Unmanned aerial vehicle distribution task real-time optimization scheduling management system

Abstract

The invention relates to the technical field of resource optimization, in particular to an unmanned aerial vehicle distribution task real-time optimization scheduling management system, which comprises a bat sound wave calculation module for performing echo positioning frequency conversion calculation on the distance between the current position of an unmanned aerial vehicle and a target distribution point, a time window and a carrying weight upper limit value through a bat algorithm, and obtaining a task sound wave waveform matrix. According to the invention, the application of the bat algorithm enables the unmanned aerial vehicle to accurately calculate the task search space when the unmanned aerial vehicle executes the distribution task, and avoids the resource waste caused by a static distribution mode. The task search vector is formed based on the comparison operation of the task waveform amplitude and the distribution distance, so that the unmanned aerial vehicle can quickly adjust the task priority in a dynamic environment, and the execution accuracy of the distribution task is improved. The task group is optimized through the subspace decomposition operation, so that the unmanned aerial vehicle can execute multiple tasks in a parallel mode, the load pressure of a single unmanned aerial vehicle is reduced, and the cooperation efficiency of the overall tasks is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of resource optimization, and particularly to a real-time optimization scheduling management system for UAV delivery tasks. Background Art

[0002] The real-time optimization scheduling management system for UAV delivery tasks is an intelligent scheduling platform based on resource optimization technology, aiming to dynamically allocate UAVs to execute delivery tasks according to delivery requirements, UAV performance parameters, and environmental variables, and optimize flight paths, scheduling sequences, and emergency response strategies.

[0003] However, the existing technology relies on static task allocation methods, making it difficult for UAVs to adapt to the dynamic changes in the delivery environment during task execution, resulting in low flexibility in task allocation. Due to the failure to form an accurate task search space, problems such as uneven task execution efficiency of UAVs are likely to occur during task allocation, and some tasks may be delayed due to the failure to match suitable UAVs. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies existing in the prior art and propose a real-time optimization scheduling management system for UAV delivery tasks.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: The real-time optimization scheduling management system for UAV delivery tasks includes:

[0006] A bat sound wave calculation module, which performs echo location frequency conversion calculations on the distance between the current position of the UAV and the target delivery point, time window, and upper limit value of the carrying weight through the bat algorithm to obtain a task sound wave waveform matrix; compares and calculates the waveform amplitude in the task sound wave waveform matrix with the delivery distance to generate a task search space vector;

[0007] A task resource allocation module, which performs subspace decomposition operations on the task search space vector through the bat algorithm to obtain a parallel task group sequence; performs constraint operations on the parallel task group sequence with the remaining battery power, load capacity, and flight speed of the UAV to generate a resource matching allocation plan;

[0008] A scheduling optimization execution module, which groups and calculates the delivery distance and time window in the resource matching allocation plan according to geographical locations to generate a multi-task parallel scheduling instruction;

[0009] An emergency response processing module, which performs emergency degree evaluation operations on the real-time order increment at the station, the available number of UAVs, and the task completion time limit to obtain an emergency task priority sequence.

[0010] Preferably, the steps for obtaining the task sound wave waveform matrix are as follows:

[0011] Set the coordinates of the current position of the drone and the target delivery point, calculate the Euclidean distance between the two points, and obtain the distance data;

[0012] Based on the distance data, calculate the echo location frequency conversion value. The calculation formula is:

[0013]

[0014] where TD is the distance between the two points, T w is the time window, W max is the upper limit of the carrying weight, K is a constant affecting the adjustment unit, F echo is the echo location frequency conversion value;

[0015] Based on the echo location frequency conversion value, simulate the sound wave frequency conversion of echo location through the bat algorithm to generate a task sound wave waveform matrix.

[0016] Preferably, the steps for obtaining the task search space vector are as follows:

[0017] Extract the amplitude of each waveform in the task sound wave waveform matrix, perform index matching on the amplitude data according to the delivery distance, screen the amplitude range that meets the delivery task, and generate the waveform amplitude data after matching;

[0018] Based on the waveform amplitude data after matching, calculate the task search space vector. The calculation formula is:

[0019]

[0020] where A 1 is the maximum value of the waveform amplitude data, A 2 is the minimum value of the waveform amplitude data, B 1 is the square multiplication term of the delivery distance, B 2 is the sine mapping term of the delivery distance, PD is the delivery distance, PC is the normalized offset term of the delivery distance and the waveform amplitude, A 3 is the average value of the waveform amplitude data, S task is the task search space vector.

[0021] Preferably, the steps for obtaining the parallel task group sequence are as follows:

[0022] Based on the task search space vector, use the bat algorithm to analyze the features, identify the potential subspace boundaries, and obtain the initial subspace feature data;

[0023] Based on the initial subspace feature data, perform subspace decomposition operation. The formula is:

[0024]

[0025] Among them, θ i represents the position angle of the i-th data point, V i represents the search space vector value of the i-th task, H groups represents the subspace decomposition result, and RN represents the total number of tasks in the task search space vector;

[0026] Based on the subspace decomposition result, group the data points according to similarity to form a parallel task group sequence.

[0027] Preferably, the steps for obtaining the resource matching and allocation scheme are as follows:

[0028] Based on the parallel task group sequence, calculate the matching degree between the task and the UAV resources. The calculation formula is:

[0029]

[0030] Among them, R match is the matching degree between the task and the UAV resources, E is the remaining battery power of the UAV, C is the minimum battery power required for the task, ZL is the current load of the UAV, S is the flight speed of the UAV, RD is the task delivery distance, and P is the maximum delivery distance of the UAV;

[0031] Allocate UAVs according to the matching degree between the task and the UAV resources to generate a resource matching and allocation scheme.

[0032] Preferably, the steps for obtaining the task space clustering result are as follows:

[0033] Extract the delivery distance and time window of each task from the resource matching and allocation scheme, classify the tasks according to geographical location, calculate the distribution density of tasks in each region, and obtain the regional task distribution data;

[0034] Based on the regional task distribution data, calculate the task grouping fitness. The calculation formula is:

[0035]

[0036] Among them, F group is the task grouping fitness, D i is the delivery distance of the i-th task, D avg is the average delivery distance of all tasks in the region, T i is the time window of the i-th task, T max is the maximum time window of tasks in the region, L i is the geographical location coordinate of the i-th task, L c is the central location coordinate of the region, and N is the total number of tasks in the current region;

[0037] Generate a multi-task parallel scheduling instruction based on the task grouping fitness.

[0038] Preferably, the steps for obtaining the emergency task priority sequence are as follows:

[0039] Collect the real-time order increment of the site, the available number of drones, and the task completion time limit to obtain the basic data of task urgency;

[0040] Based on the basic data of task urgency, calculate the emergency task priority score, and the calculation formula is:

[0041]

[0042] Among them, P emg is the emergency task priority score, Q 1 represents the current order increment, Q 2 represents the historical average order level, M represents the current available number of drones, M th represents the minimum number of drones required for the task, Q 3 represents the minimum order processing capacity required for task execution, T rem represents the remaining time of the task;

[0043] Based on the emergency task priority score, sort the tasks according to the priority to obtain the emergency task priority sequence.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0045] In the present invention, the application of the bat algorithm enables the drone to accurately calculate the task search space when performing delivery tasks, avoiding resource waste caused by static allocation methods. Based on the comparison operation between the task waveform amplitude and the delivery distance, a task search vector is formed, enabling the drone to quickly adjust the task priority in a dynamic environment and improving the execution accuracy of delivery tasks. The subspace decomposition operation optimizes the task group, enabling the drone to execute multiple tasks in parallel, reducing the load pressure on a single drone, and enhancing the overall task collaboration efficiency. By combining the battery power, load capacity, and flight speed of the drone for constraint operations, the rationality of task allocation is ensured, enabling the drone to maintain the maximum endurance when performing tasks and reducing the probability of task failure due to insufficient battery power. The delivery distance and time window are used to group tasks according to geographical location, making the delivery route more reasonable, avoiding scheduling conflicts among multiple drones in the same area, and improving the spatial distribution balance of task execution. The emergency task priority evaluation mechanism can make real-time judgments on the situation of surging orders, enabling the drone resources to be preferentially allocated to high-priority tasks, avoiding the occupation of delivery resources by ordinary tasks, and ensuring the efficient execution of key tasks. The dynamic task optimization mechanism enhances the adaptability of the drone in complex environments, makes task planning more intelligent, and reduces the additional energy consumption caused by redundant scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a system flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] In order to make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0048] Please refer to Figure 1 , the present invention provides a technical solution: The real-time optimization scheduling management system for drone delivery tasks includes:

[0049] A bat sound wave calculation module, which performs echo location frequency conversion calculations on the distance between the current position of the drone and the target delivery point, the time window, and the upper limit value of the carrying weight through the bat algorithm to obtain a task sound wave waveform matrix; performs a comparison operation between the waveform amplitude in the task sound wave waveform matrix and the delivery distance to generate a task search space vector;

[0050] A task resource allocation module, which performs subspace decomposition operations on the task search space vector through the bat algorithm to obtain a parallel task group sequence; performs constraint operations on the parallel task group sequence, the remaining battery power, load capacity, and flight speed of the drone to generate a resource matching allocation plan;

[0051] The scheduling optimization execution module groups and calculates the delivery distances and time windows in the resource matching and allocation plan according to geographical locations, and generates multi-task parallel scheduling instructions;

[0052] The emergency response processing module performs an emergency evaluation calculation on the real-time order increment at the site, the available number of drones, and the task completion time limit, and obtains an emergency task priority sequence.

