Shared aerial photography full-time air-remaining electric quantity scheduling method based on dual unmanned aerial vehicles

Abstract

The invention discloses a shared aerial photography full-time air-remaining electric quantity scheduling method based on dual unmanned aerial vehicles, and the method comprises the following steps: S1, receiving a task request of a user, and obtaining task parameters; s2, collecting current state data of the unmanned aerial vehicle, and generating task scheduling parameters; s3, predicting the power consumption of the unmanned aerial vehicle by using the long-short-term memory network model, and generating optimized scheduling parameters; s4, calculating a task response priority, and generating a task response sequence; s5, the unmanned aerial vehicles are distributed to execute tasks, and the electric quantity state and the task progress are monitored in real time; s6, the low-electric-quantity unmanned aerial vehicle returns to change the electricity, and a replacing unmanned aerial vehicle is selected from the task response sequence; and S7, updating the state data of the unmanned aerial vehicle, and optimizing a scheduling strategy and a model. Through the intelligent scheduling and energy management technology, full-time space reserving, efficient response and energy optimization of an unmanned aerial vehicle group in a shared aerial photography task are achieved, and the task completion rate and the resource utilization efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and particularly to a shared aerial photography full-time airborne power scheduling method based on dual unmanned aerial vehicles. Background Art

[0002] With the rapid development of unmanned aerial vehicle technology, its application fields have been continuously expanded. Especially in scenarios such as shared aerial photography, logistics transportation, and disaster monitoring, the collaborative task execution of unmanned aerial vehicle groups has become a hot topic in technical research. However, the existing technologies still face many challenges in the energy management and task scheduling of unmanned aerial vehicle groups, which limits their efficient application in complex dynamic environments.

[0003] Currently, in the task scheduling of unmanned aerial vehicle groups, energy management is usually achieved through simple power monitoring methods. These methods only focus on the remaining power status of unmanned aerial vehicles and lack the ability to dynamically predict the power consumption trend. Therefore, when the power of an unmanned aerial vehicle is insufficient during task execution, the task may be interrupted due to returning or losing contact, which further increases the complexity of the system and the uncertainty of task response. In addition, many existing methods fail to fully consider the time factors in the task response of unmanned aerial vehicles, such as the flight time and task execution time for the unmanned aerial vehicle to reach the task point, etc. These directly affect the task completion efficiency and user experience. In the peak task period or complex scenarios, this simple power management method is prone to cause task response delays or even failures, thus affecting the overall performance of the system.

[0004] In terms of task allocation, the existing unmanned aerial vehicle scheduling algorithms mainly adopt rule-based static designs. For example, tasks are sorted according to fixed priority rules, and the unmanned aerial vehicle closest to the task point is assigned. However, this method fails to fully consider the current comprehensive status of unmanned aerial vehicles, such as the dynamic balance of power level, task priority, and distance. Especially in an unmanned aerial vehicle group, the power status and task response capabilities of different unmanned aerial vehicles vary greatly. Simply relying on distance or task urgency for allocation may lead to waste of high-power unmanned aerial vehicle resources, or misallocation of low-power unmanned aerial vehicles to tasks, thereby increasing the power consumption and scheduling cost of the system. In addition, the ability to predict the status of unmanned aerial vehicles in task allocation is insufficient, and the deep learning technology fails to be combined to pre-judge the task requirements and optimize the response, which also results in limited adaptability of the existing scheduling system in dynamic task scenarios.

[0005] In the practical application of shared aerial photography services, in order to meet users' requirements for real-time response and efficient task completion, drones need to achieve a dynamic "full-time airborne" state. However, it is difficult for existing technologies to coordinate task switching and status updates among multiple drones while ensuring efficient energy utilization. On the one hand, after low-battery drones return, they are not replaced by high-battery drones in a timely manner, resulting in a drone-free period at the task point and affecting the continuity and stability of the task. On the other hand, existing methods lack the ability to optimize resource scheduling during peak task periods. For example, when multiple tasks are initiated simultaneously, how to reasonably allocate drone resources to maximize the task completion rate. These problems indicate that there is still much room for improvement in the task response efficiency, energy utilization efficiency, and task completion quality of existing technologies.

[0006] In addition, with the rapid development of artificial intelligence and machine learning technologies, research on applying deep learning and optimization algorithms to drone scheduling systems has gradually received attention. However, current research mainly focuses on single-task scenarios or specific algorithm optimizations, and fails to fully consider the multi-dimensional factors of the energy state and task requirements of the drone swarm for comprehensive scheduling. For example, the task allocation algorithm fails to dynamically predict the power consumption and task execution time of drones, and lacks a comprehensive consideration of the task priority allocation of drones in complex scenarios. At the same time, the research on the integration of deep learning models and traditional scheduling logics is not yet mature, and the robustness and generalization ability of the models in actual scenarios also need to be improved.

