UAV dispatching method and system based on artificial intelligence

Abstract

The present application provides a method and system for UAV scheduling based on artificial intelligence. By automatically generating a logical knowledge graph of candidate tasks and predicting knowledge points, it can quickly capture the core features and technical requirements of the flight tasks to be assigned, and provide a solid data foundation for subsequent scheduling decisions. The matching algorithm is used to quickly screen out the solution that best matches the current task from the rich prior flight scheduling solution data, which not only greatly shortens the scheduling preparation time, but also ensures the rationality and feasibility of the scheduling solution. Finally, the UAV scheduling execution route corresponding to the matched prior task logical knowledge graph is directly applied to the task to be assigned, realizing the intelligence and automation of UAV scheduling, effectively reducing errors and delays caused by human factors, and improving the overall efficiency and resource utilization of UAV operations. As a result, the flexibility and response speed of UAV applications can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based drone scheduling method and system. Background Art

[0002] With the rapid development of drone technology, drones are increasingly used in various fields, including environmental monitoring, logistics distribution, agricultural plant protection, disaster relief, etc. However, with the complexity and diversification of application scenarios, the difficulty of drone scheduling has also increased. Traditional drone scheduling methods often rely on manual experience, and have problems such as low scheduling efficiency, slow response speed, and easy errors, which make it difficult to meet the needs of efficient and accurate drone operations. Summary of the invention

[0003] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide a drone scheduling method and system based on artificial intelligence.

[0004] In a first aspect, the present application provides a drone scheduling method based on artificial intelligence, the method comprising:

[0005] Acquire a flight task to be assigned, wherein the flight task to be assigned includes a first task description and a corresponding second task description, wherein the first task description is used to summarize task feature information of the second task description;

[0006] Generate a task logic knowledge graph based on the first task description to generate a candidate first task logic knowledge graph, and predict task knowledge points based on the candidate first task logic knowledge graph to generate a first task knowledge point label;

[0007] Generate a task logic knowledge graph according to the second task description to generate a candidate second task logic knowledge graph, and predict task knowledge points according to the candidate second task logic knowledge graph to generate a second task knowledge point label;

[0008] Acquire each priori flight scheduling solution data that matches the first task knowledge point label and the second task knowledge point label respectively, wherein the priori flight scheduling solution data includes a priori task logic knowledge graph and a corresponding UAV scheduling execution route;

[0009] Extracting a target priori task logic knowledge graph matching the flight task to be assigned from each priori task logic knowledge graph based on the candidate first task logic knowledge graph and the candidate second task logic knowledge graph;

[0010] The UAV scheduling execution route corresponding to the target priori task logic knowledge graph is used as the target UAV scheduling execution route of the flight task to be assigned.

[0011] In a second aspect, an embodiment of the present application also provides an artificial intelligence-based drone scheduling system, which includes a processor and a machine-readable storage medium, in which a computer program is stored. The computer program is loaded and executed in conjunction with the processor to implement the artificial intelligence-based drone scheduling method of the first aspect above.

[0012] By adopting the technical solutions of any of the above aspects, the embodiments of the present application significantly improve the efficiency and accuracy of UAV task scheduling. By automatically generating a logical knowledge graph of candidate tasks and performing knowledge point prediction, the core features and technical requirements of the flight tasks to be assigned can be quickly captured, providing a solid data foundation for subsequent scheduling decisions. The matching algorithm is used to quickly screen out the solution that best matches the current task from the rich prior flight scheduling solution data, which not only greatly shortens the scheduling preparation time, but also ensures the rationality and feasibility of the scheduling solution. Finally, the UAV scheduling execution route corresponding to the matched prior task logical knowledge graph is directly applied to the task to be assigned, realizing the intelligentization and automation of UAV scheduling, effectively reducing errors and delays caused by human factors, and improving the overall efficiency and resource utilization of UAV operations. As a result, the flexibility and response speed of UAV applications can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained by combining these drawings without paying creative work.

[0014] Figure 1 A schematic diagram of a flow chart of a drone dispatching method based on artificial intelligence provided in an embodiment of the present application;

[0015] Figure 2 A block diagram showing the functional structure of an artificial intelligence-based drone dispatching system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following description is intended to enable one of ordinary skill in the art to implement and combine the present application, and the description is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various changes may be made to the disclosed embodiments, and that the general principles defined in the present application may be applied to other embodiments and application scenarios without departing from the principles and scope of the present application. Therefore, the present application is not limited to the described embodiments, but should be given the broadest scope consistent with the claims.

[0017] See also Figure 1 As shown, the present application provides a drone scheduling method based on artificial intelligence, comprising the following steps.

[0018] Step S110, obtaining a flight mission to be assigned, wherein the flight mission to be assigned includes a first mission description and a corresponding second mission description, wherein the first mission description is used to summarize mission feature information of the second mission description.

[0019] Specifically, the flight mission to be assigned refers to a UAV flight mission that has not yet been assigned a specific execution plan or route. These missions usually include information such as specific requirements, objectives, execution time and location of the mission.

[0020] The first task description is used to summarize the core features and objectives of the task, and is usually concise and clear. The second task description is a specific expansion of the first task description, including detailed execution steps, required equipment, execution conditions and other information.

[0021] For example, in this embodiment, it is assumed that the server receives a flight mission to be assigned from a city management department. The first mission description is: "Monitor the environment and check for abnormal conditions in a specific area of ​​the city at night", which summarizes the core features of the entire mission.

[0022] The corresponding second task description is more detailed, for example: "Use a drone equipped with a high-definition camera and infrared sensor to take off from the helipad on top of a street lamp in the city center between 8 pm and 2 am, fly along a preset route, and collect environmental data for the industrial park in the west of the city, the commercial district in the east, and the residential area in the south, including air quality and noise levels. Use image recognition technology to check for abnormal conditions such as fires and gatherings of people. During the flight, data is transmitted to the control center every 15 minutes, and when the battery level is less than 30%, it automatically returns to the nearest helipad on top of a street lamp to charge."

[0023] Step S120: Generate a task logic knowledge graph based on the first task description to generate a candidate first task logic knowledge graph, and predict task knowledge points based on the candidate first task logic knowledge graph to generate a first task knowledge point label.

[0024] In detail, the task logic knowledge graph is a graph that graphically displays the various nodes in a task (such as task steps, key elements, etc.) and the logical relationships between them, which helps to understand the overall process and key links of the task.

[0025] The candidate first task logic knowledge graph is a possible task logic knowledge graph generated based on the first task description. The candidate second task logic knowledge graph is a possible task logic knowledge graph generated based on the second task description, and usually contains more detailed steps and logical relationships.

[0026] The task knowledge point labels are key knowledge points or skill labels related to the task predicted based on the task logic knowledge graph. These task knowledge point labels are helpful for subsequent matching of appropriate drone scheduling execution plans.

[0027] That is, in this embodiment, the server first analyzes the first task description "monitoring the environment and troubleshooting abnormal situations in specific areas of the city at night", extracts keywords therein, such as "nighttime", "environmental monitoring", "anomaly troubleshooting", "specific areas", etc., and uses these keywords as nodes in a directed graph.

[0028] Using the named entity recognition algorithm, we identify entities such as "city", "industrial park", "commercial area", and "residential area", and extract the relationship between keywords and entities. For example, there is a relationship between "night" and "environmental monitoring" in terms of time and behavior, and there is a relationship between "environmental monitoring" and "specific area" in terms of object and scope.

[0029] The extracted keywords and entities are used as nodes of a directed graph, and each node contains multiple attributes. For example, the attributes of the node "night" include the word itself "night", the part of speech "noun", and the importance weight "high".

[0030] According to the results of relationship extraction, directed edges between nodes are defined, such as from "night" to "environmental monitoring". The weight of the edge represents the strength of the relationship. For example, the influence of "night" on "environmental monitoring" is higher.

[0031] The directed graph is converted into a vector representation using the adjacency matrix of the graph and the eigenvectors of the Laplacian matrix to generate a first directed graph vector.

[0032] By analyzing the direct connection relationship in the first directed graph vector, the explicit task logic is extracted, for example, “nighttime” directly determines the time range of “environmental monitoring”.

[0033] Using path analysis or community discovery algorithms, implicit task logic can be inferred from the first directed graph vector, for example, “environmental monitoring” may be indirectly related to the subsequent “data analysis and report generation”.

[0034] Integrate the explicit task logic and the implicit task logic to generate a candidate first task logic knowledge graph.

[0035] Based on this candidate first task logical knowledge graph, the server predicts task knowledge points, such as predicting knowledge points such as "night flight conditions", "environmental data collection technology", and "abnormal situation identification algorithm", and generates first task knowledge point labels.