[0053] The steps for obtaining the task sound wave waveform matrix are as follows:

[0054] Set the coordinates of the current position of the drone and the target delivery point, calculate the Euclidean distance between the two points, and obtain distance data;

[0055] Based on the distance data, calculate the echo location frequency conversion value. The calculation formula is:

[0056]

[0057] where TD is the distance between the two points, T w is the time window, W max is the upper limit of the carrying weight, K is a constant affecting the adjustment unit, and F echo is the echo location frequency conversion value;

[0058] Based on the echo location frequency conversion value, simulate the sound wave frequency conversion of echo location through the bat algorithm to generate a task sound wave waveform matrix.

[0059] Specifically, set the coordinates of the current position of the drone and the target delivery point. When preparing the positioning information, first read the real-time longitude, latitude and relative height data from the GNSS receiving device of the drone, and then combine the longitude, latitude and altitude information of the target delivery point detected by the ground calibration system. When comparing the latitude values in the two groups of coordinates, first judge whether they are within the range of -90° to 90°, and whether the longitude values are within the range of -180° to 180°. If they exceed, re-read the positioning information and check the antenna reception sensitivity. After completing the coordinate range check, record the current position coordinates of the drone and the target delivery point coordinates in numerical form respectively, and then calculate the Euclidean distance between the two groups of coordinates according to the point-to-point distance algorithm in three-dimensional space. During the calculation process, convert the geographical coordinates to metric units to obtain a more accurate distance value, and compare the converted result with the maximum flight radius obtained from the statistical data of the same type of previous flights. When the maximum flight radius threshold is exceeded, perform further steps, such as checking the flight endurance record of the drone or replacing the battery and payload to ensure that the numerical range meets the feasible range. If it does not exceed, directly store the calculated distance value as distance data. Monitor the attitude of the drone at fixed time intervals throughout the process to prevent numerical anomalies, and finally obtain the distance data.

[0060] The benefit of the formula lies in combining three key factors: distance, time window, and the upper limit of carrying weight. It calculates the echolocation frequency conversion value through the combined influence of exponential decay and cosine function, thus taking into account the flight distance limit and mission timeliness in the UAV delivery scenario.

[0061] The steps to obtain TD are as follows: This parameter represents the two-point distance between the current position of the UAV and the target delivery point. The value is calculated from the coordinate information collected by the GNSS receiver and the ground calibration system. First, the GNSS receiver of the UAV records the longitude, latitude, and altitude information of the current flight position, and calculates the three-dimensional space distance by comparing with the longitude, latitude, and altitude of the target delivery point. During this process, all the collected coordinate data are checked one by one for abnormal values beyond the recognizable range of the sensor. If an error point is encountered, the outlier is removed by comparing the distance distribution characteristics in the historical flight, and then the effective coordinates are taken for three-dimensional Euclidean distance calculation. The distance calculation formula is defined as where x 1 , y 1 , z 1 and x 2 , y 2 , z 2 are the longitude and latitude converted to metric coordinates of the two points respectively. By this method, in a specific monitoring case, TD = 12800 meters is obtained.

[0062] T w The steps to obtain it are as follows: This parameter represents the time window. The value needs to summarize the start and end times of historical orders and the current task through the delivery task planning system. First, retrieve the takeoff and delivery completion times of all completed tasks, calculate the average task cycle using statistical methods, and separately record the delays caused by sudden increase in orders or special weather. Then, form a time window reference sequence, and select the time window value that best reflects the actual delivery requirements in combination with the target delivery time limit of this task and the UAV scheduling cycle. By comparing the recent similar flight routes and the average flight duration of the UAV under similar load conditions, a time window based on actual situation matching is obtained. In a certain delivery monitoring, the time difference between the start and end times recorded is 1800 seconds, so in this case, T w = 1800.

[0063] W maxThe acquisition steps for it are as follows: This parameter is the maximum payload that the drone is allowed to carry during delivery. The actual value needs to be summarized based on the load test data of the drone manufacturer and the maximum safety load detection results of our unit for different aircraft models. First, in an experimental scenario, let the drone gradually increase the load and record the takeoff and landing stability and the power values consumed by flight control at different payloads to form comparison data of payload and flight performance. Then, repeat the above process in several typical operating environments to form multiple payload-power-flight stability curves. Finally, organize the intersection intervals of each curve. In a delivery model detection report, the maximum safety load test value is statistically 10 kilograms. Therefore, in this case, W max = 10.

[0064] The acquisition steps for K are as follows: This parameter is a constant that adjusts the influence of the unit. It is necessary to record the quantization results of the sensitivity of the supporting flight path to factors such as distance, time, and payload in multiple rounds of flight tests, and then compare the actual deviation between this sensitivity value and the echo location frequency conversion value to find the coefficient that is most suitable for balancing the influence of distance attenuation and cosine oscillation. The specific method is to conduct actual flight measurements under different driving radii and different payload conditions, and record the deviation degree between the echo location frequency conversion value and the actual execution situation in each test. Determine the numerical range through an optimization process of minimizing the deviation degree. After more than fifty tests, the results show that the matching degree is the highest when K = 2. Therefore, K is set to 2 in this case.

[0065] Calculation process:

[0066] First step, substitute TD = 12800, T w = 1800, W max = 10, and K = 2 obtained above into the formula respectively:

[0067]

[0068] Second step, first calculate the exponential part:

[0069]

[0070] Third step, calculate the cosine part:

[0071] cos(10) ≈ -0.83907

[0072] Fourth step, perform the numerator operation:

[0073] 12800 × 0.00081 ≈ 10.368, 10.368 + (-0.83907) ≈ 9.52893

[0074] Fifth step, divide the numerator result by K:

[0075]

[0076] The result shows that under the conditions of the current drone with a distance of about 12,800 meters, a time window of 1,800 seconds, and a maximum payload of 10 kilograms, the corresponding echo - location frequency conversion value is approximately 4.764465. The larger the value, the relatively lower the attenuation effect of distance on echo - location under the same time window and payload limit. When this value is significantly lower than 1, it indicates that the factors of distance and time window will have a greater impact on delivery.

[0077] Based on the echo - location frequency conversion value, when simulating the acoustic wave frequency conversion of echo - location through the bat algorithm, a batch of historical flight data needs to be selected first as training and reference, which includes echo detection records and corresponding spectrum information under different distances and payload conditions. These records are read one by one, and the amplitude peak and central frequency of the echo signal are identified to form a set of echo frequency lists for calibration. Then, the search structure composed of pulse emission rate and loudness change in the bat algorithm is selected, and the echo frequency list obtained just now is combined with the echo - location frequency conversion value calculated previously. In each iteration step, the target frequency is compared with the current echo reference frequency, the search step size is corrected, and the flight parameters of the bat group are updated one by one. Then, in each iteration, the frequency change amount corresponding to the optimal fitness is recorded, and the deviation size from the actual flight cases is analyzed synchronously. Once it exceeds the maximum deviation threshold obtained from multiple experimental statistics, the amplitude oscillation range is automatically reduced to make subsequent iterations more concentrated in a reasonable range. When the deviation is within the acceptable range, subtle corrections to the frequency conversion are maintained until all search iterations are completed. Such a processing process can continuously update and maintain a sequence of acoustic wave frequency conversion values that conform to the current delivery scenario. Finally, after all iterations are completed, the amplitude changes and time nodes in the frequency conversion sequence are item - by - item corresponded and output as a task acoustic wave waveform matrix.

[0078] The steps to obtain the task search space vector are as follows:

[0079] Extract each waveform amplitude in the task acoustic wave waveform matrix, index - match the amplitude data according to the delivery distance, screen the amplitude range that meets the delivery task, and generate the waveform amplitude data after matching;

[0080] Based on the waveform amplitude data after matching, calculate the task search space vector, and the calculation formula is:

[0081]

[0082] where A 1 is the maximum value of the waveform amplitude data, A 2 is the minimum value of the waveform amplitude data, B 1 is the square - multiplication term of the delivery distance, B 2is the sine mapping term of the delivery distance, PD is the delivery distance, PC is the normalized offset term of the delivery distance and the waveform amplitude, A 3 is the average value of the waveform amplitude data, S task is the task search space vector.

