Industrial emergency rescue decision system based on multi-agent cooperative control

Abstract

The application discloses an industrial emergency rescue decision system based on multi-agent collaborative control, comprising: a data acquisition module for acquiring rescue sensing data and environmental constraint data; an association generation module for generating an event association graph and an agent state sequence; a witness construction module for generating a collaborative witness sequence; a domain mapping module for generating a current control state, an alternative control state, a closure bearing domain, a conflict transition path, a fact frontier domain and an anti-fact frontier domain; an algorithm construction module for constructing an improved Dubois-Prade algorithm; an evidence generation module for generating an evidence focus element and a basic probability assignment; a conflict fusion module for generating a fact domain fusion state and an anti-fact domain fusion state; and a decision output module for generating a global situation result, a collaborative control instruction and an emergency rescue decision result. The application improves the conflict evidence utilization rate and the emergency rescue decision reliability.

Description

Technical Field

[0001] This invention relates to the field of industrial emergency rescue technology, and in particular to an industrial emergency rescue decision-making system based on multi-agent collaborative control. Background Technology

[0002] With the increasing demand for handling complex accidents in industrial settings and the growing applications of multi-agent collaborative rescue, integrated technologies for accident situation awareness, evidence fusion analysis, and collaborative control decision-making have received widespread attention. Existing industrial emergency rescue systems primarily rely on early warning from a single sensing source, fixed-rule scheduling, or traditional evidence fusion methods for risk assessment and task allocation. However, these systems commonly suffer from the following problems in practical applications: The collected rescue perception data and environmental constraint data are scattered and lack unified spatiotemporal benchmarks. Existing data integration methods often fail to achieve accurate correspondence between dangerous events, regional states, and agent states, resulting in incomplete correlations of accident situations and discontinuous expressions of agent collaborative states. When faced with conflicting judgments by multiple agents regarding the same dangerous event, traditional fusion methods typically rely solely on static focal element relationships to perform conflict allocation. This makes it difficult to structurally decompose conflict quality by incorporating event propagation paths, regional connectivity structures, and collaborative witness relationships, leading to insufficient interpretability of fusion results and poor consistency in spatial constraints. In the control decision-making stage, existing methods generally lack simultaneous characterization of the current control state and alternative control states, making it difficult to establish a stable correspondence between the factual and counterfactual situations. This results in insufficient adaptability of collaborative control commands to changes in the global situation.

[0003] Therefore, how to provide an industrial emergency rescue decision-making system based on multi-agent collaborative control is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose an industrial emergency rescue decision-making system based on multi-agent collaborative control. This invention employs an improved Dubois–Prade algorithm, constructing an event association graph, collaborative witness sequence, closure carrying domain, conflict transfer path, fact front domain, and counterfactual front domain around rescue perception data and environmental constraint data. It performs conflict quality calculation, structured decomposition, intra-domain projection, and dual-domain coupling solution on evidence focal elements and basic probability assignments, realizing global situational assessment and collaborative control decision-making in industrial accident scenarios. It has the advantages of strong correlation, high conflict handling accuracy, and high decision reliability.

[0005] An industrial emergency rescue decision-making system based on multi-agent cooperative control according to an embodiment of the present invention includes: The data acquisition module is used to collect rescue perception data and environmental constraint data; The association generation module is used to perform time alignment, spatial registration, and event association on rescue perception data and environmental constraint data, and generate event association graphs and agent state sequences. The witness building module is used to perform chain-like connection of confirmation records in the agent's state sequence to generate a collaborative witness sequence; The domain mapping module is used to form the current control state and alternative control state based on the agent's state sequence, map the event association graph into the closure carrying domain, map the collaborative witness sequence into the conflict transfer path, map the current control state into the fact front domain, and map the alternative control state into the counterfactual front domain. The algorithm building module is used to construct the improved Dubois–Prade algorithm based on the closure bearing domain, conflict transition path, fact front domain, and counterfactual front domain; The evidence generation module is used to perform local situation assessment on rescue perception data and generate evidence focal elements and basic probability assignments. The conflict fusion module is used to perform conflict quality calculation, structured decomposition and intra-domain projection on evidence focal elements and basic probability assignments using the improved Dubois–Prade algorithm to generate fact domain fusion state and antifact domain fusion state. The decision output module is used to perform dual-domain coupled solution on the fact domain fusion state and the antifactual domain fusion state, and generate global situation results, collaborative control commands and emergency rescue decision results.

[0006] Optionally, the data acquisition module specifically comprises: The system collects status records, location records, task records, and observation records of the rescue unit during the accident handling process, and extracts agent identifiers, collection times, spatial coordinates, task identifiers, hazardous event identifiers, and observation values ​​to form a rescue perception record set. Collect regional topology records, obstacle distribution records, hazard source records, diffusion boundary records, and traffic constraint records in the accident area, and extract regional identifiers, connection relationship identifiers, boundary coordinates, hazard level values, diffusion direction values, diffusion range values, and constraint type identifiers to form an environmental constraint record set; Data aggregation is performed on the rescue perception record set and the environmental constraint record set to generate rescue perception data and environmental constraint data.

[0007] Optionally, the association generation module specifically comprises: The rescue perception data is arranged in the order of collection time, and the arrangement results are mapped to a unified time base to form a time-aligned sequence. Read spatial coordinates from the time-corrected sequence and match the spatial coordinates with the region identifiers, boundary coordinates, and connection relationship identifiers in the environmental constraint data to form a spatial registration sequence. Read the hazard event identifier from the spatial registration sequence, and associate the hazard event identifier with the hazard level value, diffusion direction value, diffusion range value and constraint type identifier in the environmental constraint data to form an event association record set; The region identifier, connection relationship identifier, dangerous event identifier, and agent identifier in the event association record set are written into the graph structure to form an event association graph. The agent identifiers in the event association record set are arranged according to the collection time to form an agent state sequence.

