Intelligent charging scheduling method and system for unmanned aerial vehicle nest

Abstract

The application provides an intelligent charging scheduling method and system for a drone nest, which is suitable for a nest multi-airfield scene of transmission and distribution line patrol inspection. The nest is configured with multiple charging piles, and the method comprises the following steps: acquiring charging pile state data and state data of each drone associated with the nest, generating a nest-returning path and a charging priority according to the charging pile state data and the drone state data; predicting a space-time node path conflict of the nest-returning path according to the charging priority; when the prediction result is that there is a path conflict, acquiring a space-time node where the path conflict occurs and an identity of a conflict drone, and calculating conflict state data of the conflict drone; generating an avoidance path according to the conflict state data, updating the charging priority, and feeding back the avoidance path and the updated charging priority to the nest-returning drone. The application can realize dynamic allocation of charging resources, and take into account charging delay, path conflict and power consumption control.

Description

Technical Field

[0001] This application relates to the field of drone charging scheduling technology, specifically to an intelligent charging scheduling method and system for drone nests. Background Technology

[0002] With the advancement of intelligent power system construction, drones have become a core tool for inspecting transmission, transformation, and distribution equipment due to their advantages such as flexible operation, wide coverage, and high inspection efficiency. To adapt to the needs of intensive and routine inspection operations, the deployment mode of multiple drones per nest has become mainstream, and the charging scheduling efficiency of the nest directly determines the overall effectiveness of the inspection operation.

[0003] Existing drone nest charging systems mostly employ fixed-priority or first-come-first-served scheduling strategies. These strategies reveal several technical shortcomings in real-world applications with multiple drones per nest and intensive inspection tasks: First, the scheduling strategies lack dynamic adaptability. Fixed priorities or queuing rules struggle to adjust charging resource allocation based on the drone's actual operational status, battery level, and task urgency, severely impacting inspection efficiency. Second, traditional scheduling methods fail to consider the spatiotemporal heterogeneity of transmission, distribution, and operational inspection tasks. They lack accurate drone return-to-nest charging demand prediction mechanisms and fail to integrate multi-dimensional data such as drone flight trajectories, battery status, and inspection task semantics. This makes it impossible to quantify drone return times, battery demands, and charging priorities in advance, leading to resource imbalances such as excessive competition for charging resources or prolonged idleness. Furthermore, traditional scheduling methods do not consider path overlap when multiple drones return simultaneously, easily causing charging path conflicts. This not only increases flight safety risks but also exacerbates charging delays.

[0004] In summary, existing technologies suffer from problems such as rigid strategies, lack of demand forecasting, and lack of path conflict prevention. Therefore, there is an urgent need for an intelligent scheduling method and system for drone nest charging that can achieve dynamic allocation of charging resources and take into account charging delay, path conflict, and power consumption control. Summary of the Invention

[0005] In view of the aforementioned problems, this application is proposed to provide an intelligent charging scheduling method and system for unmanned aerial vehicle (UAV) nests that overcomes or at least partially solves the aforementioned problems, comprising:

[0006] A smart charging scheduling method for unmanned aerial vehicle (UAV) nests, applicable to multi-UAV nest scenarios during power transmission, distribution, and operational inspections, wherein the nest is equipped with multiple charging piles, includes the following steps:

[0007] Obtain charging pile status data and drone status data associated with the drone nest, and generate a return path and charging priority based on the charging pile status data and drone status data;

[0008] Based on the charging priority, spatiotemporal node path conflict prediction is performed on the return path;

[0009] When the prediction result indicates that a path conflict exists, the spatiotemporal node where the path conflict occurs and the identity identifier of the conflicting UAV are obtained, and the conflict status data of the conflicting UAV is calculated.

[0010] Based on the conflict status data, an avoidance path is generated and the charging priority is updated. The avoidance path and the updated charging priority are then fed back to the returning drone.

[0011] Furthermore, the step of acquiring charging pile status data and the status data of each drone associated with the drone nest, and generating a return path and charging priority based on the charging pile status data and the drone status data, specifically includes:

[0012] Acquire charging pile status data and the status data of each UAV associated with the UAV nest; the charging pile status data includes the location data and occupancy data of each charging pile in the UAV nest; the UAV status data includes the UAV's GPS flight trajectory, battery degradation curve and inspection task semantic data.

[0013] Extract the spatiotemporal features of the charging pile status data and the drone status data;

[0014] Using the pre-established spatiotemporal relationship network (STRN) prediction framework, the return path, return time probability distribution, and remaining battery power change curve of each UAV are generated.

[0015] The return-to-home charging priority is generated based on the probability distribution of the return-to-home time and the remaining power change curve.

[0016] Furthermore, the step of extracting the spatiotemporal features of the charging pile status data and the drone status data specifically includes:

[0017] Spatial features are generated by encoding the location data of the charging pile, the GPS flight trajectory of the drone, and the spatial coordinates in the semantics of the inspection task through an extended convolutional network (TCN).

[0018] The occupancy data of the charging pile, the battery degradation curve of the drone, and the timestamp information in the inspection task semantics are encoded by the Bidirectional Long Short-Term Memory (BiLSTM) network to generate time features.

[0019] The spatial features and the temporal features are fused to generate the spatiotemporal features.

[0020] Furthermore, the step of predicting spatiotemporal node path conflicts for the return path based on the charging priority specifically includes:

[0021] Based on the charging priority and the height, width, and time of the UAV's flight space, a three-dimensional spatiotemporal cube detection model is constructed;

[0022] The occupancy status of each spatiotemporal node is quantified using the spatiotemporal cube 3D detection model.

[0023] When the number of times a certain spatiotemporal node is occupied is greater than 1, it is determined that there is a path conflict.

[0024] Furthermore, the step of obtaining the spatiotemporal nodes where the path conflict occurred and the identity identifiers of the conflicting UAVs when the prediction result indicates a path conflict, and calculating the conflict state data of the conflicting UAVs, specifically includes:

[0025] When the prediction result indicates that a path conflict exists, obtain the spatiotemporal nodes where the path conflict occurred and the identification of the conflicting UAVs.

[0026] Real-time flight status parameters of each conflicting drone are extracted based on the identification identifier; the real-time flight status parameters include current position, current speed, current remaining battery power, return path, and charging priority.

[0027] Based on the real-time flight status parameters, the predicted remaining battery power of each conflicting UAV upon arrival at the spatiotemporal node is calculated.

[0028] Furthermore, the step of generating an avoidance path and updating the charging priority based on the conflict state data, and feeding back the avoidance path and the updated charging priority to the returning drone, specifically includes:

[0029] Based on the predicted remaining battery power and charging priority of the conflict drone, several candidate avoidance paths are generated with conflict avoidance as the goal.

[0030] Among the candidate avoidance paths, the avoidance path that does not conflict with the high-priority drone and has the least energy loss is selected and fed back to the returning drone.

[0031] Calculate the estimated return time of the drone based on the avoidance path;

[0032] The charging priority is updated based on the estimated return time and fed back to the returning drone.

[0033] Furthermore, the method also includes:

[0034] The optimal charging pile allocation scheme and charging sequence are calculated using a pre-trained multi-objective deep reinforcement learning model. The state space of the multi-objective deep reinforcement learning model includes the occupancy status of each charging pile in the drone nest, the updated charging priority of each drone, the current remaining power, and the return path or avoidance path. The action space includes charging pile allocation actions and charging sequence adjustment actions.

[0035] Based on the optimal charging pile allocation scheme and charging sequence, a charging strategy is generated and fed back to the charging station.

[0036] During the execution of the charging strategy in the cell, the voltage and current parameters of the cell are acquired, and the instantaneous power consumption of the cell is calculated.

[0037] When the instantaneous power consumption exceeds a preset threshold, the speed of the drone returning to its nest is dynamically adjusted or a hovering wait is introduced.

[0038] Furthermore, the step of dynamically adjusting the speed of the drone returning to its nest or initiating hovering when the instantaneous power consumption exceeds a preset threshold specifically includes:

[0039] When the instantaneous power consumption exceeds a preset threshold, obtain the real-time location, remaining power, estimated arrival time, and updated charging priority of all returning drones.