[0083] Specifically, each waveform amplitude in the task sound wave waveform matrix is extracted. Combining the matrix information obtained previously, all amplitude data are read item by item and their corresponding frequency indices are marked. Subsequently, paired operations are performed according to the delivery distance parameters and amplitude information obtained in the previous step. First, the absolute value of each amplitude is extracted from the waveform amplitude data and its distribution interval between 0 and 1 is recorded. Then, these amplitudes are segmented and matched according to the interval of the delivery distance range. Each record is compared with the pre-set valid range. For example, the amplitude data is compared with the interval between 0 and 1, and the delivery distance is compared with the interval between 0 meters and 20,000 meters to determine whether the corresponding relationship between the amplitude and the distance exceeds the design range. When it is found that the amplitude value is higher than 1 or the distance is higher than 20,000 meters, it is marked additionally in the record and whether to reject it is determined according to the empirical threshold. The way to obtain the empirical threshold is to summarize previous delivery tasks to determine the specific boundary values, count the flight distances of each delivery order and find the distribution law within the maximum feasible flight radius, and then count the frequency of the distribution of the waveform amplitude under the condition of this flight radius, so as to refine a segmented amplitude matching threshold list. Then, after each item-by-item matching is completed, all valid waveform amplitudes are re-verified to confirm that their corresponding distance terms do not exceed the set upper limit range and do not fall below the set lower threshold range. Finally, these records that meet the set standards in terms of amplitude and distance are merged into a new data set to obtain the matched waveform amplitude data.

[0084] The advantage of the formula is that it combines the difference characteristics of the waveform amplitude, the square multiplication term corresponding to the distance, and the sine mapping term, and at the same time uses the normalized offset and the average value correction term to jointly form the balance relationship between the numerator and the denominator, so as to take into account the range of the amplitude, the non-linear growth of the distance, and the sine wave fluctuation factors in the task scheduling scenario related to the delivery distance.

[0085] A 1The acquisition steps are as follows: This parameter represents the maximum value of the waveform amplitude data, which needs to be extracted from the previously obtained matched waveform amplitude data. Whenever a new set of amplitude information is recorded, this amplitude is compared successively with the current maximum value. If the current amplitude value is greater than the recorded maximum value, the maximum value is updated and the next record is continued. To prevent abnormal peaks caused by sensor noise, it is also necessary to conduct on-site verification for amplitudes higher than 1. By independently measuring and comparing the waveform data collected during the flight test of the drone, invalid or unreasonable peaks are screened out, forming a final verified maximum value record. For example, in a data sequence of 5000 records, the recorded maximum amplitude is 0.96. By referring to the data distribution graph of this amplitude at different time periods, it is confirmed that this 0.96 is a real peak.

[0086] A 2 The acquisition steps are as follows: This parameter represents the minimum value of the waveform amplitude data, which also comes from the above-mentioned matched waveform amplitude data. Each time an amplitude is read, it is compared with the existing minimum value. If the value is smaller, the minimum value is updated. When encountering abnormal values close to 0, it is judged whether it belongs to noise by comparing with the overall level of the amplitude distribution. If it is confirmed to be abnormal noise, the record is excluded. If it is confirmed to be an effective fluctuation, it is retained in the minimum value judgment queue. Cross-checks are performed on the distribution of continuously observed flight data, and the decline limit of the amplitude at different time periods is continuously monitored. Finally, the verified minimum amplitude in the same 5000-record data sequence is 0.15.

[0087] B 1 The acquisition steps are as follows: This parameter represents the square multiplication term of the delivery distance. It is necessary to quantify the correlation between the distance squared and the delivery difficulty based on previous flight data. First, in the drone flight statistics, the distance and flight power in different intervals are collected in segments. Then, the correlation coefficient between the distance squared and the power consumption is calculated, and the power consumption is compared with the difficulty index. Then, the specific value of the multiplication term in different distance intervals is obtained from the correlation assessment between the difficulty index and the flight stability. If the distance is larger, this term increases significantly with the square of the distance. The results obtained from several groups of experiments are averaged and the fluctuation range is integrated. Finally, the multiplication factor value that best reflects the task difficulty is selected. Here, it is set that B 1 = 0.000015, which is determined after multiple experimental samplings spanning a distance of 10,000 meters. By referring to the data statistics sequence of the load and energy loss of the drone during long-distance delivery, the corresponding values of the distance squared multiplication term fluctuate between 0.00001 and 0.00002. Therefore, the middle value is taken.

[0088] B 2The acquisition steps of A are as follows: This parameter represents the sine mapping term of the delivery distance, which is obtained by analyzing the waveform characteristics and the periodicity of the delivery route within a segmented distance range. First, the flight vibration data of the drone is collected within a certain mileage and the phase difference is extracted. Then, this part of the vibration data is decomposed into several sine components by Fourier analysis, and the corresponding curves between each sine component and the actual delivery distance distribution are recorded. Furthermore, the impact of the sine wave fluctuation on the delivery task is quantified. This impact shows a similar periodic change law within a certain interval. Therefore, it can be sorted into a mapping coefficient. It is found that when the distance exceeds 5000 meters, the sine property of the vibration and the flight attitude is more obvious in multiple flight routes accumulated during the test. By referring to the distribution map, the average value of the sine mapping factor in this interval is determined to be 0.035. The average value is obtained by taking the average of multiple experimental samplings to obtain B 2 = 0.035.

[0089] The acquisition steps of PD are as follows: This parameter represents the delivery distance, which is directly read from the previously obtained distance information and index-matched with the waveform amplitude data. Here, when measuring the distance during a long-range flight task, the longitude and latitude of the drone are continuously collected and converted into ground plane coordinates. Subsequently, the actual flight distance between two points is calculated. For a flight with detours or multiple segments, the segmented distances are superimposed to obtain the final effective delivery distance, and the extra path caused by the initial ground debugging of the flight is excluded. During a large-scale delivery task that crosses the city boundary, the measured distance reaches 12500 meters, which is recorded as PD = 12500.

[0090] The acquisition steps of PC are as follows: This parameter is used as the normalized offset value between the delivery distance and the waveform amplitude. It is necessary to normalize the distance and amplitude within different distribution ranges and then observe their coupling degree. Scatter plots of the normalized distance and the normalized amplitude are respectively drawn within the interval from 0 to 1 to check the dispersion generated during the entire delivery process. Then, the difference between this dispersion and the actual data is cumulatively summarized into several offset amounts, and the final reference value is determined based on the average value of the offset amounts of multiple groups of distributions. This value is further corrected and used as PC. By monitoring the fluctuation characteristics of several drones at different distances and different waveform amplitudes for a long time, the value that can best represent the overall offset range is found. Here, after multiple rounds of comparison, it is confirmed that PC = 0.85.

[0091] A 3The acquisition steps are as follows: This parameter represents the average value of the waveform amplitude data. It is necessary to add up all the amplitudes in the currently matched waveform amplitude data and then divide by the number of valid records. Each record must first undergo a validity check to exclude invalid data with excessively high or low floating values outside the minimum noise threshold, and for cases where the amplitude shows abnormal jumps, perform smoothing processing before including it in the average value calculation. Subsequently, perform a summation operation on the amplitude list obtained from the above processing and divide by the number of records. Among the same batch of 5000 data, the sum of all valid amplitudes obtained through the above processing is approximately 2500.0. Divide it by the number of valid records, 4000, to obtain A 3 = 0.625, forming an average value that can represent the overall amplitude distribution.

[0092] Calculation process:

[0093] First step, substitute the aforementioned parameters A 1 = 0.96, A 2 = 0.15, B 1 = 0.000015, B 2 = 0.035, PD = 12500, PC = 0.85, A 3 = 0.625 into the following respectively:

[0094]

[0095] Second step, first calculate the absolute value of the amplitude difference: |0.96 - 0.15| = 0.81, then take the square root:

[0096] Third step, calculate the distance squared multiplication term: 0.000015 × 12500 2 = 0.000015 × 156250000 = 2343.75.

[0097] Fourth step, calculate the sine mapping term: The value of sin(12500) is between -1 and 1. Here, the sine corresponding phase actually collected by the monitoring device is used to obtain sin(12500) ≈ -0.13235, and then multiply it by B 2 = 0.035, and the result is approximately -0.00463.

[0098] Fifth step, add the results of the third step and the fourth step: 2343.75 + (-0.00463) ≈ 2343.74537, and then add 0.9 from the second step to obtain the numerator: 2343.74537 + 0.9 = 2344.64537.

[0099] Sixth step, calculate the denominator. First, find the distance difference: |12500 - 0.625| ≈ 12499.375, then take the square root: Add it to PC = 0.85: 111.80 + 0.85 = 112.65.