[0036] Step S130, generating a task logic knowledge graph according to the second task description to generate a candidate second task logic knowledge graph, and predicting task knowledge points according to the candidate second task logic knowledge graph to generate a second task knowledge point label.

[0037] In this embodiment, the server describes the second task as "Using a drone equipped with a high-definition camera and an infrared sensor, take off from the helipad on top of a street lamp in the city center between 8 pm and 2 am, and fly along a preset route to collect environmental data for the industrial park in the west of the city, the commercial district in the east, and the residential area in the south, including air quality, noise levels, etc., and use image recognition technology to check for abnormal conditions such as fires and gatherings of people. During the flight, data is transmitted to the control center every 15 minutes, and when the battery level is less than 30%, it automatically returns to the nearest helipad on top of the street lamp to charge." The task steps are broken down.

[0038] First, it is divided into various drone mission nodes, such as "take-off preparation", "take-off from the apron", "fly according to the preset route", "data collection", "abnormal situation investigation", "data transmission", "return to the apron for charging", etc.

[0039] Then, the task node features corresponding to each drone task node are obtained. For example, the features of the "takeoff preparation" node may include "check the drone equipment status" and "confirm that the sensor is working properly", etc.; the features of the "takeoff from the apron" node may include "send the takeoff command" and "execute the takeoff action", etc.

[0040] For each UAV task node, the candidate task node features and target task node features corresponding to the candidate UAV task node are determined from the corresponding task node features. Taking the "data collection" node as an example, the candidate task node feature may be "select a suitable sensor", and the target task node features may include "determine the collection frequency" and "set the collection range".

[0041] Determine the logical connection density between the candidate task node features and the target task node features, and generate the correlation degree of each target task node. For example, the logical connection between "selecting a suitable sensor" and "determining the acquisition frequency" is close, and the correlation degree is high; the logical connection with "setting the acquisition range" is relatively weak, and the correlation degree is low.

[0042] The influence of each target task node is evaluated based on its correlation, and the influence coefficient of each target task node is generated. The target task node with high correlation has a large influence coefficient, while the target task node with low correlation has a small influence coefficient.

[0043] Determine the fusion result of each target task node influence coefficient and the corresponding target task node feature, and generate each target weighted task node feature. For example, the target task node feature with a large influence coefficient accounts for a larger proportion in the fusion result.

[0044] According to the weighted task node characteristics of each target, a node association logic subgraph corresponding to the candidate UAV task node is generated.

[0045] Poll each drone task node and repeat the above process to generate a node association logic subgraph corresponding to each drone task node.

[0046] The node-associated logic subgraphs corresponding to each UAV task node are fused to generate a candidate second task logic knowledge graph.

[0047] Based on this candidate second task logical knowledge graph, the server predicts the task knowledge points, such as "HD camera parameter settings", "infrared sensor working principle", "data transmission protocol", "apron charging process" and other knowledge points, and generates second task knowledge point labels.

[0048] Step S140, obtaining each priori flight scheduling plan data that matches the first task knowledge point label and the second task knowledge point label respectively, wherein the priori flight scheduling plan data includes a priori task logic knowledge graph and a corresponding UAV scheduling execution route.

[0049] In this embodiment, the server searches its database for prior flight scheduling plan data that matches the generated first task knowledge point labels "night flight conditions", "environmental data acquisition technology", "abnormal situation identification algorithm" and second task knowledge point labels "high-definition camera parameter settings", "infrared sensor working principle", "data transmission protocol", and "apron charging process".

[0050] Assume that the following prior flight scheduling data exists in the database:

[0051] Solution 1: The task of night patrol of urban parks has been carried out before. Its prior task logic knowledge graph includes the logic of flight route planning, image acquisition and analysis in the night environment. The corresponding drone scheduling execution route is to take off from the street lamp apron near the urban park and patrol along the roads around the park, focusing on areas where people are active.

[0052] Solution 2: The task of daytime environmental monitoring of the industrial park has been performed. The prior task logic knowledge graph covers the environmental data collection logic of the industrial park. The corresponding drone scheduling execution route is to take off from the designated take-off point in the industrial park and conduct inspections according to the factory layout.

[0053] Solution 3: Having completed the task of monitoring the flow of people in commercial areas, the prior task logic knowledge graph involves the monitoring logic of crowded areas. The corresponding drone dispatch execution route is to take off from the apron on the top of the high-rise building in the commercial center and fly around the main streets and shopping malls.

[0054] The server obtains the prior flight scheduling plan data that matches the current mission knowledge point labels, including the prior mission logic knowledge graph and the corresponding UAV scheduling execution route.

[0055] Step S150, extracting a target priori task logic knowledge graph that matches the flight task to be assigned from each priori task logic knowledge graph based on the candidate first task logic knowledge graph and the candidate second task logic knowledge graph, and using the UAV scheduling execution route corresponding to the target priori task logic knowledge graph as the target UAV scheduling execution route for the flight task to be assigned.

[0056] In this embodiment, the server first determines the task logic association between the candidate first task logic knowledge graph and each prior first task logic knowledge graph. For example, by comparing the candidate first task logic knowledge graph of the current task with the prior first task logic knowledge graph of the park night patrol task in Solution 1, it is found that the two have a certain degree of association in terms of night flight and environmental monitoring.

[0057] Similarly, the task logic correlation between the candidate second task logic knowledge graph and each prior second task logic knowledge graph is determined. For example, by comparing the candidate second task logic knowledge graph of the current task with the prior second task logic knowledge graph of the industrial park environment monitoring task in Solution 2, it is found that they have some similarities in data collection and route planning.

[0058] A weighted calculation is performed based on each priori first task logic relevance and each priori second task logic relevance. Assuming that the priori first task logic relevance of Scheme 1 is 0.7, the priori second task logic relevance is 0.6, and the weighting coefficients are 0.4 and 0.6 respectively, then the weighted priori task logic relevance is 0.66.

[0059] Poll the logical relevance of each prior task and generate the weighted relevance of all solutions.

[0060] Arrange the logical relevance of each prior task in descending order to generate a task logical relevance list. Determine the solution with the highest relevance from the list, assuming it is solution 1.

[0061] The a priori first task logic knowledge graph and the a priori second task logic knowledge graph corresponding to Scheme 1 are used as the target a priori task logic knowledge graph, and the corresponding drone scheduling execution route "take off from the street lamp helipad near the city park, patrol along the roads around the park, and focus on areas with human activities" is used as the target drone scheduling execution route for the flight mission to be assigned.

[0062] However, to further optimize the decision, the server can also take the following actions:

[0063] Arrange the logical relevance of each prior task in descending order to generate a task logical relevance list. Determine a set number (e.g., the first three) of target prior task logical relevances from the list, and obtain the prior first task logical knowledge graph and the prior second task logical knowledge graph corresponding to the three target prior task logical relevances, and generate three reference prior task logical knowledge graphs.

[0064] According to the drone dispatch execution routes corresponding to the three reference priori task logic knowledge graphs, three reference drone dispatch execution routes are obtained, and the past application data corresponding to the three reference drone dispatch execution routes are obtained. For example, the execution route of scheme one successfully detected many abnormal situations in past applications, but there was also a data transmission interruption due to weather reasons; the execution route of scheme two performed well in the accuracy of data collection, but the flight time was long; the execution route of scheme three was more efficient in crowd monitoring, but less consideration was given to environmental factors.

[0065] The target past application data is determined from the past application data corresponding to the three reference UAV scheduling execution routes, and various factors such as anomaly detection success rate, data transmission stability, flight efficiency, etc. are comprehensively considered, assuming that the overall performance of Scheme 1 is the best. The reference UAV scheduling execution route corresponding to Scheme 1 is used as the target UAV scheduling execution route for the flight mission to be assigned.

[0066] This design can effectively process the flight missions to be assigned, generate the task logic knowledge graph using the task description, obtain the matching prior flight scheduling plan data, and finally determine the UAV scheduling execution route suitable for the current task.

[0067] Based on the above steps, the embodiment of the present application significantly improves the efficiency and accuracy of UAV task scheduling. By automatically generating a candidate task logic knowledge graph and performing knowledge point prediction, the core features and technical requirements of the flight mission to be assigned can be quickly captured, providing a solid data foundation for subsequent scheduling decisions. The matching algorithm is used to quickly screen out the solution that best matches the current task from the rich prior flight scheduling solution data, which not only greatly shortens the scheduling preparation time, but also ensures the rationality and feasibility of the scheduling solution. Finally, the UAV scheduling execution route corresponding to the matched prior task logic knowledge graph is directly applied to the task to be assigned, realizing the intelligentization and automation of UAV scheduling, effectively reducing errors and delays caused by human factors, and improving the overall efficiency and resource utilization of UAV operations. As a result, the flexibility and response speed of UAV applications can be improved.

