Digital twinborn decision-making method for city-level low-altitude economy

Abstract

The invention relates to the technical field of digital twinning, in particular to a digital twinning decision-making method for city-level low-altitude economy. The method comprises the following steps: acquiring three-dimensional geographic information of a city; detecting a building three-dimensional contour in the urban three-dimensional geographic information; determining a landing area by using the three-dimensional contour of the building; performing unmanned aerial vehicle landing point optimization on the landable area by adopting a pre-trained digital twin model to generate landing point data; unmanned aerial vehicle distribution task information is acquired; determining a first distribution path based on the unmanned aerial vehicle distribution task information and the landing point data; continuously monitoring the flight state of the unmanned aerial vehicle in the first distribution path; and detecting the energy consumption of the unmanned aerial vehicle based on the flight state, and if the energy consumption of the unmanned aerial vehicle is greater than a preset energy consumption threshold, determining that the flight state is abnormal. The stability and the energy efficiency utilization rate of unmanned aerial vehicle distribution are improved based on the digital twin technology.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twinning, in particular to a digital twinning decision method for urban-level low-altitude economy. BACKGROUND

[0002] Currently, the application of digital twinning technology in urban-level low-altitude economy is gradually expanding, and has been used for tasks such as unmanned aerial vehicle scheduling, flight path planning, and urban airspace management. However, traditional digital twinning decision methods still have many shortcomings in dealing with complex urban environments, multi-source data fusion, and dynamic flight regulation. Most of them rely on static three-dimensional modeling data, lack the ability to jointly model real-time weather, dynamic airspace, and time-varying properties of buildings, resulting in a lack of adaptability and robustness in the generated flight path. Only based on geometric space modeling to make landing point or path decisions, ignoring the multi-dimensional factors such as flight safety, communication blind area, and energy consumption constraints in the urban low-altitude environment, resulting in a large deviation between the flight planning scheme and the actual execution. The decision mechanism of the existing digital twinning model mostly uses rule engines or static scoring methods, which cannot handle real-time conflict prediction, optimal obstacle avoidance, and dynamic adjustment tasks in low-altitude traffic, and lack intelligent response mechanisms. In summary, there is an urgent need for a dynamic, adjustable, and self-adaptive digital twinning decision method for urban-level low-altitude economy to achieve efficient, safe, and energy-saving flight scheduling in complex urban environments. SUMMARY

[0003] Therefore, it is necessary to provide a digital twinning decision method for urban-level low-altitude economy to solve at least one of the above technical problems.

[0004] To achieve the above-mentioned purpose, a digital twinning decision method for urban-level low-altitude economy includes the following steps: Step S1: Obtain city three-dimensional geographic information; detect the three-dimensional contour of the building in the city three-dimensional geographic information; determine the landing area using the three-dimensional contour of the building; Step S2: Use a pre-trained digital twinning model to optimize the unmanned aerial vehicle landing point in the landing area, generate landing point data; obtain unmanned aerial vehicle delivery task information; determine the first delivery path based on the unmanned aerial vehicle delivery task information and the landing point data; Step S3: Continuously monitor the flight state of the unmanned aerial vehicle in the first delivery path; detect the energy consumption of the unmanned aerial vehicle based on the flight state, if the energy consumption of the unmanned aerial vehicle is greater than the preset energy consumption threshold, it is judged as abnormal flight state; Step S4: Collect wind power parameters of abnormal flight state; exclude strong wind counter-flow sections of the first delivery path according to the wind power parameters, and obtain the second delivery path.

[0005] The present application has the following advantages: In one aspect, by acquiring multi-modal data in the target area, including three-dimensional city models, weather data and dynamic traffic information, combined with real-time division of navigable areas and flight safety zones, a digital twin environment with spatiotemporal perception capability can be constructed, effectively improving the fineness and dynamic response capability of urban airspace modeling; this process not only considers static geographic elements, but also integrates dynamic weather changes and temporary flight ban factors, which can provide more robust flight base map for UAV scheduling.

[0006] On the other hand, based on flight conflict prediction and multi-path generation mechanism, potential path overlap or intersection points can be identified in advance, and by constructing a first flight path set and distributing it to the UAV group in real time, active path avoidance in low-altitude traffic can be achieved; compared with traditional path planning methods based on static maps, the present application has adaptive ability and prediction and control mechanism, which can effectively reduce the risk of flight conflict and improve the safety and timeliness of flight tasks.

