Forest fire unmanned aerial vehicle swarm cooperative fire extinguishing planning method based on knowledge graph

Abstract

The application discloses a forest fire unmanned aerial vehicle swarm cooperative fire extinguishing planning method based on a knowledge graph, which comprises the following steps: acquiring forest fire scene data; constructing a forest fire time sequence knowledge graph, initializing an EvoKG model, and establishing a basic evolution framework; generating an intervention event sequence, injecting an event time evolution process, and generating a bidirectional evolution deduction structure; constructing a swarm cooperative situation subgraph, accessing the bidirectional evolution deduction structure, and generating a multi-scale evolution coupling structure; based on the multi-scale evolution coupling structure, constructing a task executability mapping mechanism, and generating an executable situation reasoning result; based on the executable situation reasoning result, generating an unmanned aerial vehicle swarm cooperative fire extinguishing planning scheme; acquiring execution feedback information of the planning scheme, and updating the multi-scale evolution coupling structure and the unmanned aerial vehicle swarm cooperative fire extinguishing planning scheme. The application realizes controllable deduction and high-reliability planning of fire evolution and swarm cooperative decision-making.

Description

Technical Field

[0001] This invention relates to the fields of emergency management and forest fire prevention engineering, and in particular to a knowledge graph-based method for planning collaborative fire suppression using drone swarms in forest fires. Background Technology

[0002] Forest fires are characterized by their suddenness, rapid evolution, and wide impact. Firefighting requires rapid decision-making and coordinated response under complex terrain, dynamic weather conditions, and strict airspace constraints. With the development of unmanned aerial vehicle (UAV) technology, utilizing multiple UAVs collaboratively for fire reconnaissance, firefighting, and support missions has become an important technological direction for forest fire emergency response.

[0003] In existing technologies, multi-UAV collaborative firefighting planning methods typically rely on rule-driven or static models to process fire, environment, and UAV status in stages. This makes it difficult to characterize the time dependencies and complex relationships between events during the fire's evolution. Although some methods introduce graph models or path planning algorithms, they focus more on spatial connectivity analysis and lack a unified model of the evolutionary relationship between fire events, environmental changes, and UAV behavior, resulting in limited adaptability of the planning results to dynamic changes.

[0004] Furthermore, most existing drone-based collaborative firefighting plans treat drone actions as external control commands, failing to effectively integrate actual intervention into the fire evolution reasoning process, making it difficult to assess the impact of intervention on subsequent fire development. At the same time, mission feasibility assessments are usually scattered across multiple independent modules, lacking a unified reasoning mechanism, and planning results often require repeated verification, impacting emergency response efficiency.

[0005] Therefore, how to provide a knowledge graph-based planning method for collaborative firefighting using drone swarms in forest fires is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a knowledge graph-based method for collaborative fire suppression planning using drone swarms in forest fires. This invention uses forest fire scenario data as a foundation, constructing a forest fire temporal knowledge graph and initializing an EvoKG model. It establishes a basic evolutionary framework for jointly modeling event temporal information and relational structure evolution information. Drone swarm fire suppression actions are encoded as intervention event sequences and injected into the event temporal evolution process, forming a bidirectional evolutionary inference structure. Furthermore, a swarm collaborative situational subgraph is constructed and coupled with the forest fire temporal knowledge graph to form a multi-scale evolutionary coupling structure. Based on this, relational evolutionary reasoning and task executability mapping are performed to directly generate a drone swarm collaborative fire suppression planning scheme, which is continuously updated through execution feedback event write-back. This invention can couple fire events and intervention events within a unified evolutionary framework, improving the dynamic consistency, executability, and evolutionary adaptability of collaborative planning.

[0007] The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to embodiments of the present invention includes the following steps: Acquire forest fire scene data; Based on forest fire scenario data, a forest fire time series knowledge graph was constructed, the EvoKG model was initialized, and a basic evolutionary framework was established. Based on the basic evolutionary framework, the drone swarm firefighting actions are encoded as intervention event sequences and injected into the event time evolution process of the EvoKG model to obtain a two-way evolutionary deduction structure. A swarm collaborative situational subgraph is constructed based on the forest fire time-series knowledge graph, and a two-way evolutionary inference structure is connected to form a multi-scale evolutionary coupling structure. Based on the multi-scale evolutionary coupling structure, the execution relationship evolutionary reasoning process is processed, a task executability mapping mechanism is constructed, executability-driven selection is performed, and executable situation reasoning results are generated. Based on the executable situational reasoning results, perform collaborative planning mapping processing to generate a drone swarm collaborative firefighting planning scheme. The system obtains execution feedback information of the UAV swarm collaborative fire suppression planning scheme, generates a set of feedback events, writes them back to the forest fire time series knowledge graph and inputs them into the EvoKG model, and updates the multi-scale evolution coupling structure and the UAV swarm collaborative fire suppression planning scheme.

[0008] Optionally, the acquisition and processing of the forest fire scene data includes: In forest fire scenarios, fire monitoring data, meteorological environment data, terrain data, airspace constraint data, and drone swarm status data are acquired and combined to form forest fire scenario data. The forest fire scene data is processed with unified timestamps, and each data recording unit is bound with a unified spatial identifier and organized in a structured manner.