[0068] In a possible implementation, step S120 includes:

[0069] Step S121, extracting a directed graph vector according to the first task description to generate a first directed graph vector, performing multi-level logic mapping according to the first directed graph vector to generate an explicit task logic diagram and an implicit task logic diagram.

[0070] Step S122: The explicit task logic graph and the implicit task logic graph are merged to generate the candidate first task logic knowledge graph.

[0071] In this embodiment, it is further assumed that the first task description received by the server is: "real-time traffic monitoring and congestion warning for urban transportation hubs".

[0072] The server first extracts the directed graph vector based on the first task description. It extracts keywords from the description, such as "urban transportation hub", "real-time traffic monitoring", and "congestion warning". Using the named entity recognition algorithm, it identifies the entity "city". Then it extracts the relationship between keywords and entities, such as "urban transportation hub" is the object of "real-time traffic monitoring" and "congestion warning".

[0073] The extracted keywords and entities are used as nodes of a directed graph, and each node is assigned multiple attributes. For example, the attributes of the node "urban transportation hub" may include "word itself: urban transportation hub", "part of speech: noun", and "importance weight: high".

[0074] Based on the results of relationship extraction, the directed edges between nodes are defined. For example, from "urban transportation hub" to "real-time traffic monitoring" and "congestion warning", the direction of the edge indicates the directionality of the relationship, and the weight of the edge indicates the strength of the relationship. Assuming that the relationship strength between "urban transportation hub" and "real-time traffic monitoring" is higher, the weight is set to 0.8; the relationship strength with "congestion warning" is slightly lower, and the weight is set to 0.6.

[0075] Next, the directed graph is converted into a vector representation using the adjacency matrix of the graph and the eigenvector of the Laplacian matrix to generate a first directed graph vector.

[0076] By analyzing the direct connection relationship in the first directed graph vector, the server extracts the explicit task logic. For example, "real-time traffic monitoring" directly provides data support for "congestion warning", which is a deterministic sequential relationship.

[0077] Then, using path analysis or community discovery algorithms, the server infers implicit task logic from the first directed graph vector. For example, by analyzing the indirect connections between nodes, it can be inferred that "congestion warning" may be related to the subsequent "traffic diversion plan formulation."

[0078] The explicit task logic and implicit task logic are generated into explicit task logic diagram and implicit task logic diagram respectively. In the explicit task logic diagram, the direct relationship from "real-time traffic monitoring" to "congestion warning" is clearly shown. In the implicit task logic diagram, the potential relationship from "congestion warning" to "traffic diversion plan formulation" is presented.

[0079] Finally, the explicit task logic graph and the implicit task logic graph are fused to generate a candidate first task logic knowledge graph. This fused knowledge graph comprehensively reflects the task logic relationship related to the first task description, including not only explicit direct logic but also potential indirect logic. For example, in the fused knowledge graph, "urban transportation hub" as a core node has clear and reasonable logical connections with "real-time traffic monitoring", "congestion warning" and potential "traffic diversion plan formulation".

[0080] Through such processing, the server provides a richer and more accurate basis for the subsequent prediction of task knowledge points and matching with the prior task logic knowledge graph, which helps to more efficiently determine the appropriate drone scheduling execution route for the assigned flight mission.

[0081] In a possible implementation, step S121 includes:

[0082] Step S1211, extracting keywords from the first task description, and the keywords will be used as nodes in a directed graph.

[0083] Step S1212: Use a named entity recognition algorithm to identify entities in the first task description, and extract the relationship between the keywords and the entities, where the relationship will constitute the edge of a directed graph.

[0084] Step S1213, taking the extracted keywords and entities as nodes of a directed graph, each node contains multiple attributes, and the attributes include the word itself, part of speech, and importance weight.

[0085] Step S1214, based on the relationship extraction result, define directed edges between nodes to generate a directed graph, where the direction of the edge represents the directionality of the relationship, and the weight of the edge represents the strength or confidence of the relationship.

[0086] Step S1215: Use the adjacency matrix of the graph and the eigenvector of the Laplacian matrix to convert the directed graph into a vector representation to generate a first directed graph vector.

[0087] Step S1216: extracting explicit task logic by analyzing the direct connection relationship in the first directed graph vector, wherein the explicit task logic is expressed as a deterministic instruction, condition or sequence relationship.

[0088] Step S1217: using a path analysis or a community discovery algorithm, inferring an implicit task logic from the first directed graph vector, wherein the implicit task logic is inferred through an indirect connection relationship between nodes in the first directed graph vector.

[0089] Step S1218, integrating the explicit task logic and the implicit task logic to generate an explicit task logic diagram and an implicit task logic diagram.

[0090] In this embodiment, it is further assumed that the first task description received by the server is: "Monitor the traffic flow during holidays and plan traffic diversion plans for roads around major urban attractions."

[0091] First, the server extracts keywords from the first task description, such as "major city attractions", "holidays", "traffic flow monitoring", and "traffic diversion plan planning".

[0092] Then, using the named entity recognition algorithm, the server identifies the entity "city" and extracts the relationship between keywords and entities. For example, "major attractions in the city" and "traffic flow monitoring" have an object-behavior relationship, and "holidays" have a temporal impact on "traffic flow monitoring."

[0093] Next, the extracted keywords and entities are used as nodes of a directed graph. For example, the attributes of the node "Main attractions in the city" may include "vocabulary itself: main attractions in the city", "part of speech: noun", and "importance weight: high"; the attributes of the node "holidays" may be "vocabulary itself: holidays", "part of speech: noun", and "importance weight: medium".

[0094] Based on the results of relationship extraction, we define directed edges between nodes and generate a directed graph. For example, from "Main Scenic Spots in the City" to "Traffic Flow Monitoring", the direction of the edge indicates that the former is the monitoring object of the latter, and the weight of the edge is set to 0.8, indicating that the relationship strength is strong. From "Holidays" to "Traffic Flow Monitoring", the weight is set to 0.6, indicating that holidays have a certain impact on traffic flow monitoring but it is relatively weak.

[0095] Afterwards, the directed graph is converted into a vector representation using the adjacency matrix of the graph and the eigenvector of the Laplacian matrix, thereby generating a first directed graph vector.

[0096] By analyzing the direct connection relationship in the first directed graph vector, the server extracts the explicit task logic. For example, it is clear that "traffic flow monitoring" is to provide data support for "diversion plan planning", which is a deterministic sequential relationship.

[0097] Using path analysis or community discovery algorithms, the server infers implicit task logic from the first directed graph vector. For example, through the indirect connection relationship between nodes, it is inferred that "traffic flow monitoring" may also be indirectly related to "future traffic planning adjustment."

[0098] Finally, the explicit task logic and implicit task logic are integrated. In the explicit task logic diagram, the direct logical relationship from "traffic flow monitoring" to "diversion plan planning" is clearly displayed. In the implicit task logic diagram, the potential logical relationship from "traffic flow monitoring" to "future traffic planning adjustment" is presented.

[0099] In this way, the logical relationship in the first task description can be fully and deeply understood, providing strong support for subsequent task processing and decision-making.

[0100] In a possible implementation, step S130 includes:

[0101] Step S131, decomposing the second task description into task steps, generating various drone task nodes, and obtaining task node features corresponding to the various drone task nodes.

[0102] Step S132, deriving the task logic according to the task node features corresponding to each drone task node, and generating a node association logic subgraph corresponding to each drone task node.

[0103] Step S133, merging the node association logic subgraphs corresponding to each drone task node to generate the candidate second task logic knowledge graph.

[0104] In this embodiment, it is further assumed that the second task description received by the server is: "Use a drone equipped with a high-definition camera and intelligent sensors to depart from a specific helipad in the city center and conduct an exterior inspection of buildings in the old city in the east of the city along a preset route between 9 a.m. and 11 a.m. The flight altitude is maintained at 100 meters, and photos are taken every 10 minutes, and data is transmitted to the control center in real time. When the battery level is less than 20%, go to a nearby backup charging point to charge. During the inspection, focus on cracks in the walls of the building, damage to the roof, and aging of the ancillary facilities."

[0105] The server first breaks down the second task description into task steps. The generated drone task nodes include "depart from the apron", "fly along the preset route", "photograph the exterior of the building", "real-time data transmission", "power monitoring", "go to the charging point to charge", "focus on key situations", etc.

[0106] Then, the task node features corresponding to each drone task node are obtained. For example, the features of the node "departing from the apron" may include "pre-takeoff inspection" and "obtaining takeoff instructions"; the features of "flying along the preset route" may include "following the instructions of the navigation system" and "maintaining a stable flight attitude"; the features of "shooting the exterior of the building" include "adjusting the camera angle" and "setting shooting parameters".

