Emergency decision-making fireproof isolation method for charging place based on digital twin multi-agent cooperation

Abstract

This invention relates to the field of control or regulation system technology, specifically to a digital twin multi-agent collaborative emergency decision-making and fire isolation method for charging sites. This method constructs a multi-layered digital twin topology for the charging site based on charging locations, charging piles, power supply branches, vehicle access status, fire zones, and executable control interfaces. The charging operation status, fault status, and fire monitoring status are mapped to this multi-layered topology, generating a thermal risk potential field, an electrical continuous power supply potential field, and a spatial diffusion potential field, respectively. A fire isolation boundary is collaboratively generated by a charging control agent, a power supply branch agent, a spatial zone agent, and a central control decision agent. This results in the output of a sequence of control commands for stopping charging, branch disconnection, and zone isolation, with the topology status updated based on execution feedback. This invention enables the fire isolation boundary to simultaneously cover the energy supply path of the risk source and the main propagation path of the thermal risk, improving the closed-loop consistency and boundary matching of emergency isolation control.

Description

Technical Field

[0001] This invention relates to the field of control or regulation system technology, specifically to a fire prevention and isolation method for emergency decision-making in charging locations using digital twin multi-agent collaboration. Background Technology

[0002] Charging facilities typically include multiple charging spots, charging piles, power supply branches, vehicle access points, fire protection zones, and control interfaces related to power outages, charging shutdowns, and isolation. Current emergency control systems in charging facilities primarily rely on equipment-side alarms and site fire alarms as trigger sources. Upon receiving charging pile fault codes, abnormal charging current or voltage, abnormal vehicle charging status, or temperature or smoke alarms, the control platform executes actions such as stopping charging, cutting off local power supply branches, alerting personnel, or linking fire protection facilities according to pre-set handling rules. This type of solution typically employs a fixed-point binding method, where one charging pile corresponds to one charging spot, one power supply branch corresponds to several charging piles, and one fire protection zone corresponds to several spatial areas. When an alarm is triggered by a device or zone, the control platform searches the contingency plan table based on the device number or zone number and executes the corresponding command to stop charging, cut off power, or isolate the area. Its advantages include clear logic and short execution links, but the emergency control is mainly based on the location of the alarm source and the contingency plan configuration, failing to adequately express the coupling relationships between thermal risk propagation paths, electrical power supply paths, and spatial adjacency paths within the charging facility.

[0003] In more similar technical solutions, digital twin technology is used for monitoring the operation of charging sites. By establishing site plan models, equipment ledger models, and operational status models, the online status, output power, fault status, vehicle occupancy status, and fire zone status of charging piles are mapped onto a virtual scene. Control personnel or control programs can view abnormal points in the virtual scene and select appropriate actions based on these abnormal points. Some solutions also incorporate the power supply relationships between devices into the model, such as grouping several charging piles under the same distribution box or the same feeder branch, and executing branch-level disconnection in the event of an electrical fault. This type of digital twin solution can improve the completeness of the status presentation, but its twin model is mostly used for visualization, operation records, or single-level relational queries. Even if the model contains equipment topology and spatial location, these usually serve equipment management and scene display separately, without forming a unified multi-layer topology for emergency isolation boundary calculation. Therefore, in the early stages of a fire or the initial risk of thermal runaway, the virtual model can display the location of the abnormal source, but it is difficult to directly generate the isolation range based on the combined effects of heat propagation, electrical power supply, and spatial diffusion.

[0004] Existing multi-agent collaborative control schemes are already applied in power dispatching, building control, and equipment group control. The conventional approach is to assign different control objects to different agents, each agent generating control suggestions based on its local state, and then coordinating through a central control program. In charging locations, charging pile control, power supply branch control, and fire zone control can be handled by different control logics. For example, the charging pile control logic is responsible for stopping abnormal charging sessions, the power distribution control logic is responsible for determining whether to disconnect the branch, and the fire linkage control logic is responsible for executing zone alarms or isolation. This type of scheme can break down complex control tasks into multiple local decisions, but each local decision typically operates based on data within its own control domain, lacking a unified risk potential field and a unified topology index. When charging pile control assumes that stopping a single pile output has completed risk handling, the power supply branch may still provide energy to adjacent equipment through the same branch; when fire zone control executes isolation based on spatial scope, it may extend to areas without electrical coupling to the risk source. The lack of a common computational basis for the same risk propagation process among different control domains easily leads to locally reasonable but globally uncoordinated emergency actions.

[0005] The main technical problem with existing technologies is that emergency fire isolation control in charging locations lacks a boundary generation mechanism that uniformly couples the thermal risk propagation path, electrical energy supply path, and spatial adjacency diffusion path, leading to inconsistencies between the fire isolation boundary and the actual risk evolution boundary. The technical reason for this problem is not simply the inaccuracy of a single alarm, but rather that existing solutions typically handle alarm identification, equipment power-off, and spatial isolation in different logical layers. There is no calculable unified topological relationship between alarm nodes, power supply branch nodes, fire zone nodes, and control interface nodes. When charging anomalies or fire risks occur, the control program often can only determine the scope of action based on the alarm location or fixed plans, failing to simultaneously determine whether the risk source still has upstream energy input, whether the thermal risk is spreading along adjacent charging positions, and whether the isolation action will cut off too many non-risk nodes. This results in situations where a single-point charging stop is insufficient to interrupt continuous power supply, fixed zone isolation deviates from the main propagation direction, and large-scale power outages affect status feedback, making it difficult for emergency decisions to form a closed-loop, consistent control boundary between risk source energy interruption and spatial fire isolation. Summary of the Invention

[0006] The purpose of this invention is to provide a digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites, which can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A digital twin multi-agent collaborative emergency decision-making and fire isolation method for charging sites includes: constructing a multi-layer digital twin topology of the charging site based on charging positions, charging piles, power supply branches, vehicle access status, fire zones, and executable control interfaces; acquiring charging operation status, fault status, and fire monitoring status and mapping them to the multi-layer topology; generating a thermal risk potential field, an electrical continuous power supply potential field, and a spatial diffusion potential field based on heat propagation relationships, electrical power supply relationships, and spatial adjacency relationships, respectively; collaboratively generating fire isolation boundaries by charging control agents, power supply branch agents, spatial zone agents, and overall control decision agents; outputting a sequence of control commands for stopping charging, branch disconnection, and zone isolation based on the fire isolation boundaries, and updating the multi-layer topology based on execution feedback.

[0008] Preferably, constructing the digital twin multi-layer topology of the charging site includes: setting charging positions, charging piles, power supply branches, vehicle access status, fire zones, and executable control interfaces as topology nodes; setting spatial adjacency relationships, electrical power supply relationships, thermal risk propagation relationships, and control linkage relationships as topology edges with different attributes; configuring the topology nodes with operating status, anomaly level, energy input intensity, isolation executability, and fire zone identifier; and configuring the topology edges with propagation direction, power supply direction, adjacency intensity, and control linkage sequence, so that the same risk node has a consistent node index across the spatial layer, electrical layer, and control layer.