[0009] Optionally, the generation of the forest fire temporal knowledge graph and basic evolutionary framework includes: Based on forest fire scenario data, the data is processed into event-based forms, which are then categorized into fire-related events, environment-related events, and swarm status-related events. All event representations are uniformly modeled as evolvable event nodes; A forest fire time-series knowledge graph is constructed based on evolvable event nodes, establishing the relationship structure between evolvable event nodes and characterizing the evolutionary influence relationship between them. The EvoKG model is initialized based on the forest fire time-series knowledge graph. In the EvoKG model, the evolvable event nodes are used as the evolutionary units, and the relational structure is used as the evolutionary propagation carrier. Joint modeling processing is performed to establish a joint update mechanism and form a basic evolutionary framework.

[0010] Optionally, the generation of the bidirectional evolutionary deduction structure includes: Based on the basic evolutionary framework, the firefighting actions performed by the drone swarm during the firefighting process are transformed into intervention event expressions with time attributes, and an intervention event sequence is constructed. The sequence of intervention events is used as an evolutionary control input independent of historical fire events and is fed into the EvoKG model to directly participate in the update of event time evolution parameters. During the update of event time evolution parameters, the intervention event sequence is connected to the calculation path. Under the same historical fire event conditions, evolution branch states containing the intervention event sequence and evolution branch states not containing the intervention event sequence are constructed. Event time evolution processing and relation structure evolution processing are performed to generate a bidirectional evolution inference structure.

[0011] Optionally, the generation of the multi-scale evolutionary coupling structure includes: Based on the evolvable event nodes and relational structures corresponding to the drone swarm state data in the forest fire time-series knowledge graph, we extract the set of event nodes and the set of relational structures representing the collaborative operation state of the drone swarm, and construct a swarm collaborative situation subgraph. In the basic evolutionary framework, the event node configuration in the swarm cooperative situation subgraph is consistent with the event time representation method of the forest fire temporal knowledge graph; By integrating the swarm cooperative situation subgraph into the bidirectional evolutionary deduction structure, during the event time evolution process, the cooperative events in the swarm cooperative situation subgraph and the fire events in the forest fire time series knowledge graph are included in the same event time evolution parameter update path. Based on the evolutionary state changes of collaborative events in the swarm collaborative situation subgraph, the event time evolution parameters of the corresponding event nodes in the forest fire time series knowledge graph are synchronously updated. Based on the evolutionary changes of fire events in the forest fire time-series knowledge graph, structural adjustment processing is performed on the event nodes and relationship structures in the swarm cooperative situation subgraph to generate a multi-scale evolutionary coupling structure.

[0012] Optionally, the generation of the executable situational reasoning result includes: Based on the multi-scale evolutionary coupling structure, relational evolutionary reasoning is performed to obtain a set of relational evolutionary paths constrained by the forest fire temporal knowledge graph and the bee colony collaborative situation subgraph; For the set of relationship evolution paths, based on airspace constraint data and terrain data, a safe route reachability determination process is performed on the relationship evolution paths to generate safe route reachability results; For the set of relationship evolution paths, based on the evolution status of supply-related event nodes in the forest fire time-series knowledge graph, supply reachability determination processing is performed to generate supply reachability results; For the set of relationship evolution paths, based on the coverage status of cooperative events in the bee colony cooperative situation subgraph, a cooperative coverage reachability determination process is performed to generate a cooperative coverage reachability result; Based on the reachability results of safe routes, resupply reachability results, and cooperative coverage reachability results, a joint executability determination process is performed on the set of relationship evolution paths. Unexecutable paths are marked, and the remaining relationship evolution paths are marked as executable paths, generating executable situational reasoning results.

[0013] Optionally, the generation of the drone swarm collaborative firefighting planning scheme includes: Read the relational evolution paths marked as executable paths in the executable situational reasoning results and parse them into a set of constraints for bee colony cooperative operations; Based on the set of constraints for swarm cooperative operation, the drone swarm status data is mapped and processed to generate swarm task allocation results. Based on the spatial coverage status corresponding to the task allocation results and relationship evolution paths of the bee colony, collaborative operation area mapping processing is performed on the operation area in the forest fire scenario to generate collaborative operation area division results. Based on the swarm task allocation results and the collaborative operation area division results, action sequence mapping processing is performed on the drone swarm status data to generate drone action sequences. The results of swarm task allocation, collaborative operation area division, and UAV action sequence are combined to generate a UAV swarm collaborative firefighting planning scheme.

[0014] Optionally, the model update includes: The execution feedback information of the drone swarm collaborative firefighting plan is obtained during the execution process, and the information is structured to form a set of feedback events; Based on the unified structure of event representation in the forest fire time-series knowledge graph, the set of feedback events is processed by event-based encoding, which is transformed into feedback event nodes with time attributes and incorporated into the event node set of the forest fire time-series knowledge graph. The set of feedback events is input into the EvoKG model, and the feedback event nodes are treated as evolvable event nodes in the basic evolutionary framework to participate in the joint update processing of event temporal evolution and relational structure evolution. Based on the updated forest fire temporal knowledge graph and swarm collaborative situational subgraph, a multi-scale evolutionary coupling structure is reconstructed within the basic evolutionary framework, and subsequent executable situational reasoning results and UAV swarm collaborative firefighting planning schemes are updated.

[0015] The beneficial effects of this invention are: First, this invention constructs a forest fire temporal knowledge graph and initializes an EvoKG model on it to establish a basic evolutionary framework. It integrates event temporal information and relational structure evolution information into the same evolutionary calculation process for joint modeling. This enables fire-related events, environment-related events, and swarm state-related events to be reasoned under a unified event temporal evolution parameter update path. This avoids the problem of temporal modeling and relational reasoning being separated in the prior art and improves the overall consistency and continuity of forest fire evolution characterization and situational reasoning results.

