An event-triggered recovery method and device for a post-disaster island microgrid

Abstract

The application relates to the technical field of power grid emergency scheduling, and discloses an event-triggered recovery method and device for a post-disaster island micro-grid. The method is aimed at the scenario that disasters cause the distribution network and the main network to be decoupled, equipment to be damaged, road passage to be blocked, and fault information to be gradually reported, constructs an event-triggered recovery scheduling environment, fuses fault elements, power sources, energy storage, loads, repair teams, roads and operation margin states to construct a state vector, and integrates repair dispatching, switch switching, black start, energy storage access and load recovery into a unified action space; and through an operation safety filter, actions that violate the constraints of topology, power flow, power sources, energy storage, frequency and repair resources are pre-shielded. The application can quickly update the recovery strategy when the system state substantially changes, effectively improves the recovery speed of key loads, reduces the amount of power supply failure, and significantly enhances the operation resilience of the post-disaster island micro-grid.

Description

Technical Field

[0001] This invention relates to the field of power grid emergency dispatching technology, and in particular to an event-triggered recovery method and apparatus for isolated microgrids after disasters. Background Technology

[0002] Extreme natural disasters such as typhoons, floods, and earthquakes can easily cause power grid trunk line tripping, switchgear damage, distributed power source access failures, and communication interruptions, leading to the disconnection of the distribution network from the main grid and large-scale power outages. Post-disaster recovery and dispatching of isolated microgrids is a core aspect of improving grid resilience. It requires the simultaneous coordination of multiple decisions, including faulty component repair, network topology reconfiguration, black-start power supply start-up and shutdown, energy storage charging and discharging regulation, and load tiered recovery. Under the constraints of limited repair resources and power supply capacity, it is crucial to maximize the recovery speed of critical loads and the reliability of power supply. However, existing post-disaster recovery and dispatching methods have the following main shortcomings: Insufficient adaptability of dispatch response: In post-disaster scenarios where fault information, road conditions, and equipment availability are gradually revealed, the traditional fixed-cycle dispatch mode either suffers from dispatch lag or generates a large number of invalid re-dispatch, making it difficult to dynamically match the pace of disaster evolution. High degree of decoupling in decision-making dimension: Most methods separate emergency repair path planning from microgrid operation and scheduling, failing to fully consider the coupling relationship between road traffic obstruction, emergency repair resource constraints and power balance and energy storage constraints within the island, resulting in insufficient actual feasibility of the solution; Online solution efficiency is low: Mixed integer programming optimization methods can explicitly model constraints, but the model solution time increases sharply with the scale of the fault, which cannot meet the millisecond-level rapid rescheduling requirements after the disaster. Intelligent decision-making lacks security guarantees: Existing scheduling methods based on reinforcement learning do not perform pre-screening of the action space, and are prone to outputting infeasible solutions that violate topological connectivity, power balance or frequency security, making them difficult to apply directly to engineering practice.

[0003] Therefore, those skilled in the art provide an event-triggered recovery method and apparatus for isolated microgrids after disasters to solve the problems mentioned in the background art. Summary of the Invention

[0004] This invention aims to solve the technical problems of untimely response, low decision coupling, insufficient solution efficiency, and lack of security constraints in the post-disaster recovery and scheduling of isolated microgrids. It provides an event-triggered intelligent recovery method and device to achieve rapid, safe, and highly resilient post-disaster recovery and scheduling.