[0009] Preferably, generating the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field includes: assigning initial risk values ​​to abnormal nodes based on fault status, temperature rise status, charging output changes, and fire monitoring status; calculating the thermal risk transfer amount to adjacent nodes along the thermal risk propagation relationship; calculating the continuous energy input amount from the power supply branch to the abnormal node and adjacent nodes along the electrical power supply relationship; calculating the risk diffusion amount between adjacent charging positions and fire compartments along the spatial adjacency relationship; and superimposing and calibrating the three types of potential fields according to the same node index to obtain a comprehensive node risk sequence used to determine the fire isolation boundary.

[0010] Preferably, the multi-agent collaborative generation of fire-resistant isolation boundaries includes: the charging control agent generating a target charging session termination set based on the charging output status and vehicle access status; the power supply branch agent generating a candidate branch disconnection set based on the electrical continuous power supply potential field; the spatial partitioning agent generating a candidate partition isolation set based on the thermal risk potential field and the spatial diffusion potential field; and the overall control decision agent projecting the target charging session termination set, the candidate branch disconnection set, and the candidate partition isolation set onto the multi-layer topology to form a candidate boundary combination covering the energy supply path of the risk source and the main propagation path of the thermal risk.

[0011] Preferably, configuring isolation executability for the topology node includes: obtaining the stop charging interface, branch disconnection interface, and partition isolation interface corresponding to the executable control interface, and establishing a bidirectional mapping between the interface control object and the topology node; when a control interface simultaneously affects multiple topology nodes, establishing a set of controlled nodes according to the electrical power supply relationship and control linkage relationship; when a topology node is affected by multiple control interfaces, establishing a control chain according to the sequential dependency relationship of control actions; and writing the set of controlled nodes and the control chain into the multi-layer topology so that candidate boundary combinations can correspond to executable control instruction sequences.

[0012] Preferably, obtaining the node comprehensive risk sequence includes: recording the thermal risk transmission amount, continuous energy input amount, and spatial diffusion amount for the same topological node; generating a coupling correction term based on the fire compartment to which the node belongs, the number of adjacent nodes, the power supply branch level, and the current isolation status; applying the coupling correction term to the node values ​​of the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field; sorting the risk node, adjacent nodes, and nodes to be observed according to the corrected node values, and cross-matching the sorting results with the isolation executability to generate a risk priority control list for boundary screening.

[0013] Preferably, forming the candidate boundary combination includes: extracting a local sub-topology containing an abnormal node, an upstream power supply node, an adjacent spatial node, and a corresponding control interface node from the multi-layer topology; enumerating charging termination actions, power supply branch disconnection actions, and partition isolation actions within the local sub-topology; deleting action combinations that conflict with the control linkage sequence; establishing a correspondence between the boundary node set, the boundary topology edge set, and the control command set for the remaining action combinations; and determining the action combinations that simultaneously cover the upstream power supply node and the adjacent spatial node as candidate boundary combinations.

[0014] Preferably, the screening of the candidate boundary combinations includes: calculating the remaining power supply connectivity between the abnormal node and the upstream power supply node based on the topological connectivity changes after executing each candidate boundary combination; calculating the remaining risk propagation between the abnormal node and adjacent spatial nodes based on the spatial adjacency changes after executing each candidate boundary combination; counting the number of non-abnormal nodes included in the isolation range based on the controlled node set; progressively screening the candidate boundary combinations according to the constraint order of remaining power supply connectivity, remaining risk propagation, and the number of non-abnormal node isolations, and outputting the fireproof isolation boundary with the highest ranking.

[0015] Preferably, the output of the control command sequence includes: converting the charging termination action, power supply branch disconnection action, and zone isolation action in the highest-ranked fire isolation boundary into phased commands according to the control chain; writing the corresponding topology node index, control interface identifier, execution pre-state, and execution post-feedback state into each phase command; when the same topology node belongs to multiple phase commands simultaneously, merging the commands in the order of first terminating energy output, then disconnecting the power supply branch, and finally executing zone isolation; and using the merged phase commands as the control command sequence for stopping charging, branch disconnection, and zone isolation.

[0016] Preferably, updating the multi-layer topology based on execution feedback includes: receiving charging output feedback, branch disconnection feedback, zone isolation feedback, and fire monitoring feedback corresponding to each stage instruction; comparing the feedback with the post-execution feedback status; when any feedback fails to reach the post-execution feedback status of the corresponding stage instruction, marking the associated topology nodes and topology edges as objects to be recalculated; regenerating the thermal risk potential field, electrical continuous power supply potential field, and spatial diffusion potential field based on the objects to be recalculated, and re-selecting fire isolation boundaries while retaining the executed control instructions.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a digital twin multi-layered topology for charging locations, charging piles, power supply branches, vehicle access status, fire zones, and executable control interfaces. Under the same topology index, it generates a thermal risk potential field, an electrical continuous power supply potential field, and a spatial diffusion potential field. The fire isolation boundary can be jointly determined by the risk propagation path and the energy supply path. The charging control agent, power supply branch agent, spatial zone agent, and overall control decision agent generate candidate boundary combinations based on the same multi-layered topology. This ensures that charging stoppage, branch disconnection, and zone isolation are no longer triggered by isolated rules, but rather by establishing corresponding relationships around the risk source, upstream power supply node, adjacent spatial node, and control interface node. This approach allows the isolation boundary to cover the energy supply path of the risk source and the main thermal risk propagation path, reducing residual power supply caused by stopping charging only at alarm points and range offset caused by isolating only at fixed fire zones. This creates a closed-loop control process for emergency fire isolation control in charging locations that can be fed back and recalculated.

[0018] 2. This invention configures the operating status, anomaly level, energy input intensity, isolation executability, and fire compartment identifier of topology nodes, and configures the propagation direction, power supply direction, adjacency strength, and control linkage sequence of topology edges. Control commands can establish a correspondence with the controlled node set, control chain, and stage commands. After candidate boundary combinations are progressively filtered based on remaining power supply connectivity, remaining risk propagation, and the number of isolated non-abnormal nodes, the output commands for stopping charging, branch disconnection, and zone isolation have a clear execution sequence and feedback status. When the execution feedback is inconsistent with the expected feedback status, the associated topology nodes and edges are marked as objects to be recalculated, and the potential field and isolation boundaries are regenerated and filtered while retaining the executed control commands. This closed-loop mechanism can prevent unsuccessfully executed control actions from continuing to participate in subsequent decisions as completed states, reducing the probability of control link breakage, action conflicts, and isolation range mismatch. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the digital twin multi-agent collaborative charging site emergency decision-making and fire isolation method of the present invention; Figure 2 This is a flowchart illustrating the construction and state mapping of a digital twin multi-layer topology for charging locations according to the present invention. Figure 3 This is a flowchart illustrating the generation of three types of risk potential fields and the collaborative generation of fireproof isolation boundaries by multiple agents according to the present invention. Figure 4 This is a flowchart of the candidate boundary screening, staged control instruction output, and feedback recalculation process of the present invention. Detailed Implementation