[0100] Step 7, divide the numerator by the denominator:

[0101]

[0102] The result shows that under the conditions of a delivery distance of 12,500 meters, the maximum and minimum values of the obtained waveform amplitudes are 0.96 and 0.15 respectively, the calculated numerical value of the task search space vector is approximately 20.81. When this value remains relatively high in similar task scenarios, it indicates that the impact of the delivery distance is relatively large and the amplitude difference is obvious. When it drops to 2 or lower, it usually means that the distance is short or the amplitude difference is small, and at this time, fast scheduling can be prioritized.

[0103] The steps to obtain the parallel task group sequence are as follows:

[0104] Based on the task search space vector, use the bat algorithm to analyze the features, identify the potential subspace boundaries, and obtain the initial subspace feature data;

[0105] Based on the initial subspace feature data, perform subspace decomposition operations. The formula is:

[0106]

[0107] where, θ i represents the position angle of the i-th data point, V i represents the search space vector value of the i-th task, H groups represents the subspace decomposition result, and RN represents the total number of tasks in the task search space vector;

[0108] Based on the subspace decomposition result, group the data points according to similarity to form a parallel task group sequence.

[0109] Specifically, based on the task search space vector, the bat algorithm is used to analyze features and identify potential subspace boundaries. First, prepare the sequence of task search space vectors required for the algorithm to run and confirm that each vector contains complete amplitude records and distance information. Then, during the iterative process of preparing the bat algorithm, read the search space vectors one by one and map them to the positioning reference values of flying bionic bat individuals, so that each individual has an echo pulse emission interval and amplitude. And during each iteration, update the position according to the actually detected search vector values. The corresponding ultrasonic pulse echo intensity is matched with the previously collected amplitude distribution data. By comparing the pulse emission rate and the echo reception rate, it is judged whether the individual search enters the specified feature boundary region. Then, mark all qualified vectors and record the amplitude stability shown by them in multiple iterations. For the numerical range of amplitude stability, the amplitude volatility range statistically obtained in the flight test can be referred to. The specific method is to first select several representative amplitude peak and valley sequences from the previously summarized waveform amplitude database, and compare them with the echo matching degree generated in the current bat algorithm respectively. If the difference is lower than an empirical threshold, this record is determined as a potential subspace boundary point. Regarding this empirical threshold, it can be obtained by summarizing the boundary detection frequencies that appear in different flight states at different times. For example, in the data of cross-city air routes, the high-intensity trigger frequency of echo signals in each 1000-meter distance interval is statistically analyzed, and the proportion of the number of times this frequency accumulates in multiple flights is observed, so as to locate a threshold range covering most situations. Then, accumulate and count the number of times the boundary points repeat in different batches of iterations. If this number exceeds the preset value obtained from the historical distribution analysis, this boundary point is marked as a stable boundary. All the determined subspace boundary point sets will be further processed to remove duplicates, eliminating the duplicate determination records caused by the randomness of bat individual search, and sorting them according to their relative positions in the search space. Finally, according to this order, several feature indicators related to the subspace are extracted, such as the average level of waveform amplitude, the distribution of amplitude change rate, relative time index, etc., and combined into the initial subspace feature data, obtaining the initial subspace feature data.

[0110] The benefit of the formula is that through the fusion analysis of the cosine function of the position angle and the square root of the absolute value of the search space vector, the projection features of different tasks in the multi-dimensional space are integrated into a grouping index. At the same time, the exponential function is used to balance the overall influence of the sine distribution of all position angles, so as to reflect the comprehensive effect of multiple angles.

[0111] θ iThe acquisition steps are as follows: This parameter represents the position angle of the i-th data point. The angle information needs to be obtained through on-site measurement or comparison with flight logs. Each task can be projected as a vector in the geographic coordinate system. After comparing the direction from the starting position of the UAV to the target area with the reference direction (such as magnetic north), the included angle value is recorded as the initial angle value. Then, considering the influence of altitude and terrain environment on the attitude of the task end point, the elevation or depression angle of the task end point relative to the ground horizontal plane is measured additionally. Then, these components are synthesized into the final position angle in three-dimensional space. To ensure the accuracy of the angle data, the UAV heading can be continuously monitored during flight and combined with the rotation rate collected by the inertial measurement device to obtain the dynamically corrected angle value. In a conventional flight monitoring case, 24 task nodes are selected. After angle analysis, the recorded angle range is between 0 degrees and 360 degrees, and coordinate transformation calibration is performed on records that are abnormally greater than 360 or less than 0. Finally, a set of legal angles is obtained for the corresponding θ i Assignment.

[0112] V i The acquisition steps are as follows: This parameter represents the search space vector value of the i-th task, which is derived from the task search space vector sequence obtained after amplitude screening and distance normalization. To ensure that each vector value truly reflects the UAV delivery difficulty, the flight mileage, load energy consumption distribution, and peak-valley distribution of the waveform amplitude need to be incorporated into the vector compilation. By recording multi-dimensional data on energy consumption and waveform differences in the trajectory planning, after removing noise points, a vector value is assigned to each task record. If excessive energy consumption is found during segmented flight or extremely heavy load periods within the monitoring period, additional calibration will be performed and the vector will be updated. When the corresponding vector value fluctuates between 0 and 100, it usually indicates a relatively low task difficulty, while exceeding 100 often means approaching the UAV endurance limit. In a statistical data containing 50 long-range deliveries, the search space vector value of each task is mostly between 30 and 120. After comprehensive consideration, the finally corrected vector is assigned to the corresponding V i Above.

[0113] The acquisition steps of RN are as follows: This parameter represents the total number of tasks in the task search space vector, which is the result of counting all the organized tasks in the current batch. It is necessary to complete the statistics before formal grouping. During the calculation, those invalid tasks caused by misoperations or incomplete tests are excluded. If the same task is repeatedly recorded, it needs to be merged and counted after index comparison. Only the valid and non-repeatedly calculated tasks are summed up. In a cross-regional UAV scheduling test, the number of valid tasks obtained by statistics is 50, and it is recorded as RN = 50.

[0114] The steps to obtain N are as follows: This parameter is used to limit the superposition range of the exponential function in the denominator and calibrate the number of data points for sine analysis at the position angle level. It is usually the same as or slightly smaller than RN. The specific value depends on whether there is valid angle data in each task record. To maintain the integrity of the overall analysis, records with only a small amount of missing angle data are generally interpolated and repaired before being included in the statistics. For example, among the 50 tasks in the previous example, 5 tasks had missing angles due to incomplete geographical coordinates. An average angle value can be inserted between two adjacent flight records. After statistics, N = 47 is obtained, ensuring the continuity of the analysis results.

[0115] Calculation process:

[0116] First step, assume that RN = 50 and N = 47 have been obtained, and read in θ i and V i :

[0117] θ 1 , θ 2 , …, θ 50 ; V 1 , V 2 , …, V 50

[0118] Among them, 3 records do not contain complete angles. After interpolation, calculations are performed to ensure that the statistical requirement of N = 47 is met.

[0119] Second step, calculate the numerator:

[0120]

[0121] In the collected data, assume that each θ i is converted to radians and the cosine value is calculated in sequence, and the square root is calculated one by one according to different values of V i between 30 and 120. For example, when V 10 = 100 is substituted, can be obtained. Then, according to the measured θ 10 = 50°, after conversion to radians, it is approximately 0.87266 radians, and cos(0.87266) ≈ 0.656. Then the contribution of this item is approximately 0.656 × 10 = 6.56. Thus, calculations are performed on all 50 records in sequence and the sum is taken. The result of the numerator is recorded as X sum .

[0122] Third step, calculate the exponential part in the denominator:

[0123]

[0124] Perform sine operations on 47 valid angles, and then sum these sine values, which is recorded as Ysum , since the sine value can be negative, the final sum may fluctuate in the range of -47 to 47. If all the angle data are sorted out in this example, we get Then exp(-3.25) ≈ 0.03853.

[0125] Step 4, the denominator part:

[0126]

[0127] Step 5, divide the sum of the numerators X in Step 2 sum by the denominator in Step 4 to obtain the final result H groups .

[0128] For example, the cumulative sum of the numerator operations gives X sum = 150, then:

[0129]

[0130] This result shows that among the 50 task search space vectors in the statistics, after considering 47 effective angle data, the grouped index value is about 144.43. The higher the value, the more concentrated the subspace decomposition result formed by the combined action of cosine analysis and vector amplitude. When this value drops significantly below 10, it usually means that there are significant differences in the spatial distribution among tasks.