[0016] Secondly, this invention encodes the drone swarm firefighting actions as an intervention event sequence and injects the intervention event sequence as an independent evolutionary control input into the event time evolution process, constructing a two-way evolutionary deduction structure. Under the same historical fire event conditions, it forms evolutionary branch states that include intervention event sequences and those that do not, thereby enabling the reasoning process to reflect the actual impact of drone swarm collaborative firefighting behavior on fire evolution, significantly enhancing the pertinence and controllability of the reasoning results for collaborative firefighting decision-making.

[0017] Furthermore, this invention constructs a swarm cooperative situational subgraph and forms a multi-scale evolutionary coupling structure. Based on this, it performs relational evolutionary reasoning and task executability mapping, directly mapping the relational evolution path to safe route reachability results, supply reachability results, and cooperative coverage reachability results. This directly generates a drone swarm cooperative firefighting planning scheme, and combines feedback event write-back to achieve continuous updates. This ensures that the planning results have executability constraints at the generation stage, improving the reliability and continuous adaptability of drone swarm cooperative firefighting planning in complex forest fire scenarios. Attached Figure Description

[0018] 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 an overall flowchart of the knowledge graph-based forest fire drone swarm collaborative fire suppression planning method proposed in this invention. Figure 2 This is a schematic diagram of the bidirectional evolutionary deduction structure in this invention; Figure 3 This is a schematic diagram of the multi-scale evolution coupling structure and task executability mapping mechanism in this invention. Detailed Implementation

[0019] 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.

[0020] refer to Figure 1-3 A knowledge graph-based planning method for collaborative forest fire suppression using drone swarms includes the following steps: Acquire forest fire scene data, which includes fire monitoring data, meteorological environment data, terrain data, airspace constraint data, and drone swarm status data; A forest fire time series knowledge graph is constructed based on forest fire scenario data, and an EvoKG model is initialized based on the forest fire time series knowledge graph to establish a basic evolutionary framework. The basic evolutionary framework is used to jointly model event time information and relational structure evolution information. Based on the basic evolutionary framework, the drone swarm firefighting actions are encoded into an intervention event sequence. The intervention event sequence is then injected into the event time evolution process of the EvoKG model to obtain a two-way evolutionary inference structure. This two-way evolutionary inference structure is used to couple historical fire events and intervention event sequences in the same evolutionary process. Based on the forest fire time series knowledge graph, a swarm cooperative situation subgraph is constructed and connected to a two-way evolutionary deduction structure to form a multi-scale evolutionary coupling structure. The multi-scale evolutionary coupling structure is used to couple the swarm cooperative situation subgraph and the forest fire time series knowledge graph at the event time parameter level. Based on a multi-scale evolutionary coupling structure, relational evolutionary reasoning is performed, and a task executability mapping mechanism is constructed. This mechanism maps relational evolutionary paths to safe route reachability results, resupply reachability results, and cooperative coverage reachability results. Based on these results, executability-driven selection is performed on the relational evolutionary paths, generating executable situational reasoning results. Based on the executable situational reasoning results, collaborative planning and mapping processing is performed to generate a drone swarm collaborative firefighting planning scheme. The drone swarm collaborative firefighting planning scheme includes swarm task allocation results, collaborative operation area division results, and drone action sequence. The system obtains execution feedback information of the drone swarm collaborative fire suppression planning scheme, generates a set of feedback events, writes the set of feedback events back to the forest fire time series knowledge graph and inputs it into the EvoKG model, updates the multi-scale evolution coupling structure through the basic evolution framework, and updates the drone swarm collaborative fire suppression planning scheme.

[0021] In this embodiment, the acquisition and processing of the forest fire scene data includes: In forest fire scenarios, fire monitoring data, meteorological environment data, terrain data, airspace constraint data, and drone swarm status data are acquired and combined to form forest fire scenario data. The data on forest fire scenarios are processed with unified timestamps so that all types of data form data record units with consistent time under the same time base. Based on the unified processing of timestamps, a unified spatial identifier is bound to each data recording unit, so that fire monitoring data, meteorological environment data, terrain data, airspace constraint data, and UAV swarm status data form a one-to-one correspondence at the spatial location level. Based on time-consistent data recording units and unified spatial identifiers, forest fire scene data is structured and organized so that it has temporal continuity and spatial consistency before entering the construction of the forest fire time-series knowledge graph, thus serving as a deterministic data foundation for constructing the forest fire time-series knowledge graph.