[0007] On the other hand, by introducing energy consumption monitoring and dynamic path adjustment mechanism, the flight energy consumption parameters of UAV are collected in real time, and the path re-planning is realized by combining with the path segment energy consumption model, which optimizes the power distribution and range management under the premise of ensuring task completion; this method can avoid task interruption or return failure caused by energy consumption misjudgment, and significantly improve the stability and endurance efficiency of urban-level UAV operation. BRIEF DESCRIPTION OF DRAWINGS

[0008] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings: Figure 1 A step flowchart of a digital twin decision-making method for urban-level low-altitude economy of the present application; Figure 2 A detailed step flowchart of step S1 in the present application; Figure 3 A schematic diagram of UAV flight speed energy consumption monitoring in the present application; The implementation of the purpose of the present application, functional characteristics and advantages will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0009] The technical method of the present application will be described clearly and completely below in combination with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0010] Moreover, the attached drawings are only schematic and are non-limiting. Each block denoted by a reference in the attached drawings represents at least one physical entity that can be realized in hardware or software. It is contemplated that such entities can be implemented to handle one or more of the described functionalities without deviating from the scope of the application. The numerous innovative teachings of the present application specified above can be implemented in many physical ways.

[0011] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Pronouns in the masculine form include the feminine form, and vice versa, and the singular form also includes the plural form, unless the context clearly dictates otherwise.

[0012] To achieve the above object, there is provided Figures 1 to 3 The present application provides a digital twin decision-making method for urban-level low-altitude economy, comprising the following steps: Step S1: Obtain urban three-dimensional geographic information; detect the three-dimensional contour of buildings in the urban three-dimensional geographic information; and determine the landing area using the three-dimensional contour of the buildings; In an embodiment, laser radar point cloud data and oblique photography image fusion technology are used to obtain three-dimensional geographic information of the core area of the city, with a point cloud accuracy of better than 0.1 m and a coverage range of about 2.5 km 2 Buildings are identified by a point cloud instance segmentation network based on deep learning (such as PointNet++), and the three-dimensional contour of the roof is extracted. The landing area is screened using normal vector calculation and spatial projection analysis (such as: area > 4 m 2 , inclination < 10°, no obstructions, etc.). The screened area is marked as a landing area and output in Shapefile format for subsequent path planning; In another embodiment, assuming that the point cloud data of the city area processed by the system is 1 million points, the number of buildings extracted is 820, and a total of 1472 roof areas are generated after analysis. Based on the set rules (area threshold of 5 m 2 , slope less than 8°), 312 candidate landing areas are screened, and finally 92 areas are selected as landing areas. These areas are distributed on the top of buildings, parking lots, and logistics platform roofs, providing a spatial basis for subsequent unmanned aerial vehicle path planning.

[0013] Step S2: Optimize the UAV landing point in the landable area using the pre-trained digital twin model, generate landing point data; obtain UAV delivery task information; determine the first delivery path based on the UAV delivery task information and the landing point data; In an embodiment, the digital twin model trained by a fusion neural network (CNN+MLP) is used to evaluate the landing point in the landable area. The model input includes multi-dimensional features such as the geometric parameters of the area, the occlusion risk, the building type, the surrounding wind field, etc., and the output is a score value. The point with the highest score is selected as the target landing point of the UAV. Then the system obtains the current delivery task data (including the starting point, the ending point, the power, the load, the priority, etc.), combines the landing point data, plans the first delivery path through the three-dimensional A* algorithm, and outputs the path coordinate sequence and the flight parameters; In another embodiment, assuming that the number of delivery tasks is 50, and each task contains 3 alternative landing points. Each point is evaluated by the digital twin scoring model, with a score range of [0, 1], and points with a score > 0.85 are selected as optional landing points, with an average of 2 points per task. Based on path length, power limit and obstacle distribution, multi-objective optimization is performed to calculate 50 first delivery paths, with an average path length of 2.4 km and an estimated average flight time of 8.2 minutes.