[0007] Therefore, how to provide a full-time airborne power scheduling method for shared aerial photography based on dual drones is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0008] An object of the present invention is to propose a full-time airborne power scheduling method for shared aerial photography based on dual drones. The present invention fully combines deep learning models, optimization algorithms, and reinforcement learning strategies, and details how to achieve precise power management of drones, task response priority allocation, and seamless switching scheduling of high- and low-battery drones in a complex dynamic environment, with the advantages of high energy utilization efficiency, fast task response speed, high task completion quality, and strong system robustness.

[0009] The full-time airborne power scheduling method for shared aerial photography based on dual drones according to an embodiment of the present invention includes the following steps:

[0010] S1. Receive a user task request, and obtain the spatial location, task type, task time limit, and user permission task parameters of the task point;

[0011] S2. Collect the current state data of the drone, associate the task parameters with the drone state data, and generate task scheduling parameters;

[0012] S3. Based on historical power consumption data and task scheduling parameters, use a long short-term memory network model to predict the power consumption of the drone, determine the power satisfaction of the drone's tasks, and generate optimized scheduling parameters;

[0013] S4. According to the power prediction results, combine the task priorities, the distances between the current positions of the drones and the task points, and use a deep neural network model to calculate the task response priorities and generate a task response ranking;

[0014] S5. According to the task response ranking, allocate appropriate drones to execute tasks, and monitor the power status and task progress of the drones executing tasks in real time;

[0015] S6. During the task execution process, when the power of the drone is lower than the set return threshold, trigger the drone return and battery replacement process, and select a matching drone from the task response ranking to take over the task;

[0016] S7. After the task is completed, update the status data of the drone, including remaining power, flight time record, and task execution situation, and optimize the long short-term memory network model and scheduling strategy.

[0017] Optionally, the specific content of S3 includes:

[0018] S31. Based on the historical power consumption data {E 1 , E 2 ,, E t} and the task scheduling parameters {D 1 , D 2 ,, D t} ， establish a long short-term memory network model, and define the input state vector as x t =[E t , D t ;

[0019] S32. Input the state vector x t into the LSTM unit to calculate the hidden state h t :

[0020] h t =σ(W f x t +U f h t-1 +b f )⊙c t-1 +tanh(W c x t +U c h t-1 +b c );

[0021] Among them, h tis the hidden state at time step t, W f , W c , U f , U c are weight matrices, b f , b c are bias vectors, σ is the activation function, tanh is the hyperbolic tangent activation function, ⊙ is the element-wise product, h t-1 is the hidden state at time step t - 1, c t-1 is the cell state at time step t - 1;

[0022] S33. Based on the hidden state h t , predict the power consumption result through multi-layer non-linear transformation and dynamic weighting mechanism

[0023]

[0024] where, W 1 , W 2 , W 3 are the weight matrices of the first layer, the second layer and the dynamic weighting layer respectively, b 1 , b 2 , b 3 are the bias vectors of the corresponding weight layers, ReLU is the rectified linear unit, γ is the dynamic weighting factor;

[0025] S34. Optimize the weight parameters of the LSTM model using the training data, and the loss function is defined as the mean square error MSE:

[0026]

[0027] where, y i is the actual power consumption, is the predicted power consumption, and N is the total number of samples;

[0028] S35. Based on the predicted power consumption result and the current UAV state data, including the remaining power, mission distance and mission time limit, generate optimized scheduling parameters.

[0029] Optionally, the specific steps of S4 include:

[0030] S41. Based on the generated optimized scheduling parameters P = {p 1 , p 2 ,..., p n}, where each p i includes the power of the UAV the current position L i , the distance D to the mission point i and the mission priority U i ;

[0031] S42. Define the task response priority function Φ i , which is used to measure the response ability of UAV i to the current task:

[0032]

[0033] where w 1 , w 2 , w 3 are the priority weights, and E max is the maximum battery power of the UAV;

[0034] S43. Based on the task response priorities {Φ 1 , Φ 2 ,, Φ n}, sort all UAVs according to the priority, and at the same time generate the final sorting result by combining the dynamic adjustment mechanism;

[0035]

[0036] Among them, sort from high to low according to the priority:

[0037] Priority_List = Sort({Φ′ 1 , Φ′ 2 ,, Φ n ′}, descending);

[0038] Among them, Φ i ' is the adjusted task response priority, λ is the dynamic adjustment factor, exp is the exponential function, θ is the distance influence factor, Sort is the sorting function, descending is the descending order, and Priority_List is the finally generated priority sorting list;

[0039] S44. From the priority sorting list Priority_List generated in step, select the UAV j with the highest response priority and meeting the conditions. The selection conditions are as follows:

[0040] Φ j = max({Φ 1 , Φ 2 ,, Φ n}) and

[0041] Among them, max is to take the maximum value in the set, is the predicted battery power of UAV j, E min is the minimum battery power threshold required to execute the task, and Φ j is the task response priority of UAV j;

[0042] S45. Assign the selected UAV j to the current task and update its status parameters.