[0020] refer to Figure 1In one embodiment, the digital twin multi-agent collaborative emergency decision-making and fire isolation method for charging sites is applied to charging sites with multiple charging positions, charging piles, power supply branches, vehicle access status, fire zones, and executable control interfaces. During implementation, a multi-layered digital twin topology of the charging site is used as a unified computational base map. Charging positions, charging piles, power supply branches, vehicle access status, fire zones, and executable control interfaces are mapped to nodes within the same topological space. Thermal propagation relationships, electrical power supply relationships, and spatial adjacency relationships are mapped to topological edges with different attributes. Charging operation status, fault status, and fire monitoring status are written to the corresponding nodes in real time. Abnormal nodes are jointly described by the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field. The impact range on surrounding nodes is then determined by the charging control agent, power supply branch agent, spatial partition agent, and overall control decision agent collaboratively generating a fireproof isolation boundary based on the same topology index. Subsequently, according to the fireproof isolation boundary, a sequence of control instructions for stopping charging, branch disconnection, and partition isolation is output, and the multi-layer topology is updated based on the execution feedback. This transforms emergency fireproof isolation control from being triggered by a single alarm point to closed-loop control constrained by thermal risk propagation paths, electrical energy supply paths, and spatial adjacency diffusion paths. The advantage of this embodiment is that the isolation boundary is jointly determined by multiple risk paths in the same digital twin topology, which can avoid the problem of inconsistent control boundaries caused by independent judgments of stopping charging, power outage, and partition isolation.

[0021] refer to Figure 2 Specifically, the digital twin multi-layer topology of the charging site is established through a node table, an edge table, and a status table. The node table records charging position nodes, charging pile nodes, power supply branch nodes, vehicle access nodes, fire compartment nodes, and executable control interface nodes. The edge table records the spatial adjacency, power supply direction, heat propagation direction, and control linkage sequence between nodes. The status table records the online status, anomaly level, energy input intensity, fire compartment affiliation, and executable control flag of each node. During topology initialization, multiple nodes related to the same physical object use the same object index and are configured with different level labels. For example, a charging position is simultaneously associated with adjacent charging positions in the spatial layer, charging piles in the charging layer, feeder branches in the power supply layer, and stop-charging interfaces in the control layer. When any level receives an anomaly, other levels can obtain the associated nodes through the same object index. This avoids the problem in existing solutions where alarm points, power distribution branches, and fire compartments are stored separately and difficult to compute together. The advantage of this embodiment is that the same risk object has a consistent index across the spatial, electrical, and control layers, providing a computable data foundation for subsequent potential field generation and boundary screening.

[0022] In this embodiment, the charging operation status includes charging session status, charging output status, vehicle access status, and charging pile fault status. The fire monitoring status includes the alarm status and zone status output by the existing fire monitoring equipment in the site. The fault status includes the abnormal code reported by the charging pile, branch power supply abnormality, and vehicle access abnormality. All statuses are time-aligned and object attribution confirmed before being written into the multi-layer topology. The time alignment adopts the status merging method within the same acquisition cycle. If the same object generates multiple statuses within one acquisition cycle, a status queue is generated with the status timestamp and object index, and the change direction is retained. When writing into the topology, the current value, the previous cycle value, and the change direction are simultaneously bound to the corresponding node, so that the subsequent potential field calculation can distinguish between continuous abnormalities, sudden abnormalities, and recovery abnormalities. The advantage of this embodiment is that the risk input is not an isolated alarm result, but a topology status with object attribution, time sequence, and change direction, which reduces the mismatch between abnormal statuses and control objects.

[0023] refer to Figure 3 In one embodiment, the initial risk value of an abnormal node is calculated jointly based on the normalized anomaly level, changes in charging output, fire monitoring status, and vehicle access anomalies, using the following formula: in, Indicates the first Initial risk values ​​for each topology node Indicates the normalized fault anomaly level. This represents the normalized change in charging output. This represents the normalized fire monitoring status quantity. This indicates the number of abnormal vehicle connections under normalization. , , , Represents the weight of the corresponding input quantity and satisfies For example, a certain charging node , , , ,Pick , , , At that time, the calculation yielded Its physical meaning is the local risk intensity of the node before topology propagation. The higher the value, the greater the initial driving force on the adjacent nodes when entering the potential field propagation calculation.

[0024] Furthermore, the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field are calculated from the thermal propagation edge, the electrical power supply edge, and the spatial adjacency edge, respectively. The thermal risk potential field diffuses from the abnormal node to the spatially adjacent node that satisfies the thermal propagation direction. The electrical continuous power supply potential field is transmitted along the power supply branch from the upstream power supply node to the charging pile and vehicle access node. The spatial diffusion potential field is transmitted according to the adjacency relationship between charging positions and between charging positions and fire compartments. The three types of potential fields are stored using the same topological node index. Different potential fields are not replaced by each other but are superimposed and calibrated in the comprehensive risk sequence. For example, the same node may have a low thermal risk but is still in the upstream power supply connected state, or the electrical power supply may have been cut off but is still located in the adjacent spatial diffusion path. The advantage of this embodiment is that the fire isolation boundary is not determined solely by the location of the abnormal node, but is jointly determined by whether the risk propagates, whether the energy is continuously input, and whether the space is adjacent.

[0025] The nodal values ​​of the three types of potential fields are generated according to the following formula: in, Indicates the first Thermal risk potential value of each topological node. Indicates the first Electrical continuous power supply potential field value of each topological node Indicates the first Spatial diffusion potential values ​​of each topological node , , They respectively represent the ability to send to the first A set of neighboring nodes that transmit thermal risks, power supply energy, and spatial diffusion effects to a given node. , , They represent the first The node to the first The weights passed to each node on the corresponding topological edge. Indicates the first The normalized energy input intensity of each node, such as the local energy input intensity of a node. Given two adjacent nodes with heat propagation risks of 0.4 and 0.3, corresponding to heat propagation weights of 0.5 and 0.2, then... This result indicates that the first The thermal risk potential field value of a node under the combined effect of local anomaly and adjacent heat propagation.

[0026] In one embodiment, the charging control agent receives the status related to charging pile nodes, vehicle access nodes, and charging session nodes. Based on the charging output status, fault status, vehicle access status, and initial risk value, it generates a target charging session termination set. During generation, it does not directly include all abnormal piles in the termination set. Instead, it includes the charging session corresponding to the abnormal node, adjacent high-risk sessions under the same power supply branch, and sessions adjacent to the abnormal node on the heat propagation path into the candidate range. Then, it combines the executable control interface node to determine whether there is a corresponding interface for the stop charging command. If there is a situation where the same stop charging interface controls multiple charging sessions, the controlled charging sessions are written into the candidate action as a whole set. The advantage of this embodiment is that the charging termination action can be linked with the heat propagation relationship and the power supply relationship, avoiding only stopping the alarm charging pile and omitting high-risk charging objects in the same branch or adjacent areas.