[0008] Optionally, the witness construction module specifically comprises: The confirmation records are extracted from the agent's state sequence and arranged in the order of execution according to the acquisition time to generate a confirmation arrangement sequence. Read the danger event identifier and agent identifier from the confirmation sequence, perform adjacent pairing on the confirmation records with consistent danger event identifiers and progressive collection times, and generate a confirmation connection record set; Based on the adjacent pairing relationship in the confirmed connection record set, the confirmed records are chained together, and the continuously connected confirmed records are merged into witness chain records to generate a witness chain record set; The witness chain record set is arranged in the order of execution based on the time of collection to generate a collaborative witness sequence.

[0009] Optionally, the domain mapping module specifically comprises: The agent state sequence is arranged in a dual order according to the acquisition time and agent identifier. The spatial coordinates, task identifiers and danger event identifiers corresponding to the same agent identifier are arranged according to the progressive relationship of the acquisition time to form the control state arrangement result. The task identifiers in the control state arrangement result are kept in their original configuration to generate the current control state. In the current control state, the agent identifier, data acquisition time, spatial coordinates, and danger event identifier are kept consistent. An alternative configuration orchestration is performed on the task identifier to generate an alternative control state. Perform connectivity unpacking on the region identifiers and connection relationship identifiers in the event association graph, and configure the hazard event identifiers to the corresponding region positions in the connectivity unpacking results, so that the regions with connectivity relationships and corresponding to the same hazard event form a closed association structure, and generate a closed carrying domain. The witness chain records in the collaborative witness sequence are arranged in the order of execution at the time of collection, and the link is connected according to the connection relationship between adjacent witness chain records, so that the witness chain records corresponding to the same dangerous event form a continuous transmission structure and generate a conflict transfer path; The current control state is projected onto the set of front positions according to the acquisition time and spatial coordinates to generate the factual front domain, and the alternative control state is projected onto the set of front positions according to the acquisition time and spatial coordinates to generate the counterfactual front domain.

[0010] Optionally, the algorithm construction module specifically comprises: The closed associated ranges in the closure bearing domain are segmented and organized according to their regional locations, and the regional locations and connectivity relationships corresponding to each closed associated range are arranged accordingly to form closure bearing units; The transmission links in the conflict transfer path are unfolded according to the execution order of the acquisition time, and the start position, end position and connection order of each transmission link are arranged accordingly to form a conflict transfer unit. Synchronize and align the fact frontier domain and the counterfactual frontier domain according to the acquisition time, and arrange the fact frontier position and the counterfactual frontier position corresponding to the same acquisition time to form a frontier corresponding unit; The closure-bearing unit is used as the conflict quality bearing range, the conflict transfer unit is used as the conflict quality decomposition path, and the front-end corresponding unit is used as the corresponding structure between the fact domain fusion state and the antifactual domain fusion state, thus forming the bearing structure, transfer structure and coupling structure of the improved Dubois–Prade algorithm. By mapping the evidence focal elements and basic probability assignments into the bearing structure, transfer structure, and coupling structure, the conflict quality between evidence focal elements is calculated and structurally decomposed along the conflict transfer unit, projected into the domain along the closure bearing unit, and coupled with the fact domain fusion state and antifactual domain fusion state along the front corresponding unit, thus forming the improved Dubois–Prade algorithm.

[0011] Optionally, the evidence generation module specifically comprises: The rescue perception data is arranged in the order of collection time, and then collected according to the intelligent agent identifier and the dangerous event identifier to generate a local perception record set; Spatial coordinates, task identifiers, and observation values ​​are extracted from the local sensing record set, and the spatial coordinates, task identifiers, and observation values ​​are associated and arranged according to the dangerous event identifiers to generate a local situation determination unit; The observed values ​​in the local situation assessment unit are continuously compared according to the acquisition time, and local situation assessment is performed in combination with spatial coordinates and task identifiers to generate local situation assessment results; The local situation assessment results are mapped to focal elements according to the dangerous event identifiers to generate evidence focal elements. The observation values ​​and confirmation strengths corresponding to the local situation assessment results are then configured to the evidence focal elements to generate basic probability assignments.

[0012] Optionally, the conflict fusion module specifically comprises: The evidence focal elements and basic probability assignments are mapped into the bearing structure, transfer structure and coupling structure of the improved Dubois–Prade algorithm according to the dangerous event identifier and the collection time to generate conflict calculation units. In the conflict calculation unit, the evidence focal elements with consistent dangerous event identifiers but inconsistent focal element values ​​are compared, and the conflict quality between evidence focal elements is determined based on the basic probability assignment, generating a conflict quality record set. The conflict quality record set is expanded along the conflict transfer units in the transfer structure in the order of penetration, and the conflict quality is decomposed in a structured manner according to the start and end positions to generate a conflict decomposition record set. The conflict decomposition record set is projected into the domain according to the closure bearer unit in the bearer structure, and a dual-domain assignment is performed according to the front-end corresponding unit in the coupled structure to generate the fact domain fusion state and the antifact domain fusion state.