[0107] The task logic is deduced based on the task node characteristics corresponding to each UAV task node, and the node association logic subgraph corresponding to each UAV task node is generated.

[0108] Taking the "Building Appearance Photography" node as an example, it has a close logical connection with the "Real-time Data Transmission" node. Because the photos taken need to be transmitted in a timely manner, it is determined that the logical connection density between them is high, generating a high task node association degree. Based on this association degree, the influence evaluation is performed, and it is concluded that the influence coefficient of "Real-time Data Transmission" on "Building Appearance Photography" is large. This influence coefficient is fused with the task node feature of "Real-time Data Transmission" to generate a weighted task node feature, thereby generating a node association logic subgraph corresponding to the "Building Appearance Photography" node.

[0109] This processing is performed on each drone mission node to generate the corresponding node-related logical subgraph.

[0110] Finally, the node-related logical subgraphs corresponding to each drone task node are fused. For example, the subgraphs of "departing from the apron" and "flying along the preset route" are fused together through logical association, and the related subgraphs such as "shooting the exterior of the building", "real-time data transmission" and "power monitoring" are also connected and fused with each other. The final candidate second task logical knowledge graph comprehensively and clearly displays the logical relationship and dependency relationship between each link in the entire task, providing accurate guidance for subsequent task scheduling and execution.

[0111] In a possible implementation, step S132 includes:

[0112] Step S1321, determining the candidate task node features corresponding to the candidate UAV task nodes and each target task node features from the task node features corresponding to each UAV task node.

[0113] Step S1322, determine the logical connection density between the candidate task node characteristics and the characteristics of each target task node, generate the correlation degree of each target task node, perform influence evaluation based on the correlation degree of each target task node, and generate the influence coefficient of each target task node.

[0114] Step S1323, determine the fusion result of the influence coefficient of each target task node and the corresponding target task node feature, generate each target weighted task node feature, and generate the node association logic subgraph corresponding to the candidate UAV task node based on the each target weighted task node feature.

[0115] Step S1324, polling each of the drone task nodes to generate a node association logic subgraph corresponding to each of the drone task nodes.

[0116] In this embodiment, the following is an example of a very detailed scenario of the above steps, with the server as the execution subject:

[0117] Assume the following drone mission nodes and their corresponding mission node features:

[0118] Node "Takeoff Preparation": Features include "Check battery power", "Confirm communication connection", "Calibrate sensors", etc.

[0119] Node "fly according to the planned route": features include "receive navigation instructions", "adjust flight speed", "avoid obstacles", etc.

[0120] Node "Data Collection": Features include "Start collection device", "Set collection frequency", "Store collection data", etc.

[0121] First, the candidate task node features and target task node features of the candidate UAV task nodes are determined from the task node features corresponding to each UAV task node. Assuming that "data collection" is used as a candidate UAV task node, its candidate task node feature is "starting the collection device", and the related features of "take-off preparation" and "flying according to the predetermined route" are used as target task node features.

[0122] Then, determine the logical connection density of target task node features such as "starting the acquisition device" and "checking the battery power", "confirming the communication connection", and "receiving navigation instructions". For example, the logical connection between "starting the acquisition device" and "checking the battery power" is relatively close, because insufficient battery power may cause the acquisition device to fail to start normally, so a higher task node correlation is generated. The influence is evaluated based on the correlation of each target task node. For example, "checking the battery power" has a greater impact on "starting the acquisition device", generating a higher target task node influence coefficient.

[0123] Next, determine the fusion results of the impact coefficients of each target task node and the corresponding target task node features. For example, the impact coefficient of "checking the battery power" is high, and it is fused with the feature of "checking the battery power" to generate target weighted task node features, such as "focus on ensuring sufficient battery power to support the startup of the acquisition equipment."

[0124] According to these target-weighted task node characteristics, a node-association logic subgraph corresponding to the “data collection” candidate UAV task node is generated, which shows the close logical relationship between “data collection” and other related task nodes.

[0125] Poll other UAV mission nodes, for example, take "takeoff preparation" as a candidate node, repeat the above process, determine its correlation with other node features, influence coefficient, etc., and generate the corresponding node association logic subgraph.

[0126] For "takeoff preparation", it is also closely logically connected with "flying according to the planned route". For example, "confirming the communication connection" is crucial for the subsequent receipt of navigation instructions. By calculating the correlation and influence coefficient, the corresponding weighted task node features and node association logic subgraphs are generated.

[0127] Through such polling processing, a node-associated logic subgraph is generated for each drone mission node, which accurately reflects its logical relationship with other nodes, providing a solid foundation for building a complete mission logic knowledge graph.

[0128] In another possible implementation, step S132 may further include:

[0129] Step S1325, obtaining the execution order information of each drone task node in the second task description, performing feature representation on the execution order information, and generating a task order coding vector corresponding to each drone task node.

[0130] Step S1326, obtaining the task type information corresponding to each drone task node, performing feature representation on the task type information, and generating a task type coding vector corresponding to each drone task node.

[0131] Step S1327, fusing the task node features, the task order coding vector and the task type coding vector to generate target task node features corresponding to each UAV task node.

[0132] Step S1328, deriving the task logic according to the characteristics of the target task node, and generating a node association logic subgraph corresponding to each drone task node.

[0133] In this embodiment, it is further assumed that the second task description received by the server is: "Use a drone equipped with a meteorological monitoring device to take off from a designated helipad at the edge of the city, first fly to the sky above a park in the city center along a preset arc route, then hover over the park to collect meteorological data, record data every 5 minutes during the collection process, and return to the helipad after the collection is completed."

[0134] First, the server obtains the execution order information of each drone task node in this second task description. For example, "take off from the designated helipad" is the first task to be executed, and its order is coded as 1; "fly over the park according to the preset arc route" is the second task to be executed, coded as 2; "hover over the park to collect meteorological data" is the third task to be executed, coded as 3; "return to the helipad" is the last task to be executed, coded as 4. These execution order information are represented by features to generate the task order encoding vector corresponding to each drone task node.

[0135] Next, the server obtains the task type information corresponding to each drone task node. "Take off from the designated helipad" belongs to the take-off task type, "fly over the park according to the preset arc route" belongs to the flight navigation task type, "hover over the park to collect meteorological data" belongs to the data collection task type, and "return to the helipad" belongs to the return task type. These task type information are represented by features to generate the task type encoding vector corresponding to each drone task node.

[0136] Then, the server fuses the task node features (for example, the features of the "take-off from the designated apron" node may include "check take-off equipment", "confirm take-off conditions", etc.), the task sequence coding vector, and the task type coding vector. Taking the "take-off from the designated apron" node as an example, the fused target task node features may include "check take-off equipment", "confirm take-off conditions", "sequence code 1", and "take-off task type code".

[0137] Based on the characteristics of these fused target task nodes, the server performs task logic deduction. For example, because "take off from the designated apron" is at the front of the order and belongs to the take-off task type, it has a close logical relationship with the subsequent "fly over the park according to the preset arc route" node, that is, subsequent flights can only be carried out after the take-off is successful. Based on this logical deduction, the server generates a node association logic subgraph corresponding to the "take off from the designated apron" node, which shows its logical association with other related nodes.

[0138] Each UAV mission node is processed according to these steps to generate the corresponding node association logic subgraph, thereby constructing a complete mission logic relationship network to provide accurate guidance for subsequent task scheduling and execution.

[0139] In a possible implementation manner, before step S110, the method further includes:

[0140] Step S101, obtaining each priori task description and the corresponding drone scheduling execution route.

[0141] Step S102, extracting the task logic knowledge graph corresponding to each priori task description, and generating the priori task logic knowledge graph corresponding to each priori task description.

[0142] Step S103, associating the a priori task logic knowledge graphs corresponding to the various a priori task descriptions with the corresponding UAV scheduling execution routes, to generate the various a priori flight scheduling plan data.

[0143] In this embodiment, a large amount of flight mission data has been accumulated in advance. These flight mission data include a priori mission descriptions of different types, different scenarios, and different requirements, as well as the UAV scheduling execution route corresponding to each mission.

[0144] First, the server obtains a series of prior task descriptions and corresponding drone scheduling execution routes.

[0145] For example, there is a priori task description: "Conduct daytime safety inspections and equipment status monitoring of large factories on the outskirts of the city." The corresponding drone scheduling execution route is "take off from the rooftop apron of the warehouse near the factory, fly along the factory wall first, and then inspect the production areas inside the factory in turn, focusing on flammable and explosive areas and large equipment, transmitting data to the control center every 20 minutes, and returning to the rooftop apron of the warehouse to charge when the battery level is less than 25%."