[0005] To achieve the above objectives, the present invention provides an event-triggered recovery method for isolated microgrids after a disaster, comprising the following steps: Acquire the topology, faulty component status, switch status, distributed power source status, energy storage status, load level, road access status, and emergency repair resource status of the isolated microgrid after the disaster. Construct an event-triggered recovery scheduling environment and trigger recovery strategy updates when a new fault is reported, a repair team arrives, a fault is repaired, road traffic status changes, power status changes, energy storage status changes, or critical load recovery requests change. Construct a recovery state vector based on the state of the islanded microgrid at the current event moment; Generate candidate recovery actions, including dispatching emergency repair teams, fault repair, switch switching, black start power supply activation, energy storage access, and load tiered recovery. The operation safety filter is used to eliminate candidate recovery actions that do not meet the constraints of island topology, power flow, power capacity, energy storage energy, frequency security and emergency repair resources; The recovery action is selected from the set of available actions through a deep reinforcement learning policy network; Perform the recovery action and update the operating status of the islanded microgrid until the recovery termination condition is met.

[0006] Furthermore, the recovery state vector includes the fault component state vector, switch state vector, energy storage state vector, load priority vector, emergency repair team location and availability vector, road access time vector, and islanded microgrid operating margin vector.

[0007] Furthermore, the candidate recovery actions include at least one of the following actions: dispatching a repair team to the faulty component, closing or opening a sectionalizing switch, closing or opening a tie switch, starting a black start power supply, connecting mobile energy storage to a designated node, restoring a designated load level, and cutting off non-critical loads.

[0008] Furthermore, the operational security filter is used to determine the feasibility of candidate recovery actions before the policy network selects an action, and to set the selection probability of infeasible actions to zero. The prohibited actions include actions that cause the islanded microgrid topology to fail to meet the power supply connectivity requirements, actions that cause power flow, voltage or line capacity to exceed limits, actions that cause energy storage charge state to exceed limits, actions that cause frequency deviation or frequency change rate to exceed thresholds, and actions that cause repair team task conflicts or work time to exceed limits.

[0009] Furthermore, the deep reinforcement learning policy network includes a memory state update module, an action scoring module, and an action selection module; the memory state update module is used to fuse historical event information and current state information, the action scoring module is used to generate candidate action scores, and the action selection module is used to combine a running safety filter to output the probability of feasible actions.

[0010] Furthermore, the training reward function of the deep reinforcement learning policy network includes power outage penalty, emergency repair team movement time penalty, switching action penalty, operational risk penalty, critical load restoration reward, and frequency safety margin reward.

[0011] Furthermore, the event-triggered recovery scheduling environment uses disaster intensity, equipment vulnerability, road obstruction probability, fault reporting process, and repair time distribution to generate training scenarios to simulate the recovery process as post-disaster fault information is gradually revealed.

[0012] An event-triggered recovery device for isolated microgrids after disasters includes a data acquisition module, an event recognition module, a state construction module, an action generation module, a safety screening module, an intelligent decision-making module, and an execution feedback module. The data acquisition module is used to acquire the topology, equipment status, load status, power supply status, energy storage status, road status, and emergency repair resource status of the isolated microgrid. The event recognition module is used to identify whether an event has occurred that triggers a recovery strategy update; The state construction module is used to construct a recovery state vector based on the current event time information; The action generation module is used to generate a set of candidate recovery actions; The security screening module is used to eliminate candidate recovery actions that do not meet the operational security constraints; The intelligent decision-making module is used to select recovery actions based on a deep reinforcement learning policy network. The execution feedback module is used to perform recovery actions and provide feedback on the recovery results.

[0013] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an event-triggered recovery method for post-disaster isolated microgrids.

[0014] The present invention has the following beneficial effects: 1. The present invention proposes an event-triggered recovery method and device for isolated microgrids after disasters. Compared with the prior art, it dynamically adapts to the disaster process, significantly improves scheduling efficiency, and adopts an event-triggered mechanism to replace fixed-period scheduling. The strategy is updated only when the system state changes substantially, which avoids the waste of ineffective rescheduling and can respond in real time to dynamic information such as new fault reports, changes in road accessibility, and equipment repair completion. It perfectly adapts to the characteristics of the gradual revelation of information after a disaster.