[0022] In this embodiment, the generation of the forest fire time-series knowledge graph and basic evolutionary framework includes: Based on forest fire scenario data, event-based processing is performed on fire monitoring data, meteorological environment data, terrain data, airspace constraint data, and drone swarm status data. This transforms various types of data into event expressions with time attributes, and classifies the event expressions according to the event source and event semantics, forming fire-related events, environment-related events, and drone swarm status-related events. Based on fire-related events, environment-related events, and swarm status-related events, the various event representations are uniformly modeled as evolvable event nodes, and event occurrence time information and event association identifiers are configured for the evolvable event nodes; A forest fire time-series knowledge graph is constructed based on evolvable event nodes. A relational structure between evolvable event nodes is established in the forest fire time-series knowledge graph. The relational structure is used to characterize the evolutionary influence relationship between evolvable event nodes, and the relational structure has both event time association attribute and structural association attribute. The EvoKG model is initialized based on the forest fire time-series knowledge graph. In the EvoKG model, the evolvable event nodes are used as the evolutionary units and the relational structure is used as the evolution propagation carrier. Joint modeling processing is performed on the event time information and the relational structure evolution information to establish a joint update mechanism for event time information and relational structure evolution information, forming a basic evolutionary framework for subsequent synchronous reasoning processing of event time evolution and relational structure evolution. The aforementioned basic evolutionary framework is specifically an evolutionary computation framework built within the EvoKG model to unify the constraints of event temporal evolution and relational structure evolution. This basic evolutionary framework uses the evolvable event nodes in the forest fire temporal knowledge graph as the smallest evolutionary unit, and the relational structure between event nodes as the evolutionary propagation carrier. It jointly models the event occurrence time attribute and the relational structure state during the same evolutionary computation process. Specifically, the basic evolutionary framework maintains the corresponding event occurrence time state for each evolvable event node and the corresponding evolution intensity state for each relational structure. During the evolution process, based on the propagation direction and order of the relational structure, it performs synchronous update processing on the event occurrence time state and the relational structure state, enabling changes in the event time state to propagate between event nodes through the relational structure. Simultaneously, it constrains changes in the relational structure state by the evolutionary results of the event time state. Through this joint update method, a unified event time evolution parameter update path is formed, supporting the synchronous execution of event time evolution reasoning and relational structure evolution reasoning in subsequent processing. The event time information represents a set of time attributes in the evolvable event nodes that describe the location and order of the event occurrence. The set of time attributes consists of the occurrence time, duration, and relative order identifier of the event in the event timeline, and is used to characterize the temporal evolution of fire-related events, environment-related events, and bee colony state-related events in the forest fire time series knowledge graph. The relational structure evolution information represents the state change information of the relational structure between event nodes in the forest fire time-series knowledge graph during the evolution process. The state change information consists of the validity state of the relational structure, the association strength state, and the structural adjustment state generated as the event evolves over time. It is used to characterize the structural changes of the evolutionary influence relationship between event nodes over time.

[0023] In this embodiment, the generation of the bidirectional evolutionary deduction structure includes: Based on the basic evolutionary framework, the fire-fighting actions performed by drone swarms during the fire-fighting process are transformed into intervention events with time attributes. An intervention event sequence is constructed according to the chronological order, so that the intervention event sequence is on the same time axis as the historical fire events in the forest fire time sequence knowledge graph. Specifically, within the basic evolutionary framework, for the firefighting actions performed by drone swarms during fire suppression, the execution start time, duration, and effective spatial range of each action are determined based on drone swarm status data. The execution start time and duration are mapped to event occurrence time parameters, and the effective spatial range is correlated with the fire areas recorded in fire monitoring data, thus forming a time attribute description consistent with the event time representation used in historical fire events in the forest fire time-series knowledge graph. Based on this, according to the unified structure of event representation in the basic evolutionary framework, the above time attribute descriptions are combined with the corresponding fire area association information to generate an intervention event expression form. This intervention event expression form serves as a standardized event representation of the intervention event within the basic evolutionary framework, ensuring that the intervention event adopts a consistent event time modeling method with historical fire events during event time evolution. Subsequently, according to the chronological order of the event occurrence time parameters, multiple intervention event expression forms are organized into an intervention event sequence, enabling the intervention event sequence to directly participate in the event time evolution calculation process within the basic evolutionary framework as evolutionary control input, and influencing subsequent event time evolution results and relational structure evolution results. By using the sequence of intervention events as an evolutionary control input independent of historical fire events into the EvoKG model, the sequence of intervention events is not written into the forest fire time series knowledge graph as a historical event node during the event time evolution process, but directly participates in the update processing of event time evolution parameters. To achieve direct modulation of the event time evolution process by the intervention event sequence, the evolution of event occurrence time is described using continuous-time modeling in the basic evolutionary framework. For any evolvable event node, its occurrence tendency on the time axis is characterized by an event time intensity function. This function characterizes the intensity of the event in the continuous time domain and performs update processing based on the joint input of historical event states, relational structure evolution states, and the intervention event sequence. Its expression is as follows: ; in, This represents the intensity of an event occurring at time t, representing the tendency of the event to occur in the continuous time domain. This represents the baseline event temporal intensity without considering the influence of historical events, relational structures, or intervention events. It serves as a basic reference intensity for the temporal evolution of events. This term represents the modulation of the intensity of a historical event by its temporal evolution. It is formed by the evolution of fire-related events and environmental events that have occurred in the forest fire time series knowledge graph over time, and is used to characterize the cumulative impact of historical events on the timing of current events. The term representing the modulation of the intensity of event occurrence by the evolution state of the relational structure is given by the term where... This represents the validity and correlation strength states of the relationship structure between event nodes in a forest fire time-series knowledge graph during its temporal evolution. It is used to characterize the propagation effect of the evolutionary influence relationships between event nodes on the timing of event occurrence. The term representing the modulation of the intensity of an event occurrence by the intervention event sequence, where Formed from the sequence of intervention events corresponding to the firefighting actions of drone swarms, it is used to characterize the temporal modulation effect of the proactive intervention behavior of drone swarms on the temporal evolution of events. The exponential mapping operation is used to map the evolutionary state of historical events, the evolutionary state of relational structures, and the modulation terms corresponding to the sequence of intervention events to the event time intensity space in an exponential form. This ensures that the intensity of event occurrence remains non-negative during continuous time evolution and is used to amplify or suppress the comprehensive influence of each modulation term on the event occurrence time. In the basic evolutionary framework, through the exponential mapping operation, modulation information from different sources is superimposed in the same event time intensity calculation space, thereby achieving a unified characterization of the influence of event time evolution parameters on historical events, relational structures, and intervention events. In the basic evolutionary framework, the modulation terms corresponding to the sequence of intervention events As a time modulation input independent of historical event nodes, the intervention event sequence participates in the calculation and processing of event time intensity. Instead of being written into the forest fire time series knowledge graph as a historical event node, it directly participates in the update of event time evolution parameters through the event time intensity function. Under the same historical fire event state and relation structure evolution state, the event time evolution process forms different event time evolution trajectories depending on whether the intervention event sequence is introduced, thus providing a controllable event time evolution basis for the construction of subsequent two-way evolution inference structure. During the update of event time evolution parameters, the intervention event sequence is connected to the calculation path of event time evolution parameters. Under the same historical fire event conditions, evolution branch states containing intervention event sequences and evolution branch states not containing intervention event sequences are constructed. Event time evolution processing and relation structure evolution processing are performed on each evolution branch state respectively, thereby generating a bidirectional evolution inference structure for coupling historical fire events and intervention event sequences. In the basic evolutionary framework, the intervention event sequence, as an evolutionary control input, only participates in the update of event temporal evolution parameters. It is used to modulate the evolution of event occurrence time in the temporal dimension. The intervention event sequence does not participate in the direct construction and adjustment of the relationship structure between event nodes in the forest fire temporal knowledge graph. Its impact on the evolution of the relationship structure is indirectly reflected through the event temporal evolution results. In the event temporal evolution process, the intervention event sequence preferentially acts on the evolution path of the event occurrence time state to change the occurrence position and temporal distribution of the event on the time axis. Thus, while maintaining the semantic consistency of the relationship structure, it guides the subsequent relationship structure evolution to unfold along different temporal evolution trajectories.