[0013] Optionally, the decision output module specifically comprises: Based on the hazardous event identifier and the acquisition time, the fact domain fusion state and antifactual domain fusion state are matched accordingly to generate a dual-domain coupling unit; In the dual-domain coupling unit, the state difference is calculated between the fused state of the fact domain and the fused state of the antifact domain, and the consistency judgment is performed on the fact front position and the antifact front position corresponding to the state difference, generating a coupling judgment record set; Based on the coupling judgment record set, joint merging is performed on the leading edge position, fusion strength and control order corresponding to the dangerous event identifier to generate global situation results; The leading position, fusion strength, and control order in the global situation results are mapped to the agent identifier and task identifier in the agent state sequence to generate collaborative control instructions, and emergency rescue decision results are generated based on the collaborative control instructions and the global situation results.

[0014] The beneficial effects of this invention are: This invention performs unified collection, temporal realignment, spatial registration, and event correlation on rescue perception data and environmental constraint data to construct an event correlation graph and agent state sequence. Furthermore, it forms a collaborative witness sequence, a closure-bearing domain, a conflict transfer path, a fact front domain, and a counterfactual front domain. This enables the continuous expression and correlated organization of previously dispersed perception, environmental, and control information during industrial accident handling within the same technical framework. Compared to existing technologies that suffer from dispersed data sources, inconsistent spatiotemporal benchmarks, and difficulty in stably corresponding hazardous events with agent states, this invention improves the completeness and consistency of accident scene representation. It allows for the synchronous characterization of hazardous event propagation relationships, regional connectivity relationships, and agent collaborative relationships, thereby providing a stable data foundation for subsequent evidence fusion and decision output.

[0015] This invention, based on an improved Dubois–Prade algorithm, performs conflict quality calculation, structured decomposition, intra-domain projection, and dual-domain coupling solution on the evidence focal elements and basic probability assignments. This allows conflict evidence to move beyond simple static combinations and instead be constrainedly allocated within the closure-bearing domain along conflict transfer paths, completing corresponding solutions between fact-domain fusion states and antifactual-domain fusion states. Compared to existing traditional fusion methods that struggle to effectively process conflict information by considering regional connectivity structures, collaborative witness relationships, and control state differences, this invention improves the interpretability of conflict evidence, the accuracy of fusion results, and the matching degree between collaborative control commands and accident situations, thereby enhancing the reliability and effectiveness of industrial emergency rescue decision-making. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of the industrial emergency rescue decision-making system based on multi-agent cooperative control proposed in this invention; Figure 2 This is a diagram illustrating the construction of the improved Dubois–Prade algorithm for the industrial emergency rescue decision-making system based on multi-agent cooperative control proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2 An industrial emergency rescue decision-making system based on multi-agent collaborative control includes: The data acquisition module is used to collect rescue perception data and environmental constraint data; The association generation module is used to perform time alignment, spatial registration, and event association on rescue perception data and environmental constraint data, and generate event association graphs and agent state sequences. The witness building module is used to perform chain-like connection of confirmation records in the agent's state sequence to generate a collaborative witness sequence; The domain mapping module is used to form the current control state and alternative control state based on the agent's state sequence, map the event association graph into the closure carrying domain, map the collaborative witness sequence into the conflict transfer path, map the current control state into the fact front domain, and map the alternative control state into the counterfactual front domain. The algorithm building module is used to construct the improved Dubois–Prade algorithm based on the closure bearing domain, conflict transition path, fact front domain, and counterfactual front domain; The evidence generation module is used to perform local situation assessment on rescue perception data and generate evidence focal elements and basic probability assignments. The conflict fusion module is used to perform conflict quality calculation, structured decomposition and intra-domain projection on evidence focal elements and basic probability assignments using the improved Dubois–Prade algorithm to generate fact domain fusion state and antifact domain fusion state. The decision output module is used to perform dual-domain coupled solution on the fact domain fusion state and the antifactual domain fusion state, and generate global situation results, collaborative control commands and emergency rescue decision results.

[0019] In this embodiment, the data acquisition module specifically comprises: Using rescue units as the data collection objects, the status records, location records, task records and observation records generated during the accident handling process are read according to the collection time. Records belonging to the same intelligent agent identifier are collected, and then spatial coordinates, task identifiers, danger event identifiers and observation values ​​are extracted from the collected records. The intelligent agent identifier and collection time are written into the corresponding record to form a rescue perception record set. Using the accident area as the data collection object, the regional topology records, obstacle distribution records, hazard source records, diffusion boundary records, and access constraint records are read according to the collection time. Records belonging to the same area identifier are grouped together, and then connection relationship identifiers, boundary coordinates, hazard level values, diffusion direction values, diffusion range values, and constraint type identifiers are extracted from the grouped records to form an environmental constraint record set. The rescue perception record set and the environmental constraint record set are time-aligned according to the time of collection, spatially aggregated according to the correspondence between spatial coordinates and area identifiers, and event-paired according to the correspondence between hazard event identifiers and hazard level values, diffusion direction values, and diffusion range values. The records that have completed time alignment, spatial aggregation, and event pairing are merged into a unified data structure to form rescue perception data and environmental constraint data.