[0040] Based on the charging priority from low to high, delay instructions are sent sequentially to the returning drones; the delay instructions include reducing air speed, initiating hovering and waiting in the air, or flying around along an extended path.

[0041] When the instantaneous power consumption drops to within a preset threshold, a delay cancellation command is sent to the delayed drone.

[0042] A smart charging scheduling system for drone nests includes:

[0043] The home return scheduling module is used to collect charging pile status data and the status data of each UAV associated with the home, and generate home return paths and charging priorities based on the charging pile status data and UAV status data.

[0044] The spatiotemporal conflict detection module is used to predict spatiotemporal node path conflicts of the return path based on the charging priority.

[0045] The conflict status analysis module is used to obtain the spatiotemporal nodes where the path conflict occurs and the identity identifiers of the conflicting UAVs when the prediction result indicates that a path conflict exists, and to calculate the conflict status data of the conflicting UAVs.

[0046] The obstacle avoidance scheduling module is used to generate an obstacle avoidance path and update the charging priority based on the conflict status data, and to feed back the obstacle avoidance path and the updated charging priority to the returning drone.

[0047] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of an intelligent charging scheduling method for a drone nest as described above.

[0048] This application has the following advantages:

[0049] In the embodiments of this application, addressing the technical shortcomings of existing technologies such as rigid drone nest charging scheduling strategies that often employ fixed priorities or first-come-first-served charging modes, leading to high charging delays, significant energy losses, and frequent drone return path conflicts in multi-drone scenarios, this application provides an intelligent charging scheduling method for drone nests. This method is applicable to multi-drone scenarios involving power transmission, distribution, and operational inspections. The nest is equipped with multiple charging piles. The method includes the following steps: collecting charging pile status data and the status data of each drone associated with the nest; generating a return path and charging priority based on the charging pile status data and drone status data; predicting spatiotemporal node path conflicts for the return path based on the charging priority; when the prediction result indicates a path conflict, obtaining the spatiotemporal node of the conflict and the identity identifier of the conflicting drone, and calculating the conflict status data of the conflicting drone; generating an avoidance path and updating the charging priority based on the conflict status data, and feeding back the avoidance path and the updated charging priority to the returning drone. By constructing a Spatiotemporal Relationship Network (STRN) prediction framework and integrating multimodal data such as UAV GPS flight trajectories, battery degradation curves, and inspection task semantics, the probability distribution of return-to-home time and the remaining battery power change curve of each UAV are accurately quantified. This solves the problems of existing technologies that cannot accurately predict UAV return-to-home charging needs and lack prior basis for charging resource allocation, achieving the technical effect of improving prediction accuracy and providing reliable data support for charging scheduling. Furthermore, by introducing a spatiotemporal cube 3D detection model to predict the spatiotemporal node occupancy of the return-to-home path, and combining conflict state data to generate avoidance paths and dynamically adjust charging priorities, the system further enhances the accuracy of predictions and provides reliable data support for charging scheduling. This invention addresses the problems of frequent multi-drone return path conflicts and low inspection efficiency caused by delayed conflict avoidance in existing technologies. It effectively avoids path conflicts and improves return-to-home safety and operational efficiency. By constructing a multi-objective deep reinforcement learning model to generate a dynamic charging scheduling strategy, and combining it with a sequential charging strategy to monitor the instantaneous power consumption of the drone nest, and dynamically adjust the speed of the returning drone or introduce hovering waiting when the power consumption exceeds the limit, it solves the problems of intense competition for charging resources and instantaneous power overload affecting the stability of the power supply system in existing technologies. It optimizes the allocation of charging resources, ensures the stable operation of the power supply system, and improves the overall efficiency of charging scheduling. Attached Figure Description

[0050] To more clearly illustrate the technical solution of this application, the drawings used in the description of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart illustrating the steps of an intelligent charging scheduling method for a drone nest, as provided in an embodiment of this application.

[0052] Figure 2 This is a schematic diagram comparing the performance of various scheduling algorithms in a static task scenario provided by an embodiment of this application;

[0053] Figure 3 This is a schematic diagram comparing the task completion rates of various scheduling algorithms under different task densities, provided in an embodiment of this application.

[0054] Figure 4 This is a structural block diagram of an intelligent charging scheduling system for a drone nest provided in one embodiment of this application;

[0055] Figure 5 This is a schematic diagram of the structure of a computer electronic device provided in an embodiment of this application;

[0056] 1. Computer electronic device; 2. External device; 3. Processing unit; 4. Bus; 5. Network adapter; 6. I / O interface; 7. Display; 8. Memory; 9. Random access memory; 10. Cache memory; 11. Storage system; 12. Program / utility; 13. Program module. Detailed Implementation

[0057] To make the objectives, features, and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0058] The inventors discovered through analysis of existing technologies that traditional scheduling methods do not take into account the path overlap problem when multiple drones return to their nests at the same time, which can easily lead to conflicts in the drones' return-to-base charging paths. This not only increases the flight safety risks of drones, but also further exacerbates charging delays.

[0059] Reference Figure 1This paper illustrates an intelligent charging scheduling method for a drone nest according to an embodiment of this application. It is applicable to multi-drone scenarios involving power transmission, distribution, and operational inspections. The nest is equipped with multiple charging piles, and includes the following steps:

[0060] S110. Obtain the status data of the charging pile and the status data of each UAV associated with the nest, and generate the return path and charging priority based on the status data of the charging pile and the status data of the UAV.

[0061] S120. Based on the charging priority, perform spatiotemporal node path conflict prediction on the return path;

[0062] S130. When the prediction result indicates that there is a path conflict, obtain the spatiotemporal node where the path conflict occurs and the identity identifier of the conflicting UAV, and calculate the conflict status data of the conflicting UAV.

[0063] S140. Generate an avoidance path and update the charging priority based on the conflict status data, and feed back the avoidance path and the updated charging priority to the returning drone.

[0064] In the embodiments of this application, addressing the technical shortcomings of existing technologies where drone nest charging scheduling strategies are rigid and often employ fixed priorities or first-come-first-served charging modes, leading to high charging delays, significant energy losses, and frequent drone return path conflicts in multi-drone scenarios, this application provides an intelligent charging scheduling method for drone nests. By constructing a spatiotemporal relation network (STRN) prediction framework and integrating multimodal data such as drone GPS flight trajectories, battery degradation curves, and inspection task semantics, this method accurately quantifies the probability distribution of return times and remaining battery power changes for each drone. This solves the problems of existing technologies failing to accurately predict drone return charging needs and lacking prior evidence for charging resource allocation, achieving the technical effect of improving prediction accuracy and providing reliable data support for charging scheduling. By introducing spatiotemporal... The cubic 3D detection model performs forward-looking spatiotemporal node occupancy prediction on the return path, and generates avoidance paths and dynamically adjusts charging priorities by combining conflict state data. This solves the problems of frequent conflicts on multi-drone return paths and low inspection efficiency caused by delayed conflict avoidance in existing technologies, achieving the technical effect of effectively avoiding path conflicts and improving return safety and operational efficiency. By constructing a multi-objective deep reinforcement learning model to generate dynamic charging scheduling strategies, and combining sequential charging strategies to monitor the instantaneous power consumption of the drone nest, dynamically adjusts the speed of the returning drone or introduces hovering waiting when the power consumption exceeds the limit, solves the problems of intense competition for charging resources and the impact of instantaneous power overload on the stability of the power supply system in existing technologies, achieving the technical effect of optimizing charging resource allocation, ensuring stable operation of the power supply system, and improving the overall efficiency of charging scheduling.

[0065] The following will further explain a smart charging scheduling method for a drone nest in this exemplary embodiment.

[0066] In one embodiment of this application, the specific process of step S110, "obtaining charging pile status data and the status data of each UAV associated with the UAV nest, and generating a return path and charging priority based on the charging pile status data and the UAV status data," can be further explained in conjunction with the following description.

[0067] As described in the following steps

[0068] Acquire charging pile status data and the status data of each UAV associated with the UAV nest; the charging pile status data includes the location data and occupancy data of each charging pile in the UAV nest; the UAV status data includes the UAV's GPS flight trajectory, battery degradation curve and inspection task semantic data.

