mobile emergency resource adjacency matrix A t Attention coefficients are calculated, and features are extracted based on the attention matrix. The output is based on adjacency relationships. Features The final layer employs the Dueling DQN algorithm strategy, splitting the feature information into two fully connected branches: the first branch outputs the scalar value V(o) of the state function. t A t The output status of the second branch. The action advantage value function vector The calculation of the Q value in formulas (17) and (18) is expressed as follows: In the formula, ψ t For V(o) t A t The parameters of the fully connected neural network in the branch containing θ; t for The parameters of the fully connected neural network of the branch; the mobile energy storage vehicle and the emergency repair team select routing actions based on the action value functions calculated by their respective neural networks. and 7. The strategy for optimizing scheduling of multiple types of mobile emergency resources and improving the resilience of distribution networks based on a data-model hybrid driving method as described in claim 6, characterized in that: In step S7, the mobile energy storage vehicle scheduling model, distribution network reconfiguration model, and distribution network optimal power flow model are constructed as follows: Among them, formulas (20)-(25) are the mobile energy storage vehicle scheduling models; This represents the set of charging station nodes in the transportation network; This is a binary variable representing whether the mobile energy storage vehicle is located at node n in the transportation network. tn If it is at this node, it is 1; otherwise, it is 0. and These are binary variables, representing the charging and discharging states of the m-th mobile energy storage vehicle at time t. and Let represent the charging active power, discharging active power, and reactive power at time t, respectively. and These represent the upper limits of active power and reactive power, respectively; S m,t The state of charge (SOC) of the battery at time t is represented by t. and These are the charging efficiency coefficient and the discharging efficiency coefficient, respectively. S m and These are the upper and lower limits of the SOC value, respectively; Formulas (20)-(22) limit the charging and discharging power and reactive power of the mobile energy storage vehicle; Formula (23) limits the charging and discharging behavior of the mobile energy storage vehicle, that is, the mobile energy storage vehicle can only be in a charging state or a discharging state, and if the current location of the mobile energy storage vehicle is not located at a charging station node of the transportation network, it cannot charge or discharge; Formulas (24)-(25) constrain the SOC value of the mobile energy storage vehicle; Formulas (26)-(31) are the distribution network reconfiguration models; L PDS L PDS,off L PDS,switch and Represent the set of distribution network lines, the set of damaged lines, the set of lines equipped with tie switches, and the set of distribution network source nodes, respectively; ρ(·) and φ(·) represent the set of parent nodes and the set of child nodes, respectively; α ij,t This represents the connection status of the distribution network line (i,j) at time t; 1 indicates connection and 0 indicates disconnection. PDS Indicates the number of nodes in the distribution network; If it is a binary variable, then at time t, the distribution network node n pds If it is the source node, it is 1; otherwise, it is 0. ij,t denoted as virtual power flow into distribution network node j at time t; M is a large constant; constraint (26) is a constraint on line state based on distribution network line fault settings; constraint (27) is a distribution network radiation constraint, which describes the relationship between distribution network line state and distribution network source node; constraints (28)-(31) describe the virtual power flow balance constraints of the distribution network. Formulas (32)-(42) are the optimal power flow models for the distribution network; and Let represent the active power and reactive power flowing through the distribution network line (i,j) at time t, respectively. and These represent the active and reactive power outputs of the generator at node j in the distribution network, respectively. and These represent the active and reactive power outputs of the wind turbine WT, respectively. and Let represent the active power exchange and reactive power exchange between the mobile energy storage vehicle and node j, respectively. and Both represent the load loss at node j; and These represent the load demand of node j; and These represent the maximum active power and maximum reactive power of the distribution network line (i,j), respectively. This represents the upper limit of the capacity of the distribution network line (i,j); r represents the squared voltage value of node j in the distribution network at time t; ij and x ij These represent the line resistance and reactance values ​​of the distribution network line (i,j), respectively. and Let represent the upper and lower limits of the square of the voltage at node j in the distribution network at time t; constraints (32) and (33) describe the active power balance and reactive power balance of the distribution network lines; constraints (34) and (35) describe the active and reactive power of the mobile energy storage vehicle charging or discharging at the charging station node at time t; constraint (36) restricts the active and reactive power of the mobile energy storage vehicle to 0 when it is not at the charging station node; constraints (37) and (38) limit the active and reactive power flowing through the distribution network lines at time t; constraint (39) limits the heat capacity of line (i,j); constraints (40)-(42) describe the voltage relationship between adjacent nodes in the distribution network. The emergency repair team dispatch model is shown in detail below: In the formula, W represents the assembly of emergency repair teams; and All are binary variables. If at time t the w-th emergency repair team is located at node n of the transportation network... tn Or road tn The value is 1 if it is not 1, and 0 otherwise. This represents a collection of roads in the transportation network that are damaged. This is a binary variable, representing the damage to the road caused by the w-th emergency repair team at time t. The repair decision is set to 1 if the item is repaired, and 0 otherwise. It is a binary variable representing the road at time t. The state; To repair the damaged road Required time; β w,t This is a binary variable used to determine whether the emergency repair team has reached the warehouse node; β w,t A binary variable is used to determine whether the emergency repair team that has reached the warehouse location needs to be dispatched further; constraints (43)-(48) represent the emergency repair team's road repair decisions, and constraints (49)-(50) determine whether the emergency repair team should remain at the warehouse; constraint (43) indicates that the