[0020] In this embodiment, the association generation module specifically includes: The rescue perception data is arranged in the order of collection time, and the records corresponding to the same collection time are written to the same time position. Then, each time position is mapped to the corresponding position in the unified time base, so that the rescue perception data forms a continuous arrangement result in the unified time base, forming a time-aligned sequence. In the time-corrected sequence, the spatial coordinates corresponding to each time position are read one by one, and the spatial coordinates are matched with the region identifier corresponding to the region range, the boundary coordinate corresponding to the boundary position, and the connection relationship identifier corresponding to the connection position in the environmental constraint data. This makes each spatial coordinate obtain the corresponding region position and connection position in the environmental constraint data, forming a spatial registration sequence. In the spatial registration sequence, the hazard event identifiers corresponding to each time location are read one by one, and the hazard event identifiers are associated with the hazard level value, diffusion direction value, diffusion range value and constraint type identifier in the environmental constraint data according to the same time location and the same area location, so that the hazard event identifiers are associated with the hazard level value, diffusion direction value, diffusion range value and constraint type identifier, forming an event association record set; The region identifier, connection relationship identifier, dangerous event identifier, and agent identifier in the event association record set are written into the nodes and connecting edges of the graph structure according to the association relationship to form an event association graph. The agent identifiers in the event association record set are arranged in order of time position corresponding to the collection time to form an agent state sequence.

[0021] In this embodiment, the witness construction module specifically includes: Read each state item with the acknowledgment attribute from the agent's state sequence as an acknowledgment record, and write them into the same arrangement structure from front to back according to the collection time, so that each acknowledgment record corresponds to a time position in the arrangement structure, forming an acknowledgment arrangement sequence; In the confirmation sequence, the danger event identifier and agent identifier corresponding to each time position are read one by one. The previous confirmation record and the next confirmation record are established to maintain the same danger event identifier and the collection time is in a progressive relationship. The pairing start position, pairing end position, danger event identifier and agent identifier are written into the same recording unit to form a confirmation connection record set. Based on the adjacent pairing relationships in the confirmation connection record set, the confirmation records with continuous time connection relationships are sequentially linked to ensure that the pairing termination position of the previous confirmation record is consistent with the pairing start position of the next confirmation record. The confirmation records that have completed continuous linking are merged into witness chain records, and each witness chain record is written into the witness chain record set. The witness chain records in the witness chain record set are arranged in the execution order according to the starting time position corresponding to the collection time, so that the witness chain records are arranged sequentially along the time direction to generate a collaborative witness sequence.

[0022] In this embodiment, the domain mapping module specifically comprises: The agent state sequence is first arranged from front to back according to the acquisition time, and then merged according to the agent identifier, so that the state items corresponding to the same agent identifier form a continuous unfolding result along the time direction. Spatial coordinates, task identifiers and hazard event identifiers are extracted from the continuous unfolding result, and the spatial coordinates, task identifiers and hazard event identifiers are matched item by item according to the progressive relationship of acquisition time to obtain the control state arrangement result. The original configuration of task identifiers is maintained in the control state arrangement result to generate the current control state. By arranging the time first and then merging the identifiers, the current control state can maintain the temporal continuity and task continuity of the same agent in the accident handling process, thereby providing a continuous state basis for the formation of the fact front domain. In the current control state, the agent identifier, acquisition time, spatial coordinates, task identifier, and danger event identifier corresponding to each time position are read one by one. The agent identifier, acquisition time, spatial coordinates, and danger event identifier are kept consistent. Only the task identifier is replaced with alternative configuration orchestration, so that the task configuration at the same time position is changed from the current task configuration to the alternative task configuration, generating an alternative control state. By replacing only the task identifier while keeping the spatial coordinates consistent with the danger event identifier, the alternative control state and the current control state maintain a unified spatiotemporal reference, which facilitates the corresponding comparison between the fact frontier domain and the counterfactual frontier domain. The region identifiers and connection relationship identifiers in the event association graph are expanded along the connection direction. First, the direct connection relationship between region identifiers is determined based on the connection relationship identifiers. Then, continuous connected segments are determined based on the direct connection relationship. Dangerous event identifiers are configured to the corresponding region positions in the continuous connected segments, so that the region positions corresponding to the same dangerous event within the continuous connected segments form a closed association range, generating a closure carrying domain. By restricting dangerous event identifiers to the continuous connected segments for aggregation, the closure carrying domain can maintain the consistency between the event propagation range and the region connection range, thereby improving the concentration of conflict quality projection positions. The witness chain records in the collaborative witness sequence are arranged in the order of execution of the collection time. The connection relationship between adjacent witness chain records is then compared. The termination time position of the previous witness chain record is connected with the start time position of the next witness chain record. The witness chain records corresponding to the same dangerous event are continuously connected according to the connection determination result to form a transmission link extending in the time direction and generate a conflict transfer path. By performing connection determination and continuous connection on the witness chain records, the conflict transfer path can reflect the continuation relationship of the confirmation records in the time direction, thereby providing link constraints for the directional transfer after the conflict quality decomposition. The spatial coordinates in the current control state are mapped to the frontier position set according to the acquisition time, and the spatial coordinates corresponding to the same acquisition time are arranged continuously according to the time position to form the fact frontier domain. Similarly, the spatial coordinates in the alternative control state are mapped to the frontier position set according to the acquisition time, and the spatial coordinates corresponding to the same acquisition time are arranged continuously according to the time position to form the antifactual frontier domain. By mapping the current control state and the alternative control state using the same frontier position set, the fact frontier domain and the antifactual frontier domain maintain a unified correspondence in both time and space, which facilitates the dual-domain coupling solution of the fact domain fusion state and the antifactual domain fusion state in the improved Dubois–Prade algorithm.