[0024] In this embodiment, the generation of the multi-scale evolution coupling structure includes: Based on the evolvable event nodes and relational structures corresponding to the state data of UAV swarms in the forest fire time-series knowledge graph, we extract the set of event nodes and the set of relational structures to represent the collaborative operation state of UAV swarms, construct a swarm collaborative situation subgraph, and limit the swarm collaborative situation subgraph to a local evolutionary substructure of the forest fire time-series knowledge graph. In the basic evolutionary framework, the event nodes in the swarm cooperative situation subgraph are configured with the same event time representation as the forest fire time series knowledge graph, so that the cooperative events in the swarm cooperative situation subgraph and the fire events in the forest fire time series knowledge graph have a unified event time parameter space. By integrating the swarm cooperative situation subgraph into the bidirectional evolutionary deduction structure, during the event time evolution process, the cooperative events in the swarm cooperative situation subgraph and the fire events in the forest fire time series knowledge graph are included in the same event time evolution parameter update path, thus completing the alignment configuration of the event time evolution parameter update path. During the event time evolution process, based on the evolution state changes of collaborative events in the bee colony collaborative situation subgraph, the event time evolution parameters of the corresponding event nodes in the forest fire time series knowledge graph are synchronously updated. During the event time evolution, based on the evolution status changes of fire events in the forest fire time sequence knowledge graph, structural adjustment processing is performed on the event nodes and relationship structure in the swarm collaborative situation subgraph to generate a multi-scale evolution coupling structure that achieves bidirectional association at the event time parameter level.

[0025] In this embodiment, the generation of the executable situational reasoning result includes: Based on the multi-scale evolutionary coupling structure, relational evolutionary reasoning is performed to obtain a set of relational evolutionary paths constrained by the forest fire temporal knowledge graph and the swarm collaborative situation subgraph. The relational evolutionary paths are composed of event node sequences and node relational structure sequences, which are used to characterize the feasible evolutionary paths of UAV swarms in the firefighting mission execution process. The specific relation evolution reasoning process involves reading the evolvable event nodes and their relational structures corresponding to fire-related events, environment-related events, and swarm state-related events in the forest fire time-series knowledge graph based on a multi-scale evolution coupling structure, and reading the event nodes and their relational structures corresponding to collaborative events in the swarm collaborative situation subgraph. During the event time evolution process, based on the joint update mechanism of event occurrence time attributes and relational structure state changes in the basic evolution framework, time consistency update processing is performed on the event occurrence time attributes of the evolvable event nodes, and structural association consistency update processing is performed on the relational structures to form evolutionary state inputs for reasoning calculations. Based on the evolutionary state inputs, multi-hop relation propagation calculations are performed along the directed propagation direction of the relational structure in the EvoKG model to obtain the relation evolution propagation state for the preset target event node. Path backtracking processing is performed on the relation evolution propagation state to extract candidate relation evolution paths composed of event node sequences and relational structure sequences between nodes. Event time sequence consistency verification processing and relational structure validity verification processing are performed on the candidate relation evolution paths. The candidate relation evolution paths that pass the verification are retained and organized into a relation evolution path set. For the set of relationship evolution paths, based on airspace constraint data and terrain data, a safe route reachability determination process is performed on the relationship evolution paths. The safe route reachability determination process is used to determine whether the relationship evolution paths satisfy the existence of continuous routes, spatial constraint consistency and temporal order consistency within the evolution time range, and generate a safe route reachability result. For the set of relationship evolution paths, based on the evolution status of supply-related event nodes in the forest fire time-series knowledge graph, supply reachability determination processing is performed on the relationship evolution paths. The supply reachability determination processing is used to determine whether there is a valid supply path that satisfies the supply time constraint and the supply location constraint within the corresponding evolution time window, and to generate supply reachability results. For the set of relationship evolution paths, based on the coverage status of collaborative events in the bee colony collaborative situation subgraph, a collaborative coverage reachability determination process is performed on the relationship evolution paths. The collaborative coverage reachability determination process is used to determine whether the relationship evolution path meets the continuous coverage condition of the bee colony collaborative operation area within the evolution time range, and generates a collaborative coverage reachability result. Based on the results of safe route reachability, resupply reachability, and cooperative coverage reachability, a joint executability determination process is performed on the set of relationship evolution paths. When the safe route reachability, resupply reachability, or cooperative coverage reachability of any relationship evolution path is in an unreachable state, the relationship evolution path is marked as an unexecutable path, and the remaining relationship evolution paths are marked as executable paths. An executable situation reasoning result is generated based on the executable paths. In the mission executability mapping mechanism, the executability determination of the relationship evolution path is performed in a fixed order, sequentially completing the safe route reachability determination, resupply reachability determination, and cooperative coverage reachability determination. During the determination process, if the relationship evolution path fails any of the preceding reachability determinations, the subsequent determination processing of the relationship evolution path is terminated, and the relationship evolution path is marked as an unexecutable path. Cooperative coverage reachability determination is only performed if the safe route reachability determination and resupply reachability determination are passed. Through the above fixed determination order, the executability determination process of the relationship evolution path has a definite determination process and a consistent determination benchmark.