emergency repair team has not reached the damaged road in the transportation network. When the road repair decision variable is 0, the road repair decision variable will be restricted to 0; constraint (44) describes the relationship between the emergency repair team's repair decision variable and the state of the damaged roads in the traffic network. If the road network at time t is 0, the road repair decision variable will be 0. Still damaged, the emergency repair team repaired it; Constraint (45) Constrain the road status of the traffic network based on the road fault settings of the traffic network; Constraint (46) indicates that each road in the transportation network is damaged. Only one emergency repair team can repair the damaged road; constraint (47) indicates that the emergency repair team can only repair the damaged road. Repaired After that time, the road Traffic resumes at time t; constraint (48) assumes that the damaged roads in the transportation network will not be damaged again after they are repaired; constraint (49) calculates whether the emergency repair team has reached the warehouse node; constraint (50) then constrains the subsequent scheduling of the emergency repair team by using the status of all damaged roads and the location of the emergency repair team at time t, that is, calculates the scheduling status of the emergency repair team. The objective function is constructed based on the tasks of load restoration at distribution network nodes and road repair for damaged transportation networks, as shown below: In the formula, c pds and c tn Let n be the distribution network node. pds Unit load loss cost and transportation network roads Damage cost per unit time; Represents node n pds Importance coefficient; Objective function (51) represents minimizing the power-transportation coupling network loss cost; Formula (52) is used to calculate the distribution network node load loss cost; Formula (53) is used to calculate the transportation network road damage cost and emergency repair team dispatch cost; During the calculation, constraints (34) and (35) were linearized using the Big M method, and constraint (39) was linearized using the second-order cone relaxation method; constraints (46) and (47) of the emergency repair team scheduling problem were linearized using the McCormick envelope method. The model-driven part uses formulas (51)-(53) as the objective function and formulas (20)-(50) as constraints to construct a mixed integer second-order cone programming model (MISOCP), which is solved using the Gurobi solver. The solution yields the load of the distribution network nodes restored by the mobile energy storage vehicle, and the reward function of the mobile energy storage vehicle is calculated using formula (54). In the formula, Let represent the set of distribution network nodes whose load is supported by the m-th mobile energy storage vehicle at time t; Let t be the distribution network node The amount of load recovery; For distribution network nodes The load level weighting coefficient; the reward value for the mobile energy storage vehicle during its movement between charging stations in the transportation network is 0; The road maintenance strategy or dispatch strategy of the emergency repair team is obtained by solving MISOCP, and the reward function of the emergency repair team is calculated by formulas (55)-(61). In the formula, This represents the travel time from the starting node b to the destination node p in the transportation network. For road l tn Free passage time; α tn and All are road delay coefficients; The reward value obtained by the emergency repair team upon completing the task at time t; The road repaired by the emergency repair team Contribution reward value; This is the penalty value; and All are weighting coefficients; The emergency repair team completed the repairs on the damaged roads in the transportation network at the designated time. The reward obtained from the repair is set to a constant. The reward value for the emergency repair team to return to the initial warehouse node at time t after all the damaged roads have been repaired is set to a constant. The variable is a binary variable, which is 1 if the road selected at time t is the same as the road selected at time t-1, and 0 otherwise; c1 and c2 are both constants; Formula (59) calculates the number of damaged roads in the traffic network repaired by the wth emergency repair team at time t. The proportion of the number of vehicles to the total number of vehicles; the three penalty terms in formula (60) represent: the penalty when the emergency repair team chooses the road it passed through in the previous moment as the next road it will enter, the penalty when the emergency repair team repairs the road but fails to return to its repair station node in the last moment, and the penalty when it fails to repair the road and fails to return to the warehouse in the last moment.

8. The strategy for optimizing scheduling of multiple types of mobile emergency resources and improving the resilience of distribution networks based on a data-model hybrid driving method as described in claim 7, characterized in that: In step S10, in the mobile energy storage vehicle image reinforcement learning memory unit D... M Samples N were collected using an importance sampling method. s And the importance weight of each sample is calculated using formula (62): In the formula, N D Represents memory unit D M The total number of samples in the sample; This indicates the priority of the sample; β is a hyperparameter. Combining the strategy of the Double DQN algorithm, the loss function is calculated using formulas (63) and (64): In the formula, and These are the current network parameters and the target network parameters of the mobile energy storage vehicle, respectively; γ is a discount factor reflecting the impact of the future Q value on the current action, with a value of [0,1]. The neural network parameters of the mobile energy storage vehicle graph reinforcement learning algorithm are updated based on the stochastic gradient descent method and sample importance weights, as specifically expressed as follows: In the formula, α M The learning rate of the gradient descent algorithm for the current network of the mobile energy storage vehicle is set; and after a fixed number of steps N... up Then, the parameters of the target network for the mobile energy storage vehicle are processed in the following manner. Update: The TD-error value is calculated according to formula (63). Calculate the new priority for each sample Where ∈ is a constant used to prevent the priority from being zero; in the mobile energy storage vehicle graph reinforcement learning memory unit D M The priority of each sample is updated in the process; The parameters of the graph reinforcement learning algorithm neural network for emergency repair teams are updated.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the multi-type mobile emergency resource optimization scheduling and distribution network resilience enhancement strategy based on the data-model hybrid driving method as described in any one of claims 1-8.

Images