[0023] In this embodiment, the algorithm construction module is specifically as follows: Each closed associated range in the closure carrying domain is identified segment by segment according to its regional location. First, the regional location covered by each closed associated range is determined. Then, the connectivity relationship between each regional location and its adjacent regional locations is determined. The regional locations within the same closed associated range are arranged into the same arrangement unit according to the connectivity relationship, so that each arrangement unit corresponds to a set of regional locations with clear boundaries and continuous connectivity, thus forming a closure carrying unit. Each transmission link in the conflict transfer path is unfolded sequentially according to the acquisition time. The start position, end position and connection sequence of each transmission link are determined in turn. The connection directions between the start position and the end position are arranged continuously according to the connection sequence, so that each transmission link forms a traceable connection sequence in the time direction, thus forming a conflict transfer unit. The fact frontier domain and the counterfactual frontier domain are synchronized and aligned according to the acquisition time. The fact frontier position and the counterfactual frontier position corresponding to the same acquisition time are matched one-to-one. The position matching results corresponding to each acquisition time are arranged according to the acquisition time, so that the fact frontier domain and the counterfactual frontier domain form a stable correspondence under the same time reference, forming a frontier correspondence unit. The closure bearing unit is used to limit the spatial bearing position of conflict quality, the conflict transfer unit is used to limit the transmission direction and decomposition order of conflict quality, and the front correspondence unit is used to limit the coupling position of the fact domain fusion state and the antifactual domain fusion state, so that spatial bearing, path transfer and dual domain correspondence are mutually constrained in the same operation relationship, forming bearing structure, transfer structure and coupling structure. The evidence focal elements and basic probability assignments are configured into the bearing structure, transfer structure, and coupling structure. The conflict quality calculation position between evidence focal elements is determined according to the transfer structure. The structured decomposition position of conflict quality is determined according to the connection order of conflict transfer units. The intradomain projection position of conflict quality is determined according to the closure bearing unit. The coupling operation position of fact domain fusion state and antifactual domain fusion state is determined according to the front-end corresponding unit. This ensures that conflict quality calculation, structured decomposition, intradomain projection, and coupling operation are continuously associated, forming the improved Dubois–Prade algorithm.

[0024] This invention addresses the limitations of the traditional Dubois-Prade algorithm in industrial emergency rescue scenarios. It can only perform static union allocation of conflicting evidence, failing to reflect spatial connectivity constraints, temporal transit relationships, and differences in control strategies. The invention restructures and improves the algorithm for multi-agent collaborative control. Specifically, it maps the event association graph into the closure-bearing domain, ensuring that conflict quality is no longer confined to abstract propositional spatial allocation but is constrained by the connectivity of the accident area and the association of dangerous events, and limited to a bearable closed association range. It also maps the collaborative witness sequence into the conflict transfer path, transforming the original centralized calculation of evidence conflict into a structured decomposition process unfolding along the confirmation record transmission link, thus reflecting the temporal sequence and transfer direction of conflict sources. Finally, it maps the current control state and the alternative control state into the fact frontier. The invention incorporates a domain and a counterfactual frontier domain, enabling the algorithm to simultaneously incorporate the frontier position relationship between the actual scheduling state and the alternative scheduling state during the fusion phase. This expands the single-domain evidence combination into a dual-domain coupled solution. Simultaneously, the closure-bearing unit, conflict transfer unit, and frontier-corresponding unit are organized into a bearing structure, transfer structure, and coupling structure. This ensures that conflict quality calculation, structured decomposition, intra-domain projection, and dual-domain coupling are continuously correlated within the same computational framework. Through these improvements, the invention enhances the interpretability and constrainability of high-conflict evidence in complex accident scenarios, avoids the situational divergence caused by excessively broad conflict quality allocation in traditional algorithms, strengthens the consistency between evidence fusion results and regional connectivity, event propagation, and control states, and improves the accuracy of global situational results, the pertinence of collaborative control commands, and the reliability of emergency rescue decision-making results.

[0025] In this embodiment, the evidence generation module specifically comprises: The rescue perception data is sorted by time according to the collection time, and records are merged according to the agent identifier at each collection time. Then, the records corresponding to the agent identifier are classified and organized according to the danger event identifier, so that the records corresponding to the same collection time, the same agent identifier, and the same danger event identifier form a unified recording unit and generate a local perception record set. Spatial coordinates, task identifiers, and observation values ​​are extracted one by one from the local sensing record set. Correspondence is established around the dangerous event identifier, so that the spatial coordinates represent the location of the dangerous event, the task identifier represents the corresponding handling task of the dangerous event, and the observation value represents the corresponding observation status of the dangerous event. Then, the spatial coordinates, task identifiers, and observation values ​​with the same dangerous event identifier are grouped into the same judgment unit to generate a local situation judgment unit. The observed values ​​in the local situation assessment unit are continuously compared according to the acquisition time to determine the change relationship between adjacent acquisition times. The change relationship is then combined with the positional distribution relationship corresponding to the spatial coordinates and the task status relationship corresponding to the task identifier to form the status assessment result of the dangerous event in the local area and generate the local situation assessment result. The local situation assessment results are mapped to focal elements according to the dangerous event identifiers, so that each dangerous event identifier corresponds to an evidence focal element. The observation values ​​and confirmation strengths corresponding to the local situation assessment results are configured to the evidence focal elements, so that the observation values ​​represent the degree of evidence support and the confirmation strengths represent the degree of evidence confirmation, thus generating a basic probability assignment.