[0043] Optionally, the S6 specifically includes:

[0044] S61. Monitor the status of the UAV executing the task in real time and obtain the status parameters Where:

[0045]

[0046] Among them, is the current remaining battery power of UAV j, L j is the current position of UAV j, T j is the task execution time, C j is the battery power consumed by the UAV for the task, k 1 is the battery consumption coefficient for the flight distance, k 2 is the battery consumption coefficient for the task execution time, (x j , y j ) is the current position of the UAV, (x t , y t ) are the coordinates of the task point;

[0047] S62. Judge the current UAV battery power whether it is lower than the return battery threshold E min :

[0048]

[0049] Among them, D j,home is the distance from UAV j to the return point, k 3 is the basic battery consumption coefficient for the flight return, γ is the dynamic environment factor, exp(-α·T j ) is the attenuation function of the task execution time T j α is the time influence factor, R is the battery reserve safety threshold, ρ is the prediction error influence factor, is the average value of the predicted battery powers of all current UAVs, is the current remaining battery power of UAV i, n is the number of UAV groups;

[0050] S63. If the condition is met Immediately trigger the return process, and at the same time, among the priority sorting results, select the UAV k with the second highest response priority and meeting the conditions:

[0051] Φ k′ = Φ k + η·exp(-β·D k ) + ε·W task ;

[0052]

[0053] Among them, Φ k ' is the task response priority after dynamic adjustment, Φ k is the basic priority, η is the distance dynamic adjustment factor, exp(-β·D k ) is the distance attenuation function, β is the distance influence coefficient, D k is the distance from the UAV k to the task point, ε is the task type influence factor, W task is the task type weight, argmax i is to find the UAV number k that maximizes the dynamic priority Φ ' among all UAVs that meet the power constraint k ;

[0054] S64. After selecting the replacement UAV, update the task execution status, the status of the returning UAV, and the status of the replacement UAV in real time:

[0055]

[0056] Among them, the status of the returning UAV

[0057]

[0058] Among them, the status of the replacement UAV

[0059]

[0060] Among them, S task is the comprehensive status of the task and the UAV, T executed is the time that the current task has been executed, T total is the total expected time of the task, D j,home is the distance from the UAV j to the return point, v j is the flight speed of the UAV, k 1 , k 2 are the power consumption coefficients of the flight distance and the task execution time respectively, D k is the distance from the replacement UAV to the task point, T remaining is the remaining execution time of the task;

[0061] S65. After the replacement UAV k successfully completes the task, update the task status and the UAV status, generate the final task completion data, and feedback it to the scheduling system.

[0062] Optionally, the specific content of the S7 includes:

[0063] S71. Record the status data of the drone after the task is completed, including the remaining battery power The current location L j 、The task execution time T j And the power consumption E for task execution used , and update the status representation:

[0064]

[0065] Among them, the total power consumption E of the drone for task execution used :

[0066] E used = k 1 ·D m + k 2 ·T j ;

[0067] Among them, S j Is the status data vector of drone j, k 1 Is the power consumption coefficient for flight distance, k 2 Is the power consumption coefficient for task execution time, D m Is the task flight distance;

[0068] S72. Update the task completion status and execution record to the historical task database. The task record includes:

[0069] R task =(T completion , E used , L start , L end );

[0070] Among them, T completion Is the task completion time, L start And L end Are the coordinates of the task start point and end point respectively, R task Is the task completion record;

[0071] S73. Use the historical task records and drone status data to optimize the scheduling model M, with the goal of minimizing the model error:

[0072]

[0073] Among them, y i Is the actual task result, Is the model prediction result, N is the number of historical task records, λR(M) is the model regularization term, argmin M Is to find the parameter M to minimize the objective function, M new Is the optimized scheduling model parameter set;

[0074] S74. Optimize the task scheduling strategy based on reinforcement learning, and define the cumulative reward function as follows:

[0075]

[0076] where π is the scheduling strategy, is the discount factor, ω is the trajectory, and R t is the immediate reward at time step t, J(π) is the cumulative reward function of the scheduling strategy, and E ω～π is the expected value, and T is the total number of time steps of the task;

[0077] S75. Update the optimized scheduling model M new and the scheduling strategy to the scheduling system, and feedback the UAV status data.