[0026] In this embodiment, the generation of the drone swarm collaborative firefighting planning scheme includes: Read the relation evolution path marked as an executable path in the executable situation reasoning result, and parse the corresponding event node sequence and relation structure sequence in the relation evolution path into a set of swarm cooperative operation constraints. The set of swarm cooperative operation constraints is used to limit the task association relationship and time sequence relationship of the UAV swarm in the fire-fighting mission execution process. Based on the set of constraints for swarm collaborative operation, the swarm status data of UAVs is mapped and processed to generate swarm task allocation results. The swarm task allocation results are used to characterize the task undertaking relationship and execution sequence relationship of each UAV in the collaborative firefighting task. Based on the spatial coverage state corresponding to the swarm task allocation result and the relationship evolution path, a collaborative operation area mapping process is performed on the operation area in the forest fire scenario to generate a collaborative operation area division result. The collaborative operation area division result is used to limit the spatial operation range of the drone swarm during the firefighting process. Based on the swarm task allocation results and the collaborative operation area division results, action sequence mapping processing is performed on the drone swarm status data to generate drone action sequences. The drone action sequences are used to characterize the action order and time arrangement of each drone in the firefighting task execution process. The results of swarm task allocation, collaborative operation area division, and UAV action sequence are combined to generate a UAV swarm collaborative firefighting plan, which is then used as the direct input for subsequent execution and feedback processing.

[0027] In this embodiment, the model update includes: The execution feedback information of the UAV swarm collaborative firefighting plan is obtained during the execution process. The execution feedback information includes the actual action status of the UAV, the mission completion status, the actual coverage status of the collaborative operation area and the time execution deviation information. The execution feedback information is then structured to form a set of feedback events. Based on the unified structure of event representation in the forest fire time-series knowledge graph, the feedback event set is processed by event-based encoding, which transforms the feedback event set into feedback event nodes with time attributes. The feedback event nodes are then incorporated into the event node set of the forest fire time-series knowledge graph and stored as part of the evolutionary history. During the write-back process of the feedback event set, the feedback event nodes are incorporated into the event timeline of the forest fire time series knowledge graph as newly added event nodes. Together with existing historical fire events, environmental events, and swarm state-related events, they form a complete evolutionary history sequence. The feedback event nodes do not cover, replace, or roll back existing historical event nodes. Instead, they participate in the subsequent event time evolution and relational structure evolution in the form of newly added events. This enables the forest fire time series knowledge graph to form a continuously expanding evolutionary history in the time dimension, which supports the continuous updating of the subsequent multi-scale evolutionary coupling structure. The set of feedback events is input into the EvoKG model. In the basic evolutionary framework, the feedback event nodes are treated as evolvable event nodes and participate in the joint update process of event temporal evolution and relational structure evolution to update the event temporal state and relational structure state of event nodes in the forest fire time series knowledge graph. Based on the updated forest fire temporal knowledge graph and the swarm collaborative situational subgraph, a multi-scale evolutionary coupling structure is reconstructed within the basic evolutionary framework. The subsequent executable situational reasoning results and UAV swarm collaborative firefighting planning schemes are then updated based on the updated multi-scale evolutionary coupling structure.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a drone swarm collaborative firefighting planning task in a forest fire emergency response scenario. In this scenario, fires are characterized by rapid development, dynamic changes in spatial scope, complex terrain, and limited airspace. A single drone cannot complete continuous monitoring and firefighting operations within an effective time window. Traditional methods based on static rules or single-path planning have significant shortcomings in terms of task executability, collaborative stability, and dynamic adaptability, and are prone to problems such as route conflicts, supply mismatches, and discontinuous collaborative coverage.

[0029] In practical applications, the system continuously acquires forest fire scenario data, including fire monitoring data, meteorological data, terrain data, airspace constraint data, and UAV swarm status data. Fire monitoring data includes the location of fire points, the rate of fire spread, and changes in combustion intensity. Meteorological data includes changes in wind speed, wind direction, and ambient humidity. Terrain data reflects the elevation variations and obstacle distribution within the operational area. Airspace constraint data describes no-fly zones and flight altitude restrictions. UAV swarm status data includes the remaining battery power, payload capacity, and current pose of each UAV. After organizing this data under the same temporal and spatial references, it is used to construct a forest fire temporal knowledge graph, thereby ensuring that subsequent evolutionary reasoning processes have a foundation of temporal continuity and spatial consistency.