[0015] 2. The present invention proposes an event-triggered recovery method and device for isolated microgrids after disasters, which integrates five types of decisions—emergency repair dispatch, network reconfiguration, black start, energy storage access, and load restoration—into the same action space, thereby achieving coordinated decision-making between emergency repair resource scheduling and microgrid operation optimization, avoiding scheme deviations caused by decoupled scheduling, and maximizing the overall post-disaster recovery benefits.

[0016] 3. The present invention proposes an event-triggered recovery method and device for isolated microgrids after disasters. It sets up an operation safety filter in front of the intelligent policy network to pre-verify the feasibility of actions from six dimensions: topology, power flow, power supply, energy storage, frequency, and emergency repair resources. It directly blocks infeasible solutions and avoids the risk of reinforcement learning outputting illegal actions from a mechanism perspective, ensuring that the recovery plan meets the requirements of power grid operation safety. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the event-triggered islanded microgrid recovery and scheduling process described in this invention; Figure 2 This is a schematic diagram of the collaborative structure of the security filter and the intelligent policy network described in this invention. Detailed Implementation

[0018] The technical solutions of the specific embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described specific embodiments are only a part of the specific embodiments of the present invention, and not all of them. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0019] Reference Figure 1-2 The present invention provides a specific implementation method: Post-disaster isolated microgrid recovery object modeling: Following a disaster, some power distribution lines, switching equipment, distributed power supply access equipment, and load connection points may fail, and the upstream power grid may lose its power supply capacity. At this time, areas with local power sources, energy storage, and controllable loads can switch to islanded microgrid operation mode. The objects of post-disaster islanded microgrid recovery are represented as: ; Where N is the set of nodes, Let S be the set of controllable switches. Let P be the set of repairable faulty components, and let P be the set of controllable power supplies within the island. Let D be the set of energy storage and emergency power supplies, D be the set of loads, and U be the set of emergency repair teams. The indices for nodes, lines, switches, power supplies, energy storage, loads, and emergency repair teams are denoted as follows: ; The moment when the event triggers the decision is denoted as The time when the previous event was triggered is recorded as The next state time after the recovery action is recorded as This is an indicator function; it takes the value 1 if the condition inside the parentheses is true, and 0 otherwise. Line status: For the line These represent its first and last nodes, respectively; Indicates the line at the time of the event. Available status, Indicates availability. Indicates unavailable; Faulty component status: For repairable faulty components Indicates its damaged state. This indicates that the item is damaged and has not been repaired. Indicates availability; after repair, the damage status is updated from 1 to 0; Switch status: For switches Indicates its switching state. Indicates closure. Indicates disconnection; the switch state is used to change the power supply path, enabling network reconfiguration and load restoration; Controllable power supply status: For controllable power supplies Indicates startup status. This indicates that power has been supplied and is operational. This indicates that the system is not started or is unavailable. To contribute one's efforts To define the upper and lower limits of output, power sources with black-start capability can serve as initial power sources when the main grid loses power. The set of started power sources is defined as: Pτ(on) = {g∈P|µg,τ=1}; Energy storage status: For energy storage or mobile emergency power supplies Indicates the access status. This indicates that the connection has been established. Indicates no connection; In a charged state, These are the charging and discharging power, respectively. These are the upper and lower limits of the state of charge. Rated energy capacity, These are the charging and discharging efficiencies, respectively. Load status: For load For the purpose of merit, The weights are assigned based on importance; the higher the weight, the higher the recovery priority. This represents the actual power supply status. This indicates that power has been restored. Indicates no recovery. Load recovery action. The update relationship is satisfied with the power supply status: χdτ + = max(χd,τ , λd(ld),τ); ∀d∈D; If load d has been powered at event time τ or has been restored at that event time, it will remain powered at the next state time. Emergency repair team status: For the emergency repair team Current position It is in an available state. Indicates that the space is available for dispatch. This indicates that a task is being performed. The travel time for the repair team to travel from its current location to the faulty component r. Repair time for faulty components; Road conditions: Indicates the moment of the event The passage time from road node m to n is updated with the event when the road is blocked or the emergency repair passage is restored. The emergency repair passage time is dynamically calculated based on this. In summary, the above events The recovery state of an isolated microgrid can be represented as: ; This state provides a unified input for event recognition, action generation, security screening, and intelligent decision-making.