[0146] There is also a priori task description: "To conduct fire hazard inspections and ecological environment monitoring in forests in mountainous areas." The corresponding drone dispatch execution route is "to take off from the fire station helipad at the foot of the mountain, fly along the preset uphill route, monitor the tree density, humidity, temperature, etc. in the forest, pay special attention to any signs of smoke or open flames, transmit data every 10 minutes, and go to the temporary charging point halfway up the mountain to charge when the battery level is less than 35%."

[0147] Another a priori task description is: "To conduct nighttime supervision and statistics on ships and cargo in coastal ports." The corresponding drone dispatch execution route is "to take off from the rooftop apron of the port management office, patrol the port's divisions in turn, record the number, model, type and quantity of ships, and monitor whether there are abnormal floating objects on the sea surface. Data will be sent to the port control center every 15 minutes, and it will automatically return to the rooftop apron for charging when the battery power is less than 20%."

[0148] The server starts to extract the task logic knowledge graph corresponding to each prior task description.

[0149] For the task description of "conducting daytime safety inspections and equipment status monitoring of large factories outside the city", the server first extracts keywords, such as "outskirts of the city", "large factories", "daytime", "safety inspection", "equipment status monitoring", etc. These keywords become nodes in the directed graph. Then, using the named entity recognition algorithm, entities such as "city" and "factory" are identified and the relationship between them is extracted. For example, "outskirts of the city" and "large factories" have a location relationship, and "daytime" has a time and behavior relationship with "safety inspection" and "equipment status monitoring".

[0150] Next, the server uses these keywords and entities as nodes of a directed graph. Each node contains rich attributes. For example, the attributes of the node "city suburbs" include "vocabulary itself: city suburbs", "part of speech: noun phrase", and "importance weight: medium". Based on the results of relationship extraction, the directed edges between nodes are defined. For example, from "city suburbs" to "large factory", the weight of the edge represents the strength of the relationship, which is assumed to be 0.7 here. Use the adjacency matrix of the graph and the eigenvector of the Laplace matrix to convert the directed graph into a vector representation to generate the first directed graph vector.

[0151] By analyzing the direct connection relationship in the first directed graph vector, the server extracts the explicit task logic, such as "safety inspection" depends on the results of "equipment status monitoring". Using path analysis or community discovery algorithms, the implicit task logic is inferred from the first directed graph vector, such as "equipment status monitoring" may be related to the subsequent "maintenance plan formulation".

[0152] Integrate the explicit task logic and the implicit task logic to generate the task logic knowledge graph of this priori task description, that is, the priori first task logic knowledge graph.

[0153] For the task description of "carry out fire hazard inspection and ecological environment monitoring in forests in mountainous areas", the server also performs similar processing. It extracts keywords such as "mountainous area", "forest", "fire hazard inspection", and "ecological environment monitoring", identifies the entity "mountainous area", and extracts the relationship between keywords and entities. These are used as nodes of a directed graph, and directed edges between nodes are defined to generate directed graph vectors, extract explicit and implicit task logic, and finally generate a priori first task logic knowledge graph.

[0154] The task description of "monitoring and counting ships and cargoes at coastal ports at night" is also processed according to the same process to generate the corresponding a priori first task logical knowledge graph.

[0155] After generating the a priori first task logical knowledge graphs corresponding to the various a priori task descriptions, the server continues to process the a priori second task descriptions.

[0156] For example, for the above-mentioned task of "conducting daytime safety inspections and equipment status monitoring of large factories on the outskirts of the city", the corresponding prior second task description may be "using a drone equipped with a high-definition thermal imager and a gas sensor to take off from the rooftop helipad of a warehouse near the factory between 9 a.m. and 5 p.m., maintaining a flight speed of 10 meters per second, and conduct detailed scans of key areas within the factory, conduct real-time analysis of the collected data, and immediately issue an alarm if any abnormal situation is found."

[0157] The server breaks down the task steps of this a priori second task description. First, it breaks it down into drone task nodes such as "takeoff preparation", "fly at a set speed", "scan key areas", "real-time data analysis", and "abnormal alarm". Then it obtains the characteristics of each task node. For example, the characteristics of the "takeoff preparation" node are "checking the drone components" and "confirming that the sensor is normal", etc.; the characteristics of the "fly at a set speed" node are "receiving flight instructions" and "controlling flight speed".

[0158] For each drone task node, the server determines the candidate task node characteristics and the target task node characteristics. Taking the "key area scanning" node as an example, the candidate task node characteristics may be "select scanning accuracy", and the target task node characteristics may be "determine the scope of key areas" and "set scanning frequency". Determine the logical connection density between the candidate task node characteristics and the characteristics of each target task node. For example, "select scanning accuracy" and "determine the scope of key areas" are closely related, generating a higher task node correlation. Perform influence assessment based on the correlation and obtain the influence coefficient of the target task node. Fusion the influence coefficient with the target task node characteristics to generate weighted task node characteristics, thereby generating a node association logic subgraph corresponding to the "key area scanning" node.

[0159] Poll each drone task node and repeat the above process to generate the node association logic subgraph corresponding to each drone task node. These subgraphs are merged to generate the candidate second task logic knowledge graph.

[0160] For other prior task descriptions, the corresponding prior second task logical knowledge graphs are generated in a similar manner.

[0161] The server associates the generated a priori first task logic knowledge graph and a priori second task logic knowledge graph with the corresponding drone scheduling execution routes respectively.

[0162] For example, the a priori first task logic knowledge graph of "conducting daytime safety inspections and equipment status monitoring of large factories on the outskirts of the city" and the a priori second task logic knowledge graph are associated with the drone scheduling execution route of "taking off from the rooftop helipad of a warehouse near the factory, flying along the factory wall first, and then inspecting the production areas inside the factory in turn, focusing on flammable and explosive areas and large equipment, transmitting data to the control center every 20 minutes, and returning to the rooftop helipad of the warehouse to charge when the battery level is less than 25%" to form a complete a priori flight scheduling plan data.

[0163] For task descriptions such as "inspecting fire hazards and monitoring the ecological environment in forests in mountainous areas" and "monitoring and counting ships and cargoes in coastal ports at night", their a priori first task logic knowledge graph and a priori second task logic knowledge graph are also associated with the corresponding UAV scheduling execution routes to generate their respective a priori flight scheduling plan data.

[0164] In this way, the processing of each prior task description is completed, and a series of prior flight scheduling plan data is generated, which provides rich reference and reference for subsequent new flight tasks to be assigned. When receiving a new flight task to be assigned, the server can analyze and process it based on these prior data and quickly generate a suitable drone scheduling execution route.

[0165] In a possible implementation, the a priori task description includes a a priori first task description and a a priori second task description.

[0166] Step S102 includes:

[0167] Step S1021, extracting the task logic knowledge graph of the a priori first task description to generate the a priori first task logic knowledge graph, and extracting the task logic knowledge graph of the a priori second task description to generate the a priori second task logic knowledge graph.

[0168] Step S1022: Generate the a priori task logic knowledge graph according to the a priori first task logic knowledge graph and the a priori second task logic knowledge graph.

[0169] In this embodiment, let's first look at a priori task description, where the first prior task description is: "Monitor the flow of people and consumption behavior in the commercial districts in the city center."

[0170] The server first processes this a priori first task description. It extracts the keywords "city center", "commercial district", "traffic flow", "consumption behavior", and "monitoring". Using the named entity recognition algorithm, it identifies entities such as "city" and "commercial district". Then, it extracts the relationship between these keywords and entities. For example, "city center" and "commercial district" have a location relationship, and "traffic flow" and "consumption behavior" are the objects of "monitoring".

[0171] These keywords and entities are used as nodes of a directed graph, and each node contains detailed attributes, such as the attributes of "city center" include "vocabulary itself: city center", "part of speech: noun phrase", and "importance weight: high". According to the results of relationship extraction, the directed edges between nodes are defined, and the weight of the edge reflects the strength of the relationship. Then, the directed graph is converted into a vector representation through the graph-related algorithm to generate the first directed graph vector.

[0172] Analyze the direct connection relationship in the first directed graph vector to extract the explicit task logic, such as "monitoring" is to obtain data on "traffic flow" and "consumption behavior". Use path analysis or community discovery algorithms to infer implicit task logic, such as the monitoring results may be related to the subsequent "marketing strategy adjustment".

[0173] Integrate explicit and implicit task logic to generate a priori first task logic knowledge graph for this priori first task description.

[0174] Let’s look at the a priori second task description of this a priori task description: “Use a drone equipped with a high-definition camera and intelligent analysis software to take off from the apron on the top of an office building in a commercial district, fly along a preset patrol route between 10 a.m. and 8 p.m., collect data every 30 minutes and transmit it in real time, and automatically return to the apron to charge when the battery level drops below 40%.”