[0014] Step S3: Continuously monitor the flight state of the UAV in the first delivery path; detect the energy consumption of the UAV based on the flight state, and if the energy consumption of the UAV is greater than the preset energy consumption threshold, it is judged as an abnormal flight state; In an embodiment, during the flight of the UAV, flight speed, attitude, battery current and voltage, and GPS trajectory data are collected every second, and the unit distance energy consumption (unit: Wh / km) is calculated in real time. Set the energy consumption threshold to 125% of the standard working condition of the UAV (such as 0.4 Wh / km x 1.25 = 0.5 Wh / km), and once the unit energy consumption of a certain path exceeds the threshold for 5 seconds in a row, it is marked as "abnormal flight state". The system records the location of the abnormality and prompts that there may be air flow disturbance or equipment abnormality; In another embodiment, assuming that the system collects 1200 flight state samples during a 20-minute delivery process, the unit distance energy consumption is calculated to be between 0.35-0.7 Wh / km. Set the preset threshold to 0.5 Wh / km, and statistics show that there are 164 sample points exceeding the threshold, of which 78 consecutive samples are marked as abnormal sections, corresponding to 3 main areas in the flight path, prompting the existence of meteorological disturbance or terrain rising area.

[0015] Step S4: Collect wind force parameters of the abnormal flight state; exclude the strong wind reverse flight section of the first delivery path according to the wind force parameters, and obtain the second delivery path; In an embodiment, after detecting the area of abnormal flight state, the system obtains the city-level wind field data (provided by wind radar or meteorological API) of the location in real time, including wind speed and wind direction. If the wind speed > 6 m / s and the angle between the wind direction and the flight direction > 120°, the section is determined as a strong wind counter-flying section. The system uses a path re-planning algorithm based on cost re-weighting (such as improved Dijkstra) to avoid all counter-wind sections and recalculate the second delivery path. The path preferentially selects channels with lower wind speed to ensure flight safety and energy efficiency; In another embodiment, assuming that the first path is divided into 50 flight sections, 12 of which have abnormal energy consumption, and after wind data analysis, it is determined that 7 of them are strong wind counter-flying sections with an average wind speed of 7.8 m / s. After path re-planning, these 7 sections are replaced to generate a new second delivery path. After simulation evaluation, the new path increases the total distance by 4.5%, but the energy consumption decreases by 12%, the flight time is prolonged by about 1.2 minutes, and the stability is significantly improved.

[0016] Preferably, step S1 comprises the following steps: Step S11: Obtain city three-dimensional geographic information; In an embodiment, a fixed-wing unmanned aerial vehicle equipped with a high-precision laser radar (such as RIEGL VUX-1) and an RTK-GNSS system is used to scan the city core area (about 2.5 km 2 ) at low altitude to obtain about 72 million original three-dimensional geographic information point cloud data, with a point spacing accuracy of better than 0.05 m, covering city elements such as buildings, roads, and greenery.

[0017] Step S12: Establish a grid structure, import the original point cloud of city three-dimensional geographic information into the grid structure, and output a city vertical point cloud set; construct a vertical slice sequence along the Z-axis direction of the city vertical point cloud set; In an embodiment, the city area is divided into 1 m × 1 m two-dimensional grid units, and the points in each unit are aggregated along the Z-axis direction to form a city vertical point cloud set. Then, with a slice thickness of every 0.2 m, the vertical slice sequence is divided from the ground up, generating a total of about 150 layers of vertical slices; In another embodiment, the grid unit is set to 0.5 m × 0.5 m, all point clouds are mapped into a two-dimensional grid structure, and the vertical direction slices are constructed with an interval of 0.3 m, generating a total of 96 slices. Each layer contains an average of about 35,000 points, facilitating the identification of the spatial evolution characteristics of building structures in the subsequent steps.

[0018] Step S13: Calculate the normal vector of the vertical slice sequence; identify the face structure based on the vertical slice sequence; if the normal vector points to the face structure, record it as a potential building area; In an embodiment, the PCA algorithm is used to locally fit the point cloud neighborhood in each slice, and the normal vector is extracted. It is determined whether the normal direction between adjacent layers is continuous. If the planarity and normal consistency conditions (the included angle is less than 15°, and the fitting error is less than 0.01m) are met, it is determined that it is a facet structure, and is marked as a potential building area; In another embodiment, assuming that the normal vector is calculated using a 16-neighbor point set for each slice, when the average normal included angle between two consecutive slices is less than 12°, and the point density exceeds 25 points per square meter, it is considered as a building facet. About 540 potential building areas are identified, and the preliminary coverage rate is 88.6%.