[0020] Event triggering mechanism; The set of triggering events for a post-disaster isolated microgrid is defined as follows: ; in Report new faults or power outages; To allow the repair team to reach the target location; Equipment repair complete; Changes in road traffic hours; This refers to the change in the available state of a black-start power supply, energy storage, or mobile power source. Changes to critical load recovery requests; when the conditions are met The recovery strategy is updated when the time is right; otherwise, the recovery arrangement from the previous moment is maintained to avoid frequent and ineffective decisions.

[0021] Recover the state vector construction; At the moment of the event Construct the recovered state vector for the policy network input: ; in For the state vector of the faulty component, For the switch state vector, Let be the energy storage state vector. For load priority vector, This is a vector representing the location and availability of the repair team. This is the road travel time vector. This is the operating margin vector for islanded microgrids. The operating margin vector describes the current safety margin of the islanded microgrid, including power balance margin, voltage margin, line capacity margin, energy storage margin, and frequency safety margin. Restore the action space definition; At the moment of the event The set of recovery actions is defined as follows: ; These correspond to five types of actions: emergency repair dispatch, switch switching, black start power supply activation, energy storage / mobile power supply connection, and load tiered restoration. Emergency repair dispatch actions: Indicates the emergency repair team Was it dispatched to the faulty component? Set 1 to dispatch, and 0 to no dispatch; Switching action: Indicates whether to switch Execute the switching operation; a value of 1 indicates execution, and a value of 0 indicates non-execution. Black Start Action: Indicates whether to start the black boot power supply. Set 1 to start, and 0 to not start; Energy storage integration actions: Indicate whether to include energy storage / mobile power supply To connect to an isolated microgrid, set 1 to connect and 0 to not connect. Load recovery action: Indicates whether it is at the time of the event. Post-load recovery Power supply: 1 indicates restoration, 0 indicates no restoration; Run a security filter design; The operational safety filter, located between the candidate action generation module and the intelligent policy network, determines whether an action meets the operational safety requirements of an islanded microgrid before execution. For any candidate recovery action... The next state induced after its execution is denoted as If the subsequent state satisfies all safety constraints, the action is retained; otherwise, it is masked. Define a feasibility flag for the candidate action: in, Representing logic, These are the feasibility assessment conditions for topology, power flow, power supply, energy storage, frequency, and emergency repair resources; when all conditions are met... Actions can be selected by the policy network; otherwise, actions set to 0 are blocked. Island topology constraints; Topological constraints are used to determine whether, after a candidate action is executed, the load whose power supply has been restored can be connected to the already started power source through available lines and closed switches. Let... For the path after the action is executed The charged state, For nodes The power supply status must meet the following requirements: ; ; ; in To configure a set of circuits with controllable switches, The corresponding switch for the circuit. The node where the load is located. This indicates that the faulty circuit should not be energized. This indicates that the circuit with the switch can be energized only when the switch is closed. This indicates that the load can only be restored when the node where the load is located is powered. The feasibility criteria for island topology are: .

[0022] Current safety constraints; Power flow safety constraints are used to verify power balance, voltage safety, and line capacity limitations. Active power balance constraints are as follows: in For isolated network losses; The node voltage constraint is: ;in These are the upper and lower limits of the allowable node voltage, respectively. The line capacity constraint is: ;in This is the upper limit of the allowable power flow for the line; when the line is not energized, the power flow is forcibly set to 0. The conditions for determining the feasibility of tidal current safety are: (The above three formulas are true) Power capacity constraints; The power supply capacity constraint verifies the controllable power output range and black-start availability. The power output constraint is as follows: ; The power output of a non-activated power supply is forced to 0, while the power output of an activated power supply is within the allowable range. Black start availability constraints are: ;in This is the black start usable state for the power supply; only power supplies that meet the black start conditions can be started. The conditions for determining the feasibility of power supply capacity are: .