[0175] The server breaks down the a priori second task description into drone task nodes such as "takeoff preparation", "flying according to the route", "data collection and transmission", "power monitoring and return", etc. The features of each task node are obtained. For example, the features of "takeoff preparation" include "checking drone equipment" and "confirming that the software is normal".

[0176] Determine the candidate task node characteristics and target task node characteristics of each drone task node. For example, for the "data collection and transmission" node, the candidate task node characteristics are "selecting the appropriate data collection frequency", and the target task node characteristics are "determining the transmission protocol" and "ensuring data security". Calculate the logical connection density between the candidate task node characteristics and the target task node characteristics, obtain the task node association, and then evaluate the influence to obtain the influence coefficient. Fusion the influence coefficient with the target task node characteristics to generate weighted task node characteristics, thereby generating a node association logic subgraph.

[0177] Poll each drone task node to generate the corresponding node-related logic subgraphs. Fuse these subgraphs to generate the prior second task logic knowledge graph.

[0178] With the prior first task logic knowledge graph and the prior second task logic knowledge graph, the server further integrates them. For example, it is found that the "monitoring" requirement in the prior first task logic knowledge graph is closely related to the "data collection and transmission" link in the prior second task logic knowledge graph, and this association is clearly reflected in the integrated prior task logic knowledge graph.

[0179] Let’s look at another prior task description. The prior first task description is: “Conduct environmental quality and production safety inspections on industrial parks around the city.”

[0180] The server repeats the above extraction, analysis and integration process to generate the corresponding a priori first task logical knowledge graph.

[0181] The corresponding a priori second task description is: "Use drones equipped with multiple sensors to depart from the dedicated helipad in the park, patrol along the planned route within a specific time period, analyze the collected data in real time, and promptly report an emergency and return home."

[0182] The server also processes the a priori second task description to generate the a priori second task logic knowledge graph, and then integrates it with the a priori first task logic knowledge graph to obtain a complete a priori task logic knowledge graph.

[0183] By processing the descriptions of numerous prior missions in this way, a rich and comprehensive prior mission logic knowledge graph system is constructed, which provides strong support and reference for the analysis and scheduling of subsequent new flight missions.

[0184] In a possible implementation, the a priori task logic knowledge graph includes an a priori first task logic knowledge graph and an a priori second task logic knowledge graph.

[0185] Step S150 includes:

[0186] Step S151, determining the task logic associations between the candidate first task logic knowledge graph and each priori first task logic knowledge graph, and generating each priori first task logic association.

[0187] Step S152, determining the task logic associations between the candidate second task logic knowledge graph and each priori second task logic knowledge graph, and generating each priori second task logic association.

[0188] Step S153 , performing weighted calculation according to each of the a priori first task logical associations and each of the a priori second task logical associations to generate each of the a priori task logical associations.

[0189] Step S154, arrange the various prior task logical associations in descending order to generate a task logical association list, determine the target prior task logical associations of the target list order from the task logical association list, use the prior first task logical knowledge graph and the prior second task logical knowledge graph corresponding to the target prior task logical associations of the target list order as the target prior first task logical knowledge graph and the target prior second task logical knowledge graph, and generate the target prior task logical knowledge graph based on the target prior first task logical knowledge graph and the target prior second task logical knowledge graph.

[0190] In this embodiment, the server receives a flight mission to be assigned, and its candidate first task logic knowledge graph is described as "monitoring the construction progress and safety status of construction sites in new urban areas", and the candidate second task logic knowledge graph describes the detailed execution method, such as "using a drone equipped with specific monitoring equipment, taking off from a designated helipad during a specific time period, flying along a scheduled route, and transmitting monitoring data in real time, etc."

[0191] The server first compares the candidate first task logic knowledge graph with each prior first task logic knowledge graph in the database. Assume that the following prior first task logic knowledge graphs exist in the database:

[0192] 1. "Monitoring the construction progress and safety status of renovation sites in old urban areas"

[0193] 2. "Construction progress and quality monitoring of factory construction sites in suburban areas"

[0194] The server determines the task logic association between the candidate first task logic knowledge graph and each prior first task logic knowledge graph by analyzing the similarity and correlation of elements such as keywords, entities, and relationships. The association with the first prior graph may be determined to be 0.7, and the association with the second prior graph may be 0.6, thereby generating each prior first task logic association.

[0195] Next, the server compares the candidate second task logic knowledge graph with each prior second task logic knowledge graph in the database. Assume that the following prior second task logic knowledge graph exists:

[0196] 1. "Use similar monitoring equipment to monitor construction sites in old urban areas, with the same flight and data transmission methods"

[0197] 2. "Using different monitoring equipment to monitor factory sites in suburban areas, the flight and data transmission methods are different"

[0198] Similarly, through detailed analysis, the task logic correlation between the candidate second task logic knowledge graph and each prior second task logic knowledge graph is determined, for example, the correlation with the first prior graph is 0.8, and the correlation with the second prior graph is 0.5, and each prior second task logic correlation is generated.

[0199] Then, the server performs weighted calculation on the a priori first task logical relevance and the a priori second task logical relevance according to the set weighting coefficients. Assuming the weighting coefficients are 0.4 and 0.6 respectively, for the above example, the weighted relevance of the first a priori task is 0.7*0.4+0.8*0.6=0.76, and the weighted relevance of the second a priori task is 0.6*0.4+0.5*0.6=0.54, thereby generating the logical relevance of each a priori task.

[0200] The server sorts the logical relevance of each prior task in descending order and generates a task logical relevance list. Assume that the sorted list is: [0.76, 0.54].

[0201] From the task logic relevance list, the server determines the target list order, such as selecting the one with the highest relevance first, and uses the corresponding a priori first task logic knowledge graph and a priori second task logic knowledge graph as the target a priori first task logic knowledge graph and the target a priori second task logic knowledge graph.

[0202] Finally, the final target priori task logic knowledge graph is generated based on the target priori first task logic knowledge graph and the target priori second task logic knowledge graph, which provides reference and reference for the flight mission to be assigned and helps determine the appropriate scheduling execution route.

[0203] In a possible implementation, step S153 includes:

[0204] Step S1531 , determining a candidate first task logical association degree from the a priori first task logical association degrees, and determining a candidate second task logical association degree corresponding to the candidate first task logical association degree from the a priori second task logical association degrees.

[0205] Step S1532, performing weighted calculation based on the candidate first task logical association and the candidate second task logical association to generate a weighted task logical association, and using the weighted task logical association as the prior task logical association corresponding to the candidate first task logical association and the candidate second task logical association.

[0206] Step S1533 , polling each of the a priori first task logical associations and the corresponding a priori second task logical associations to generate each of the a priori task logical associations.

[0207] This embodiment has calculated a series of a priori first task logical associations and a priori second task logical associations.

[0208] Assume that the a priori logical relevance of the first task is 0.8, 0.7, 0.6, and 0.5, respectively, and the corresponding a priori logical relevance of the second task is 0.7, 0.8, 0.6, and 0.7, respectively.

[0209] First, the server determines a candidate first task logical association degree from each priori first task logical association degree, for example, first selects 0.8. Then, from each priori second task logical association degree, the server finds a corresponding candidate second task logical association degree, that is, 0.7.

[0210] Next, the server sets weighting coefficients, assuming that the a priori weighting coefficient of the first task logic relevance is 0.6, and the a priori weighting coefficient of the second task logic relevance is 0.4.

[0211] Based on these two weighting coefficients, the candidate first task logical relevance of 0.8 and the candidate second task logical relevance of 0.7 are weighted calculated:

[0212] Weighted task logic relevance = 0.8 × 0.6 + 0.7 × 0.4 = 0.48 + 0.28 = 0.76

[0213] The weighted task logic relevance of 0.76 is used as the priori task logic relevance corresponding to the candidate first task logic relevance of 0.8 and the candidate second task logic relevance of 0.7.

[0214] Then, the server polls other a priori first task logical associations and corresponding a priori second task logical associations.

[0215] For example, the next candidate first task is selected with a logical relevance of 0.7, and the corresponding candidate second task has a logical relevance of 0.8.

[0216] The same weighted calculation is performed:

[0217] Weighted task logic correlation & = 0.7 × 0.6 + 0.8 × 0.4 = 0.42 + 0.32 = 0.74

[0218] By analogy, such weighted calculation is performed on each group of candidate a priori first task logical associations and a priori second task logical associations to generate all a priori task logical associations.

[0219] In this way, the correlation between the a priori first task and the a priori second task and the current flight task to be assigned can be comprehensively considered, providing an accurate basis for the subsequent screening of the most matching target a priori task logical knowledge graph.