[0019] Step S14: identifying the boundary points of the potential building area, connecting the boundary points in the contour order to generate a three-dimensional contour of the building; In an embodiment, the boundary points of the edge of each potential building area are extracted, the contour line is extracted based on the alpha shape algorithm, and the three-dimensional contour of the building is connected in the Z-axis direction. For complex contour areas, Bezier curve interpolation smoothing processing is used to generate a continuous surface model; In another embodiment, a total of 470 potential building areas are extracted, and the average contour point of each area is about 75. The boundary point set is extracted by the edge tracking algorithm, and the closed contour is reconstructed according to the connection order to construct a final three-dimensional contour grid body, and a total of 431 three-dimensional models of buildings are generated for subsequent projection analysis.

[0020] Step S15: reading the three-dimensional contour of the building, constructing a projection plane, obtaining a drop site data, and mapping the drop site data to the projection plane to determine a landable area; In an embodiment, the three-dimensional contour is extracted based on the top slice of the building, projected onto the XY plane, and a projection map is constructed. The GPS drop point (WGS84 coordinates) provided by the low-altitude logistics system is projected onto the plane after coordinate conversion and affine mapping, and it is determined whether it is located within any building contour. The area satisfying the conditions of area > 3m 2 , no occlusion, and surface slope < 5° is determined as a landable area; In another embodiment, assuming that a total of 60 drop sites are received, 42 of which are found to fall within the contour area after mapping to the projection plane. Further screening excludes areas with too large height difference and areas with occlusion, and finally 32 areas meeting the landing standard are identified. This result can support end-of-line distribution or emergency drop tasks, and has good geographical adaptability and precision guarantee.

[0021] Preferably, the step S2 of using a pre-trained digital twin model to optimize the UAV landing point of the landable area comprises: The pre-trained digital twin model is used to optimize the UAV landing point of the landable area; In an embodiment, a digital twin model pre-trained based on a city-level low-altitude environment is loaded, which takes a three-dimensional contour map of buildings, a landing area projection map, and site space constraint information as input. In the model input layer, the image information in the projection plane is aligned and encoded into a tensor format with geographical coordinates; the initial layer uses a three-level feature extraction module (convolution kernel size of 1x1, 3x3, and 5x5) to extract key features such as spatial texture, shadow edge, and landing openness of different size building surfaces.

[0022] In another embodiment, assuming that the input is a city area projection map of size 128x128, and the site density is 20 points per square kilometer, the initial layer of the model uses dilated convolution with different receptive fields (dilated rate = 1, 2, 4), which extracts 3 groups of feature maps (size 128x128x32, 64x64x64, and 32x32x128) in total. Each group of feature maps captures micro texture, local occlusion pattern, and overall geometric contour information, providing a multi-scale information basis for subsequent fusion.

[0023] The digital twin model includes an input layer, multiple initial layers, a fusion layer, and a decision layer; the initial layer is used to extract multi-scale landing area features, the fusion layer is used to upsample and weightedly fuse the landing area features to obtain a fused feature map; and in the decision layer, the fused feature map is comprehensively scored to select the optimal landing point.

[0024] In an embodiment, the initial layer uses a multi-branch convolution structure (Inception-like), which performs 1x1, 3x3, and 5x5 convolutions on the input image respectively, and uses batch normalization and ReLU activation function, outputting three scale feature maps. This process takes into account local texture information and large-scale shape structure, and finally obtains a 128x128x96-dimensional multi-scale landing area feature map with uniform scale.

[0025] In another embodiment, assuming that the original projection map size is 256x256, three feature maps with sizes of 256x256x32, 128x128x64, and 64x64x128 are extracted, respectively. The low-resolution map focuses on the overall terrain shape, the medium-resolution map captures the building top surface partition, and the high-resolution map focuses on the surface texture and occlusion boundary, providing a basic representation for the fusion layer.

[0026] It is to be noted that after the low and medium resolution feature maps are respectively up-sampled to 128x128 by bilinear interpolation, the different scale features are fused by pixel-wise channel weighting. A learnable weight vector is set for each scale channel, and after Softmax normalization, the weighted sum is obtained, and finally a unified size fusion feature map F(128x128x96) is generated, which has stronger semantic and structural expression ability.