[0131] Based on the subspace decomposition results, group the data points according to similarity. To compare multiple extracted subspace decomposition records, it is necessary to calculate the similarity of the subspace feature data attached to each record first. The specific method is to read the decomposition values of each data point and the corresponding search space vector amplitudes, and then form several comparison items with these amplitudes and decomposition values. By calculating the difference within the set distance weight and amplitude weight ranges, the threshold for the difference can be determined from the summary of previous flight data. For example, first count in a large range that when the amplitude difference is less than 5 and the decomposition value difference does not exceed 10, it is regarded as the same similar area. Then compare this difference pairwise between different data points, calculate the grouped similarity matrix, and compare the similarity values in the grouped similarity matrix with an empirical threshold. This empirical threshold is obtained by summarizing the cases where the distance is greater than 3000 meters or the load amplitude difference exceeds 2 kilograms after repeatedly dispatching the drone in the experiment. Finally, a threshold range that can distinguish significant differences is sorted out. If the similarity value between any two data points is lower than this threshold, they are divided into different groups. If the similarity value is greater than or equal to this threshold, they are merged into the same group. After the preliminary grouping, check the dispersion degree of the decomposition values within each group again. If the dispersion degree is too large, use the grouped similarity matrix to screen out the data points with large differences from most records and separately summarize them into a new group. Finally, a series of parallel task group sequences are obtained, forming a parallel task group sequence.

[0132] The steps to obtain the resource matching and allocation scheme are as follows:

[0133] Based on the parallel task group sequence, calculate the matching degree between the task and the UAV resources. The calculation formula is:

[0134]

[0135] Among them, R match is the matching degree between the task and the UAV resources, E is the remaining battery power of the UAV, C is the minimum power required for the task, ZL is the current load capacity of the UAV, S is the flight speed of the UAV, RD is the task delivery distance, and P is the maximum deliverable distance of the UAV;

[0136] Allocate UAVs according to the matching degree between the task and the UAV resources to generate a resource matching and allocation scheme.

[0137] Specifically, the advantage of the formula is that it comprehensively considers factors such as the current remaining battery power, flight speed, load capacity, and delivery distance of the UAV, and through the exponential function form, makes the different distance differences present a non-linear influence in the denominator, so as to more comprehensively measure the mutual relationship between the remaining battery power difference, load, flight speed and task requirements when allocating UAV resources.

[0138] Steps to obtain the E parameter:

[0139] E represents the remaining battery power of the UAV, which is a value closely related to the endurance performance of the UAV and is usually detected by an on-board power management device. This device will monitor information such as the output voltage, current, and discharge rate of the battery in real time during the startup and flight of the UAV, and calculate according to the nominal capacity and cycle performance parameters provided by the manufacturer to form a percentage representation of the remaining battery power. To determine the value of E, it is necessary to first record the battery capacity of the UAV in units such as milliampere-hours or watt-hours, and then obtain the power consumption trend by comparing the voltage decay curve tested in multiple time periods with the corresponding time cumulative value. Finally, combine these data to calculate the remaining battery power at the current moment. For example, in a certain medium-range flight test, the initial battery power was recorded as 100%, and multiple measurements were taken during the flight of 15 minutes and the load of 3 kilograms in the middle, and the remaining battery power was between 65% and 75%. Subsequently, the comprehensive observation value of 70% was selected as the reference value of E for subsequent resource matching degree calculations.

[0140] Steps to obtain the C parameter:

[0141] C represents the minimum power required for a mission, which is a critical value obtained by evaluating elements such as flight distance, load conditions, and obstacle environment during mission planning. It is necessary to ensure that the drone can complete a round trip or at least a one-way flight under the minimum requirements of the mission. To obtain C, the actual power consumption of this model under different distances and load conditions is usually pre-statistically analyzed. Through multiple flight tests using on-board energy consumption monitoring equipment or ground experiments, the corresponding power consumption data is recorded. Then, by multiplying the minimum power required to complete the mission by the corresponding time or distance, the minimum power can be quantified. For example, when a certain model of drone carries 3 kilograms of goods and executes a 10-kilometer route, after multiple test flights, the minimum power requirement is approximately 20%. This data is marked as C = 20%.

[0142] Steps to obtain the ZL parameter:

[0143] ZL represents the current load of the drone, and the value is obtained through ground weighing or hanging measurement. The specific process includes first weighing and recording the empty weight of the drone, and then weighing and stacking each piece of goods or packages required for this mission one by one. During the loading process, the balance in the cabin or the hanging point and the weight of the actual auxiliary equipment installed also need to be tracked, and then the total load value at the current moment is summarized and recorded as ZL. If the goods are replaced or added during flight, the ZL value needs to be weighed and updated again on the ground or at an intermediate station. For example, in a regional sorting and distribution test, the total weight of the hanging objects when the drone takes off is 2.8 kilograms. After flying to a temporary station, part of the goods are unloaded and new goods are loaded. After recording on the ground and recalculating the load, it becomes 3.2 kilograms. Finally, ZL = 3.2 is used as the input value for subsequent calculations.

[0144] Steps to obtain the S parameter:

[0145] S represents the flight speed of the drone, which can be monitored by the speedometer or inertial navigation equipment of the on-board flight control system. Usually, the horizontal and vertical speed components are recorded during flight, and the three-dimensional flight speed is obtained through trigonometric synthesis, or the instantaneous speed value is directly read from a high-precision flight control system. To ensure accuracy, instantaneous acceleration and deceleration caused by environmental factors, such as sudden changes in wind force or non-steady flight during takeoff and landing, are generally excluded. Therefore, the sampling can be selected during a relatively stable cruising stage. If there are multiple speed intervals during the flight, different intervals need to be averaged or weighted to obtain a more realistic S value. For example, during the test period, the speed of the drone in most cruising intervals remains between 8 meters and 12 meters per second. After data analysis, the current stage speed S = 10 m / s is obtained.

[0146] Steps to obtain the RD parameter:

[0147] RD represents the task delivery distance, which refers to the one-way or round-trip distance of the UAV from the starting point to the destination. Usually, a predetermined flight path is first generated in the mission planning system based on geographical coordinate information, and then the total mileage is confirmed by means of map measurement or real-time measurement of the on-board navigation trajectory. The distance of unnecessary hovering or waiting segments during the test is deducted, and only the effective mileage of the actual delivery flight segment is retained. If there are multiple intermittent stops during the flight path, the distances of each segment can also be added up to obtain the total delivery distance. For example, for an in-city delivery route, the navigation calibration result shows that the one-way distance is about 10 kilometers, and the round-trip is 20 kilometers in total. During a certain actual flight, it is slightly corrected according to the flight track recorded by the flight control system. Finally, RD = 10000 meters.

[0148] Steps to obtain the P parameter:

[0149] P represents the maximum deliverable distance of the UAV, which refers to the longest flight distance that the UAV can execute on the premise that factors such as battery power, load, and safety redundancy meet the standards. To calculate P, it is necessary to measure the extreme navigation mileage of the UAV under full battery takeoff and normal load conditions in the test environment multiple times, and at the same time exclude unnecessary invalid flight segments, such as the mileage consumed during low-altitude hovering or simulated fault testing. Finally, the measured results of several extreme flight mileages are statistically analyzed, and the maximum mileage that can stably complete the flight is selected as P. The average value of the extreme distances measured during multiple flight tests of a certain type of UAV when carrying 3 kilograms of goods is 12000 meters, so P = 12000 meters is set.

[0150] Calculation process:

[0151] First step, take the values obtained above: E = 70%, C = 20%, ZL = 3.2 kg, S = 10 m / s, RD = 10000 m, P = 12000 m.

[0152] Second step, first calculate the part in the numerator

[0153] If both E and C are regarded as percentages, they can be first converted to numerical values: E = 70, C = 20;

[0154] |E - C| = |70 - 20| = 50,

[0155] Third step, the numerator also contains ZL·S:

[0156] ZL·S = 3.2×10 = 32

[0157] Fourth step, the sum of the numerator:

[0158] 7.07 + 32 = 39.07

[0159] Fifth step, calculate the denominator:

[0160] 1 + exp(-|RD - P|) = 1 + exp(-|10000 - 12000|) = 1 + exp(-2000)

[0161] Here, exp(-2000) is extremely close to 0. As can be seen using a high-precision calculator, it is basically on the order of about 0. Therefore, the denominator is approximately 1 + 0 ≈ 1.

[0162] Step 6, divide the result of the numerator by the denominator:

[0163]

[0164] This result indicates that for the scenario with a remaining battery level of 70% and a minimum battery level required for the task of 20%, the matching degree value of the current drone under the conditions of a load of 3.2 kg, a speed of 10 m / s, a delivery distance of 10000 m, and a maximum deliverable distance of 12000 m is approximately 39.07. Generally, the larger the value, the relatively higher the executable degree of the drone in this task. If the value drops below 5, it means there is a large gap between the remaining battery level and the task requirements, or factors such as load and speed will seriously affect the execution ability, and re-scheduling or replacing the aircraft model is required.