[0069] Extract the spatiotemporal features of the charging pile status data and the drone status data;

[0070] Using the pre-established spatiotemporal relationship network (STRN) prediction framework, the return path, return time probability distribution, and remaining battery power change curve of each UAV are generated.

[0071] The return-to-home charging priority is generated based on the probability distribution of the return-to-home time and the remaining power change curve.

[0072] It should be noted that charging pile status data refers to the real-time operational status information of each charging pile in the drone nest, including the location of the charging pile, whether it is currently occupied, the identification of the occupied drone, the charging duration, and the estimated remaining charging time. Drone status data refers to the multi-dimensional operational information of each drone associated with the drone nest, including GPS flight trajectory, battery degradation curve, and inspection task semantic data. Specifically, the GPS flight trajectory records the historical position sequence of the drone in three-dimensional space; the battery degradation curve reflects the change in the drone's remaining power over time or flight distance; and the inspection task semantic data includes unstructured information such as task type (e.g., power line inspection, substation inspection), operating radius, inspection area, and task urgency level. The Spatiotemporal Relationship Network (STRN) prediction framework is a multimodal deep neural network model used to fuse the above multi-source heterogeneous data to output the drone's return path, return time probability distribution, and remaining power change curve. The model extracts spatial features through a dilated convolutional network (TCN) to capture the spatial dependence between local obstacles and global hotspots; it extracts temporal features through a bidirectional long short-term memory network (BiLSTM) to model the long-term pattern of battery wear; finally, the spatiotemporal features are fused and input into a quantile regression layer to output the probability distribution of homing time, in order to cope with the prediction bias caused by environmental variables.

[0073] The training process of the STRN model is as follows: First, historical flight data is collected as training samples. Each sample contains the drone's GPS trajectory sequence, battery degradation sequence, task semantic labels, and the corresponding actual return time and remaining battery power. Multimodal data is mapped to the feature space through a unified encoding layer. After decoupling spatial and temporal features, they are input into the TCN and BiLSTM branches respectively, and features are extracted and then fused. The network parameters are iteratively updated using the quantile loss function and mean squared error loss function as optimization objectives, through backpropagation, until the prediction error converges.

[0074] Charging priority is a dimensionless, normalized value, typically ranging from 0 to 1. It characterizes the urgency and priority of a drone obtaining charging resources within the current scheduling cycle. A higher value indicates a higher priority, meaning it should be allocated charging resources or have its charging process initiated more quickly. Charging priority is determined by a weighted fusion of three core factors: remaining battery power (lower battery power, higher priority), mission urgency (more urgent mission, higher priority), and return-to-home distance (closer distance, higher priority, to expedite charging and deployment to the next mission). Specifically, charging priority... .in The normalized remaining charge (0~1, 1 represents full charge). To normalize the task urgency (0~1, 1 represents the most urgent). Normalized return distance (0~1, 1 represents the farthest). These are the weighting coefficients, and Based on actual operational experience, it is usually set as follows: =0.5 (Power safety is the top priority). =0.3, =0.2. The remaining battery power change curve output by the STRN prediction framework provides accurate... The probability distribution of return-to-home time can be mapped to return-to-home distance information, and the "task urgency level" in the inspection task semantics directly provides this information. The resulting charging priority represents the overall charging urgency of the drone in the current system state, which is used by subsequent deep reinforcement learning scheduling and conflict avoidance mechanisms.

[0075] As an example, the methods for generating return paths and charging priorities can be flexibly implemented according to actual application scenarios: Based on the STRN prediction framework, the spatiotemporal probability distribution of each return path is directly output, and priorities are generated based on the remaining power threshold; combined with the real-time occupancy status of the charging piles, the path planning is dynamically adjusted so that the drones prioritize the return paths corresponding to the idle charging piles; a task urgency weighting factor is introduced into the STRN prediction framework so that drones with urgent inspection tasks receive higher charging priorities; a multi-model fusion strategy is adopted to weight and fuse the STRN prediction results with the battery energy consumption estimation results based on the physical model to improve the prediction robustness.

[0076] The weighting coefficients for generating charging priorities can be flexibly adjusted according to different scenarios. For example, in emergency inspection scenarios, the weight of task urgency can be increased (e.g., ...). =0.5); In extreme power scenarios, the weight of remaining power can be increased (e.g., =0.8); for long-range missions, the weight of return distance can be appropriately increased. In addition, dynamic correction factors, such as the drone's historical mission completion rate and battery health status, can be introduced to fine-tune the priority.

[0077] In one specific implementation, taking a fixed drone nest as an example, the nest is equipped with two charging piles, and currently four drones are performing power transmission line inspection tasks. The system collects the GPS trajectory (longitude, latitude, altitude), battery degradation curve (remaining power percentage and discharge rate) of each drone every 10 seconds, as well as the semantics of the inspection task (task type: "power transmission line inspection", operating radius: 3km, task urgency level: "high"). The above data is input into the STRN prediction framework. The TCN layer extracts the spatial correlation features between the current location of the drone and the power transmission line towers, and the BiLSTM layer predicts the power change in the next 30 seconds based on the battery degradation curve. The quantile regression layer outputs the probability distribution of the return time of each drone. For example, the predicted return time of drone U1 is 10% quantile 80 seconds, 50% quantile 90 seconds, and 90% quantile 105 seconds. Based on the fact that U1's remaining battery power is 23% (0.23 after normalization), the task urgency is "high" (0.9 after normalization), and the predicted return distance is 1.2km, a return charging priority is generated for U1, and a detour route is planned to avoid the dense tower area.

[0078] In one embodiment of the present invention, the specific process of "extracting the spatiotemporal characteristics of the charging pile status data and the drone status data" can be further explained in conjunction with the following description.

[0079] Spatial features are generated by encoding the location data of the charging pile, the GPS flight trajectory of the drone, and the spatial coordinates in the semantics of the inspection task through an extended convolutional network (TCN).

[0080] The occupancy data of the charging pile, the battery degradation curve of the drone, and the timestamp information in the inspection task semantics are encoded by the Bidirectional Long Short-Term Memory (BiLSTM) network to generate time features.

[0081] The spatial features and the temporal features are fused to generate the spatiotemporal features.

[0082] It should be noted that the spatiotemporal features are joint feature vectors that fuse the spatial location distribution and temporal state changes of drones and charging piles, and are the core input of the STRN prediction framework; the dilated convolutional network TCN is used to capture the spatial dependencies of spatial obstacles, drone nest locations, and drone trajectories in the inspection area, expanding the receptive field without sacrificing resolution, and adapting to the wide-area spatial scenario of power transmission, distribution, and camp inspection; the bidirectional long short-term memory network BiLSTM is used to bidirectionally model the long-term evolution of battery degradation, charging pile occupancy, and task timing, solving the problem of long-term dependencies in time-series data; feature fusion adopts a combination of feature concatenation and weighted fusion to map spatial features and temporal features to the same feature space, eliminating the heterogeneity of multi-source data.

[0083] In a specific implementation, taking the transmission line inspection scenario as an example, the drone nest is equipped with 2 charging piles and associated with 4 inspection drones. The TCN is used to encode the charging pile coordinates (X1,Y1) and (X2,Y2), the drone GPS trajectory point set, and the spatial coordinates of the inspection area, outputting a spatial feature vector containing spatial location and obstacle distribution. The BiLSTM is used to encode the charging pile occupancy time sequence, drone battery degradation time series data, and task timestamp information, outputting a temporal feature vector containing time series changes. The two types of features are fused by channel concatenation to generate a 256-dimensional spatiotemporal feature vector, which is input into the quantile regression layer of the STRN prediction framework for subsequent return time and remaining power prediction.

[0084] In one embodiment of the present invention, the specific process of "predicting spatiotemporal node path conflicts for the return path based on the charging priority" in step S120 can be further explained in conjunction with the following description.

[0085] Based on the charging priority and the height, width, and time of the UAV's flight space, a three-dimensional spatiotemporal cube detection model is constructed;

[0086] The occupancy status of each spatiotemporal node is quantified using the spatiotemporal cube 3D detection model.

[0087] When the number of times a certain spatiotemporal node is occupied is greater than 1, it is determined that there is a path conflict.