[0078] The beneficial effects of the present invention are as follows:

[0079] By introducing a deep learning model and a reinforcement learning algorithm, and combining the energy state and task requirements of the UAV swarm, the present invention realizes the dynamic scheduling of UAV task response and power management, significantly improving the efficiency and reliability of the shared aerial photography service. Compared with the prior art, the present invention can accurately predict the power consumption and task completion time of UAVs, avoiding the situation of task interruption or delay caused by insufficient power in the traditional method. By optimizing the task response priority allocation strategy, the present invention can efficiently allocate high-power UAVs to replace low-power UAVs in dynamic task scenarios, ensuring full-time hovering and continuity at task points, so as to meet the requirements of real-time task response in complex scenarios.

[0080] In addition, the scheduling strategy proposed by the present invention combines multi-dimensional UAV status data and task requirement characteristics, making the scheduling process more intelligent and accurate. By dynamically adjusting task allocation through a deep neural network, the problems of resource waste and response delay caused by static rules in the prior art are solved. At the same time, the introduction of the reinforcement learning algorithm enables the scheduling strategy to be continuously optimized during actual operation, maximizing the task completion rate and improving the resource utilization efficiency of the system. The efficient energy management and task replacement strategy of UAVs significantly reduces the power consumption during task execution, enabling the system to have higher continuous operation ability.

[0081] The present invention also precisely controls the timing of the UAV's return and handover through an optimized algorithm, enabling the UAV swarm to achieve reasonable resource allocation during the peak mission period and effectively avoiding the problem of response timeout caused by excessive mission requirements. In addition, the feedback of historical mission data and the continuous update of the model make the scheduling system more adaptable and robust, capable of handling mission requirements in various complex dynamic environments. These beneficial effects not only significantly improve the quality of the UAV shared aerial photography service but also provide an innovative solution for UAV swarm scheduling and energy optimization management technology, with broad application prospects and economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0083] Figure 1 is a flowchart of the full-time airborne power scheduling method for shared aerial photography based on dual UAVs proposed by the present invention;

[0084] Figure 2 is a schematic structural diagram of the UAV power prediction model of the full-time airborne power scheduling method for shared aerial photography based on dual UAVs proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0085] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0086] Refer to Figure 1-2 , the full-time airborne power scheduling method for shared aerial photography based on dual UAVs includes the following steps:

[0087] S1. Receive the user's mission request, and obtain the spatial position, mission type, mission time limit, and user permission mission parameters of the mission point;

[0088] S2. Collect the current state data of the UAV, associate the mission parameters with the UAV state data, and generate mission scheduling parameters;

[0089] S3. Based on the historical power consumption data and mission scheduling parameters, use the long short-term memory network model to predict the power consumption of the UAV, determine the power consumption satisfaction of the UAV's mission, and generate optimized scheduling parameters;

[0090] S4. According to the power prediction result, combined with the mission priority and the distance between the current position of the UAV and the mission point, use the deep neural network model to calculate the mission response priority and generate a mission response ranking;

[0091] S5. Sort according to task response, allocate appropriate drones to execute tasks, and monitor the power status and task progress of the drones executing tasks in real time;

[0092] S6. During the task execution, when the drone's power is lower than the set return threshold, trigger the drone return and battery replacement process, and select a matching drone from the task response sorting to take over the task;

[0093] S7. After the task is completed, update the status data of the drone, including remaining power, flight time record, and task execution status, and optimize the long short-term memory network model and scheduling strategy.

[0094] In this embodiment, the S3 specifically includes:

[0095] S31. Based on the historical power consumption data {E 1 ,E 2 ,,E t} and task scheduling parameters {D 1 ,D 2 ,,D t}, establish a long short-term memory network model, and define the input state vector as x t = [E t ,D t :

[0096] S32. Input the state vector x t into the LSTM unit to calculate the hidden state h t :

[0097] h t = σ(W f x t + U f h t-1 + b f ) ⊙ c t-1 + tanh(W c x t + U c h t-1 + b c );

[0098] Among them, h t is the hidden state at time step t, W f ,W c ,U f ,U c are weight matrices, b f ,b c are bias vectors, σ is the activation function, tanh is the hyperbolic tangent activation function, ⊙ is the element-wise product, h t-1 is the hidden state at time step t-1, and c t-1 is the cell state at time step t-1;

[0099] S33. According to the hidden state h t , predict the power consumption result through multi-layer non-linear transformation and dynamic weighting mechanism