[0023] Energy storage energy constraints; Energy storage energy constraint verification includes access status, charge / discharge power, and state of charge limits. The charge / discharge power constraint is as follows: in These are the maximum charging and discharging power, respectively; charging and discharging are not possible without energy storage. The state of charge update and upper and lower bound constraints are as follows: in The time interval between the current event and the next event. The conditions for determining the feasibility of energy storage are: ; Frequency security constraints; Frequency security constraints are used to verify the frequency support capability under maximum power disturbances. Assume... This is the estimated maximum frequency deviation under the preset disturbance. This is an estimate of the maximum rate of change of frequency; To set the maximum allowable frequency deviation threshold, To define the maximum permissible rate of change threshold, a frequency safety margin is defined: ; in This is the frequency safety margin after the action is executed; when This indicates that both the maximum frequency deviation and the maximum frequency change rate have not exceeded the allowable thresholds; when This indicates that the candidate actions may pose a frequency security risk; The conditions for determining the feasibility of frequency security are as follows: ; Resource constraints for emergency repairs; The emergency repair resource constraint verification checks the number of dispatched resources, task conflicts, and operation time limits. Each emergency repair team can be dispatched to a maximum of one fault point at a time. ; Each faulty component can receive a maximum of one repair team at a time, and only the damaged component requires repair: ; Repair operations must not exceed the maximum allowable operation time for the current scheduling phase to avoid dispatching repair teams to fault locations that cannot be reached and repaired within the current scheduling phase. ; Therefore, the conditions for determining the feasibility of emergency repair resources can be expressed as: ; In summary, running the safety filter will group all candidate actions that satisfy Mτ (Λτ) = 1 into a set of actionable actions: Aτ(saf e) = {Λτ∈Aτ|Mτ (Λτ )=1}; where Aτ is the set of candidate actions at event time τ, and Aτ(saf e) is the set of actionable actions obtained after running the safety filter. The intelligent policy network selects the final recovery action only from Aτ(saf e) to ensure that the output action meets the basic safety requirements for post-disaster recovery of the islanded microgrid.

[0024] Design of deep reinforcement learning strategy networks; This invention employs a deep reinforcement learning strategy network with historical memory to generate recovery actions. Let... As a characteristic of historical memory, the memory update process is as follows: ;in For the previous event moment, For parameters The memory update network can adopt a GRU or LSTM structure to fuse historical event sequences with current state information; The policy network, combined with the running security filter, outputs the corrected action probability: ; in The action scoring function output by the policy network. The probability of choosing the corresponding action; the cause of the inability to act. The probability of being directly set to zero means that the policy network makes decisions only within the set of possible actions.

[0025] Restore the profit function and the training objective; To balance critical load restoration, power outage reduction, repair efficiency, and safe operation, the event timing is defined. Single-step recovery benefits: in The amount of power lost during the event interval. For the time required for the repair team to move, The number of times the switch is activated. The above four items are penalty items for operational risks; For critical load recovery, For frequency safety margin, the above two items are bonus items; to The weighting coefficients can be adjusted according to the recovery target; The training objective of the policy network is to maximize the cumulative discount gain throughout the recovery process. in To provide the mathematical expectation of disaster scenarios, fault evolution, road changes, and strategy trajectories. As a discount factor, This is the time when the recovery process ends.