[0088] It should be noted that the spatiotemporal cube 3D collision detection model is a collision detection model that constructs a 3D mesh using the height dimension h, width dimension w, and time dimension t of the UAV's flight space. This model discretizes the UAV's return path into a series of spatiotemporal nodes. Each node represents a specific spatial location occupied by a drone at a specific moment. By quantifying the number of occupants at each spatiotemporal node, it is possible to intuitively determine whether there are path conflicts where multiple drones converge at the same spatiotemporal node. The conflict detection function is defined as: The number of times a certain spatiotemporal node is occupied. When a path conflict is detected, it is determined to exist. This model requires no training and is a rule-based deterministic detection model. Its core lies in constructing accurate spatiotemporal node mapping relationships.

[0089] As an example, the implementation of spatiotemporal node path conflict prediction can include: dividing the flight space into a 10m×10m×10s grid based on a fixed-resolution three-dimensional spatiotemporal cube, and directly counting the occupancy of each grid; adopting a dynamic resolution strategy, increasing the spatiotemporal resolution in the vicinity of the drone nest (e.g., 5m×5m×5s) and decreasing the resolution in the far-field region, balancing detection accuracy and computational efficiency; introducing a sliding time window mechanism to predict only the spatiotemporal node occupancy status within the next T seconds, reducing computational overhead; and weighting the conflict determination based on the drone's charging priority, giving high-priority drones priority in conflict detection.

[0090] In a specific implementation, taking U1 and U2 as examples, the return path of U1 is P1: node sequence [(100,200,0s)→(100,190,10s)→(100,180,20s)], and the return path of U2 is P2: node sequence [(110,190,0s)→(105,185,10s)→(100,180,20s)]. The spatiotemporal cube resolution is 10m×10m×10s, and the height h is fixed at 50m. At time window t=20s, the node (100,180,20s) appears in the paths of both U1 and U2. The system determines that the number of times this spatiotemporal node is occupied is 2, which satisfies the condition Occup(h,w,t) ≥ 2 in the collision detection function CollisionFlag. Therefore, a path conflict is determined to exist. The detection results output the coordinates of the conflicting nodes and the identification identifiers of the conflicting drones, U1 and U2.

[0091] In one embodiment of the present invention, the specific process of step S130, "when the prediction result indicates the existence of a path conflict, obtain the spatiotemporal node where the path conflict occurs and the identity identifier of the conflicting UAV, and calculate the conflict status data of the conflicting UAV," can be further explained in conjunction with the following description.

[0092] When the prediction result indicates that a path conflict exists, obtain the spatiotemporal nodes where the path conflict occurred and the identification of the conflicting UAVs.

[0093] Real-time flight status parameters of each conflicting drone are extracted based on the identification identifier; the real-time flight status parameters include current position, current speed, current remaining battery power, return path, and charging priority.

[0094] Based on the real-time flight status parameters, the predicted remaining battery power of each conflicting UAV upon arrival at the spatiotemporal node is calculated.

[0095] It should be noted that conflict state data refers to a comprehensive information matrix used to characterize the severity and avoidance priority of conflicting drones at spatiotemporal nodes. This includes the charging priority of each conflicting drone, the expected time window for reaching the conflict node, the predicted remaining battery power upon arrival, and path occupancy intentions. The predicted remaining battery power is calculated based on the drone's current remaining battery power, speed, path length, and a battery degradation model, and is used to assess whether the drone has sufficient energy to complete the avoidance maneuver. When constructing the conflict state data, a battery safety constraint is set: This means that the drone's remaining battery level must not be lower than the minimum safe battery level. If this level is reached, the drone must be forced to return to its nest to recharge, ensuring that avoidance operations do not cause the drone's battery level to fall below the safe threshold.

[0096] As an example, the calculation method for conflict state data of conflict-prone drones may include: estimating the remaining battery power when reaching the conflict node based on the current flight speed and path length using a linear energy consumption model; dynamically predicting the remaining battery power by considering environmental factors such as wind speed and load using an energy consumption prediction model trained based on historical flight data; introducing a time window overlap coefficient to calculate the degree of overlap of the time windows of each drone's expected arrival at the conflict node, as an indicator of conflict severity; and combining the drone's mission urgency level with the mission urgency level as an additional dimension of the conflict state data.

[0097] In one specific implementation, continuing with the conflict detection results above, the real-time flight status parameters of U1 and U2 are obtained: U1's current position (100, 200, 50m), flight speed 15m / s, remaining battery power 23%, charging priority 0.92; U2's current position (110, 190, 50m), flight speed 12m / s, remaining battery power 18%, charging priority 0.65. Based on the path length calculation, U1 needs to fly approximately 20m to reach the conflict node (100, 180, 20s), taking approximately 1.3 seconds, with an estimated remaining battery power of 22.5%; U2 needs to fly approximately 22m to reach the same node, taking approximately 1.8 seconds, with an estimated remaining battery power of 17.2%. A conflict status data matrix is ​​constructed: U1 (priority 0.92, arrival time 20s, remaining battery power 22.5%), U2 (priority 0.65, arrival time 20s, remaining battery power 17.2%). This matrix shows that U1 has higher priority and more power, which meets the power safety constraints (assuming...). U1 has 23% remaining battery power, which meets the requirements, while U2 has 18% remaining battery power, which is below the safety threshold and must be forced to return to the nest. Therefore, it is more reasonable for U2 to be the one to give way.

[0098] In one embodiment of the present invention, the specific process of step S140, which involves "generating an avoidance path and updating the charging priority based on the conflict state data, and feeding back the avoidance path and the updated charging priority to the returning drone," can be further described in conjunction with the following description.

[0099] Based on the predicted remaining battery power and charging priority of the conflict drone, several candidate avoidance paths are generated with conflict avoidance as the goal.

[0100] Among the candidate avoidance paths, the avoidance path that does not conflict with the high-priority drone and has the least energy loss is selected and fed back to the returning drone.

[0101] Calculate the estimated return time of the drone based on the avoidance path;

[0102] The charging priority is updated based on the estimated return time and fed back to the returning drone.

[0103] It should be noted that candidate avoidance paths refer to several alternative return paths generated to avoid drones. These paths must meet constraints related to battery safety, charging station capacity, and path accessibility. Specifically, the path accessibility constraint is... This means that there must be no obstacles on the drone's return path. The energy-minimizing avoidance path is the path that results in the lowest drone flight energy consumption among all candidate paths that meet the avoidance conditions.

[0104] As an example, the implementation of generating avoidance paths may include: searching for the shortest path to avoid conflict spatiotemporal nodes in a three-dimensional spatial grid based on the A* algorithm; generating multiple random avoidance paths in continuous space based on the Fast Exploratory Random Tree (RRT) algorithm, and then selecting the optimal path; using the dynamic window method to adjust the UAV's heading and speed in real time to achieve local avoidance; and combining genetic algorithms to encode and iteratively optimize multiple candidate paths to select the path that is optimal in terms of both energy consumption and conflict risk. If no candidate avoidance path exists to circumvent the conflict, the following operations are performed: The drone with the lowest charging priority among the conflicting drones is identified as the drone to be adjusted, its charging priority is temporarily lowered to the preset lowest level, and its estimated return time is recalculated; the updated charging priority of the drone to be adjusted is fed back to the charging scheduling queue of the drone nest, delaying its charging order; a hovering or return-to-home command is sent to the drone to be adjusted, suspending its return-to-home process until a higher-priority drone passes the conflict node and resumes its return-to-home process; if multiple drones cannot generate an effective avoidance path, delay commands are issued to the drones sequentially according to their charging priority from low to high, until the spatiotemporal occupancy of the conflict node is resolved.