[0100]

[0101] where W 1 , W 2 , W 3 are the weight matrices of the first layer, the second layer and the dynamic weighting layer respectively, b 1 , b 2 , b 3 are the bias vectors of the corresponding weight layers respectively, ReLU is the rectified linear unit, and γ is the dynamic weighting factor;

[0102] S34. Optimize the weight parameters of the LSTM model using the training data, and the loss function is defined as the mean square error MSE:

[0103]

[0104] where y i is the actual power consumption, is the predicted power consumption, and N is the total number of samples;

[0105] S35. Based on the predicted power consumption result and the current UAV state data, including remaining power, mission distance and mission time limit, generate optimized scheduling parameters.

[0106] In this embodiment, the specific steps of S4 are as follows:

[0107] S41. Based on the generated optimized scheduling parameters P = {p 1 , p 2 ,..., p n}, where each p i includes the power of the UAV the current position L i , the distance D i to the mission point and the mission priority U i ;

[0108] S42. Define the mission response priority function Φ i , which is used to measure the response ability of UAV i to the current mission:

[0109]

[0110] where w 1 , w 2 , w 3 are the priority weights, Emax is the maximum power of the drone;

[0111] S43. Based on the task response priorities {Φ 1 , Φ 2 ,, Φ n}, sort all drones according to the priorities, and at the same time generate the final sorting result by combining the dynamic adjustment mechanism;

[0112]

[0113] Among them, sort from high to low according to the priorities:

[0114] Priority_List = Sort({Φ 1 ′, Φ 2 ′,, Φ n ′}, descending);

[0115] Among them, Φ i ' is the adjusted task response priority, λ is the dynamic adjustment factor, exp is the exponential function, θ is the distance influence factor, Sort is the sorting function, descending is in descending order, and Priority_List is the finally generated priority sorting list;

[0116] S44. From the priority sorting list Priority_List generated in step, select the drone j with the highest response priority and meeting the conditions. The selection conditions are as follows:

[0117] Φ j = max({Φ 1 , Φ 2 ,, Φ n}) and

[0118] Among them, max is to take the maximum value in the set, is the predicted power of drone j, E min is the minimum power threshold required to execute the task, Φ j is the task response priority of drone j;

[0119] S45. Assign the selected drone j to the current task and update its status parameters.

[0120] In this embodiment, the S6 specifically includes:

[0121] S61. Real-time monitor the status of the task execution drone and obtain the status parameters Among them:

[0122]

[0123] Among them, is the current remaining power of UAV j, L j is the current position of UAV j, T j is the mission execution time, C j is the power consumed by the UAV mission, k 1 is the power consumption coefficient for flight distance, k 2 is the power consumption coefficient for mission execution time, (x j , y j ) is the current position of the UAV, (x t , y t ) are the coordinates of the mission point;

[0124] S62. Determine whether the current UAV power is lower than the return power threshold E min :

[0125]

[0126] Among them, D j,home is the distance from UAV j to the return point, k 3 is the basic power consumption coefficient for flight return, γ is the dynamic environment factor, exp(-α·T j ) is the attenuation function of the mission execution time T j where α is the time impact factor, R is the power margin safety threshold, ρ is the prediction error impact factor, is the average value of the predicted power of all current UAVs, is the current remaining power of UAV i, and n is the number of UAV groups;

[0127] S63. If the condition is met immediately trigger the return process. At the same time, among the priority sorting results, select the UAV k with the second highest response priority and meeting the conditions:

[0128] Φ k ′ = Φ k +η·exp(-β·D k )+ε·W task ;

[0129]

[0130] Among them, Φ k ' is the dynamically adjusted mission response priority, Φ k is the basic priority, η is the distance dynamic adjustment factor, exp(-β·D k ) is the distance attenuation function, β is the distance impact coefficient, D k is the distance from UAV k to the mission point, and ε is the mission type impact factor, Wtask is the task type weight, argmax i is to find, among all the drones that meet the power constraint the drone number k that maximizes the dynamic priority Φ k ';

[0131] S64. After selecting the replacement drone, update the task execution status, the status of the returning drone, and the status of the replacement drone in real time:

[0132]

[0133] Among them, the status of the returning drone

[0134]

[0135] Among them, the status of the replacement drone

[0136]

[0137] Among them, S task is the comprehensive status of the task and the drone, T executed is the time elapsed for the current task execution, T total is the total estimated time of the task, D j,home is the distance of drone j to the return point, v j is the flight speed of the drone, k 1 , k 2 are the power consumption coefficients for the flight distance and the task execution time respectively, D k the distance of the replacement drone to the task point, T remaining is the remaining execution time of the task;

[0138] S65. After the replacement drone k successfully completes the task, update the task status and the drone status, generate the final task completion data, and feedback it to the scheduling system.