[0165] Drones are allocated according to the matching degree between tasks and drone resources. First, read the remaining power and maximum deliverable distance of each drone model or each individual drone in the drone queue, and retrieve information such as load capacity and flight speed from previous records. Compare them with the minimum power requirement and task distance data of the delivery task respectively. If it is found during the preliminary comparison that the remaining power of a certain drone is lower than the minimum power requirement or the distance difference caused by the maximum deliverable distance being lower than the task requirement, then mark this model as not meeting the current task requirements. Then, read the information of all drones that meet the minimum requirements one by one and substitute them into the matching degree formula for item-by-item calculation. Compare the values calculated each time with all the calculated values to form a comparison table and rank them according to the actual flight route and load conditions. Manually verify the models with too low values, for example, check whether there are obvious fluctuations in their load records or whether the power has decayed rapidly due to external factors. For the models ranked at the top, immediately summarize their matching result values and correspond them to the task characteristics one by one to ensure that the execution time and flight radius of the current batch are consistent with the performance of this model. In this way, sort the original list of drones assigned as targets in descending order of matching degree. If it is found that the matching degrees of multiple drones are relatively high, then conduct a detailed allocation based on the remaining available quantity and the duration of subsequent tasks. Record the data in the allocation list and set a separate identification number to identify which task unit this drone belongs to. For all successfully allocated drones, also update the current number of tasks to be executed, the remaining power value, and the actual flight radius available for the next call in the resource table. Then summarize and file these allocation information for easy subsequent replenishment or retrieval. When the drone starts the task, if its endurance or load status changes, relevant data needs to be dynamically revised according to this list and the real-time allocation situation needs to be maintained. Finally, output the corresponding execution detail data based on the records after all allocations are completed to form a resource matching and allocation plan.

[0166] The steps to obtain the task space clustering results are as follows:

[0167] Extract the delivery distance and time window of each task from the resource matching and allocation plan, classify the tasks according to geographical location, calculate the distribution density of tasks in each region, and obtain the regional task distribution data;

[0168] Based on the regional task distribution data, calculate the task grouping fitness, and the calculation formula is:

[0169]

[0170] where, F group is the task grouping fitness, D i is the delivery distance of the i-th task, D avg is the average delivery distance of all tasks in the region, T i is the time window of the i-th task, T maxis the maximum time window of tasks in the region, L i is the geographical location coordinates of the ith task, L c is the center coordinate of the area, and N is the total number of tasks in the current area;

[0171] Generate multi-task parallel scheduling instructions based on task grouping adaptability.

[0172] Specifically, the delivery distance and time window of each task are obtained by reading the resource matching allocation plan. First, all valid records are screened out from the list of assigned tasks and compared one by one with the geographical location information marked therein. Then, a regional division standard is set according to the longitude and latitude range of the geographical coordinates. For example, the range of coordinate latitude between 30° and 35° and longitude between 110° and 115° is defined as Zone A, and the surrounding area is defined as Zone B or Zone C. The corresponding relationship between the area where the coordinate data of each task is located and the above-mentioned division standard is compared to confirm the partition to which it belongs. In order to ensure the comparability of the distance difference between different areas, the route positioning accuracy data previously counted in the flight log can be referred to and an allowable error value, such as a longitude and latitude range of ±0.001 degrees, can be set to correct the coordinate drift. When it is found that the coordinate value of a task exceeds this range, its coordinate information needs to be measured and recorded for multiple times, and then these divided areas are divided. The distribution density of all tasks in the domain is calculated. In order to achieve this calculation, a threshold is first determined to distinguish between sparse and dense concepts. The threshold can be derived from the statistical results of the peak task volume in the same area in the similar time period in the past and the distribution mean is taken as a reference. Then, each task record is traversed and compared with the threshold. The areas with high density are included in the subsequent key monitoring areas, and the areas with low distribution density are also marked accordingly to distinguish the data characteristics. In this way, the distribution of the number of tasks in each area in the current period is obtained and a relative density value is calculated per unit area. If the value exceeds an empirical threshold, it means that the area has more task aggregation. If the value is lower than another empirical threshold, it means that the area has a relatively small amount of tasks. Through the coordinated judgment of this high and low thresholds, a regional task density map can be formed. Finally, these calculated distribution densities are recorded in correspondence with the geographical locations of each region and summarized to obtain regional task distribution data.

[0173] The benefit of the formula is that it conducts a comprehensive analysis of factors such as delivery distance and time window. On the one hand, the numerator reflects the task characteristics through the logarithmic form of absolute distance difference and time window difference. On the other hand, the denominator uses the exponential suppression effect of geographic coordinate difference to balance the distribution differences within the region, thereby achieving a measure of the adaptability of task groups in the same region.

[0174] D iThe acquisition steps for it are as follows: This parameter represents the delivery distance of the \(i\)-th task, and the actual flight mileage of each task needs to be retrieved from the resource matching and allocation plan or the scheduling system. When measuring the mileage, the coordinates of the starting point and the ending point are often converted in combination with the geographic information system, and the total flight distance is accumulated through the route data recorded on the ground or on board the aircraft. To remove outliers and invalid detour segments, the mileage of non-effective operation segments such as climbs and returns will be retrieved from the flight log and excluded. Subsequently, the remaining mileage that meets the effective delivery segment is used as \(D\). i It is retained in the final record. In a typical batch of urban logistics flight tests, after analyzing the flight track of each task, \(D\) is obtained as 1 = 8.2 km, \(D\) is 2 = 5.6 km, etc. By traversing all tasks, the respective delivery distances can be obtained.

[0175] The acquisition steps for \(D\) are as follows: This parameter represents the average delivery distance of all tasks in the region, that is, all \(D\) values in the current region are added up and then divided by the total number of tasks \(N\). When obtaining it, it is necessary to first determine the task list belonging to the same region and exclude the task entries for which no valid distance data has been collected, and then perform the accumulation summation and division operations to obtain it. For example, in the above urban logistics flight test, assuming that a total of 10 valid task distance data have been collected in this region, and the total mileage is about 76.4 km, dividing it by 10 gives \(D\) as avg = 7.64 km. Other users can obtain \(D\) by simply replacing their own task quantity and mileage summary value in the same statistical process. i avg = 7.64 km, and other users can obtain \(D\) by simply replacing their own task quantity and mileage summary value in the same statistical process. avg

[0176] The acquisition steps for \(T\) are as follows: This parameter represents the time window of the \(i\)-th task, usually referring to the maximum duration or limited time period allowed for the task to be executed from the start to the completion, and needs to be jointly determined in combination with elements such as the delivery time requirement and the receiving time period. When obtaining it, generally, the system automatically assigns an executable time period for each task when dispatching the task, or records the start and end time points of the task during the actual operation of the drone, and after calculating the effective usage time, compares it with the constraints set in advance by the scheduling system to confirm the actual available time window. In some scenarios, delays caused by uncontrollable factors such as weather or control are also excluded from the time window, and then the final executable time length is registered in \(T\). i i For example, in a group of delivery statistics, the available time window for a certain task is 1800 seconds, and for another task it is 2400 seconds, both of which can be recorded as \(T\) for subsequent calculation. i

[0177] The acquisition steps for \(T\) are as follows: This parameter represents the maximum time window of the tasks in the region, and all tasks \(T\) in the same region need to be traversed. max ​​​​​i After that, select the one with the largest value among them or take the same value for multiple equal maximum values. To ensure data rationality, some extreme values caused by incorrect records should be excluded, such as unconventional data less than 100 seconds or greater than 100,000 seconds. After comparing the correct records, select the maximum value and assign this value to T max . For example, in a statistics of multi-machine collaboration, after comparing the time windows of 12 tasks in the same area, it is found that the maximum available duration is 3,600 seconds, then T max = 3,600.

[0178] L i The acquisition steps of L are as follows: This parameter represents the geographical location coordinates of the i-th task. It cannot be directly input into the formula in the form of a string, etc., but needs to be first converted into quantifiable coordinate values. The specific methods include: when obtaining the longitude and latitude of each task, use a unified coordinate system such as WGS84 or GCJ02 for calibration, convert the longitude and latitude into plane coordinates or calculate in meters to facilitate subsequent absolute difference processing. If the task contains altitude information, it should also be included in the three-dimensional coordinates. In a multi-UAV urban flight test, in order to unify the dimension, the longitude and latitude will first be projected by Mercator projection and converted into meter coordinates, and then L will be obtained item by item i For example, (12,500, 9,820) means that the x and y coordinates are 12,500 meters and 9,820 meters respectively.

[0179] L c The acquisition steps of L are as follows: This parameter represents the central position coordinates of the area, and the same coordinate system as L i needs to be used for quantification. Generally, it is obtained by averaging the coordinates of multiple task points or directly reading the central area coordinates of this area from the geographical database to get L c . If the area belongs to a large range, the center will be calculated weighted according to the distribution position of the specific task points. For example, when counting the coordinates of all task points in the current area, the x and y coordinates are added respectively and divided by the total number of points to obtain the center coordinates (x c , y c ), and then it is recorded as L c . In some cases, if the geographical center of this area is pre-planned as (13,000, 10,000), this coordinate can be directly set as L c .

[0180] The acquisition steps of N are as follows: This parameter represents the total number of tasks in the current area. Since it is necessary to ensure that each task has available delivery distance, time window, and coordinate information, when counting, the task records with missing or abnormal data should be excluded first, and then the remaining tasks are counted to obtain N. In the above case, if there are 10 records in this area that meet the conditions, then N = 10.