[0105] In one specific implementation, to address U2's obstacle avoidance requirements, a path replanning engine is invoked. Starting from U2's current position (110, 190, 50m) and targeting the nest position (100, 100, 50m), with the conflict node (100, 180, 20s) as the restricted area, the A* algorithm is run in a 3D spatial grid to generate three candidate avoidance paths: Candidate Path 1, detouring eastward via (115, 185) → (115, 170m). Candidate 1: Route 1 → (100, 150), adding approximately 40m of distance and increasing energy consumption by approximately 5%; Candidate 2: Detour westward via (95, 185) → (95, 170) → (100, 150), adding approximately 38m of distance and increasing energy consumption by approximately 4.5%; Candidate 3: Decelerate and wait, maintaining the original route but reducing speed from 12m / s to 8m / s, delaying the arrival time at the conflict node to 25s, avoiding U1's occupied window, increasing energy consumption by approximately 2%. The system selects the avoidance path with no conflict with U1 and the least energy loss—Candidate 3—and generates a deceleration command to send to U2, adjusting its speed to 8m / s. Based on the new path, the estimated return time for U2 is increased from the original 85 seconds to 105 seconds. Accordingly, the system lowers U2's charging priority and updates it to the charging scheduling queue. This process satisfies the charging pile capacity constraint. That is, charging piles at the same time The number of charging drones shall not exceed their maximum capacity.

[0106] In an additional embodiment of the present invention, the intelligent charging scheduling method for a drone nest further includes, in conjunction with the following description:

[0107] The optimal charging pile allocation scheme and charging sequence are calculated using a pre-trained multi-objective deep reinforcement learning model. The state space of the multi-objective deep reinforcement learning model includes the occupancy status of each charging pile in the drone nest, the updated charging priority of each drone, the current remaining power, and the return path or avoidance path. The action space includes charging pile allocation actions and charging sequence adjustment actions.

[0108] Based on the optimal charging pile allocation scheme and charging sequence, a charging strategy is generated and fed back to the charging station.

[0109] During the execution of the charging strategy in the cell, the voltage and current parameters of the cell are acquired, and the instantaneous power consumption of the cell is calculated.

[0110] When the instantaneous power consumption exceeds a preset threshold, the speed of the drone returning to its nest is dynamically adjusted or a hovering wait is introduced.

[0111] It should be noted that the multi-objective deep reinforcement learning model is a reinforcement learning model based on deep Q-networks or policy gradient methods. It is used to learn the optimal charging scheduling policy in a complex state space, achieving the optimization objective of simultaneously minimizing UAV charging delay, flight energy loss, and the risk of return path conflict. The optimization objective function is: Where: N is the total number of drones to be scheduled. Delay the charging time for the i-th drone. Let i be the flight energy consumption of the i-th drone. For drones and The path conflict indicator function is defined (1 for conflict, 0 for no conflict). The state space of the multi-objective deep reinforcement learning model includes the occupancy status of each charging station in the aviary (idle / occupied / estimated release time), the charging priority of each drone, the current remaining battery power, and the return path or avoidance path. The action space includes charging station allocation actions (assigning a drone to a charging station) and charging order adjustment actions (adjusting the charging queue order of each drone). The composite reward function is designed as follows: Where: η, μ, v, ξ are weighting coefficients. This is a reward item for task completion efficiency. As a task priority reward item, For energy consumption incentive items, This is a path conflict penalty term. The composite reward function guides the multi-objective deep reinforcement learning model to learn multi-objective optimization strategies and coordinate conflicting optimization objectives. The pre-training process of the multi-objective deep reinforcement learning model is divided into two stages: offline training and online fine-tuning. In the offline training stage, an experience replay pool is built using historical scheduling data. The deep reinforcement learning network is trained by randomly sampling experience tuples (state, action, reward, next state), and the Adam optimizer is used to update the network parameters until the reward function converges. In the online fine-tuning stage, scheduling results are continuously collected during actual operation, and the network weights are updated periodically using incremental learning to adapt the scheduling strategy to dynamically changing inspection scenarios.

[0112] As an example, the implementation of calculating the optimal charging pile allocation scheme and charging sequence using a pre-trained multi-objective deep reinforcement learning model can include: using a deep Q-network (DQN) to discretize the action space and output the Q-value of the optimal charging pile allocation action; optimizing the PPO based on a proximal policy and outputting the charging sequence adjustment amount in the continuous action space; using multi-agent reinforcement learning, treating each charging pile as an independent agent to collaboratively optimize the global scheduling objective; and employing hierarchical reinforcement learning, where higher-level policies determine charging pile allocation and lower-level policies determine the charging sequence, thereby improving training efficiency.

[0113] In a specific implementation, taking a nest configuration with 2 charging piles and 4 drones as an example, the current state is: charging pile 1 is occupied (U3, estimated release time 60s), charging pile 2 is idle; U1 (priority 0.92, remaining power 23%, return path P1), U2 (priority 0.55, remaining power 17%, avoidance path P2), U3 (priority 0.88, charging), U4 (priority 0.72, remaining power 25%, return path P4). The state vector is input into a pre-trained deep Q-network, and the network outputs the Q-value of each action combination. Calculation shows that the action "assign U1 to charging pile 2, U4 queues, U2 delays charging" has the highest Q-value. The system executes this action and generates a charging strategy: U1 immediately connects to charging pile 2 after returning to the nest, U4 queues after U1, and U2 waits for power consumption to recover before charging. This scheduling scheme aims to minimize charging delay and energy loss, while reducing the risk of path conflicts.

[0114] In one embodiment of the present invention, the specific process of "dynamically adjusting the speed of the UAV returning to its home or introducing hovering waiting when the instantaneous power consumption is greater than a preset threshold" can be further explained in conjunction with the following description.

[0115] When the instantaneous power consumption exceeds a preset threshold, obtain the real-time location, remaining power, estimated arrival time, and updated charging priority of all returning drones.

[0116] Based on the charging priority from low to high, delay instructions are sent sequentially to the returning drones; the delay instructions include reducing air speed, initiating hovering and waiting in the air, or flying around along an extended path.

[0117] When the instantaneous power consumption drops to within a preset threshold, a delay cancellation command is sent to the delayed drone.

[0118] It should be noted that instantaneous power consumption refers to the real-time power consumption of the satellite's power supply system at the current moment, calculated by summing the charging power of each charging station. The preset threshold is typically set based on the rated power and capacity of the satellite's power supply system, generally 80% to 90% of the rated power, to avoid overload risks. Delay commands include reducing airspeed, initiating hovering, or flying around along an extended path. The purpose is to delay the drone's return time, staggering the time window of multiple charging stations simultaneously charging at high power, thereby smoothing out instantaneous power consumption peaks. By monitoring the voltage and current of the satellite's power supply system in real time, when the instantaneous power consumption approaches the threshold, charging of low-priority drones is temporarily suspended, prioritizing the charging of high-priority drones.

[0119] As an example, the implementation methods for dynamically adjusting the speed of a drone returning to its nest or introducing hovering waiting when the instantaneous power consumption exceeds a preset threshold may include: dynamically adjusting the reduction of the drone's speed based on a PID control algorithm according to the power consumption deviation; using priority-based queue scheduling to delay drones in order of priority from low to high; introducing a power consumption prediction model to predict future power consumption peaks in advance and proactively delaying the drone's return to its nest; and allowing short-term power consumption exceeding the limit when the energy storage system is sufficient, with the energy storage system supplementing the power supply, in conjunction with the status of the nest energy storage system.

[0120] In one specific implementation, during the execution of the charging strategy at the drone nest, voltage and current parameters are monitored in real time, and instantaneous power consumption is calculated. Currently, charging pile 1 is charging U3 at 3kW, and charging pile 2 is charging U1 at 3kW, with a total power consumption of 6kW. The preset safety threshold is 5kW, and the instantaneous power consumption exceeds the threshold. The status of all returning drones is obtained: U2 is expected to arrive in 10 seconds, and U4 is expected to arrive in 15 seconds, with charging priorities of 0.55 and 0.72 respectively. Based on the order of priority from low to high, a delay command is sent to U2: reducing its speed from 12m / s to 6m / s, and postponing its expected arrival time to 30 seconds later. The power consumption prediction model shows that U3 will complete charging and release charging pile 1 in 20 seconds. At this time, U2 has not yet arrived, and its future power consumption will drop to 3kW, below the threshold. After U3 releases, the system sends a delay release command to U2, restoring its original speed. This process realizes a sequential charging strategy that starts the charging process according to charging priority and suspends the charging of lower-priority drones when the instantaneous power consumption approaches the threshold.