[0027] In one embodiment, the power supply branch intelligent agent receives the electrical power supply relationship between the power supply branch node, upstream power supply node, charging pile node, and control interface node. Based on the electrical continuous power supply potential field, it generates a candidate branch disconnection set. During generation, it first searches for the upstream power supply path of the abnormal node in the multi-layer topology, then identifies the power supply branch that still provides energy input to the abnormal node or adjacent risk node under the same power supply path. Subsequently, it determines the impact range of the disconnection action based on the mapping relationship between the branch disconnection interface and the controlled node set. If a branch disconnection interface affects multiple non-abnormal charging pile nodes at the same time, the number of affected nodes and the remaining power supply connectivity are written into the candidate branch disconnection set. The advantage of this embodiment is that the branch disconnection is not simply performed according to the branch number, but is selected around the continuous power supply path between the abnormal node and the upstream power supply node.

[0028] In one embodiment, the spatial partitioning agent receives fire zone nodes, charging station nodes, and spatial adjacent edges. Based on the thermal risk potential field and the spatial diffusion potential field, it generates a candidate partition isolation set. During generation, it extracts adjacent charging stations, adjacent fire zones, and corresponding partition isolation interfaces starting from the charging station where the abnormal node is located. It determines the main propagation path of the risk based on the direction of heat propagation and then writes the fire zone nodes and spatial boundary nodes that can block the main propagation path of the risk into the candidate set. In the case where there are multiple connected edges between adjacent partitions, the boundary coverage order is established according to the spatial diffusion potential field value from high to low. The advantage of this embodiment is that the partition isolation can be carried out around the main propagation path of the thermal risk, avoiding the inconsistency between the fixed partition isolation and the actual diffusion direction.

[0029] In one embodiment, the central control decision-making agent projects the target charging session termination set, the candidate branch disconnection set, and the candidate zone isolation set onto a multi-layer topology to form a candidate boundary combination. The candidate boundary combination is jointly represented by the boundary node set, the boundary topology edge set, and the control instruction set. The boundary node set records the isolated or retained charging position nodes, charging pile nodes, power supply branch nodes, and fire compartment nodes. The boundary topology edge set records the cut-off power supply edges, blocked spatial adjacency edges, and triggered control linkage edges. The control instruction set records the control interface identifiers and execution order corresponding to stopping charging, branch disconnection, and zone isolation. The advantage of this embodiment is that the fire isolation boundary is not a single control action, but a set of control objects jointly defined by the topology boundary and executable instructions.

[0030] In a preferred embodiment, the data structure of the multi-layer topology can be configured according to Table 1, which describes the node attributes, edge attributes, and state write content of different objects in the digital twin multi-layer topology, so that a unified risk calculation base map can be established based on the existing objects in the charging site during implementation.

[0031] Table 1 describes the spatial location and affiliation of charging station nodes with fire compartments. Charging position Charging point index, fire compartment, adjacent charging points Spatial adjacent edge, heat propagation edge Occupancy status, adjacent risk status Partition isolation interface association charging pile Charging station index, bind charging location, bind session Power supply side, control linkage side Output status, fault status Stop charging port association Power supply branch Branch line index, upstream power supply node, downstream charging pile Electrical power supply side, control linkage side Disconnection status, power supply status Branch disconnect interface association Vehicle access status Access object index, access charging pile, access time period Charging session edge, risk-related edge Access error, session change Charging stop control association Fire compartment Partition index, overlay charging position, adjacent partition Spatial adjacent edge, heat propagation edge Partition alarms, partition status Partition isolation interface association Executable control interface Interface index, controlled object, execution conditions Control the linked edge, control the chain edge Execution feedback, status return Instruction sequence association In this embodiment, the charging position nodes in Table 1 are used to describe the attribution relationship between spatial location and fire compartment, the charging pile nodes are used to describe the attribution relationship between charging output and charging session, the power supply branch nodes are used to describe the power supply relationship between upstream power supply nodes and downstream controlled nodes, the vehicle access status nodes are used to describe the time correlation between vehicle access and charging session, the fire compartment nodes are used to describe spatial isolation boundaries, and the executable control interface nodes are used to describe the executable entry points for stopping charging, branch disconnection, and compartment isolation. Each node establishes a connection with the topology edge through object index, so that the same abnormal event can be simultaneously located to the charging session, power supply path, and spatial adjacency range. The advantage of this embodiment is that the data organization method is consistent with the risk path required for fire isolation control, which can support subsequent multi-agent calculations on the same base map.

[0032] Preferably, when configuring the isolation executability of topology nodes, the stop charging interface, branch disconnection interface, and zone isolation interface corresponding to the executable control interface are obtained, and a bidirectional mapping is established between the interface control object and the topology node. The bidirectional mapping includes finding the controlled node from the control interface and finding the available control interface from the topology node. If a control interface affects multiple topology nodes at the same time, a set of controlled nodes is established according to the electrical power supply relationship and control linkage relationship. If a topology node is affected by multiple control interfaces, a control chain is established according to the sequential dependency relationship of control actions. For example, an abnormal charging pile can interrupt the session through the stop charging interface, block energy input through the upstream branch disconnection interface, and block spatial diffusion through the fire zone isolation interface. The three types of interfaces are written into the multi-layer topology according to the control chain. The advantage of this embodiment is that the candidate boundary combination can be converted into a real executable control instruction sequence, avoiding the generation of only virtual isolation ranges that cannot be implemented.

[0033] Furthermore, the controlled node set includes directly controlled nodes and indirectly controlled nodes when it is established. Directly controlled nodes are nodes whose states change after the control interface action, while indirectly controlled nodes are nodes affected by power supply relationships, spatial adjacency relationships, or control linkage relationships. When establishing the controlled node set, the directly controlled nodes are searched along the control linkage edge starting from the control interface node, and then the indirectly controlled nodes are expanded along the electrical power supply edge and spatial adjacency edge. The control source, control level, and control action type are recorded for each controlled node. When the control chain is established, the actions are arranged according to the dependency order of stopping energy output, blocking continuous power supply, and performing spatial isolation. If the precondition of a certain action is not met, the action will not enter the current stage instruction. The advantage of this embodiment is that the control actions can be consistent with the control range and execution order, reducing the overlap of actions or the lack of execution conditions.

[0034] In one embodiment, the comprehensive risk sequence is generated by three types of potential fields and a coupling correction term, calculated as follows: in, Indicates the first The overall risk value of each topology node. , , They represent the first The thermal risk potential field value, the electrical continuous power supply potential field value, and the spatial diffusion potential field value of each topological node. , , Let each represent the weight of the three types of potential fields and satisfy the following conditions: , This represents the coupling correction term generated jointly by the fire compartment to which the node belongs, the number of adjacent nodes, the power supply branch level, and the current isolation status. The weights of the coupling correction term, for example, a node. , , , ,Pick , , , At that time, the calculation yielded This value is used to sort risk nodes, adjacent nodes, and nodes to be observed.