[0030] After the forest fire time-series knowledge graph is constructed, the system initializes the EvoKG model based on this graph and establishes a basic evolutionary framework. In this framework, fire-related events, environment-related events, and swarm state-related events are uniformly modeled as evolvable event nodes. The relationship structure between these event nodes is used to characterize the evolutionary influence between fire development, environmental changes, and swarm state. Event time information and relationship structure evolution information are jointly updated during the same evolutionary process, enabling the fire evolution trend and swarm state changes to be reflected synchronously in both the temporal and structural dimensions.

[0031] During firefighting operations, the firefighting actions of the drone swarm are encoded as intervention event sequences and injected into the event time evolution process. These intervention event sequences are not written into the forest fire time-series knowledge graph as historical event nodes, but rather directly participate in the updating of event time evolution parameters as evolutionary control inputs. In this way, the system can simultaneously deduce evolutionary branch states containing and without intervention event sequences under the same historical fire conditions, thus forming a bidirectional evolutionary deduction structure that characterizes the impact of drone firefighting behavior on the fire evolution path and relational structure evolution results.

[0032] Based on this, the system constructs a swarm cooperative situational awareness subgraph to characterize the cooperative relationships, cooperative coverage status, and cooperative operation constraints among UAVs. The swarm cooperative situational awareness subgraph and the forest fire temporal knowledge graph achieve bidirectional correlation at the event time parameter level, forming a multi-scale evolutionary coupling structure. This allows changes in the swarm cooperative state to feed back and influence the fire evolution inference results, while the fire evolution results can also trigger structural adjustments in the swarm cooperative situational awareness.

[0033] Based on a multi-scale evolutionary coupling structure, the system performs relational evolutionary reasoning processing, generates a set of relational evolutionary paths, and constructs a task executability mapping mechanism on this basis. For each relational evolutionary path, the system calculates the reachability results of safe routes, resupply, and cooperative coverage, and generates an executable situational reasoning result through joint determination. This reasoning result directly serves as the basis for generating a UAV swarm cooperative firefighting planning scheme, eliminating the need for additional secondary feasibility screening and thus reducing information loss between planning and reasoning.

[0034] In experimental verification, the method of this invention was compared with the traditional multi-UAV cooperative planning method based on static rules. The planning results under multiple consecutive fire evolution cycles were selected for statistical analysis. During the experiment, indicators such as planning generation success rate, executable path ratio, cooperative coverage continuity index, number of route conflicts, and average planning update time were recorded, and the following comparison results were obtained.

[0035] Table 1. Performance Comparison of Different Methods in Forest Fire Drone Swarm Cooperative Firefighting Planning Tasks

[0036] As shown in Table 1, traditional rule-based planning methods exhibit a high proportion of unexecutable paths, insufficient collaborative coverage continuity, and a high frequency of route conflicts in scenarios with rapid fire evolution. The method of this invention, by introducing a forest fire temporal knowledge graph and a multi-scale evolutionary coupling structure, improves the planning success rate by approximately 8.3 percentage points and the proportion of executable paths by approximately 12.4 percentage points, indicating that the generated planning scheme is more stable in terms of safety and executability. The collaborative coverage continuity index increases from 0.72 to 0.83, indicating improved coverage stability of the swarm collaborative operation area in the time dimension. The number of route conflicts decreases significantly, mainly because the task executability mapping mechanism constrains and filters the accessibility of safe routes during the planning stage. The average planning update time increases slightly but remains within a reasonable range, indicating that the introduction of evolutionary reasoning has not caused unacceptable impacts on the system's real-time performance.

[0037] 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. A knowledge graph-based planning method for collaborative forest fire suppression using drone swarms, characterized in that, Includes the following steps: Acquire forest fire scene data; Based on forest fire scenario data, a forest fire time series knowledge graph was constructed, the EvoKG model was initialized, and a basic evolutionary framework was established. Based on the basic evolutionary framework, the drone swarm firefighting actions are encoded as intervention event sequences and injected into the event time evolution process of the EvoKG model to obtain a two-way evolutionary deduction structure. A swarm collaborative situational subgraph is constructed based on the forest fire time-series knowledge graph, and a two-way evolutionary inference structure is connected to form a multi-scale evolutionary coupling structure. Based on the multi-scale evolutionary coupling structure, the execution relationship evolutionary reasoning process is processed, a task executability mapping mechanism is constructed, executability-driven selection is performed, and executable situation reasoning results are generated. Based on the executable situational reasoning results, perform collaborative planning mapping processing to generate a drone swarm collaborative firefighting planning scheme. The system obtains execution feedback information of the UAV swarm collaborative fire suppression planning scheme, generates a set of feedback events, writes them back to the forest fire time series knowledge graph and inputs them into the EvoKG model, and updates the multi-scale evolution coupling structure and the UAV swarm collaborative fire suppression planning scheme.

2. The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to claim 1, characterized in that, The acquisition and processing of the forest fire scene data includes: In forest fire scenarios, fire monitoring data, meteorological environment data, terrain data, airspace constraint data, and drone swarm status data are acquired and combined to form forest fire scenario data. The forest fire scene data is processed with unified timestamps, and each data recording unit is bound with a unified spatial identifier and organized in a structured manner.