[0026] In this embodiment, the conflict fusion module specifically comprises: Evidence focal elements and basic probability assignments are matched in a dual manner according to the danger event identifier and the collection time. First, evidence focal elements are collected within the same danger event identifier range. Then, basic probability assignments are paired at the same collection time position. The results of collection and pairing are respectively configured to the corresponding positions of the bearing structure, transfer structure and coupling structure, so that evidence focal elements and basic probability assignments under the same danger event identifier are established in the same operation framework to generate conflict calculation units. In the conflict calculation unit, evidence focal elements with consistent hazard event identifiers but inconsistent focal element values ​​are read. The basic probability assignments corresponding to each evidence focal element are paired and calculated to determine the conflict quality between each group of evidence focal elements. The conflict quality of each group is then associated with the corresponding hazard event identifier, collection time, and focal element value relationship to ensure that the conflict quality corresponds to the conflict source and to generate a conflict quality record set. The conflict quality record set is expanded along the sequence of conflict transfer units. The transfer segment of the conflict quality is determined according to the start and end positions of each conflict transfer unit. Then, the conflict quality is segmented and split according to the arrangement of each transfer segment in the sequence, so that the conflict quality forms a decomposition result corresponding to the start and end positions in different transfer segments, and a conflict decomposition record set is generated. Each decomposition result in the conflict decomposition record set is mapped to the region covered by the closure bearing unit, so that each decomposition result completes intradomain projection within the range defined by the closure bearing unit. Each decomposition result that has completed intradomain projection is mapped to the fact front position and the antifact front position in the front corresponding unit, so that each decomposition result forms a corresponding allocation relationship between the fact front position and the antifact front position, generating the fact domain fusion state and the antifact domain fusion state.

[0027] In this embodiment, the decision output module is specifically as follows: Based on the hazard event identifier and the acquisition time, the fact domain fusion state and antifactual domain fusion state are retrieved accordingly, so that the fact domain fusion state and antifactual domain fusion state corresponding to the same hazard event identifier at the same acquisition time are paired one by one, and the pairing results are arranged in the order of acquisition time to generate a dual-domain coupling unit. In the dual-domain coupling unit, the state values ​​of the fact domain fusion state and the antifactual domain fusion state are read item by item. The difference calculation is performed on the state values ​​under the same pairing relationship to obtain the state difference. Then, the position consistency comparison is performed between the fact front position and the antifactual front position corresponding to the state difference to determine the position matching result corresponding to each state difference and generate a coupling decision record set. Based on the state difference and position matching results in the coupled judgment record set, the fact front position, antifactual front position, state difference and position matching results corresponding to the same hazard event identifier are jointly merged, and the merged results are organized into a unified result reflecting the overall state of the hazard event to generate a global situation result; The leading edge position, fusion strength, and control order in the global situational awareness results are mapped to the agent identifier and task identifier in the agent state sequence. The leading edge position determines the position of the control action, the fusion strength determines the degree of control action, and the control order determines the order of control execution. The results after mapping are organized into collaborative control instructions. The collaborative control instructions are then combined with the global situational awareness results to generate emergency rescue decision results.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to an industrial accident response scenario including a tank area, a pipe gallery area, a loading and unloading area, and an emergency access route. The initial accident state was a flammable medium leak at the connection point of the pipe gallery area. The temperature around the leak source continued to rise, the diffusion boundary moved towards the outer edge of the tank area, and a temporary obstacle on one side of the loading and unloading area partially blocked the emergency access route. Ground inspection units, firefighting units, aerial detection units, and mobile support units were deployed on-site. Each rescue unit continuously generated status records, location records, task records, and observation records. The accident area continuously generated regional topology records, obstacle distribution records, hazard source records, diffusion boundary records, and access constraint records. In this scenario, different rescue units showed significant differences in their assessment of the same hazardous event. The ground inspection unit obtained a higher concentration observation value near the leak point, the aerial detection unit identified a shift in the high-temperature zone, and the mobile support unit experienced a delay in arrival due to the obstructed access route. Traditional methods are prone to problems such as fluctuating hazard location, overlapping resource deployment, and imbalanced control order.

[0029] In the application of this invention, rescue perception data and environmental constraint data are first aggregated. Then, time correction is performed on the rescue perception data, spatial registration is performed on spatial coordinates with area identifiers, boundary coordinates, and connection relationship identifiers, and event association is performed on hazard event identifiers with hazard level values, diffusion direction values, diffusion range values, and constraint type identifiers to form an event association graph and an agent state sequence. Confirmation records are then extracted from the agent state sequence to generate a collaborative witnessing sequence. Subsequently, the spatial coordinates, task identifiers, and hazard event identifiers in the agent state sequence are arranged accordingly to form the current control state and alternative control state; connectivity expansion is performed on the event association graph to form a closure carrying domain; link connection is performed on the collaborative witnessing sequence to form a conflict transfer path; and frontier projection is performed on the current control state and alternative control state to form a factual frontier domain and a counterfactual frontier domain. Based on this, an improved Dubois-Prade algorithm is constructed according to the closure bearing domain, conflict transfer path, fact front domain, and counterfactual front domain. The local situation assessment results are then mapped to evidence focal elements and basic probability assignments to complete conflict quality calculation, structured decomposition, intra-domain projection, and dual-domain coupling solution. Finally, the system outputs global situation results, collaborative control commands, and emergency rescue decision results. The operational results show that the system determined the core location of the leak in the utility tunnel area to be in a high-risk closure area after the connection is expanded. Priority was given to deploying the aerial detection unit and ground inspection unit to the edge of the utility tunnel area. The fire extinguishing unit was given the task of isolating the outer edge of the tank area, and the mobile support unit was given the task of clearing obstacles in the loading and unloading area. This avoided control overlap caused by two ground units simultaneously pressing against the outer edge of the tank.