[0220] In another possible implementation, step S153 may further include:

[0221] Step S1534, determining a candidate first task logical association degree from the a priori first task logical association degrees, and determining a candidate second task logical association degree corresponding to the candidate first task logical association degree from the a priori second task logical association degrees.

[0222] Step S1535, compare the candidate first task logical association degree and the candidate second task logical association degree, when the candidate first task logical association degree is greater than the candidate second task logical association degree, use the candidate first task logical association degree as the prior task logical association degree corresponding to the candidate first task logical association degree and the candidate second task logical association degree.

[0223] Step S1536: when the candidate first task logical association is not greater than the candidate second task logical association, the candidate second task logical association is used as the priori task logical association corresponding to the candidate first task logical association and the candidate second task logical association.

[0224] Step S1537 , polling each of the a priori first task logical associations and the corresponding a priori second task logical associations to generate each of the a priori task logical associations.

[0225] This embodiment has derived a series of a priori first task logical associations and a priori second task logical associations.

[0226] Assume that the a priori logical relevance of the first task is 0.85, 0.72, 0.68, and 0.55, respectively, and the corresponding a priori logical relevance of the second task is 0.78, 0.82, 0.65, and 0.70, respectively.

[0227] First, the server determines the candidate first task logical association degree from each prior first task logical association degree, for example, 0.85 is selected first, and then finds the corresponding candidate second task logical association degree from each prior second task logical association degree, that is, 0.78.

[0228] Next, the server compares the two correlations. Since 0.85 is greater than 0.78, 0.85 is used as the prior task logic correlation corresponding to this set of candidate correlations.

[0229] Then, the server continues to poll, and selects the next candidate first task logic relevance of 0.72, and the corresponding candidate second task logic relevance of 0.82. Because 0.72 is less than 0.82, 0.82 is used as the priori task logic relevance of this group.

[0230] Then select 0.68 as the candidate first task logic relevance, and the corresponding 0.65 as the candidate second task logic relevance. Since 0.68 is greater than 0.65, the corresponding prior task logic relevance of this group is 0.68.

[0231] By analogy, the server performs such comparison and determination on each set of a priori first task logical associations and the corresponding a priori second task logical associations, and finally generates all a priori task logical associations.

[0232] In this way, the server can accurately determine the prior task logical correlation corresponding to each combination based on the size relationship between the prior first task logical correlation and the prior second task logical correlation, providing an important basis for subsequent screening of the target prior task logical knowledge graph that best matches the flight mission to be assigned.

[0233] In a possible implementation, the method further includes:

[0234] Step A110, arranging the a priori task logic relevances in descending order to generate a task logic relevance list.

[0235] Step A120, determine a set number of target priori task logical associations from the task logical association list, and obtain the a priori first task logical knowledge graph and the a priori second task logical knowledge graph corresponding to the set number of target priori task logical associations, and generate a set number of reference priori task logical knowledge graphs.

[0236] Step A130, obtaining each reference drone scheduling execution route according to the drone scheduling execution routes corresponding to the set number of reference priori task logic knowledge graphs, and obtaining past application data corresponding to each reference drone scheduling execution route.

[0237] Step A140, determining target past application data from the past application data corresponding to the reference UAV scheduling execution routes, and using the reference UAV scheduling execution route corresponding to the target past application data as the target UAV scheduling execution route for the flight mission to be assigned.

[0238] In this embodiment, multiple priori task logical associations have been calculated, for example: 0.85, 0.78, 0.92, 0.65, and 0.72.

[0239] The server first arranges these prior task logic relevances in descending order, and generates a task logic relevance list as follows: 0.92, 0.85, 0.78, 0.72, 0.65.

[0240] Assuming the set number is 3, the server determines the first three target prior task logical associations from this task logical association list, which are 0.92, 0.85, and 0.78 respectively.

[0241] Then, the server obtains the a priori first task logic knowledge graph and the a priori second task logic knowledge graph corresponding to the three target a priori task logical associations, and generates three reference a priori task logic knowledge graphs.

[0242] Next, the server obtains each reference drone scheduling execution route based on the drone scheduling execution routes corresponding to the three reference priori task logic knowledge graphs. Assume that the three reference drone scheduling execution routes are:

[0243] Route 1: Take off from location A and fly along route X, transmitting data every 15 minutes. When the battery level drops below 30%, return to location A to charge.

[0244] Route 2: Take off from location B and fly along route Y, transmitting data every 20 minutes. When the battery level drops below 25%, go to location C to charge.

[0245] Route 3: Take off from point D and fly along route Z, transmitting data every 10 minutes. When the battery level drops below 40%, return to point D to charge.

[0246] The server obtains the past application data corresponding to the three reference drone dispatch execution routes. For example:

[0247] Past application data of Route 1 showed that data transmission was stable during the execution of this route, but it encountered bad weather which extended the flight time.

[0248] Past application data of Route 2 shows that it performs well in data collection accuracy, but there was a minor collision with an obstacle during the flight.

[0249] Past application data of Route 3 shows that its flight efficiency is high, but there was a mistake in power management that caused it to return early.

[0250] The server comprehensively analyzes these past application data and determines the target past application data from the past application data corresponding to each reference drone scheduling execution route. For example, after comprehensive consideration, it is considered that although route 1 encountered bad weather, the overall performance was relatively balanced and stable.

[0251] Finally, the server uses the reference UAV scheduling execution route (i.e., route 1) corresponding to the target past application data as the target UAV scheduling execution route for the flight mission to be assigned.

[0252] Further, Figure 2 FIG. 1 is a schematic diagram of the hardware structure of a drone dispatching system 100 based on artificial intelligence for implementing the method provided in the embodiment of the present application. Figure 2 As shown, the artificial intelligence-based drone scheduling system 100 includes a memory 111, a storage controller 112, a processor 113, a peripheral interface 114, an input and output unit 115, an audio unit 116, a display unit 117 and a radio frequency unit 118.

[0253] The memory 111, the storage controller 112, the processor 113, the peripheral interface 114, the input / output unit 115, the audio unit 116, the display unit 117, and the radio frequency unit 118 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines.

[0254] The memory 111 may be, but not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electric erasable programmable read-only memory (EEPROM), etc. The memory 111 is used to store the application 210, and the processor 113 executes the application 210 after receiving the execution instruction. The processor 113 and other possible components may access the memory 111 under the control of the storage controller 112.

[0255] The processor 113 may be an integrated circuit chip with signal processing capabilities. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0256] The peripheral interface 114 couples various input / output devices to the processor 113 and the memory 111. In some embodiments, the peripheral interface 114, the processor 113 and the memory controller 112 can be implemented in a single chip. In other embodiments, they can be implemented by separate chips.

[0257] The input-output unit 115 is used to provide input data to the user to enable the user to interact with the artificial intelligence-based drone scheduling system 100. The input-output unit 115 can be, but is not limited to, a mouse and a keyboard.

[0258] The audio unit 116 provides an audio interface to the user and may include one or more microphones, one or more speakers, and an audio circuit.

[0259] The display unit 117 provides an interactive interface (e.g., a user operation interface) between the artificial intelligence-based drone dispatching system 100 and the user or is used to display image data. In this embodiment, the display unit 117 can be a liquid crystal display or a touch display. If it is a touch display, it can be a capacitive touch screen or a resistive touch screen that supports single-point and multi-point touch operations. Supporting single-point and multi-point touch operations means that the touch display can sense touch operations generated from one or more positions on the touch display, and the sensed touch operations are handed over to the processor for calculation and processing.

[0260] The radio frequency unit 118 is used to receive and send radio wave signals (such as electromagnetic waves) to achieve mutual conversion between radio waves and electrical signals, thereby realizing communication between the artificial intelligence-based drone scheduling system 100 and the network or other communication equipment.

[0261] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of the embodiments of the present application are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed according to an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0262] Each embodiment in the embodiments of the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above different embodiments, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0263] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program. The above program may be stored in a computer-readable storage medium, and the above storage medium may be a read-only memory, a disk or an optical disk, etc.