[0121] In an additional embodiment of the present invention, the intelligent charging scheduling method for a drone nest further includes, in conjunction with the following description:

[0122] Construct a spatiotemporal pheromone field to discretize the UAV flight space into a grid;

[0123] Based on the grid's historical conflict frequency, current occupancy status, and task urgency, a corresponding pheromone concentration is generated;

[0124] The pheromone concentration of the grid where the conflict space node is located is increased by a preset amount, and the updated pheromone concentration is fed back to the returning drone; the returning drone is used to adjust its flight direction in real time based on the sensed pheromone concentration gradient during the return process.

[0125] It should be noted that the drone flight space is discretized into three-dimensional grid cells, each maintaining a dynamically changing pheromone concentration value. The pheromone concentration is constructed using three dimensions: historical conflict frequency, indicating the number of times the grid has been simultaneously occupied by multiple drones within a past period, reflecting the inherent conflict risk of the area; current occupancy status, indicating whether the grid is occupied or pre-occupied by a high-priority drone within the current time window; and task urgency, indicating the urgency level of the task of the drone about to pass through the grid. Grids with urgent tasks will receive a lower pheromone concentration (representing passage priority). The initial pheromone concentration is generated by weighted fusion of the above three dimensions, and the formula can be expressed as: ,in Let (h,w,t) be the pheromone concentration at the spatiotemporal node (h,w,t). This is a normalized value for the frequency of historical conflicts. This indicates the current occupancy status (1 for occupied, 0 for idle). The urgency level is normalized, and α, β, and γ are weighting coefficients. The pheromone field update mechanism is as follows: the pheromone concentration of the grid containing the conflict node is increased by a preset increment. (For example, increasing it by 0.3) raises the pheromone concentration in that area, thus serving as an avoidance signal in subsequent path decisions; the pheromone concentration naturally decays over time, with the decay function being... Where λ is the attenuation coefficient and Δt is the time interval, to avoid the influence of historical conflict information on current decisions. During the return process, low-priority UAVs acquire the pheromone concentration values ​​of each grid on the path ahead in real time through the onboard perception module and calculate the pheromone concentration gradient. The concentration gradient points in the direction of the fastest increase in pheromone concentration, representing the direction of increased conflict risk. Based on the gradient direction, the drone actively adjusts its flight path to avoid high-concentration areas and chooses a lower-concentration path to return to its nest. This mechanism achieves decentralized path coordination, delegating global conflict avoidance decision-making to individual drone autonomous decision-making, reducing the computational burden on the nest scheduling center.

[0126] As an example, the implementation of constructing a spatiotemporal pheromone field and adjusting the UAV's flight direction in real time based on the pheromone concentration gradient can include: 1. Employing a global pheromone field broadcasting mechanism, where the pheromone field is uniformly maintained and updated at the pod end, and the pheromone concentration of each grid is periodically broadcast to all UAVs. The UAVs plan local paths based on the received pheromone field data. 2. Employing a distributed pheromone field maintenance mechanism, where each UAV broadcasts its occupied spatiotemporal node information to surrounding UAVs through inter-UAV communication during flight. Each UAV constructs its own local pheromone field, achieving avoidance coordination between adjacent UAVs. 3. Introducing a pheromone threshold sensitivity mechanism, where path adjustment is triggered only when the UAV senses that the pheromone concentration on the path ahead exceeds a preset threshold (e.g., 0.7), avoiding overreaction to low-concentration areas. 4. Differentiating the sensitivity of pheromone concentration based on charging priority, with higher path adjustment thresholds for high-priority UAVs and lower thresholds for low-priority UAVs, making low-priority UAVs more inclined to actively avoid high-priority UAVs. 5. A multimodal pheromone field fusion strategy is adopted to weight and fuse the spatiotemporal pheromone field with the spatiotemporal cube collision detection results. The pheromone field provides long-term statistical risk distribution, while the spatiotemporal cube provides real-time collision nodes. The two together guide UAV decision-making.

[0127] In a specific implementation, taking a fixed drone nest as an example, this nest is associated with 5 drones performing inspection tasks. First, the 2km×2km flight space around the nest is discretized into a 20m×20m×10s grid, with a total of 100×100×60 spatiotemporal nodes. When initializing the pheromone field, the conflict frequency of each grid is statistically analyzed based on the flight data of the past 30 days. The area near the corner tower of the power transmission line (coordinates (500,600) to (550,650)) has the highest historical conflict frequency, with a normalized value of 0.85. At the current moment, high-priority drone U1 (priority 0.95) is about to pass through this area along path P1, occupying the state. Set to 1; U1's task urgency is "urgent". =0.9, then the pheromone concentration is calculated as τ=0.3×0.85+0.4×1+0.3×(1-0.9)=0.255+0.4+0.03=0.685. At this time, the low-priority U5 (priority 0.35) originally planned to pass through the same area along path P5. When U5 senses that the pheromone concentration of the grid ahead is 0.685, which is significantly higher than the average concentration of 0.2 in the surrounding area, it calculates the concentration gradient pointing to that grid, and U5 actively triggers path adjustment. Based on the direction of the pheromone gradient, three avoidance paths are generated: detouring to the northeast (avoiding the high concentration area, adding 120m of distance, and expected to increase energy loss by 3.5%), detouring to the southwest (adding 95m of distance, and expected to increase energy loss by 2.8%), and hovering in the air for 10 seconds (increasing energy consumption by 1.5%). U5 chose the hovering and waiting scheme with minimal energy consumption, entering a hovering state near its original path. After U1 passed through the conflict area, the pheromone concentration in that area decreased to below 0.3, and U5 resumed flight and returned to its nest along the original path. Throughout the entire process, the UAV's autonomous and coordinated avoidance was achieved solely through the dynamic updating of the pheromone field, without imposing any additional computational load on the nest scheduling center. Simultaneously, the historical conflict frequency characteristics of the pheromone field enabled the system to proactively avoid high-risk areas, transforming path conflict detection from a passive response to active prevention, effectively reducing the probability of path conflicts occurring.

[0128] Reference Figure 2 This diagram illustrates the performance comparison of various scheduling algorithms in a static task scenario according to an embodiment of this application. Static task benchmark tests were conducted on the scheduling method of this application to verify its scheduling performance. The tests compared GA, MOEA / D, DQN, DRL, ST-Greedy, GNN-Scheduler, and the STRN-DRL algorithm of this application, with metrics covering Total Task Completion Rate (TCR), Charging Conflict Reduction Rate (HPCR), Average Task Latency, Energy Consumption per Task, and Number of Path Conflicts. Test results show that the STRN-DRL scheduling algorithm of this application achieves a TCR of 0.937, an HPCR of 0.915, an average task latency as low as 79.2s, energy consumption per task reduced to 0.32kWh / task, and only 4 path conflicts. All performance metrics are superior to traditional scheduling algorithms, fully verifying the effectiveness of the multi-objective optimization and spatiotemporal-aware scheduling scheme of this application.

[0129] Reference Figure 3This diagram illustrates the comparison of task completion rates of various scheduling algorithms under different task densities, as provided in an embodiment of this application. Dynamic load stress testing was conducted on the scheduling method of this application to verify its stability under high task density. The horizontal axis represents task density, and the vertical axis represents the total task completion rate (TR). As the task density gradually increases from 3 to 8, the task completion rate of traditional scheduling algorithms shows a rapid downward trend. However, the STRN-DRL algorithm of this application, relying on spatiotemporal prediction to pre-allocate charging resources and dynamically adjust the drone's speed and return-to-home strategy, consistently maintains a task completion rate above 85%. Even in high task density scenarios, it can stably output the optimal scheduling strategy, effectively solving the technical problems of charging resource competition, frequent path conflicts, and power overload in scenarios with multiple drones per nest and dense inspection tasks. It is fully adaptable to the task density requirements of the entire transmission, transformation, and distribution camp inspection scenario.

[0130] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0131] Reference Figure 4 This application illustrates an embodiment of an intelligent charging scheduling system for a drone nest, specifically including the following modules:

[0132] Specifically, it includes:

[0133] The home return scheduling module 210 is used to collect charging pile status data and the status data of each UAV associated with the home, and generate home return paths and charging priorities based on the charging pile status data and UAV status data.