[0027] It is to be noted that assuming that the fusion layer uses transpose convolution (kernel=4, stride=2) to up-sample the 64x64 and 32x32 feature maps twice to restore them to 128x128 size; when fusing, an initial weight is assigned to each scale, i.e., a1=0.5, a2=0.3, and a3=0.2, which is automatically adjusted after model training. The final fusion feature map has a dimension of 128x128x64, and the position with the strongest response is the high-confidence landing point candidate region identified by the model.

[0028] In an embodiment, a three-layer fully connected neural network is used to score the fusion feature map at the region level. Spatial pooling and flattening are performed on the F feature map to obtain a one-dimensional feature vector; then FC1 (dimension 256), FC2 (dimension 64), and FC3 (dimension 1) are sequentially passed through to output the comprehensive landing score of each candidate point. The point with the highest score is selected as the optimal landing point, which is then mapped back to the original building projection plane for scheduling and deployment.

[0029] It is to be noted that assuming that the final generated fusion feature map is 128x128x64, a 64-dimensional vector is obtained after global average pooling, which is input into the decision layer for scoring and output. The comprehensive score is activated by sigmoid to output a confidence value between 0 and 1, and a threshold of 0.85 is set. If the maximum score point exceeds the threshold, it is considered as a "safe landing point"; if it is lower than the threshold, it is prompted to switch to a backup landing strategy (such as transferring to an adjacent site). This method improves the safety and adaptability of landing point determination.

[0030] Preferably, the up-sampling and weighted fusion of multi-scale features include: Receiving the landable area feature map from the initial layer and calculating the resolution of the landable area feature; using the resolution to filter low-resolution feature maps; In an embodiment, the initial layer outputs three groups of landable area feature maps of different scales, with sizes of 128x128, 64x64, and 32x32, and corresponding channel numbers of 32, 64, and 128. The system automatically calculates the spatial resolution (in pixels per square meter) of each group of feature maps according to their width and height dimensions, taking the 128x128 feature map as the benchmark high resolution. Feature maps with a resolution lower than the benchmark Figure 1Half of the feature maps (i.e. 64x64 and 32x32) are filtered as low resolution feature maps for subsequent upsampling and fusion processing.

[0031] In another embodiment, assuming the actual area of the input region is 16m x 16m, the corresponding feature map size is 128x128 (resolution: 1 pixel = 12.5cm), 64x64 (1 pixel = 25cm), and 32x32 (1 pixel = 50cm). The system sets the resolution filtering threshold to be ≥ 25cm (i.e. lower than the baseline Figure 1 half), so the 64x64 and 32x32 feature maps are identified as low resolution feature maps.

[0032] Each pixel of the low resolution feature map is copied into a larger area of surrounding pixels to obtain an upsampled feature map; In an embodiment, the nearest neighbor interpolation method is used to upsample the low resolution feature map. The 64x64 feature map is upsampled by 2 times (copying 1 pixel to a 2x2 neighborhood), and the 32x32 feature map is upsampled by 4 times (copying 1 pixel to a 4x4 neighborhood), and their sizes are unified to 128x128. During the copying process, the channels remain unchanged, and only the spatial dimension is expanded to ensure alignment with the baseline resolution map for pixel-level fusion.

[0033] In another embodiment, the 64x64x64 feature map is upsampled to 128x128x64, and the 32x32x128 feature map is upsampled to 128x128x128. The value of each pixel is copied as is to the corresponding 2x2 or 4x4 neighborhood area to generate a new upsampled feature map. This operation can be simply understood as pixel expansion in space, so that each scale feature has the same size.

[0034] A weight coefficient is assigned to the upsampled feature map, and the pixel value of the upsampled feature map is multiplied by the corresponding weight coefficient to obtain a fused feature map.

[0035] In an embodiment, learnable weight coefficients a1, a2, a3 (satisfying a1+a2+a3=1) are assigned to upsampled feature maps of different scales, respectively, corresponding to the original high resolution map, the 64→128 map, and the 32→128 map. By introducing a weight mechanism, the fused feature map at each position is obtained by weighted summation: wherein , represents the upsampled feature map. This fused map integrates multi-scale information and automatically adjusts the importance weight of each scale through training.