[0026] Offline training and online application process; Offline training process: Based on disaster intensity, equipment vulnerability, road obstruction probability, fault reporting delay distribution, and repair time distribution, batch generation of post-disaster fault evolution training scenarios; Simulate the event triggering process step by step in various scenarios, and train the policy network using reinforcement learning algorithms (such as PPO, DQN); The safety filter operates throughout the training process, guiding the policy network to learn safe and feasible recovery decisions. After training converges, the policy network parameters are saved for online deployment. The online application algorithm process is as follows: Input the post-disaster isolated microgrid topology, equipment status, load level, power supply status, energy storage status, and emergency repair resource status; Initialize event clock Memory state and the set of faults to be recovered; If critical loads have not been restored or critical components have not been repaired, execute the following cyclically: 1. Proceed to the next event moment. ; 2. Update fault reporting, road conditions, power availability, energy storage status, and emergency repair team status; 3. Construct the restored state vector ; 4. Update memory status ; 5. Generate a set of candidate recovery actions; 6. Calculate action feasibility indicators and filter the set of feasible actions; 7. Select the optimal recovery action through the policy network. ; 8. Perform emergency repair dispatch, switch switching, power startup, energy storage connection, and load restoration actions;

[0027] 9. Calculate revenue and update system status; After the cycle ends, the output includes the islanded microgrid recovery sequence, critical load recovery time, total power loss, and emergency repair resource utilization rate. Example 2

[0028] like Figure 1-2 As shown, taking a typical islanded microgrid containing distributed photovoltaics, energy storage, diesel generators, mobile emergency power supplies and multi-level loads as an example, after a disaster occurs, the upstream grid interconnection switch is disconnected, and the microgrid switches to islanded operation; some feeders, sectionalizing switches and load access points fail; hospitals, communication base stations, water supply pumping stations and emergency shelters are set as first-level critical loads; In the initial recovery phase, the system prioritizes starting diesel generators with black-start capabilities. Based on the energy storage charge status, it determines the first batch of critical loads that can be restored. Subsequently, when the repair team arrives at the fault point, the faulty equipment is repaired, a new fault is reported, or road conditions change, the intelligent recovery strategy is updated. The operational safety filter first eliminates actions that may lead to islanded power imbalance, voltage exceeding limits, excessive energy storage discharge, insufficient frequency margin, or conflicting repair team tasks. The strategy network then selects the recovery plan with the highest overall benefit from the remaining available actions. The implementation steps are as follows: Event-triggered intelligent recovery for isolated microgrids: Step 1: Input the post-disaster isolated microgrid topology, equipment status, load level, power supply status, energy storage status, and emergency repair resource status; Step 2: Initialize the event clock The memory state h0 and the set of faults to be recovered; Step 3: while there are unrecovered critical loads or unrepaired critical components do; Step 4: Proceed to the next event moment ; Step 5: Update fault reporting, road status, power availability, energy storage status, and emergency repair team status; Step 6: Construct the restored state vector ; Step 7: According to Update memory status ; Step 8: Generate a set of candidate recovery actions ; Step 9: According to the formula: Calculate the feasibility indicators of the action; Step 10: According to the formula

[0029] Select recovery action ; Step 11: Execute emergency repair dispatch, switch switching, power startup, energy storage connection, and load restoration actions; Step 12: Calculate the benefit Jτ based on the recovery effect. And update the system status; Step 13: Output the islanded microgrid recovery sequence, critical load recovery time, total power loss, and emergency repair resource utilization rate; The following table compares different recovery methods:

[0030] The following points should be noted in this article: 1. The accompanying drawings of the embodiments disclosed herein only relate to the structures involved in the embodiments disclosed herein; other structures can be referred to in general design.

[0031] 2. Where there is no conflict, the embodiments of this disclosure and the features in the embodiments can be combined with each other to obtain new embodiments.