[0035] In this embodiment, the coupling correction term is generated according to the fire zone to which the node belongs, the number of adjacent nodes, the power supply branch level, and the current isolation status. The fire zone is used to express whether the area where the node is located is in the same or adjacent zone as the abnormal node. The number of adjacent nodes is used to express the complexity of the spatial diffusion channel. The power supply branch level is used to express the length and break point of the power supply path between the node and the upstream power supply node. The current isolation status is used to express whether the node has been isolated by the control action of the previous stage. After generating the comprehensive risk sequence, the nodes are arranged into risk nodes, adjacent nodes, and nodes to be observed according to the comprehensive risk value. The sorting result is cross-matched with the isolation executability. Only nodes that have a high comprehensive risk value and are associated with an executable control interface are included in the risk priority control list. The advantage of this embodiment is that the risk sorting reflects the propagation risk, power supply risk, and controllable conditions at the same time, avoiding the selection of unexecutable boundary nodes only based on the size of the abnormal value.

[0036] Preferably, the formation process of candidate boundary combinations is carried out within a local sub-topology. The local sub-topology is centered on the abnormal node and includes the abnormal node, upstream power supply node, adjacent spatial nodes, and corresponding control interface nodes. When extracting the local sub-topology, the power supply branch node is searched upstream along the electrical power supply edge, the adjacent charging position and adjacent fire compartment are searched around along the spatial adjacency edge, and the charging stop, branch disconnection and compartment isolation interfaces are searched along the control linkage edge. Then, the charging termination action, power supply branch disconnection action and compartment isolation action are enumerated in the local sub-topology respectively. Action combinations that conflict with the control linkage sequence are deleted, and the correspondence between the boundary node set, the boundary topology edge set and the control command set is established for the remaining action combinations. The advantage of this embodiment is that the boundary search is limited to the local range where there is a power supply, spatial or control relationship with the abnormal node, which can reduce the entry of irrelevant nodes into the emergency isolation calculation.

[0037] Furthermore, when deleting conflicting action combinations, a control chain consistency check is used. If an action combination requires partition isolation to be performed first and then the charging session of the risk source to be stopped, and the partition isolation will cut off the control feedback path of the charging session, then the action combination is deleted. If an action combination requires branch disconnection but does not include the charging stop confirmation status of the abnormal charging session under the corresponding branch, then the action combination is marked as needing to supplement the charging stop pre-state. If an action combination contains repeated triggers of the same control interface, then the repeated triggers are merged into one stage action and the earliest execution time is retained. After the above checks, only action combinations that simultaneously cover the upstream power supply node and the adjacent spatial node are retained as candidate boundary combinations. The advantage of this embodiment is that the candidate boundary combinations satisfy the control order and interface executable relationship during the generation stage, reducing the occurrence of mutually exclusive actions in subsequent instruction sequences.

[0038] refer to Figure 4 In one embodiment, the selection of candidate boundary combinations is performed according to the constraints of remaining power connectivity, remaining risk propagation, and the number of isolated non-abnormal nodes. The selection calculation formula is as follows: in, Represents a candidate boundary combination. This represents the screening value for candidate boundary combinations. This represents the remaining power supply connectivity between the abnormal node and the upstream power supply node after performing candidate boundary combinations. This represents the remaining risk propagation amount between the anomalous node and its adjacent spatial nodes after performing candidate boundary combinations. This represents the normalized value indicating the number of non-abnormal nodes included in the isolation scope. , , This represents the weights of the corresponding constraints, all of which are non-negative, for example, candidate boundary combination A. , , ,Pick , , hour, During the screening process, the combination with the smaller screening value is selected from the candidate combinations that meet the upper limit of the remaining energy supply connectivity and the upper limit of the remaining risk propagation.

[0039] In this embodiment, the remaining power supply connectivity is calculated by the connectivity of electrical power supply edges in the multi-layer topology after executing candidate boundary combinations. If there is still an unbroken power supply edge between the abnormal node and the upstream power supply node, the corresponding connectivity contribution is retained. If the power supply edge is cut off by the branch disconnection action, the corresponding connectivity contribution is set to the disconnected state. The remaining risk propagation is calculated by the connectivity of spatial adjacent edges and thermal propagation edges after executing candidate boundary combinations. If there is still an unblocked thermal propagation path between the abnormal node and the adjacent spatial node, the corresponding propagation contribution is retained. The number of non-abnormal node isolations is obtained by statistical analysis of the controlled node set. During screening, candidate combinations that do not meet the requirements for remaining power supply connectivity are first excluded, and then candidate combinations that do not meet the requirements for remaining risk propagation are excluded. Subsequently, the number of non-abnormal node isolations is compared among the remaining candidate combinations. The advantage of this embodiment is that the boundary screening follows the technical logic of first blocking continuous power supply, then blocking spatial propagation, and then controlling non-risk areas.

[0040] In a preferred embodiment, the data records for the candidate boundary combination screening process can be configured according to Table 2, which describes the data fields required for each candidate boundary combination in the calculation, screening, instruction generation and feedback recalculation stages, so that the source and execution status of each fire isolation boundary can be traced during implementation.

[0041] Table 2 shows the set of boundary nodes, which indicates the specific topological nodes involved in the candidate boundary. Boundary node set Local subtopology Abnormal node, upstream power supply node, adjacent spatial node, control interface node Candidate boundary generation Write instruction object index Boundary topology edge set Multi-layer topology edge table Power supply cut-off edge, space blocked edge, control linkage edge Candidate boundary screening Calculate the connectivity residue control command set Control interface mapping Stop charging command, branch disconnect command, zone isolation command Instruction sequence generation Convert to stage instructions Return state set Execution feedback data Charging output feedback, branch status feedback, zone status feedback, fire monitoring feedback Closed-loop calibration Mark objects to be recalculated Constraint record set Filtering calculation results Remaining power supply connectivity, remaining risk propagation capacity, and number of isolated non-abnormal nodes. Candidate Combination Ranking Save filtering criteria In this embodiment, the boundary node set in Table 2 is used to indicate the specific topological nodes involved in the candidate boundary, the boundary topological edge set is used to indicate the relationship to be cut off or blocked by the candidate boundary, the control command set is used to convert the boundary into an executable control action, the feedback state set is used to record the state changes after the control action is executed, and the constraint record set is used to save the calculation basis in the candidate combination screening process. If the feedback state is inconsistent with the command expectation after execution, the mismatch node is located by the feedback state set, and the affected propagation edge and power supply edge are located by the boundary topological edge set. The advantage of this embodiment is that the candidate boundary combination has continuous data records from generation to recalculation, which can support closed-loop calibration.

[0042] Preferably, the control command sequence is obtained by converting the highest-ranked fire isolation boundary. During the conversion, the charging termination action, power supply branch disconnection action, and zone isolation action are converted into phased commands according to the control chain. Each phase command is written with the topology node index, control interface identifier, execution pre-state, and execution post-return state. If the same topology node belongs to multiple phase commands at the same time, the commands are merged in the order of first terminating energy output, then disconnecting the power supply branch, and finally executing zone isolation. For example, if an abnormal charging pile node is both in the charging stop phase and in the branch disconnection influence range, the charging stop phase command is generated first and the charging output decrease state is requested to be returned, and then the branch disconnection phase command is generated and the branch disconnection state is requested to be returned. The advantage of this embodiment is that the control commands are not issued in parallel, but rely on the topology relationship and the execution pre-state to form a phased control chain.