[0036] ​In another embodiment, the initial weights are set to α1=0.5, α2=0.3, and α3=0.2, indicating that high-resolution features contribute the most. For each pixel, the values ​​at corresponding positions in the three feature maps are weighted and summed. For example, at coordinates (50, 80), the values ​​in the three feature maps are 0.6, 0.4, and 0.7 respectively, so the fusion value is F(50,80) = 0.5×0.6 + 0.3×0.4 + 0.2×0.7 = 0.55. The final fused feature map has a size of 128×128×C (C is the sum of the channels), providing strong regional semantic features for subsequent landing point recognition.

[0037] Preferably, step S2, which determines the first delivery route based on the drone delivery mission information and landing point data, includes: Based on the analysis of drone delivery mission information, the takeoff point coordinates and delivery mission sequence are determined. In one embodiment, the system receives drone delivery task information, including takeoff point coordinates (e.g., warehouse location), multiple landing point coordinates (e.g., user pickup points), and the execution order of the delivery tasks (e.g., according to order time). The system automatically extracts the takeoff point coordinates P0=(x0, y0, z0) from the task information and sequentially obtains each landing point Pi=(x0, y0, z0) according to the task order. i , y i , z i ), i=1,2,...,N, construct the set of points for the path to be planned.

[0038] In another embodiment, assuming the takeoff point P0 has coordinates of (100, 200, 30), the delivery mission includes three landing points: P1=(130, 230, 35), P2=(120, 220, 40), and P3=(150, 210, 28); the mission sequence is P1→P2→P3. The system stores the coordinates of each point in this order for subsequent flight path generation.

[0039] The lateral difference is obtained by subtracting the lateral coordinates of the takeoff point from the lateral coordinates of the landing point. In one embodiment, the system calculates the lateral difference Δx between each landing point Pi and the takeoff point P0. i = x i –x0 is used to estimate the horizontal offset. For example, for P1, Δx1 = 130 – 100 = 30; for P2, Δx2 = 120 – 100 = 20; for P3, Δx3 = 150 – 100 = 50.

[0040] The longitudinal difference is obtained by subtracting the longitudinal coordinates of the takeoff point from the longitudinal coordinates of the landing point. In one embodiment, the longitudinal difference Δy is calculated. i = y i – y0 reflects the horizontal offset of the drone perpendicular to the horizontal axis. For example, Δy1 = 230 – 200 = 30; Δy2 = 220 – 200 = 20; Δy3 = 210 – 200 = 10.

[0041] Subtract the altitude of the takeoff point from the altitude of the landing point to obtain the altitude difference; In one embodiment, the system calculates the height difference Δz i = z i – z0 represents the vertical flight change, used for subsequent spatial distance calculations. For example, Δz1 = 35 – 30 = 5; Δz2 = 40 – 30 = 10; Δz3 = 28 – 30 = –2.

[0042] Square the lateral difference, longitudinal difference, and height difference respectively, and take the squared result; add the squared results and solve for the positive square root, which is the straight-line flight distance; In one embodiment, for each landing point, the three-dimensional straight-line distance D between it and the takeoff point is calculated sequentially. i = √(Δx i 2 + Δy i 2 + Δz i 2 ).For example: ; ; ; An ordered flight path sequence is generated using the delivery task order and straight-line flight distance to determine the first delivery path.

[0043] In one embodiment, the system combines the preset order of the delivery tasks and the flight distance Dᵢ of each segment into a first delivery path sequence: P0→ P1→ P2→ P3, and can further calculate indicators such as cumulative path length and estimated flight time for task scheduling and energy consumption prediction.

[0044] In another embodiment, based on the takeoff point P0 and the sequence P1→P2→P3, a flight path sequence [(P0,P1, 42.72m), (P1, P2, D] is generated. 12 ), (P2, P3, D 23 )], where D 12 and D 23The same calculation can be performed. The final path planning structure is clear and can be directly used in the UAV flight path execution module.

[0045] Preferably, generating an ordered flight path sequence using the delivery task order and straight-line flight distance includes: Extract the location information of the delivery task sequence and arrange them according to the delivery task sequence to construct an initial task point sequence; In one embodiment, the system first receives drone delivery task information, including the spatial coordinates of multiple delivery target points and the corresponding task execution order (e.g., based on order time and priority level). The system automatically parses the task content and extracts the location coordinates Pi = (x... i , y i , z i ), and arrange them in a predetermined order to construct the initial task point sequence [P1, P2, ..., P n ], used for subsequent path optimization calculations.