[0181] Calculation process:

[0182] In the first step, substitute the aforementioned parameters into the example: D 1 = 8.2, D 2 = 5.6,..., D avg = 7.64, T 1 = 1800, T 2 = 2400,..., T max = 3600, L 1 = (12500, 9820), L c = (13000, 10000), N = 10. First, calculate the numerator part:

[0183]

[0184] For the i = 1 term, when D 1 = 8.2, then |8.2 - 7.64| = 0.56, and T 1 = 1800, 3600 - 1800 = 1800, ln(1800) ≈ 7.495, and the sum of the two is approximately 8.243. Calculate and accumulate the results of the remaining terms in the same way to obtain the sum of the numerator.

[0185] Second step, the exponential part in the denominator:

[0186]

[0187] If the difference between L 1 = (12500, 9820) and L c = (13000, 10000) is used to measure the Euclidean distance or coordinate difference, the measurement method needs to be defined first. If the Euclidean distance is taken, then

[0188]

[0189] For the remaining L 2 , L 3 ,... can also be calculated one by one to obtain the distance relative to L c and accumulated, denoted as S L . Substitute this value into exp(-S L ) and then add 1 to get the denominator value.

[0190] Third step, divide the numerator result by the denominator to obtain F group .

[0191] In a demonstration scenario, for example, the accumulated result of the numerator is approximately 85.36, and the denominator is approximately 1 + exp(-2380.5) ≈ 1 after calculation. Then:

[0192]

[0193] The results show that the current region has a relatively concentrated distribution of delivery distance differences and time window differences. At the same time, the denominator suppression caused by the overall distribution difference between geographic coordinates is not large, so the grouping adaptability is at a high level. When the value drops significantly to around 10, it often indicates that the distance distribution difference is too large or the time window is severely differentiated, resulting in poor consistency of tasks in the same region.

[0194] Based on the task grouping fitness, first record the grouping fitness that has been counted in each area, and then refer to the above grouping fitness calculation results item by item in the task list of the same area and make a unified comparison. If it is found that there is a large gap between the delivery distance and time window of multiple tasks, these tasks will be regarded as dispersed groups. If the grouping fitness is high, the batch of tasks will be regarded as a group with strong aggregation. Then, use the scheduling tool pre-deployed in the area to read all the divided groups and archive them according to the order of instruction execution. During this period, further resource allocation will be carried out for the groups whose grouping fitness exceeds a certain empirical threshold to balance the load and check the geographical location of each group member. Or whether the time data conforms to the valid range recorded before. If it is found that the coordinates of the abnormal position exceed the reference interval established during the comparison, for example, it is greater than the longest radius of the same area or exceeds the limit of the specified distribution radius, the task point will be temporarily removed and recorded as a separate update. In order to avoid the distortion of fitness caused by delayed statistics in high-concurrency scheduling during peak hours, a continuous monitoring mechanism can also be set up to retrieve the latest data every few minutes to refresh the group fitness. After completing the above operations, the task indexes under these legal groups are organized into a sequence to be executed, and entered into the overall multi-task management list according to the final order, and finally summarized into a multi-task parallel scheduling instruction.

[0195] The steps to obtain the emergency task priority sequence are:

[0196] Collect the site's real-time order surge, available drone quantity, and mission completion deadline to obtain basic mission urgency data;

[0197] Based on the basic data of task urgency, the priority score of emergency tasks is calculated using the following formula:

[0198]

[0199] Among them, P emg Score the priority of emergency tasks, Q 1 represents the current order surge, Q 2 represents the historical average order level, M represents the current number of available drones, and M th represents the minimum number of drones required for the mission, Q 3Represents the minimum order processing capacity required for task execution, T rem Represents the remaining time of the task;

[0200] Based on the emergency task priority score, the tasks are sorted according to the priority to obtain the emergency task priority sequence.

[0201] Specifically, collect the real-time order increment at the site, the available number of drones, and the task completion time limit. First, select the number of new orders per minute from the site's data records and compare it with the average order processing volume within a day to complete the extraction of the increment and summarize the number of new orders per minute to obtain the real-time order increment in the current time period. Then, confirm the available number of drones. Determine whether a drone can be added to the emergency dispatch by checking its current status, including remaining power, payload mounting status, and task completion degree one by one. Classify the available drones according to their models and executable capabilities. Then, view the task completion time limit information, record the deadline or the latest delivery period required for each task. If there is a task that has exceeded the preset time period, mark it specially. If the task is still within the effective execution range, include it in the subsequent calculation process. Immediately, summarize the order increment, the available number of drones, and the time limit data of each task and establish a list of basic emergency data. Attach a timestamp to each record in the basic data list for subsequent tracing whether any abnormal values have been missed or misdetected. When all the data corresponding to each record is matched, compare it with the pre-established effective range. For example, compare the increment value with the range of 0 to 1000 orders, compare the number of drones with the range of 0 to 50, and compare the task completion time limit with the range of 0 seconds to 86400 seconds. Handle the records that are higher than the upper limit or lower than the lower limit for abnormal situations to avoid statistical deviations. Then, extract the current increment, the number of drones, and the task completion time limit from all the data records within the effective range respectively, write them into a parameter table that can be called for emergency calculation, and finally obtain this task emergency basic data.

[0202] The benefit of the formula is that it combines the difference between the order increment and the historical order level to reflect the current order pressure, and also uses the contrast between the available number of drones and the minimum demand number of tasks to map to the arctangent function, so as to express the relative gap between demand and supply in the numerator. In addition, it incorporates the denominator in the form of a cosine function combined with the remaining time difference, so that the difference between the time constraint and the order processing capacity is moderately amplified or suppressed, thus giving a more flexible emergency task priority score.

[0203] Q 1The acquisition steps for are as follows: This parameter represents the current order increment, which needs to be refined from the real-time order addition records at the site. The specific method is to first count the number of newly added orders generated within each minute or shorter time period in the site data center, filter out possible duplicate orders or cancelled orders, and then accumulate and summarize these cleaned newly added order numbers to obtain the increment within a certain period. If the order surge amplitudes in different time periods vary greatly, segmented statistics can be performed based on multiple time periods and the sudden increase peaks can be marked separately. For example, in a peak period monitoring scenario, the site added a total of 235 orders within 10 minutes. After checking each order for anomalies, 220 valid orders were confirmed. Therefore, at this time Q 1 = 220.

[0204] Q 2 The acquisition steps for are as follows: This parameter represents the historical average order level, which is usually extracted from the site's historical order database and comprehensively compared with the order data of multiple time periods. After excluding abnormal peak periods such as holidays or special promotions, the remaining time periods are summed and divided by the number of days or hours in that time period to obtain a relatively stable average value. To improve accuracy, it can be statistically analyzed separately for morning, afternoon, evening, or weekdays and weekends, and then weighted and synthesized. For example, in the annual statistical data, the total number of orders on normal days excluding large promotion seasons is 300,000, and then divided by 200 days to get an average of 1,500 orders per day. If refined to the 1-hour dimension, it needs to be further distributed over 24 hours to form a more accurate average value. At this time, through subdivision, Q 2 = 1,500.

[0205] The acquisition steps for M are as follows: This parameter represents the current available number of drones. It is necessary to clarify on-site or within the dispatching system which drones meet the conditions for participating in emergency tasks, including the remaining battery power not being lower than a certain percentage, having sufficient load capacity, and being in a standby or idle state. When obtaining, the fleet status list can be retrieved. First, exclude the drones that are in flight or in the maintenance period, and then exclude the airframes that are in an unavailable state due to faults or wear of key components. The remaining ones are regarded as the available quantity. If it is found that the deployment areas of drones are different under high-concurrency dispatching, it is necessary to synchronously check the matching degree between the drones and the task geographical areas in the dispatching system to ensure that each drone can execute the task within a reasonable distance range. After such item-by-item comparison, the current available number of drones is obtained. For example, a certain period of statistics shows that there are 10 models in standby, then M = 10.

[0206] M thThe steps to obtain are: This parameter represents the minimum number of drones required for the task. When planning emergency tasks, it is necessary to pre-evaluate the overall order volume, delivery time and flight distance, so as to calculate how many drones must be dispatched in parallel to meet the requirements within a certain period of time. The calculation process is usually combined with the scale of the task. If the order volume and spatial distribution are large, the required fleet size will also increase relatively. In order to obtain more accurate values, the latest delivery time limit of the task, the average number of orders that can be completed by each drone, and the scheduling and changeover time of each drone can be included in the calculation. If it is confirmed after a detailed analysis that at least 6 drones are required to work simultaneously to complete N orders, M can be recorded in the scheduling list. th =6.