[0043] Furthermore, before issuing the stage instructions, node index consistency verification and interface executability verification are performed. The node index consistency verification is used to confirm that the charging pile nodes, power supply branch nodes, and fire compartment nodes in the instructions are still consistent with the object indexes in the current multi-layer topology. The interface executability verification is used to confirm that the stop charging interface, branch disconnection interface, and partition isolation interface are still in an executable state. If there are nodes in the topology that have been changed by the control actions of the previous stage, the execution pre-state in the next stage instructions is updated. If a control interface has returned to an unexecutable state before execution, the corresponding topology node and control linkage edge are marked as objects to be recalculated and the candidate boundary screening process is returned. The advantage of this embodiment is that the stage instructions can be updated with the executed actions, reducing the need to continue issuing instructions in the old state.

[0044] In one embodiment, the process of performing feedback update of the multi-layer topology is determined based on the feedback residual, calculated as follows: in, Indicates fire-resistant isolation boundary Post-execution feedback residuals Indicates fire-resistant isolation boundary The set of return nodes involved, Indicates the first The feedback weight of each return node. Indicates the first The status of each return node after execution. Indicates the first The expected feedback status of each feedback node, for example, a boundary involves three nodes: charging output feedback, branch disconnection feedback, and zone isolation feedback. Both are 1 / 3, if the 3 nodes If the values ​​are 0, 1, and 0 respectively, then... This indicates that there is an inconsistency between the branch break feedback and the expected state, and the associated nodes and topological edges need to be marked as objects to be recalculated.

[0045] In this embodiment, the feedback update receives the charging output feedback, branch disconnection feedback, zone isolation feedback, and fire monitoring feedback corresponding to each stage instruction, and compares the feedback status with the post-execution feedback status. When any feedback does not reach the post-execution feedback status of the corresponding stage instruction, the corresponding charging pile node, power supply branch node, fire zone node, and the connected power supply edge, spatial adjacency edge, and control linkage edge are marked as objects to be recalculated. The objects to be recalculated do not directly use the previous state in the next round of potential field calculation, but instead reread the current state and retain the influence of the executed control instructions on the topological connectivity. The advantage of this embodiment is that incomplete or unconfirmed control actions will not be mistakenly treated as completed boundaries and participate in subsequent calculations.

[0046] Furthermore, when re-screening fire isolation boundaries while retaining executed control instructions, the multi-layer topology first freezes the control edge changes that have been confirmed to be successfully executed. For example, the charging output edge corresponding to a successfully stopped charging session remains disconnected, the power supply edge corresponding to a successfully disconnected power supply branch remains disconnected, and the spatial adjacency edge corresponding to a successfully executed partition isolation remains blocked. For nodes and edges that have not been confirmed or have failed to execute, the thermal risk potential field, electrical continuous power supply potential field, and spatial diffusion potential field are recalculated, and a new risk priority control list and candidate boundary combinations are generated. If the new candidate boundary combinations overlap with the executed control instructions, only the uncovered control interfaces and boundary nodes are supplemented. The advantage of this embodiment is that the recalculation process can inherit the already formed isolation state and avoid repeating the already confirmed control actions.

[0047] In a preferred embodiment, when there is a delay in status feedback, a status queue within the acquisition cycle is used for processing. Each feedback node stores the status value, status time, and control instruction number within the most recent several acquisition cycles. If the time of a certain feedback status is earlier than the issuance time of the corresponding stage instruction, it is not used as a feedback status after execution, but is only retained as a historical status. If the time of a certain feedback status is later than the issuance time of the stage instruction and the object index is consistent, it is written into the current feedback status set. If the same node receives multiple feedback statuses within one acquisition cycle, the feedback corresponding to the current stage instruction is determined according to the control instruction number and time order. The advantage of this embodiment is that the feedback update can distinguish between old status, current status, and duplicate status, reducing erroneous recalculation caused by feedback delay.

[0048] In one embodiment, the multi-agent collaborative process employs a shared topology index and a controlled range locking mechanism. Before generating a candidate set, each agent reads the same version of the multi-layer topology, recording the topology version number and state version number during the reading. After generating the candidate set, each agent submits the control object index, control interface identifier, and controlled node set. When combining the candidate set, the overall control decision agent verifies whether the content submitted by each agent originates from the same topology version. If the versions are inconsistent, the agent is required to regenerate the candidate set based on the latest topology. If multiple agents simultaneously select the same control interface, the earlier action in the control chain is used as the main action, and other actions are incorporated into the controlled range record. The advantage of this embodiment is that local judgments in different control domains can be combined under the same data version, reducing boundary conflicts caused by different state versions.

[0049] Furthermore, the risk priority control list is reordered after each round of control. During the reordering, confirmed isolated nodes are not deleted. Instead, the power supply connectivity and spatial propagation status of confirmed isolated nodes are written into the current isolation status in the comprehensive risk sequence. Nodes that are not isolated and still have power supply connectivity or spatial propagation connectivity with abnormal nodes continue to participate in the reordering. If the thermal risk potential field of an adjacent node decreases but the electrical continuous power supply potential field remains connected, then the node remains in the range of nodes to be observed. If the electrical power supply of a node is disconnected but the spatial diffusion potential field is still connected to the abnormal node, then the node still enters the candidate partition isolation set. The advantage of this embodiment is that subsequent control will not ignore other risk paths due to the decrease of a single type of potential field.

[0050] In a preferred embodiment, the digital twin multi-layer topology can be updated in an event-driven manner. When an event occurs, such as a charging failure, output change, vehicle access anomaly, or fire monitoring status change, the event evidence unit is written to the node corresponding to the object index and triggers a local sub-topology update. The local sub-topology update only involves the abnormal node, adjacent spatial nodes, upstream power supply nodes, and control interface nodes, without needing to rebuild the complete site topology. If the event evidence unit shows that the abnormal node has recovered but the adjacent nodes still have residual risk propagation, the local sub-topology is maintained and the feedback residual confirmation is awaited. If the abnormal node recovers and both the residual power supply connection and the residual risk propagation meet the termination conditions, the boundary recalculation of the current local sub-topology is stopped. The advantage of this embodiment is that the topology update range is consistent with the risk association range, reducing the interference of irrelevant area status changes on the current fire isolation boundary calculation.

[0051] In this embodiment, the boundary termination condition is determined jointly by the current comprehensive risk sequence and the feedback residual, and the calculation formula is as follows: in, Indicates fire-resistant isolation boundary The closed-loop judgment quantity. Represents the set of return nodes The maximum value of the comprehensive risk value, Indicates fire-resistant isolation boundary Post-execution feedback residuals This represents the weight of the feedback residual in the closed-loop judgment. For example, if the maximum comprehensive risk value within a certain boundary backhaul node set is 0.28 and the feedback residual is 0.10, then... hour, If the closed-loop control condition requires the judgment quantity to not exceed the set termination boundary, then the boundary can enter the monitoring state; otherwise, recalculation and supplementary isolation will continue.