[0046] In another embodiment, assume the task involves a takeoff point P0=(100, 200, 30) and three delivery points P1=(120, 210, 35), P2=(140, 220, 28), and P3=(130, 205, 33). The system reads the task in the order P1→P2→P3 and organizes it into an initial task point sequence: [P0, P1, P2, P3].

[0047] In the initial task point sequence, the relative flight cost between task points is calculated using the straight-line flight distance to obtain flight cost data; the flight cost data is then used to sequentially connect the initial task point sequence to generate an ordered flight path sequence to determine the first delivery path.

[0048] In one embodiment, the three-dimensional Euclidean distance is calculated for any two adjacent points Pi and Pi+1 in the initial task point sequence. The system integrates the costs of all adjacent task pairs into a flight cost data table for path cost evaluation and ranking.

[0049] In another embodiment, according to the aforementioned task point sequence: ; ; ; In another embodiment, the system sorts the initial waypoints based on flight cost data in terms of connectivity, and can use heuristic greedy algorithms, dynamic programming, or TSP (Traveling Salesman) variants to optimize the paths. Finally, a minimum total cost path sequence is generated, which is used to control the drone to complete the delivery task along the low-cost path; the resulting order is the first delivery path. In another embodiment, the system directly accumulates and sorts the flight cost data to confirm that the original mission sequence (P0→P1→P2→P3) is the approximately optimal path. The total cost is calculated as D. 01 + D 12 + D 23 ≈ 22.91 + 23.43 + 18.71 ≈ 65.05 meters. The system uses this path as the first delivery route and transmits it to the backend navigation module for track planning and task scheduling.

[0050] Preferably, step S3 includes the following steps: Step S31: Continuously monitor the flight status of the drone in the first delivery route; In one embodiment, a wireless communication protocol (such as a 4G / 5G link or a self-organizing mesh network) between the ground station system and the UAV is used to acquire the UAV's flight status data in real time along the first delivery path. The flight status data includes, but is not limited to: current position (three-dimensional coordinates), flight speed, flight attitude (heading angle, pitch angle, roll angle), thruster speed, current, and voltage. The system is set to a refresh rate of 10Hz, meaning the UAV status is collected every 0.1 seconds. In the continuous state stream, a sliding window mechanism (e.g., 10 frames per second) is used to perform short-term stability analysis to eliminate transient jitter interference and ensure that subsequent energy consumption assessments have temporal consistency and robustness. In another embodiment, assuming the UAV is equipped with a flight control module that supports state sampling at a sampling frequency of 10Hz, the state sequence recorded during the execution of the first delivery route task is as follows: Time point t1: speed 5.2m / s, yaw angle 2.1°, current 2.8A; Time point t2: speed 5.4m / s, yaw angle 2.0°, current 3.0A; ... This is recorded continuously for 60 seconds, collecting a total of 600 flight state records. This data will serve as the basis for subsequent calculations of propulsion energy consumption and velocity energy consumption.

[0051] Step S32: Determine propulsion energy consumption based on flight status; In one embodiment, propulsion energy consumption is primarily determined by the output power of the UAV motors. Based on the collected real-time current I(t) and voltage U(t), the instantaneous power P(t) = U(t) × I(t) at each moment can be calculated. Integrating the instantaneous power over a unit of time (e.g., 1 second) yields the propulsion energy consumption for that time period. Propulsion energy consumption is also affected by attitude adjustments; for example, during wind resistance adjustments, some propellers may need to temporarily increase their rotational speed, leading to increased propulsion energy consumption. The system introduces an attitude compensation factor f(θ,φ,ψ) to adjust the accuracy of energy consumption estimation, thereby more accurately reflecting the actual energy output of the propulsion module.