[0207] Q 3 The steps to obtain are: This parameter represents the minimum order processing capacity required for task execution, which refers to the site or system processing capacity required to ensure timely processing of concurrent orders within a certain time window, usually measured in "orders / minute" or "orders / hour". When obtaining, you can first view the parallel order volume in different time periods from the task overview, and then combine the dispatchable manpower and equipment processing speed to minimize the processing rate of each link. If you need to process 300 orders within 30 minutes, the minimum order processing capacity must reach 300 / 30=10 orders / minute. If it is detected that some peak periods will be higher, you need to record the peak in the test log and round it or leave a certain buffer. A certain monitoring shows that the global average processing rate needs to be maintained at 12 orders / minute to avoid delays, so take Q 3 =12, which is convenient for quick judgment in emergency tasks.

[0208] T rem The steps to obtain t are as follows: This parameter represents the remaining time of the task. It is necessary to continuously update the length of the period from the current time to the scheduled end or deadline of the task after the task is issued. A timer can be set for each task in the scheduling system and decremented in real time in seconds or minutes. If the task has been completed or terminated, the timer is stopped. If there are 1800 seconds available from the time the task is assigned to the current statistical time, then T rem =1800, and exclude expired or timed records.

[0209] Calculation process:

[0210] Step 1, example value: Q 1 =220, Q 2 =1500, M=10, M th =6, Q 3 =12, T rem =1800.

[0211] The second step is to calculate the

[0212] |220 - 1500| = 1280,

[0213] Then calculate tan -1 (M - M th ):

[0214] M - M th = 10 - 6 = 4, tan -1 (4) ≈ 1.3258

[0215] Therefore, add the numerator parts:

[0216] 35.78 + 1.3258 = 37.1058

[0217] In the third step, calculate the denominator: First, calculate |Q 3 - T rem |:

[0218] |12 - 1800| = 1788

[0219] Then calculate cos 2 (M), where M = 10. 10 needs to be regarded as radians or degrees for the cosine operation. If 10 is regarded as radians, cos(10) ≈ -0.83907, cos 2 (10) ≈ 0.7040, then:

[0220]

[0221] So, the fractional part in the denominator:

[0222]

[0223] Add 1, that is:

[0224] 1 + 1370.11 = 1371.11

[0225] In the fourth step, divide the numerator by the denominator:

[0226]

[0227] This result indicates that when the current order increment is quite different from the historical average order level and there is a certain margin in the available number of drones compared to the minimum demand, the final priority score is still relatively low due to the difference between the remaining time and the minimum order processing capacity. A value of around 0.02706 means that the emergency urgency is not extremely high. Once this value rises to 0.5 or greater, it indicates that the contradiction between the urgency of order processing in the current scenario and the shortage of drone resources is more prominent, and higher-priority handling is required immediately.

[0228] Based on the emergency task priority score, sort the tasks according to the priority. First, check the priority score value corresponding to each task and record the timestamp at the time of evaluation in a temporary sorting list. Compare the score values with an empirical threshold respectively. This threshold can be obtained by backtracking the historical emergency task data. For example, by counting that among the past 50 urgent orders, the cases with a score greater than 0.3 are all in a critical state. Then, mark the tasks with a score greater than or equal to 0.3 as high-priority level, and those with a score lower than the threshold are regarded as regular emergency level. Sort all the high-priority level tasks in descending order of the score. For tasks with the same score, the one with shorter remaining time is ranked first. Similarly, arrange the score values of the tasks at the regular emergency level. Subsequently, merge the two sorted queues to form a marked priority sequence. If there are significant changes in the remaining time or the order increment of a certain task in the subsequent time period, recalculate its score and update its position in this sequence. Continuously maintain the priority order through this dynamic iterative method. Finally, display an emergency task priority sequence sorted from high to low according to the emergency score in the scheduling management interface.

Claims

1. Real-time optimization and dispatching management system for drone delivery tasks, characterized by: The system comprises: The bat sound wave calculation module uses the bat algorithm to perform echolocation frequency conversion calculation on the distance between the current position of the drone and the target delivery point, the time window, and the upper limit of the carrying weight to obtain the task sound wave waveform matrix; the waveform amplitude in the task sound wave waveform matrix is ​​compared with the delivery distance to generate a task search space vector; The task resource allocation module performs subspace decomposition operation on the task search space vector through the bat algorithm to obtain a parallel task group sequence; performs constraint operation on the parallel task group sequence and the remaining power, load capacity and flight speed of the drone to generate a resource matching allocation plan; The scheduling optimization execution module calculates the delivery distance and time window in the resource matching allocation plan in groups according to geographical locations, and generates multi-task parallel scheduling instructions; The emergency response processing module performs urgency assessment on the site's real-time order surge, the number of available drones, and the task completion deadline to obtain an emergency task priority sequence.

2. The real-time optimization and dispatching management system for drone delivery tasks according to claim 1 is characterized in that: The steps for obtaining the task sound wave waveform matrix are as follows: Set the coordinates of the current position of the drone and the target delivery point, calculate the Euclidean distance between the two points, and obtain the distance data; Based on the distance data, the echolocation frequency conversion value is calculated, and the calculation formula is: Where TD is the distance between two points, T w is the time window, W max is the upper limit of the carrying weight, K is the constant for adjusting the unit influence, F echo Convert values ​​for echolocation frequency; Based on the echolocation frequency conversion value, the sound wave frequency conversion of echolocation is simulated by a bat algorithm to generate a task sound wave waveform matrix.

3. The real-time optimization and dispatching management system for drone delivery tasks according to claim 1 is characterized in that: The steps for obtaining the task search space vector are: Extracting each waveform amplitude in the task acoustic wave waveform matrix, indexing and matching the amplitude data according to the delivery distance, screening the amplitude range that meets the delivery task, and generating matched waveform amplitude data; Based on the matched waveform amplitude data, the task search space vector is calculated using the following formula: Among them, A1 is the maximum value of the waveform amplitude data, A2 is the minimum value of the waveform amplitude data, B1 is the square multiplication term of the delivery distance, B2 is the sine mapping term of the delivery distance, PD is the delivery distance, PC is the normalized offset term of the delivery distance and the waveform amplitude, A3 is the average value of the waveform amplitude data, S task Search space vector for the task.

4. The real-time optimization and dispatching management system for drone delivery tasks according to claim 1 is characterized in that: The steps for obtaining the parallel task group sequence are: Based on the task search space vector, a bat algorithm is used to analyze features, identify potential subspace boundaries, and obtain initial subspace feature data; Based on the initial subspace feature data, a subspace decomposition operation is performed, and the formula is: Among them, θ i Represents the position angle of the i-th data point, V i represents the search space vector value of the i-th task, H groups represents the subspace decomposition result, RN represents the total number of tasks in the task search space vector; Based on the subspace decomposition result, the data points are grouped according to similarity to form a parallel task group sequence.

5. The real-time optimization and dispatching management system for drone delivery tasks according to claim 1 is characterized in that: The steps for obtaining the resource matching allocation scheme are: Based on the parallel task group sequence, the matching degree between the task and the UAV resources is calculated, and the calculation formula is: Among them, R match is the matching degree between the task and the UAV resources, E is the remaining power of the UAV, C is the minimum power required for the task, ZL is the current load of the UAV, S is the flight speed of the UAV, RD is the task delivery distance, and P is the maximum delivery distance of the UAV; The drones are allocated according to the matching degree between the tasks and the drone resources, and a resource matching allocation plan is generated.

6. The real-time optimization and dispatching management system for drone delivery tasks according to claim 1 is characterized in that: The steps for obtaining the task space clustering results are as follows: Extract the delivery distance and time window of each task from the resource matching allocation plan, classify the tasks by geographical location, calculate the distribution density of tasks in each area, and obtain regional task distribution data; Based on the regional task distribution data, the task grouping adaptability is calculated using the following formula: Among them, F group is the task grouping adaptability, D i is the delivery distance of the ith task, D avg is the average delivery distance of all tasks in the region, T i is the time window of the ith task, T max is the maximum time window of tasks in the region, L i is the geographical location coordinates of the ith task, L c is the center coordinate of the area, and N is the total number of tasks in the current area; Based on the task grouping adaptability, a multi-task parallel scheduling instruction is generated.

7. The real-time optimization and dispatching management system for drone delivery tasks according to claim 1 is characterized in that: The steps for obtaining the emergency task priority sequence are: Collect the site's real-time order surge, available drone quantity, and mission completion deadline to obtain basic mission urgency data; Based on the basic data of task urgency, the priority score of the emergency task is calculated, and the calculation formula is: Among them, P emg The priority of the emergency task is scored, Q1 represents the current order surge, Q2 represents the historical average order level, M represents the current number of available drones, and M represents the current number of available drones. th represents the minimum number of drones required for the mission, Q3 represents the minimum order processing capacity required for mission execution, and T rem Represents the remaining time of the task; Based on the emergency task priority score, the tasks are sorted according to the priority to obtain an emergency task priority sequence.

Images