[0052] Preferably, no new isolation actions are added while maintaining the monitoring state, but the charging output feedback, branch status feedback, zone status feedback, and fire monitoring feedback are continued to be read. The topology edge changes caused by the executed control commands are retained in the multi-layer topology. If the comprehensive risk value of any node increases again or the feedback residual increases again during the monitoring period, the monitoring state is released and a new local sub-topology is extracted. If only the confirmed isolation node remains in the risk priority control list during the monitoring period and there is no remaining power supply connection or risk propagation, the corresponding fire isolation boundary is marked as a stable boundary and the current emergency control chain is terminated. The advantage of this embodiment is that the control process can switch between adding isolation and maintaining monitoring, avoiding the continued expansion of the isolation range when the boundary has stabilized.

[0053] In a preferred embodiment, when an abnormal node is located at the boundary of two fire zones, the spatial partitioning agent does not directly include both fire zones in the isolation range. Instead, it extracts the boundary topology edges connected to the abnormal node based on spatial adjacency edges and thermal propagation edges, and generates a candidate partition isolation set by combining the spatial diffusion potential field values ​​of adjacent charging positions. If the comprehensive risk value of adjacent nodes in one fire zone is low and there is no remaining power supply connection, then that zone only enters the range of nodes to be observed. If there is a high spatial diffusion potential field value in another fire zone and there is a direct adjacency edge with the abnormal node, then that zone enters the candidate partition isolation set. The advantage of this embodiment is that the isolation range in cross-partition abnormal scenarios is jointly limited by topology edges and potential fields, avoiding mechanical expansion of isolation according to partition number.

[0054] In a preferred embodiment, when there are multiple power supply paths between an abnormal node and an upstream power supply node, the power supply branch agent calculates the continuous energy input of each power supply path in the electrical power supply potential field, and records the branch disconnection interface corresponding to each path in the candidate branch disconnection set. If a branch disconnection action can cut off multiple power supply paths at the same time, the remaining power supply connection corresponding to the action is calculated according to the remaining path after the disconnection. If a branch disconnection action will cut off a large number of non-abnormal nodes but still cannot block the main power supply path of the abnormal node, the action is placed later in the candidate screening. The advantage of this embodiment is that in the scenario of multiple power supply paths, branch disconnection actions can be screened around the actual energy input path.

[0055] In a preferred embodiment, when the same control interface affects both abnormal and non-abnormal nodes simultaneously, the overall control decision agent cross-matches the set of controlled nodes of the control interface with the risk priority control list. If there is a heat propagation edge or spatial adjacency edge between the non-abnormal node and the abnormal node in the set of controlled nodes, it is treated as an adjacent risk node. If there is no heat propagation, power supply, or spatial adjacency relationship between the non-abnormal node and the abnormal node in the set of controlled nodes, it is included in the number of non-abnormal node isolations and participates in the candidate boundary combination screening. If there are multiple control interfaces that can cover abnormal nodes, the interface combination with the remaining power supply connectivity and risk propagation capacity that both meet the requirements and the number of non-abnormal node isolations is selected first. The advantage of this embodiment is that the control interface coverage can enter the boundary screening logic, reducing the number of irrelevant nodes included in the isolation range.

[0056] In a preferred embodiment, when multiple abnormal nodes report simultaneously at a charging site, the multi-layer topology generates an initial risk value for each abnormal node and extracts its respective local sub-topology. If there are overlapping power supply branch nodes, fire compartment nodes, or control interface nodes among the multiple local sub-topologies, the central control decision-making agent merges the overlapping nodes into a joint candidate boundary. During merging, the thermal risk potential field, electrical continuous power supply potential field, and spatial diffusion potential field corresponding to each abnormal node are retained, and a first-stage command is established for the common control interface. If the main propagation directions of multiple abnormal nodes are opposite and they share the same fire compartment boundary, the compartment isolation action is generated according to the boundary topology edge that can simultaneously block multiple main propagation paths. The advantage of this embodiment is that it can merge boundaries in the shared topology in multi-source abnormal scenarios, avoiding the repeated issuance of conflicting control commands to the same interface.

[0057] Furthermore, when screening the joint candidate boundaries of multiple anomalous nodes, the remaining power supply connectivity is accumulated according to the remaining connectivity between each anomalous node and its upstream power supply node, the remaining risk propagation is accumulated according to the remaining propagation between each anomalous node and its adjacent spatial nodes, and the number of isolated non-anomalous nodes is counted according to the merged set of controlled nodes. If a candidate combination only blocks the power supply path of one anomalous node while retaining the main power supply path of other anomalous nodes, it is excluded under the power supply connectivity constraint. If a candidate combination can cut off the main power supply path of all anomalous nodes but still retains the main heat propagation edge, it is excluded under the risk propagation constraint. The advantage of this embodiment is that multiple anomalous sources are included in the same topology evaluation, and the isolation boundary can cover the common risk path.

[0058] In a preferred embodiment, if the fire monitoring feedback shows that the risk location has shifted during the execution of the control command sequence, the digital twin multi-layer topology will keep the original abnormal node as a controlled risk node and write the new high-risk node into the risk priority control list. The executed control edges of the controlled risk node will remain disconnected. The new high-risk node will recalculate the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field. If the new risk node shares a power supply branch or fire compartment with the original fire isolation boundary, candidate boundary combinations will be added based on the original boundary. If the new risk node has no topological association with the original boundary, a new local sub-topology will be created and the control chain of the two boundaries will be maintained by the central control decision-making agent. The advantage of this embodiment is that it can retain the existing isolation results and supplement the control of the newly added risk path when the risk shifts.

[0059] In one embodiment, the aforementioned method can be executed by a computer program on a charging site control platform, edge control device, or data processing device with control authority. During program execution, the program reads the charging operation status, fault status, fire monitoring status, and control execution feedback, and maintains a multi-layer topology table, a state queue list, a candidate boundary table, an instruction sequence table, and a feedback residual table in the storage medium. The charging control agent, the power supply branch agent, the spatial partition agent, and the overall control decision agent can be implemented as parallel running control processes or control tasks running in a scheduled order. Each control task reads and writes the same data structure through a shared topology index. The output commands for stopping charging, branch disconnection, and partition isolation are issued through an executable control interface. The advantage of this embodiment is that the software implementation process can interface with existing control objects in the charging site, and an emergency fire prevention isolation closed loop can be formed without changing the physical structure of the charging site.

[0060] Preferably, the data table maintained by the computer program generates a new topology version after each event is triggered, and each candidate boundary generation, filtering, instruction issuance, and feedback update is mapped to the same version record. If a feedback update triggers recalculation, the new topology version inherits the control edge changes that have been successfully executed in the previous version, and at the same time rewrites the node state and edge state of the object to be recalculated. The candidate boundary table retains the reasons for the exclusion of the excluded combination, including insufficient remaining power supply connectivity, insufficient remaining risk propagation, conflicting control linkage sequence, and the number of isolated non-abnormal nodes being too late. The instruction sequence table retains the pre-execution state and post-execution feedback state of each stage of instruction. The advantage of this embodiment is that the control process can be reproduced through version records, which facilitates the determination of the calculation basis and feedback basis for each isolation boundary.