Claims

1. A drone dispatching method based on artificial intelligence, characterized in that: The method comprises: Acquire a flight task to be assigned, wherein the flight task to be assigned includes a first task description and a corresponding second task description, wherein the first task description is used to summarize task feature information of the second task description; Generate a task logic knowledge graph according to the first task description to generate a candidate first task logic knowledge graph, and predict task knowledge points according to the candidate first task logic knowledge graph to generate a first task knowledge point label; Generate a task logic knowledge graph according to the second task description to generate a candidate second task logic knowledge graph, and predict task knowledge points according to the candidate second task logic knowledge graph to generate a second task knowledge point label; Acquire each priori flight scheduling solution data that matches the first task knowledge point label and the second task knowledge point label respectively, wherein the priori flight scheduling solution data includes a priori task logic knowledge graph and a corresponding UAV scheduling execution route; Extracting a target priori task logic knowledge graph matching the flight task to be assigned from each priori task logic knowledge graph based on the candidate first task logic knowledge graph and the candidate second task logic knowledge graph; The UAV scheduling execution route corresponding to the target priori task logic knowledge graph is used as the target UAV scheduling execution route of the flight task to be assigned; The step of generating a task logic knowledge graph according to the first task description to generate a candidate first task logic knowledge graph includes: Extracting a directed graph vector according to the first task description to generate a first directed graph vector, and performing multi-level logic mapping according to the first directed graph vector to generate an explicit task logic diagram and an implicit task logic diagram; Merging the explicit task logic graph and the implicit task logic graph to generate the candidate first task logic knowledge graph; The steps of extracting a directed graph vector according to the first task description to generate a first directed graph vector, performing multi-level logic mapping according to the first directed graph vector, and generating an explicit task logic diagram and an implicit task logic diagram include: Extracting keywords from the first task description, wherein the keywords are used as nodes in a directed graph; Using a named entity recognition algorithm to identify entities in the first task description, and extracting relationships between the keywords and the entities, where the relationships will constitute edges of a directed graph; The extracted keywords and entities are used as nodes of a directed graph, each node containing multiple attributes, including the word itself, part of speech, and importance weight; According to the result of relationship extraction, directed edges between nodes are defined to generate a directed graph, wherein the direction of the edge represents the directionality of the relationship, and the weight of the edge represents the strength or confidence of the relationship; Convert the directed graph into a vector representation using the adjacency matrix of the graph and the eigenvector of the Laplacian matrix to generate a first directed graph vector; By analyzing the direct connection relationship in the first directed graph vector, an explicit task logic is extracted, wherein the explicit task logic is expressed as a deterministic instruction, condition or sequence relationship; Inferring implicit task logic from the first directed graph vector by using a path analysis or a community discovery algorithm, wherein the implicit task logic is inferred through an indirect connection relationship between nodes in the first directed graph vector; Integrating the explicit task logic and the implicit task logic to generate an explicit task logic diagram and an implicit task logic diagram; The step of generating a task logic knowledge graph according to the second task description to generate a candidate second task logic knowledge graph includes: Decomposing the second task description into task steps, generating each drone task node, and obtaining task node features corresponding to each drone task node; Deducing task logic according to the task node features corresponding to each drone task node, and generating a node association logic subgraph corresponding to each drone task node; The node association logic subgraphs corresponding to each drone task node are merged to generate the candidate second task logic knowledge graph; The step of performing task logic deduction according to the task node features corresponding to each drone task node to generate a node association logic subgraph corresponding to each drone task node includes: Determine the candidate task node features and each target task node features corresponding to the candidate UAV task node from the task node features corresponding to each UAV task node; Determine the logical connection density between the candidate task node characteristics and the characteristics of each target task node, generate the correlation degree of each target task node, perform influence evaluation based on the correlation degree of each target task node, and generate the influence coefficient of each target task node; Determine the fusion result of the influence coefficient of each target task node and the corresponding target task node feature, generate each target weighted task node feature, and generate a node association logic subgraph corresponding to the candidate UAV task node according to each target weighted task node feature; Polling each of the drone task nodes to generate a node association logic subgraph corresponding to each of the drone task nodes; The step of performing task logic deduction according to the task node features corresponding to each drone task node to generate a node association logic subgraph corresponding to each drone task node includes: Obtaining execution order information of each drone task node in the second task description, performing feature representation on the execution order information, and generating task order encoding vectors corresponding to each drone task node; Obtaining task type information corresponding to each drone task node, performing feature representation on the task type information, and generating a task type encoding vector corresponding to each drone task node; The task node feature, the task order coding vector and the task type coding vector are merged to generate target task node features corresponding to each UAV task node; Deducing the task logic according to the target task node characteristics, and generating a node association logic subgraph corresponding to each drone task node; Before obtaining the flight mission to be assigned, the method further includes: Obtain each prior task description and the corresponding drone scheduling execution route; Extracting the task logic knowledge graph corresponding to each priori task description, and generating the priori task logic knowledge graph corresponding to each priori task description; Associating the a priori task logic knowledge graphs corresponding to the a priori task descriptions with the corresponding UAV scheduling execution routes respectively, to generate the a priori flight scheduling plan data; The a priori task description includes a priori first task description and a priori second task description; The extracting the task logic knowledge graph corresponding to each priori task description and generating the priori task logic knowledge graph corresponding to each priori task description includes: Extracting the task logic knowledge graph of the a priori first task description to generate the a priori first task logic knowledge graph, and extracting the task logic knowledge graph of the a priori second task description to generate the a priori second task logic knowledge graph; Generating the a priori task logic knowledge graph according to the a priori first task logic knowledge graph and the a priori second task logic knowledge graph; The a priori task logic knowledge graph includes a a priori first task logic knowledge graph and a a priori second task logic knowledge graph; The step of extracting a target priori task logic knowledge graph matching the to-be-assigned flight task from each priori task logic knowledge graph based on the candidate first task logic knowledge graph and the candidate second task logic knowledge graph includes: Determine the task logic association degree between the candidate first task logic knowledge graph and each priori first task logic knowledge graph, and generate each priori first task logic association degree; Determine the task logic association degree between the candidate second task logic knowledge graph and each priori second task logic knowledge graph, and generate each priori second task logic association degree; Performing weighted calculation according to each a priori first task logical association degree and each a priori second task logical association degree to generate each a priori task logical association degree; Arrange the various prior task logical associations in descending order to generate a task logical association list, determine the target prior task logical associations in the target list order from the task logical association list, use the prior first task logical knowledge graph and the prior second task logical knowledge graph corresponding to the target prior task logical associations in the target list order as the target prior first task logical knowledge graph and the target prior second task logical knowledge graph, and generate the target prior task logical knowledge graph based on the target prior first task logical knowledge graph and the target prior second task logical knowledge graph.

2. The artificial intelligence-based drone dispatching method according to claim 1, characterized in that: The step of performing weighted calculation based on the a priori first task logical associations and the a priori second task logical associations to generate the a priori task logical associations includes: Determine a candidate first task logical association degree from the respective a priori first task logical association degrees, and determine a candidate second task logical association degree corresponding to the candidate first task logical association degree from the respective a priori second task logical association degrees; Performing weighted calculation according to the candidate first task logical association degree and the candidate second task logical association degree to generate a weighted task logical association degree, and using the weighted task logical association degree as the priori task logical association degree corresponding to the candidate first task logical association degree and the candidate second task logical association degree; Polling each of the a priori first task logical associations and the corresponding a priori second task logical associations to generate each of the a priori task logical associations; Alternatively, a candidate first task logical association degree is determined from each of the a priori first task logical association degrees, and a candidate second task logical association degree corresponding to the candidate first task logical association degree is determined from each of the a priori second task logical association degrees; Comparing the candidate first task logical association degree with the candidate second task logical association degree, when the candidate first task logical association degree is greater than the candidate second task logical association degree, using the candidate first task logical association degree as the priori task logical association degree corresponding to the candidate first task logical association degree and the candidate second task logical association degree; When the candidate first task logical relevance is not greater than the candidate second task logical relevance, taking the candidate second task logical relevance as the priori task logical relevance corresponding to the candidate first task logical relevance and the candidate second task logical relevance; Polling each of the a priori first task logical associations and the corresponding a priori second task logical associations to generate each of the a priori task logical associations; Wherein, the method further comprises: Arrange the logical relevance of each priori task in descending order to generate a task logical relevance list; Determine a set number of target a priori task logical associations from the task logical association list, obtain a priori first task logical knowledge graphs and a priori second task logical knowledge graphs corresponding to the set number of target a priori task logical associations, and generate a set number of reference a priori task logical knowledge graphs; Obtaining each reference drone scheduling execution route according to the drone scheduling execution routes corresponding to the set number of reference priori task logic knowledge graphs, and obtaining past application data corresponding to each reference drone scheduling execution route; The target past application data is determined from the past application data corresponding to the reference UAV scheduling execution routes, and the reference UAV scheduling execution route corresponding to the target past application data is used as the target UAV scheduling execution route of the flight mission to be assigned.

3. An artificial intelligence-based drone dispatching system, characterized in that: The artificial intelligence-based drone scheduling system includes a processor and a machine-readable storage medium, in which machine-executable instructions are stored. The machine-executable instructions are loaded and executed by the processor to implement the artificial intelligence-based drone scheduling method described in any one of claims 1-2.

Images