[0139] In this embodiment, the specific content of the S7 includes:

[0140] S71. Record the status data of the drone after the task is completed, including the remaining power the current position L j the task execution time T j and the power consumption E for the task execution used , and update the status representation:

[0141]

[0142] Among them, the total power consumption E of the drone for task execution used :

[0143] E used = k 1 ·D m + k 2 ·T j ;

[0144] Among them, S j is the state data vector of UAV j, k 1 is the power consumption coefficient of the flight distance, k 2 is the power consumption coefficient of the mission execution time, D m is the mission flight distance;

[0145] S72. Update the mission completion status and execution record to the historical mission database. The mission record includes:

[0146] R task = (T completion , E used , L start , L end );

[0147] Among them, T completion is the mission completion time, L start and L end are the coordinates of the mission start point and end point respectively, and R task is the mission completion record;

[0148] S73. Use the historical mission record and UAV state data to optimize the scheduling model M. The goal is to minimize the model error:

[0149]

[0150] Among them, y i is the actual mission result, is the model prediction result, N is the number of historical mission records, λR(M) is the model regularization term, argmin M is to find the parameter M to minimize the objective function, and M new is the optimized scheduling model parameter set;

[0151] S74. Optimize the task scheduling strategy based on reinforcement learning. Define the cumulative reward function as:

[0152]

[0153] Among them, π is the scheduling strategy, is the discount factor, ω is the trajectory, R t is the immediate reward at time step t, and J(π) is the cumulative reward function of the scheduling strategy, E ω～πis the expected value, T is the total time steps of the task;

[0154] S75, the optimized scheduling model M new The dispatching system is updated with dispatching strategies and the drone status data is fed back.

[0155] Embodiment 1:

[0156] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a famous tourist attraction, which has a strong demand from tourists and has very high requirements for high-quality real-time aerial photography services. However, due to the vast area and complex terrain of the scenic area, as well as the surge in task demand during periods of concentrated tourist demand, traditional drone scheduling solutions are difficult to meet the needs of real-time task response, especially during peak mission periods, which are prone to task interruptions due to insufficient drone power, or waste of resources due to improper scheduling. The present invention designs a full-time energy scheduling method for shared aerial photography based on a dual-swarm of drones. Through intelligent scheduling and power management technology, it solves the problems of untimely drone task response, low energy utilization efficiency, and low quality of task completion.

[0157] In this scenario, we deployed two drone groups, each containing 20 drones, a total of 40 drones, covering the aerial photography mission points of the entire scenic area. The main task of the drone is to conduct all-weather aerial photography of the five key scenic spots in the scenic area to meet the needs of tourists for real-time viewing and video live broadcast. The mission execution time for each mission point ranges from 15 minutes to 30 minutes, and the flight distance from the starting position of the drone is between 2 kilometers and 5 kilometers. The initial power of the drone is fully charged (100%), the maximum battery life is 50 minutes, and the peak mission period is mainly concentrated between 10 am and 2 pm.

[0158] In actual applications, the scheduling system receives real-time task requests and assigns priorities based on factors such as drone status, task urgency, and distance. For example, when task point A initiates a task request, the scheduling system uses the task response priority model proposed by the present invention to comprehensively consider the distance between the drone and the task point, the current power status, and the urgency of the task, and selects the optimal drone B to go to task point A to perform the task. At the same time, the system uses a deep learning model to predict the power consumption of drone B. If it is found that its power may not be enough to complete the task, a high-power drone C is arranged in advance as a replacement. Through such a dynamic scheduling and replacement mechanism, the drone group can achieve seamless switching during the peak period of the task to ensure that the task point is vacant all the time.

[0159] To verify the beneficial effects of the present invention, we selected the peak period from 10:00 am to 2:00 pm as the experimental time period, and recorded and analyzed data such as the UAV task completion rate, task response time, power consumption, and the number of UAV replacements. The experimental site was the above-mentioned scenic area. During the experiment, the task requirements at 5 task points were simulated. The total number of task requests was 100, and the tasks were randomly distributed among the 5 task points. The experimental data was compared with the traditional scheduling method.

[0160] Table 1 Performance comparison table between the present invention and the traditional method in shared aerial photography tasks

[0161]

[0162] According to the table data, the advantages of the present invention in terms of task completion rate, response time, power consumption, and the number of UAV replacements can be summarized.