[0032] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing specific embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An event-triggered recovery method for isolated microgrids after a disaster, characterized in that: The process includes the following steps: Acquire the topology, faulty component status, switch status, distributed power source status, energy storage status, load level, road access status, and emergency repair resource status of the isolated microgrid after the disaster. Construct an event-triggered recovery scheduling environment and trigger recovery strategy updates when a new fault is reported, a repair team arrives, a fault is repaired, road traffic status changes, power status changes, energy storage status changes, or critical load recovery requests change. Construct a recovery state vector based on the state of the islanded microgrid at the current event moment; Generate candidate recovery actions, including dispatching emergency repair teams, fault repair, switch switching, black start power supply activation, energy storage access, and load tiered recovery. The operation safety filter is used to eliminate candidate recovery actions that do not meet the constraints of island topology, power flow, power capacity, energy storage energy, frequency security and emergency repair resources; The recovery action is selected from the set of available actions through a deep reinforcement learning policy network; Perform the recovery action and update the operating status of the islanded microgrid until the recovery termination condition is met.

2. The event-triggered recovery method for isolated microgrids after disasters, as described in claim 1, is characterized in that: The recovery state vector includes the fault component state vector, switch state vector, energy storage state vector, load priority vector, emergency repair team location and availability vector, road access time vector, and islanded microgrid operating margin vector.

3. The event-triggered recovery method for isolated microgrids after disasters, as described in claim 1, is characterized in that: The candidate recovery actions include at least one of the following actions: dispatching a repair team to the faulty component, closing or opening a sectionalizing switch, closing or opening a tie switch, starting a black start power supply, connecting mobile energy storage to a designated node, restoring a designated load level, and cutting off non-critical loads.

4. The event-triggered recovery method for isolated microgrids after disasters, as described in claim 1, is characterized in that: The operational security filter is used to determine the feasibility of candidate recovery actions before the policy network selects an action, and to set the selection probability of infeasible actions to zero. The prohibited actions include actions that cause the islanded microgrid topology to fail to meet the power supply connectivity requirements, actions that cause power flow, voltage or line capacity to exceed limits, actions that cause energy storage charge state to exceed limits, actions that cause frequency deviation or frequency change rate to exceed thresholds, and actions that cause repair team task conflicts or work time to exceed limits.

5. The event-triggered recovery method for isolated microgrids after disasters, as described in claim 1, is characterized in that: The deep reinforcement learning policy network includes a memory state update module, an action scoring module, and an action selection module; the memory state update module is used to fuse historical event information and current state information, the action scoring module is used to generate candidate action scores, and the action selection module is used to output the probability of feasible actions in conjunction with a running safety filter.

6. The event-triggered recovery method for isolated microgrids after disasters, as described in claim 1, is characterized in that: The training reward function of the deep reinforcement learning policy network includes power outage penalty, emergency repair team movement time penalty, switching action penalty, operational risk penalty, critical load restoration reward, and frequency safety margin reward.

7. The event-triggered recovery method for isolated microgrids after a disaster, as described in claim 1, is characterized in that: The event-triggered recovery scheduling environment uses disaster intensity, equipment vulnerability, road obstruction probability, fault reporting process, and repair time distribution to generate training scenarios to simulate the recovery process as post-disaster fault information is gradually revealed.

8. An event-triggered recovery device for isolated microgrids after disasters, comprising a data acquisition module, an event recognition module, a state construction module, an action generation module, a security screening module, an intelligent decision-making module, and an execution feedback module, characterized in that: The data acquisition module is used to acquire the topology, equipment status, load status, power supply status, energy storage status, road status, and emergency repair resource status of the isolated microgrid. The event recognition module is used to identify whether an event has occurred that triggers a recovery strategy update; The state construction module is used to construct a recovery state vector based on the current event time information; The action generation module is used to generate a set of candidate recovery actions; The security screening module is used to eliminate candidate recovery actions that do not meet the operational security constraints; The intelligent decision-making module is used to select recovery actions based on a deep reinforcement learning policy network. The execution feedback module is used to perform recovery actions and provide feedback on the recovery results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the event-triggered recovery method for post-disaster isolated microgrids as described in any one of claims 1 to 7.

Images