[0030] To ensure clear comparison results, three schemes were set up under the same accident scenario, the same number of rescue units, the same data input scale, and the same initial task configuration. The traditional rule-based scheduling scheme uses threshold triggering and fixed plan matching, without constructing event association graphs, collaborative witness sequences, closure-bearing domains, conflict transfer paths, fact front domains, and counterfactual front domains; the ordinary Dubois-Prade fusion scheme generates evidence focal elements and basic probability assignments and performs conventional Dubois-Prade fusion, without performing closure-bearing constraints, link decomposition constraints, and dual-domain coupling solutions; the system of this invention adopts the complete processing flow defined in the claims. After completing thirty rounds of scenario reproduction experiments, the statistical average results are shown in Table 1: Table 1 Comparison of the Effects of Collaborative Rescue Decisions

[0031] As shown in Table 1, the traditional rule-based scheduling scheme, which allocates tasks solely based on single observation thresholds and fixed plans, fails to effectively utilize conflict evidence and cannot reflect regional connectivity and control state differences. Consequently, it performs the worst in terms of hazardous event location error, number of erroneous scheduling attempts, and critical channel recovery time. While the ordinary Dubois–Prade fusion scheme improves evidence fusion capabilities, it lacks constraints such as closure-bearing domain constraints, conflict transfer path constraints, and dual-domain coupling constraints between the fact front and counterfactual front domains. Consequently, conflict quality still struggles to achieve directional decomposition and stable projection within the accident area. In contrast, the system of this invention integrates the event association graph, collaborative witness sequence, closure carrying domain, conflict transfer path, fact front domain, and counterfactual front domain into the improved Dubois–Prade algorithm, reducing the location error of dangerous events to 4.7 meters, increasing the effective utilization rate of conflict evidence to 89.6%, improving the matching accuracy of collaborative control commands to 94.1%, and shortening the recovery time of critical channels to 11.6 minutes. This demonstrates that the present invention can effectively solve the problems of unstable data association, difficulty in utilizing conflict evidence, and insufficient reliability of decision results in industrial emergency rescue scenarios, and achieve higher situational judgment accuracy and collaborative control effect.

[0032] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An industrial emergency rescue decision-making system based on multi-agent collaborative control, characterized in that, include: The data acquisition module is used to collect rescue perception data and environmental constraint data; The association generation module is used to perform time alignment, spatial registration, and event association on rescue perception data and environmental constraint data, and generate event association graphs and agent state sequences. The witness building module is used to perform chain-like connection of confirmation records in the agent's state sequence to generate a collaborative witness sequence; The domain mapping module is used to form the current control state and alternative control state based on the agent's state sequence, map the event association graph into the closure carrying domain, map the collaborative witness sequence into the conflict transfer path, map the current control state into the fact front domain, and map the alternative control state into the counterfactual front domain. The algorithm building module is used to construct the improved Dubois–Prade algorithm based on the closure bearing domain, conflict transition path, fact front domain, and counterfactual front domain; The evidence generation module is used to perform local situation assessment on rescue perception data and generate evidence focal elements and basic probability assignments. The conflict fusion module is used to perform conflict quality calculation, structured decomposition and intra-domain projection on evidence focal elements and basic probability assignments using the improved Dubois–Prade algorithm to generate fact domain fusion state and antifact domain fusion state. The decision output module is used to perform dual-domain coupled solution on the fact domain fusion state and the antifactual domain fusion state, and generate global situation results, collaborative control commands and emergency rescue decision results.

2. The industrial emergency rescue decision-making system based on multi-agent collaborative control according to claim 1, characterized in that, The data acquisition module is specifically: The system collects status records, location records, task records, and observation records of the rescue unit during the accident handling process, and extracts agent identifiers, collection times, spatial coordinates, task identifiers, hazardous event identifiers, and observation values ​​to form a rescue perception record set. Collect regional topology records, obstacle distribution records, hazard source records, diffusion boundary records, and traffic constraint records in the accident area, and extract regional identifiers, connection relationship identifiers, boundary coordinates, hazard level values, diffusion direction values, diffusion range values, and constraint type identifiers to form an environmental constraint record set; Data aggregation is performed on the rescue perception record set and the environmental constraint record set to generate rescue perception data and environmental constraint data.

3. The industrial emergency rescue decision-making system based on multi-agent collaborative control according to claim 1, characterized in that, The association generation module specifically includes: The rescue perception data is arranged in the order of collection time, and the arrangement results are mapped to a unified time base to form a time-aligned sequence. Read spatial coordinates from the time-corrected sequence and match the spatial coordinates with the region identifiers, boundary coordinates, and connection relationship identifiers in the environmental constraint data to form a spatial registration sequence. Read the hazard event identifier from the spatial registration sequence, and associate the hazard event identifier with the hazard level value, diffusion direction value, diffusion range value and constraint type identifier in the environmental constraint data to form an event association record set; The region identifier, connection relationship identifier, dangerous event identifier, and agent identifier in the event association record set are written into the graph structure to form an event association graph. The agent identifiers in the event association record set are arranged according to the collection time to form an agent state sequence.

4. The industrial emergency rescue decision-making system based on multi-agent collaborative control according to claim 1, characterized in that, The witness construction module is specifically as follows: The confirmation records are extracted from the agent's state sequence and arranged in the order of execution according to the acquisition time to generate a confirmation arrangement sequence. Read the danger event identifier and agent identifier from the confirmation sequence, perform adjacent pairing on the confirmation records with consistent danger event identifiers and progressive collection times, and generate a confirmation connection record set; Based on the adjacent pairing relationship in the confirmed connection record set, the confirmed records are chained together, and the continuously connected confirmed records are merged into witness chain records to generate a witness chain record set; The witness chain record set is arranged in the order of execution based on the time of collection to generate a collaborative witness sequence.