[0163] In terms of the task completion rate, the present invention is significantly superior to the traditional method. The total completion rate reaches 97%, which is 14 percentage points higher than the traditional method. Especially at task points B and D, the completion rate of the traditional method is relatively low, and the present invention is increased to 96% and 94% respectively, fully demonstrating the high efficiency of intelligent scheduling.

[0164] In terms of the task response time, the average response time of the present invention is 26 seconds, which is 40% shorter than the 43 seconds of the traditional method. For example, the response time at task point D is shortened from 50 seconds to 30 seconds, showing the excellent performance of the priority allocation model in reducing delays.

[0165] In terms of power consumption, the present invention reduces the average power consumption from 68% to 60%, achieving a saving of 8 percentage points. The power consumption at task points A and B is reduced to 58% and 63% respectively, indicating the ability of the present invention in efficiently managing the energy of UAVs.

[0166] The number of UAV replacements has also been optimized. The present invention controls the total number of replacements within 17 times through an intelligent replacement mechanism, greatly reducing resource waste and ensuring task continuity.

[0167] In summary, the present invention shows significant advantages in terms of task completion rate, response time, power consumption, and the number of replacements, especially demonstrating strong stability during the task peak period, providing an efficient solution for UAV swarm scheduling.

[0168] The above is only the preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A full-time free-space power scheduling method based on shared aerial photography of two unmanned aerial vehicles, characterized in that: The steps include: S1. Receive user task request, obtain the spatial location of the task point, task type, task time limit, user authority task parameters; S2, collect the current status data of the drone, associate the task parameters with the drone status data, and generate task scheduling parameters; S3. Based on the historical power consumption data and task scheduling parameters, the long short-term memory network model is used to predict the power consumption of the UAV, determine the power satisfaction of the UAV's mission, and generate optimized scheduling parameters; S4. Based on the power prediction results, combined with the task priority and the distance between the current position of the drone and the task point, the task response priority is calculated using the deep neural network model to generate the task response ranking; S5. Assign appropriate drones to perform tasks according to the task response order, and monitor the power status and task progress of the drones performing the tasks in real time; S6. During the mission execution, when the UAV battery level is lower than the set return threshold, the UAV return and battery replacement process is triggered, and a matching UAV is selected from the mission response sequence to take over the mission; S7. After the mission is completed, update the drone's status data, including remaining power, flight time record and mission execution status, and optimize the long-short-term memory network model and scheduling strategy.

2. The method for scheduling full-time free space electricity based on shared aerial photography of two unmanned aerial vehicles according to claim 1 is characterized in that: The S3 specifically includes: S31, based on the historical power consumption data {E1, E2,, E t } and task scheduling parameters {D1,D2,,D t }, establish a long short-term memory network model, and define the input state vector as x t =[E t ,D t ]: S32, the state vector x t Input LSTM unit and calculate hidden state h t : h t =σ(W f x t +U f h t-1 +b f )⊙c t-1 +tanh(W c x t +U c h t-1 +b c ); Among them, h t is the hidden state at time step t, W f ,W c ,U f ,U c is the weight matrix, b f ,b c is the bias vector, σ is the activation function, tanh is the hyperbolic tangent activation function, ⊙ is the element-by-element product, and h t-1 is the hidden state at time step t-1, c t-1 is the cell state at time step t-1; S33, according to the hidden state h t , predicting power consumption results through multi-layer nonlinear transformation and dynamic weighting mechanism Among them, W1, W2, W3 are the weight matrices of the first layer, the second layer and the dynamic weighted layer respectively, b1, b2, b3 are the bias vectors of the corresponding weighted layers respectively, ReLU is the rectified linear unit, and γ is the dynamic weighting factor; S34. Use training data to optimize the LSTM model weight parameters. The loss function is defined as the mean square error MSE: Among them, y i is the actual power consumption, To predict power consumption, N is the total number of samples; S35, based on the predicted power consumption result Combined with the current drone status data, including remaining battery power, mission distance, and mission time limit, the optimized scheduling parameters are generated.