[0134] The spatiotemporal conflict detection module 220 is used to predict spatiotemporal node path conflicts of the return path based on the charging priority.

[0135] The conflict status analysis module 230 is used to obtain the spatiotemporal nodes where the path conflict occurs and the identity identifiers of the conflicting UAVs when the prediction result indicates that a path conflict exists, and to calculate the conflict status data of the conflicting UAVs.

[0136] The obstacle avoidance scheduling module 240 is used to generate an obstacle avoidance path and update the charging priority based on the conflict status data, and to feed back the obstacle avoidance path and the updated charging priority to the returning drone.

[0137] In one embodiment of the present invention, the homing scheduling module 210 includes:

[0138] The first return-to-home scheduling submodule is used to collect charging pile status data and the status data of each UAV associated with the UAV nest;

[0139] The second return-to-home scheduling submodule is used to extract the spatiotemporal features of the charging pile status data and the drone status data;

[0140] The third return-to-home scheduling submodule is used to generate the return-to-home path, return-to-home time probability distribution, and remaining power change curve for each UAV through a pre-established spatiotemporal relationship network (STRN) prediction framework.

[0141] The fourth return-to-home scheduling submodule is used to generate return-to-home charging priorities based on the return-to-home time probability distribution and the remaining power change curve.

[0142] In one embodiment of the present invention, the first homing scheduling submodule includes:

[0143] The charging pile status data acquisition unit is used to collect the location data and occupancy data of each charging pile in the drone nest; the occupancy data includes whether the charging pile is currently occupied, the identity of the occupied drone, the charging time, and the estimated remaining charging time.

[0144] The UAV status data acquisition unit is used to collect GPS flight trajectories, battery degradation curves, and inspection task semantic data of each UAV associated with the UAV nest.

[0145] In one embodiment of the present invention, the spatiotemporal conflict detection module 220 includes:

[0146] The first spatiotemporal conflict detection submodule is used to construct a spatiotemporal cube three-dimensional detection model based on the charging priority and the height, width and time of the UAV's flight space.

[0147] The second spatiotemporal conflict detection submodule is used to quantify the occupancy status of each spatiotemporal node through the spatiotemporal cube three-dimensional detection model.

[0148] The third spatiotemporal conflict detection submodule is used to determine that a path conflict exists when the number of occupancy of a certain spatiotemporal node is greater than 1.

[0149] In one embodiment of the present invention, the conflict state analysis module 230 includes:

[0150] The first conflict state analysis submodule is used to obtain the spatiotemporal nodes where the path conflict occurred and the identity identifiers of the conflicting UAVs when the prediction result indicates that there is a path conflict.

[0151] The second conflict status analysis submodule is used to extract the real-time flight status parameters of each conflicting UAV based on the identity identifier; the real-time flight status parameters include current position, current speed, current remaining battery power, return path and charging priority;

[0152] The third conflict state analysis submodule is used to calculate the predicted remaining battery power of each conflicting UAV upon arrival at the spatiotemporal node based on the real-time flight state parameters.

[0153] In one embodiment of the present invention, the avoidance scheduling module 240 includes:

[0154] The first avoidance scheduling submodule is used to generate several candidate avoidance paths based on the predicted remaining battery power and charging priority of the conflicting drone, with the goal of conflict avoidance.

[0155] The second obstacle avoidance scheduling submodule is used to select the obstacle avoidance path that does not conflict with the high-priority drone and has the least energy loss from the candidate obstacle avoidance paths, and then feed it back to the returning drone.

[0156] The third avoidance scheduling submodule is used to calculate the estimated return time of the drone based on the avoidance path;

[0157] The fourth avoidance scheduling submodule is used to update the charging priority based on the expected return time and feed it back to the returning drone.

[0158] In an additional embodiment of the invention, the system further includes:

[0159] The charging scheduling decision module is used to calculate the optimal charging pile allocation scheme and charging sequence through a pre-trained multi-objective deep reinforcement learning model. The state space of the multi-objective deep reinforcement learning model includes the occupancy status of each charging pile in the drone nest, the updated charging priority of each drone, the current remaining power, and the return path or avoidance path. The action space includes charging pile allocation actions and charging sequence adjustment actions.

[0160] The charging strategy generation module is used to generate a charging strategy based on the optimal charging pile allocation scheme and charging sequence and feed it back to the charging station.

[0161] The cell power consumption monitoring module is used to acquire the voltage and current parameters of the cell and calculate the instantaneous power consumption of the cell during the execution of the charging strategy.

[0162] The dynamic power consumption control module is used to dynamically adjust the speed of the drone returning to its nest or to introduce hovering waiting when the instantaneous power consumption is greater than a preset threshold.

[0163] In one embodiment of the present invention, the dynamic power consumption control module includes:

[0164] The first dynamic power consumption control submodule is used to obtain the real-time location, remaining power, estimated arrival time and updated charging priority of all returning drones when the instantaneous power consumption is greater than a preset threshold.

[0165] The second dynamic power consumption control submodule is used to send delay instructions to the returning drone in order of charging priority from low to high; the delay instructions include reducing air speed, introducing hovering and waiting in the air, or flying around along an extended path.

[0166] The third dynamic power consumption control submodule is used to send a delay cancellation command to the delayed drone when the instantaneous power consumption drops to within a preset threshold.

[0167] In an additional embodiment of the invention, the system further includes:

[0168] The pheromone field construction module is used to construct a spatiotemporal pheromone field, discretizing the UAV flight space into a grid.

[0169] The initial concentration generation module is used to generate the corresponding pheromone concentration based on the grid's historical conflict frequency, current occupancy status, and task urgency.

[0170] The conflict concentration enhancement module is used to increase the pheromone concentration of the grid where the conflict space node is located by a preset amount, and feed the updated pheromone concentration back to the returning drone; the returning drone is used to adjust its flight direction in real time based on the sensed pheromone concentration gradient during the return process.

[0171] Reference Figure 5 The diagram illustrates a computer electronic device for implementing a smart charging scheduling method for a drone nest according to the present invention, which may specifically include the following:

[0172] The aforementioned computer device 1 is in the form of a general-purpose computing device. The components of the computer device 1 may include, but are not limited to: one or more processors or processing units 3, memory 8, and a bus 4 connecting different system components (including memory 8 and processing unit 3).

[0173] Bus 4 represents one or more of several bus architectures, including memory buses or memory controllers, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Audio / Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0174] Computer device 1 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device 1, including volatile and non-volatile media, removable and non-removable media.

[0175] Memory 8 may include computer system readable media in the form of volatile memory, such as random access memory 9 and / or cache memory 10. Computer device 1 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 11 may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as a "hard disk drive"). Although Figure 5 As not shown, a disk drive for reading and writing to a removable non-volatile disk (such as a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (such as a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 4 via one or more data media interfaces. The memory may include at least one program product having a set (e.g., at least one) of program modules 13 configured to perform the functions of the embodiments of this application.

[0176] A program / utility 12 having a set (at least one) of program modules 13 may be stored, for example, in memory. Such program modules 13 include—but are not limited to—an operating system, one or more application programs, other program modules 13, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 13 typically perform the functions and / or methods described in the embodiments of this application.

[0177] Computer device 1 can also communicate with one or more external devices 2 (e.g., keyboard, pointing device, monitor 7, camera, etc.), and with one or more devices that enable an operator to interact with computer device 1, and / or with any device that enables computer device 1 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through I / O interface 6. Furthermore, computer device 1 can also communicate with one or more networks (e.g., local area network (LAN)), wide area network (WAN), and / or public networks (e.g., the Internet) through network adapter 5. Figure 5 As shown, network adapter 5 communicates with other modules of computer device 1 via bus 4. It should be understood that, although... Figure 5 Not shown, it can be combined with computer device 1 to use other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing unit 3, external disk drive array, RAID system, tape drive and data backup storage system 11, etc.

[0178] The processing unit 3 executes various functional applications and data processing by running programs stored in memory 8, such as implementing an intelligent charging scheduling method for a drone nest provided in the embodiments of this application.