5. The industrial emergency rescue decision-making system based on multi-agent collaborative control according to claim 1, characterized in that, The domain mapping module is specifically: The agent state sequence is arranged in a dual order according to the acquisition time and agent identifier. The spatial coordinates, task identifiers and danger event identifiers corresponding to the same agent identifier are arranged according to the progressive relationship of the acquisition time to form the control state arrangement result. The task identifiers in the control state arrangement result are kept in their original configuration to generate the current control state. In the current control state, the agent identifier, data acquisition time, spatial coordinates, and danger event identifier are kept consistent. An alternative configuration orchestration is performed on the task identifier to generate an alternative control state. Perform connectivity unpacking on the region identifiers and connection relationship identifiers in the event association graph, and configure the hazard event identifiers to the corresponding region positions in the connectivity unpacking results, so that the regions with connectivity relationships and corresponding to the same hazard event form a closed association structure, and generate a closed carrying domain. The witness chain records in the collaborative witness sequence are arranged in the order of execution at the time of collection, and the link is connected according to the connection relationship between adjacent witness chain records, so that the witness chain records corresponding to the same dangerous event form a continuous transmission structure and generate a conflict transfer path; The current control state is projected onto the set of front positions according to the acquisition time and spatial coordinates to generate the factual front domain, and the alternative control state is projected onto the set of front positions according to the acquisition time and spatial coordinates to generate the counterfactual front domain.

6. The industrial emergency rescue decision-making system based on multi-agent collaborative control according to claim 1, characterized in that, The algorithm construction module is specifically as follows: The closed associated ranges in the closure bearing domain are segmented and organized according to their regional locations, and the regional locations and connectivity relationships corresponding to each closed associated range are arranged accordingly to form closure bearing units; The transmission links in the conflict transfer path are unfolded according to the execution order of the acquisition time, and the start position, end position and connection order of each transmission link are arranged accordingly to form a conflict transfer unit. Synchronize and align the fact frontier domain and the counterfactual frontier domain according to the acquisition time, and arrange the fact frontier position and the counterfactual frontier position corresponding to the same acquisition time to form a frontier corresponding unit; The closure-bearing unit is used as the conflict quality bearing range, the conflict transfer unit is used as the conflict quality decomposition path, and the front-end corresponding unit is used as the corresponding structure between the fact domain fusion state and the antifactual domain fusion state, thus forming the bearing structure, transfer structure and coupling structure of the improved Dubois–Prade algorithm. By mapping the evidence focal elements and basic probability assignments into the bearing structure, transfer structure, and coupling structure, the conflict quality between evidence focal elements is calculated and structurally decomposed along the conflict transfer unit, projected into the domain along the closure bearing unit, and coupled with the fact domain fusion state and antifactual domain fusion state along the front corresponding unit, thus forming the improved Dubois–Prade algorithm.

7. The industrial emergency rescue decision-making system based on multi-agent collaborative control according to claim 1, characterized in that, The evidence generation module specifically comprises: The rescue perception data is arranged in the order of collection time, and then collected according to the intelligent agent identifier and the dangerous event identifier to generate a local perception record set; Spatial coordinates, task identifiers, and observation values ​​are extracted from the local sensing record set, and the spatial coordinates, task identifiers, and observation values ​​are associated and arranged according to the dangerous event identifiers to generate a local situation determination unit; The observed values ​​in the local situation assessment unit are continuously compared according to the acquisition time, and local situation assessment is performed in combination with spatial coordinates and task identifiers to generate local situation assessment results; The local situation assessment results are mapped to focal elements according to the dangerous event identifiers to generate evidence focal elements. The observation values ​​and confirmation strengths corresponding to the local situation assessment results are then configured to the evidence focal elements to generate basic probability assignments.

8. The industrial emergency rescue decision-making system based on multi-agent collaborative control according to claim 1, characterized in that, The conflict fusion module is specifically as follows: The evidence focal elements and basic probability assignments are mapped into the bearing structure, transfer structure and coupling structure of the improved Dubois–Prade algorithm according to the dangerous event identifier and the collection time to generate conflict calculation units. In the conflict calculation unit, the evidence focal elements with consistent dangerous event identifiers but inconsistent focal element values ​​are compared, and the conflict quality between evidence focal elements is determined based on the basic probability assignment, generating a conflict quality record set. The conflict quality record set is expanded along the conflict transfer units in the transfer structure in the order of penetration, and the conflict quality is decomposed in a structured manner according to the start and end positions to generate a conflict decomposition record set. The conflict decomposition record set is projected into the domain according to the closure bearer unit in the bearer structure, and a dual-domain assignment is performed according to the front-end corresponding unit in the coupled structure to generate the fact domain fusion state and the antifact domain fusion state.

9. The industrial emergency rescue decision-making system based on multi-agent cooperative control according to claim 1, characterized in that, The decision output module is specifically as follows: Based on the hazardous event identifier and the acquisition time, the fact domain fusion state and antifactual domain fusion state are matched accordingly to generate a dual-domain coupling unit; In the dual-domain coupling unit, the state difference is calculated between the fused state of the fact domain and the fused state of the antifact domain, and the consistency judgment is performed on the fact front position and the antifact front position corresponding to the state difference, generating a coupling judgment record set; Based on the coupling judgment record set, joint merging is performed on the leading edge position, fusion strength and control order corresponding to the dangerous event identifier to generate global situation results; The leading position, fusion strength, and control order in the global situation results are mapped to the agent identifier and task identifier in the agent state sequence to generate collaborative control instructions, and emergency rescue decision results are generated based on the collaborative control instructions and the global situation results.

Images