3. The method for scheduling full-time free space electricity based on shared aerial photography of two unmanned aerial vehicles according to claim 1 is characterized in that: The S4 specifically includes: S41, based on the generated optimized scheduling parameters P = {p1, p2,, p n }, where each p i Including the battery of the drone Current locationL i , the distance to the task point D i and task priority U i ; S42. Define the task response priority function Φ i , used to measure the responsiveness of the drone i to the current task: Among them, w1, w2, w3 are priority weights, E max The maximum power of the drone; S43, based on the task response priority {Φ1, Φ2,, Φ n }, sort all drones according to their priorities, and generate the final sorting results in combination with the dynamic adjustment mechanism; Among them, sorted from high to low priority: Priority_List=Sort({Φ1’,Φ2’,,Φ n ’},descending); Among them, Φ i ' is the adjusted task response priority, λ is the dynamic adjustment factor, exp is the exponential function, θ is the distance influence factor, Sort is the sorting function, descending is descending, and Priority_List is the final generated priority sorting list; S44. Select the drone j with the highest response priority and meeting the conditions from the priority sorting list Priority_List generated in step 44. The selection conditions are as follows: Φ j = max({Φ1, Φ2,, Φ n}) and Among them, max is the maximum value in the set. is the predicted power of drone j, E min is the minimum power threshold required to perform the task, Φ j is the mission response priority of UAV j; S45. Assign the selected UAV j to the current task and update its state parameters.

4. The method for scheduling full-time free space electricity based on shared aerial photography of two unmanned aerial vehicles according to claim 1 is characterized in that: The S6 specifically includes: S61. Real-time monitoring of the status of the mission execution drone and obtaining status parameters in: in, is the current remaining power of drone j, L j is the current position of UAV j, T j is the task execution time, C j is the power consumed by the UAV mission, k1 is the power consumption coefficient of the flight distance, k2 is the power consumption coefficient of the mission execution time, (x j ,y j ) is the current position of the drone, (x t ,y t ) is the coordinate of the task point; S62. Determine the current battery level of the drone Is the battery level lower than the return-to-home threshold E? min : Among them, D j,home is the distance from UAV j to the return point, k3 is the basic power consumption coefficient of flight return, γ is the dynamic environmental factor, exp(-α·T j ) is the task execution time T j The attenuation function is: α is the time impact factor, R is the power reserve safety threshold, ρ is the prediction error impact factor, The average power value predicted for all current drones. is the current remaining power of drone i, and n is the number of drones in the group; S63. If the conditions are met The return process is triggered immediately. At the same time, the drone k with the second highest response priority and that meets the conditions is selected from the priority sorting results: F k '=Φ k +η·exp(-β·D k )+ε·W task ; Among them, Φ k ' is the task response priority after dynamic adjustment, Φ k is the basic priority, η is the distance dynamic adjustment factor, exp(-β·D k ) is the distance attenuation function, β is the distance influence coefficient, D k is the distance from UAV k to the mission point, ε is the mission type influencing factor, W task is the task type weight, argmax i To satisfy the power constraints Among the drones, find the dynamic priority Φ k 'The largest drone number k; S64. After selecting the replacement drone, update the mission execution status, return drone status and replacement drone status in real time: Among them, the return drone status Among them, taking over the drone status Among them, S task is the combined status of the mission and the UAV, T executed is the execution time of the current task, T total The estimated total time for the task, D j,home is the distance from UAV j to the return point, v j is the flight speed of the drone, k1 and k2 are the power consumption coefficients of the flight distance and mission execution time respectively, and D k The distance from the replacement drone to the mission point, T remaining The remaining execution time of the task; S65. After the replacement UAV k successfully completes the task, the task status and UAV status are updated, the final task completion data is generated, and fed back to the scheduling system.

5. The method for scheduling full-time free space electricity based on shared aerial photography of two unmanned aerial vehicles according to claim 1 is characterized in that: The S7 specifically includes: S71. Record the status data of the drone after the mission is completed, including the remaining battery power Current locationL j , Task execution time T j And the power consumption E of task execution used , the updated status means: Among them, the total power consumption of the drone to perform the task is E used : E used =k1·D m +k2·T j ; Among them, S j is the state data vector of drone j, k1 is the power consumption coefficient of flight distance, k2 is the power consumption coefficient of mission execution time, D m The flight distance for the mission; S72, updating the task completion status and execution record to the historical task database, the task record includes: R task =(T completion ,E used ,L start ,L end ); Among them, T completion is the task completion time, L start and L end are the coordinates of the starting point and the end point of the task, R task Recording task completion; S73. Use historical mission records and drone status data to optimize the scheduling model M, with the goal of minimizing the model error: Among them, y i For the actual task results, is the model prediction result, N is the number of historical task records, λR(M) is the model regularization term, argmin M To find the parameter M that minimizes the objective function, M new is the optimized scheduling model parameter set; S74. Based on reinforcement learning to optimize the task scheduling strategy, the cumulative reward function is defined as: Among them, π is the scheduling strategy, θ is the discount factor, ω is the trajectory, and R t is the instantaneous reward at time step t, J(π) is the cumulative reward function of the scheduling strategy, and E ω～π is the expected value, T is the total time steps of the task; S75, the optimized scheduling model M new The dispatching system is updated with dispatching strategies and the drone status data is fed back.

Images