[0061] In this embodiment, the working principle of the method is to convert alarm objects, power supply objects, spatial objects, and control objects in the charging site into the same computing structure through a unified digital twin multi-layer topology. Then, three types of potential fields describe thermal risk propagation, continuous electrical power supply, and spatial adjacency diffusion, respectively. Subsequently, multiple agents generate candidate actions for the control domain under the same topology version, and the overall control decision agent combines the candidate actions into fire protection isolation boundaries. After the fire protection isolation boundaries are filtered by the remaining amount of power supply connectivity, the remaining amount of risk propagation, and the number of isolations of non-abnormal nodes, they are converted into staged control commands. The execution feedback then updates the topology nodes and topology edges in reverse. The advantage of this embodiment is that the control input, boundary calculation, command output, and feedback recalculation form a continuous closed loop, which can provide an implementable technical solution to the problem that fixed plans in the background technology cannot simultaneously cover energy paths and spatial propagation paths.

Claims

1. A digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites, characterized in that, include: A multi-layer digital twin topology for charging sites is constructed based on charging locations, charging piles, power supply branches, vehicle access status, fire protection zones, and executable control interfaces. Acquire charging operation status, fault status, and fire monitoring status and map them to the multi-layer topology; Based on the heat propagation relationship, the electrical power supply relationship, and the spatial adjacency relationship, a thermal risk potential field, an electrical continuous power supply potential field, and a spatial diffusion potential field are generated, respectively. Fireproof isolation boundaries are generated collaboratively by charging control agents, power supply branch agents, spatial partitioning agents, and overall control decision-making agents. Based on the fireproof isolation boundary, output a sequence of control commands to stop charging, disconnect branches, and isolate zones, and update the multi-layer topology based on execution feedback.

2. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 1, characterized in that, Constructing the digital twin multi-layer topology of the charging site includes: Set the charging position, charging pile, power supply branch, vehicle access status, fire zone and executable control interface as topology nodes respectively; Spatial adjacency, electrical power supply, thermal risk propagation, and control linkage are set as topological edges with different attributes; Configure the operating status, anomaly level, energy input intensity, isolation executability, and fire compartment identifier of the topology node; Configure the propagation direction, power supply direction, adjacency strength and control linkage sequence for the topological edge so that the same risk node has a consistent node index across the spatial layer, electrical layer and control layer.

3. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 2, characterized in that, Generating the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field includes: Based on the fault status, temperature rise status, charging output changes, and fire monitoring status, initial risk values ​​are assigned to abnormal nodes. Calculate the amount of thermal risk transfer between adjacent nodes along the aforementioned thermal risk propagation relationship; Calculate the continuous energy input of the power supply branch to the abnormal node and adjacent nodes along the electrical power supply relationship; Calculate the amount of risk diffusion between adjacent charging positions and fire compartments along the spatial adjacency relationship; The three types of potential fields are superimposed and calibrated according to the same node index to obtain a node comprehensive risk sequence for determining the fire isolation boundary.

4. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 3, characterized in that, The multi-agent collaborative generation of the fire-resistant isolation boundary includes: The charging control agent generates a target charging session termination set based on the charging output status and vehicle access status. The power supply branch intelligent agent generates a candidate branch disconnection set based on the continuous electrical power supply potential field. The spatial partitioning agent generates a candidate partition isolation set based on the thermal risk potential field and the spatial diffusion potential field. The overall control decision-making agent projects the target charging session termination set, candidate branch disconnection set, and candidate partition isolation set onto the multi-layer topology to form a candidate boundary combination covering the energy supply path of the risk source and the main propagation path of the thermal risk.

5. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 4, characterized in that, Configuring isolation executability for the topology nodes includes: Obtain the stop charging interface, branch disconnection interface, and partition isolation interface corresponding to the executable control interface, and establish a bidirectional mapping between the interface control object and the topology node; When a control interface affects multiple topology nodes simultaneously, a set of controlled nodes is established according to the electrical power supply relationship and control linkage relationship. When a topology node is affected by multiple control interfaces, a control chain is established according to the sequential dependency of the control actions. The controlled node set and control chain are written into the multi-layer topology so that candidate boundary combinations can be mapped to executable control instruction sequences.

6. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 3, characterized in that, The resulting node comprehensive risk sequence includes: Record the thermal risk transfer, continuous energy input, and spatial diffusion for the same topological node; Generate a coupling correction term based on the fire zone to which the node belongs, the number of adjacent nodes, the power supply branch level, and the current isolation status; The coupling correction term is applied to the nodal values ​​of the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field; The risk nodes, adjacent nodes, and nodes to be observed are sorted according to the corrected node values, and the sorting results are cross-matched with the isolation executability to generate a risk priority control list for boundary screening.

7. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 5, characterized in that, Forming the candidate boundary combination includes: Extract a local sub-topology from the multi-layer topology, which includes abnormal nodes, upstream power supply nodes, adjacent spatial nodes, and corresponding control interface nodes; Within the local sub-topology, the charging termination action, the power supply branch disconnection action, and the partition isolation action are enumerated respectively. Delete the action combination that conflicts with the control linkage sequence; Establish a correspondence between the set of boundary nodes, the set of boundary topology edges, and the set of control commands for the remaining action combinations; The action combination that simultaneously covers the upstream power supply node and the adjacent spatial node is determined as the candidate boundary combination.

8. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 7, characterized in that, Filtering the candidate boundary combinations includes: Based on the topological connectivity changes after executing each candidate boundary combination, calculate the remaining power supply connectivity between the abnormal node and the upstream power supply node. Based on the spatial adjacency changes after executing each candidate boundary combination, calculate the remaining risk propagation amount between the abnormal node and its adjacent spatial nodes. The number of non-abnormal nodes included in the isolation range is counted based on the set of controlled nodes; Based on the constraints of remaining power supply connectivity, remaining risk propagation capacity, and the number of isolated non-abnormal nodes, the candidate boundary combinations are progressively screened, and the highest-ranked fireproof isolation boundary is output.

9. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 8, characterized in that, The output of the control command sequence includes: The charging termination action, power supply branch disconnection action, and zone isolation action in the highest-ranked fire isolation boundary are converted into phased instructions according to the control chain. Each stage instruction includes the corresponding topology node index, control interface identifier, pre-execution state, and post-execution return state. When the same topology node belongs to multiple stage instructions at the same time, the instructions are merged in the order of first terminating energy output, then disconnecting power supply branches, and finally performing partition isolation. The merged stage instructions will be used as the control instruction sequence for stopping charging, branch disconnection, and zone isolation.

10. The digital twin multi-agent collaborative emergency decision-making and fire prevention isolation method for charging sites according to claim 9, characterized in that, Updating the multi-layer topology based on execution feedback includes: Receive charging output feedback, branch disconnection feedback, zone isolation feedback and fire monitoring feedback corresponding to each stage of instruction; Compare the consistency between the returned data and the returned data after execution; When any feedback fails to reach the post-execution feedback state of the corresponding stage instruction, the associated topology node and topology edge are marked as objects to be recalculated. Based on the object to be recalculated, the thermal risk potential field, the electrical continuous power supply potential field, and the spatial diffusion potential field are regenerated, and the fire isolation boundary is re-selected while retaining the executed control instructions.

Images