[0179] That is, when the above-mentioned processing unit 3 executes the above-mentioned program, it performs the following: collecting charging pile status data and the status data of each UAV associated with the nest; generating a return path and charging priority based on the charging pile status data and the UAV status data; predicting spatiotemporal node path conflicts for the return path based on the charging priority; when the prediction result indicates that there is a path conflict, obtaining the spatiotemporal node where the path conflict occurred and the identity identifier of the conflicting UAV, and calculating the conflict status data of the conflicting UAV; generating an avoidance path and updating the charging priority based on the conflict status data, and feeding back the avoidance path and the updated charging priority to the returning UAV.

[0180] In this application embodiment, the application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements an intelligent charging scheduling method for a drone nest as provided in all embodiments of the application.

[0181] That is, when the program is executed by the processor, it performs the following: collects charging pile status data and the status data of each UAV associated with the nest; generates a return path and charging priority based on the charging pile status data and UAV status data; predicts spatiotemporal node path conflicts for the return path based on the charging priority; when the prediction result indicates that a path conflict exists, obtains the spatiotemporal node of the path conflict and the identity of the conflicting UAV, and calculates the conflict status data of the conflicting UAV; generates an avoidance path and updates the charging priority based on the conflict status data, and feeds back the avoidance path and the updated charging priority to the returning UAV.

[0182] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that includes or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0183] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0184] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the operator's computer, partially on the operator's computer, as a standalone software package, partially on the operator's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the operator's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider). The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably.

[0185] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0186] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0187] The above provides a detailed description of the intelligent charging scheduling method and system for unmanned aerial vehicle (UAV) nests provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A smart charging scheduling method for unmanned aerial vehicle (UAV) nests, applicable to multi-UAV nest scenarios during power transmission, distribution, and operational inspections, wherein the nest is equipped with multiple charging piles, characterized in that... Including the following steps: The system acquires charging pile status data and the status data of each drone associated with the drone nest. Based on the charging pile status data and drone status data, it generates return-to-home paths and charging priorities. The charging pile status data includes the location and occupancy data of each charging pile in the drone nest. The drone status data includes the drone's GPS flight trajectory, battery degradation curve, and inspection task semantic data. Specifically, the system acquires charging pile status data and the status data of each drone associated with the drone nest; extracts the spatiotemporal features of the charging pile status data and drone status data; and generates the return-to-home path, return-to-home time probability distribution, and remaining battery power change curve for each drone using a pre-established Spatiotemporal Relationship Network (STRN) prediction framework. Based on the return-to-home time probability distribution and remaining battery power change curve, it generates return-to-home charging priorities. Based on the charging priority, spatiotemporal node path conflict prediction is performed on the return path; When the prediction result indicates that a path conflict exists, the spatiotemporal node where the path conflict occurs and the identity identifier of the conflicting UAV are obtained, and the conflict status data of the conflicting UAV is calculated. Based on the conflict status data, an avoidance path is generated and the charging priority is updated, and the avoidance path and the updated charging priority are fed back to the returning drone. The optimal charging pile allocation scheme and charging sequence are calculated using a pre-trained multi-objective deep reinforcement learning model. The state space of the multi-objective deep reinforcement learning model includes the occupancy status of each charging pile in the drone nest, the updated charging priority of each drone, the current remaining power, and the return path or avoidance path. The action space includes charging pile allocation actions and charging sequence adjustment actions. Based on the optimal charging pile allocation scheme and charging sequence, a charging strategy is generated and fed back to the charging station. During the execution of the charging strategy in the cell, the voltage and current parameters of the cell are acquired, and the instantaneous power consumption of the cell is calculated. When the instantaneous power consumption exceeds a preset threshold, the flight speed of the returning drones is dynamically adjusted or hovering is introduced. Specifically, when the instantaneous power consumption exceeds the preset threshold, the real-time location, remaining battery power, estimated arrival time, and updated charging priority of all returning drones are obtained. Based on the charging priority from low to high, delay instructions are sent to the returning drones in sequence. The delay instructions include reducing flight speed or introducing hovering. When the instantaneous power consumption drops to within the preset threshold, a delay cancellation instruction is sent to the delayed drones.

2. The intelligent charging scheduling method for UAV nests according to claim 1, characterized in that, The step of extracting the spatiotemporal features of the charging pile status data and the drone status data specifically includes: Spatial features are generated by encoding the location data of the charging pile, the GPS flight trajectory of the drone, and the spatial coordinates in the semantics of the inspection task through a dilated convolutional network (TCN). The occupancy data of the charging pile, the battery degradation curve of the drone, and the timestamp information in the inspection task semantics are encoded by a bidirectional long short-term memory network (BiLSTM) to generate time features. The spatial features and the temporal features are fused to generate the spatiotemporal features.

3. The intelligent charging scheduling method for UAV nests according to claim 1, characterized in that, The step of predicting spatiotemporal node path conflicts for the return path based on the charging priority specifically includes: Based on the charging priority and the height, width, and time of the UAV's flight space, a three-dimensional spatiotemporal cube detection model is constructed. The occupancy status of each spatiotemporal node is quantified using the spatiotemporal cube 3D detection model. When the number of times a certain spatiotemporal node is occupied is greater than 1, it is determined that there is a path conflict.

4. The intelligent charging scheduling method for UAV nests according to claim 1, characterized in that, The steps of obtaining the spatiotemporal nodes where the path conflict occurred and the identity identifiers of the conflicting UAVs when the prediction result indicates a path conflict, and calculating the conflict state data of the conflicting UAVs, specifically include: When the prediction result indicates that a path conflict exists, obtain the spatiotemporal nodes where the path conflict occurred and the identification of the conflicting UAVs. Real-time flight status parameters of each conflicting drone are extracted based on the identification identifier; the real-time flight status parameters include current position, current speed, current remaining battery power, return path, and charging priority. Based on the real-time flight status parameters, the predicted remaining battery power of each conflicting UAV upon arrival at the spatiotemporal node is calculated.

5. The intelligent charging scheduling method for UAV nests according to claim 4, characterized in that, The steps of generating an avoidance path and updating the charging priority based on the conflict state data, and feeding back the avoidance path and the updated charging priority to the returning drone, specifically include: Based on the predicted remaining battery power and charging priority of the conflict drone, several candidate avoidance paths are generated with conflict avoidance as the goal. Among the candidate avoidance paths, the avoidance path that does not conflict with the high-priority drone and has the least energy loss is selected and fed back to the returning drone. Calculate the estimated return time of the drone based on the avoidance path; The charging priority is updated based on the estimated return time and fed back to the returning drone.

6. An intelligent charging scheduling system for unmanned aerial vehicle (UAV) nests, characterized in that, The steps for implementing the intelligent charging scheduling method for unmanned aerial vehicle (UAV) nests as described in any one of claims 1-5 include: The home return scheduling module is used to collect charging pile status data and the status data of each UAV associated with the home, and generate home return paths and charging priorities based on the charging pile status data and UAV status data. The spatiotemporal conflict detection module is used to predict spatiotemporal node path conflicts of the return path based on the charging priority. The conflict status analysis module is used to obtain the spatiotemporal nodes where the path conflict occurs and the identity identifiers of the conflicting UAVs when the prediction result indicates that a path conflict exists, and to calculate the conflict status data of the conflicting UAVs. The obstacle avoidance scheduling module is used to generate an obstacle avoidance path and update the charging priority based on the conflict status data, and to feed back the obstacle avoidance path and the updated charging priority to the returning drone. The charging scheduling decision module is used to calculate the optimal charging pile allocation scheme and charging sequence through a pre-trained multi-objective deep reinforcement learning model. The state space of the multi-objective deep reinforcement learning model includes the occupancy status of each charging pile in the drone nest, the updated charging priority of each drone, the current remaining power, and the return path or avoidance path. The action space includes charging pile allocation actions and charging sequence adjustment actions. The charging strategy generation module is used to generate a charging strategy based on the optimal charging pile allocation scheme and charging sequence and feed it back to the charging station. The cell power consumption monitoring module is used to acquire the voltage and current parameters of the cell and calculate the instantaneous power consumption of the cell during the execution of the charging strategy. The dynamic power consumption control module is used to dynamically adjust the speed of the drone returning to its nest or to introduce hovering waiting when the instantaneous power consumption is greater than a preset threshold.

7. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images