3. The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to claim 1, characterized in that, The generation of the forest fire temporal knowledge graph and basic evolutionary framework includes: Based on forest fire scenario data, the data is processed into event-based forms, which are then categorized into fire-related events, environment-related events, and swarm status-related events. All event representations are uniformly modeled as evolvable event nodes; A forest fire time-series knowledge graph is constructed based on evolvable event nodes, establishing the relationship structure between evolvable event nodes and characterizing the evolutionary influence relationship between them. The EvoKG model is initialized based on the forest fire time-series knowledge graph. In the EvoKG model, the evolvable event nodes are used as the evolutionary units, and the relational structure is used as the evolutionary propagation carrier. Joint modeling processing is performed to establish a joint update mechanism and form a basic evolutionary framework.

4. The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to claim 1, characterized in that, The generation of the bidirectional evolutionary deduction structure includes: Based on the basic evolutionary framework, the firefighting actions performed by the drone swarm during the firefighting process are transformed into intervention event expressions with time attributes, and an intervention event sequence is constructed. The sequence of intervention events is used as an evolutionary control input independent of historical fire events and is fed into the EvoKG model to directly participate in the update of event time evolution parameters. During the update of event time evolution parameters, the intervention event sequence is connected to the calculation path. Under the same historical fire event conditions, evolution branch states containing the intervention event sequence and evolution branch states not containing the intervention event sequence are constructed. Event time evolution processing and relation structure evolution processing are performed to generate a bidirectional evolution inference structure.

5. The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to claim 1, characterized in that, The generation of the multi-scale evolutionary coupling structure includes: Based on the evolvable event nodes and relational structures corresponding to the drone swarm state data in the forest fire time-series knowledge graph, we extract the set of event nodes and the set of relational structures representing the collaborative operation state of the drone swarm, and construct a swarm collaborative situation subgraph. In the basic evolutionary framework, the event node configuration in the swarm cooperative situation subgraph is consistent with the event time representation method of the forest fire temporal knowledge graph; By integrating the swarm cooperative situation subgraph into the bidirectional evolutionary deduction structure, during the event time evolution process, the cooperative events in the swarm cooperative situation subgraph and the fire events in the forest fire time series knowledge graph are included in the same event time evolution parameter update path. Based on the evolutionary state changes of collaborative events in the swarm collaborative situation subgraph, the event time evolution parameters of the corresponding event nodes in the forest fire time series knowledge graph are synchronously updated. Based on the evolutionary changes of fire events in the forest fire time-series knowledge graph, structural adjustment processing is performed on the event nodes and relationship structures in the swarm cooperative situation subgraph to generate a multi-scale evolutionary coupling structure.

6. The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to claim 1, characterized in that, The generation of the executable situational reasoning result includes: Based on the multi-scale evolutionary coupling structure, relational evolutionary reasoning is performed to obtain a set of relational evolutionary paths constrained by the forest fire temporal knowledge graph and the bee colony collaborative situation subgraph; For the set of relationship evolution paths, based on airspace constraint data and terrain data, a safe route reachability determination process is performed on the relationship evolution paths to generate safe route reachability results; For the set of relationship evolution paths, based on the evolution status of supply-related event nodes in the forest fire time-series knowledge graph, supply reachability determination processing is performed to generate supply reachability results; For the set of relationship evolution paths, based on the coverage status of cooperative events in the bee colony cooperative situation subgraph, a cooperative coverage reachability determination process is performed to generate a cooperative coverage reachability result; Based on the reachability results of safe routes, resupply reachability results, and cooperative coverage reachability results, a joint executability determination process is performed on the set of relationship evolution paths. Unexecutable paths are marked, and the remaining relationship evolution paths are marked as executable paths, generating executable situational reasoning results.

7. The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to claim 1, characterized in that, The generation of the drone swarm collaborative firefighting planning scheme includes: Read the relational evolution paths marked as executable paths in the executable situational reasoning results and parse them into a set of constraints for bee colony cooperative operations; Based on the set of constraints for swarm cooperative operation, the drone swarm status data is mapped and processed to generate swarm task allocation results. Based on the spatial coverage status corresponding to the task allocation results and relationship evolution paths of the bee colony, collaborative operation area mapping processing is performed on the operation area in the forest fire scenario to generate collaborative operation area division results. Based on the swarm task allocation results and the collaborative operation area division results, action sequence mapping processing is performed on the drone swarm status data to generate drone action sequences. The results of swarm task allocation, collaborative operation area division, and UAV action sequence are combined to generate a UAV swarm collaborative firefighting planning scheme.

8. The knowledge graph-based forest fire drone swarm collaborative fire suppression planning method according to claim 1, characterized in that, The model update includes: The execution feedback information of the drone swarm collaborative firefighting plan is obtained during the execution process, and the information is structured to form a set of feedback events; Based on the unified structure of event representation in the forest fire time-series knowledge graph, the set of feedback events is processed by event-based encoding, which is transformed into feedback event nodes with time attributes and incorporated into the event node set of the forest fire time-series knowledge graph. The set of feedback events is input into the EvoKG model, and the feedback event nodes are treated as evolvable event nodes in the basic evolutionary framework to participate in the joint update processing of event temporal evolution and relational structure evolution. Based on the updated forest fire temporal knowledge graph and swarm collaborative situational subgraph, a multi-scale evolutionary coupling structure is reconstructed within the basic evolutionary framework, and subsequent executable situational reasoning results and UAV swarm collaborative firefighting planning schemes are updated.

Images