[0052] Step S33: Determine flight speed energy consumption based on flight status; In one embodiment, flight speed energy consumption is used to measure the additional energy consumed by the UAV in maintaining its current flight speed. This energy consumption is proportional to air resistance and can be derived from the square of the velocity v(t) and the UAV's projected area A and air density ρ: F_drag = 0.5 × ρ × A × v 2 By combining this with time, the energy consumed per unit time to maintain speed can be estimated. For different speed ranges, the system can establish an empirical mapping table and use the table lookup method to quickly obtain the energy consumption at flight speed, thereby avoiding real-time calculations. In another embodiment, it is assumed that the drone's flight speed is 6.0 m / s and its lateral projected area is 0.04 m². 2 The air density ρ = 1.225 kg / m³ 3 Air resistance F_drag ≈ 0.5 × 1.225 × 0.04 × (6.0) 2 =0.882N. Assuming this speed is maintained for 10 seconds, the energy consumption for flight speed is: Work = F_drag × s = 0.882N × 60m = 52.92J = 0.0147Wh. If the speed increases to 8.0m / s, the corresponding drag energy consumption is approximately 0.026Wh. The energy consumption increases non-linearly with increasing speed.

[0053] Step S34: Integrate propulsion energy consumption and flight speed energy consumption into UAV energy consumption; In one embodiment, the system weighted and merges the propulsion energy consumption obtained in step S32 with the flight speed energy consumption obtained in step S33 to obtain the total energy consumption of the UAV per unit time. Let the weight of propulsion energy consumption be α and the weight of speed energy consumption be β, typically α=0.7 and β=0.3 to reflect the dominant contribution of the propulsion system to the overall energy consumption. Total energy consumption E_total = α × E_propulsion + β × E_velocity. This total energy consumption will be accumulated over time to assess whether the current flight mission exhibits energy consumption anomalies.

[0054] Step S35: If the drone's energy consumption is greater than the preset energy consumption threshold, it is judged as an abnormal flight state; In one embodiment, the system sets different energy consumption thresholds based on different flight modes (hovering, cruise, climb, descent). For example, the maximum energy consumption threshold within 10 seconds in cruise mode is 0.7Wh. If the actual energy consumption during a certain period exceeds this threshold, an abnormal state judgment will be triggered. After the judgment, the system will record the time of the abnormal period, flight altitude, attitude angle, and ambient wind speed for subsequent fault analysis, path optimization, or battery health assessment. In another embodiment, assuming the drone's energy consumption limit is set to 0.7Wh for 10 seconds during cruise, while the actual monitored energy consumption is 0.82Wh. Combined with the flight data at this time: flight speed 9.5m / s, current 3.6A, pitch angle 5.2°, and wind speed 3.8m / s, the system determines that the increased energy consumption is caused by "propulsion anomaly" due to moderate wind speed, immediately triggering a deceleration command and pushing a flight log alarm. If the anomaly continues to accumulate for three windows, the flight mission is automatically suspended and the drone returns to base.

[0055] Preferably, step S32 includes the following steps: Step S321: Extract flight control data of flight status; In one embodiment, flight log data of a certain type of quadcopter UAV was acquired during flight testing in an urban environment. The extracted flight control data includes: attitude angle data (pitch angle, roll angle, yaw angle), GPS position, flight altitude, velocity vector, flight mode identifiers (such as altitude hold, point hold, mission path, etc.), and control command data at corresponding times (such as ascent, descent, forward, and steering control variables). Each control data entry is accompanied by a millisecond-level timestamp, achieving a complete record of the entire flight process. The extracted flight control data provides a foundation for subsequent feature analysis and energy consumption modeling.

[0056] Step S322: Extract attitude adjustment commands from flight control data and record the current, voltage, and propeller speed parameters driven by the attitude adjustment commands to form a drive response data sequence aligned with timestamps; In one embodiment, based on the attitude change range in the flight control data, the actual attitude adjustment commands issued by the flight control system (e.g., stable pitch, leveling, rapid left turn, etc.) are identified, and the changes in current, voltage, and propeller speed of each motor during the execution of these commands are extracted from the synchronized sensor data stream. These three types of response signals are aligned with a unified timestamp to form a complete drive response data sequence. This data is used to analyze the dynamic response characteristics of the system under control commands.

[0057] Step S323: Set a fixed time window in the drive response data sequence, extract the current signal segment, calculate the first-order difference value of the current signal segment, and record it as a current rate sequence; In one embodiment, a fixed-length time window (e.g., 500 ms, corresponding to approximately 50 sampling points) is set for the current data in the drive response sequence, and a current sub-segment is extracted within each window. A first-order difference operation is performed on each sub-segment to calculate the rate of change of current values ​​between adjacent sampling points, resulting in a current rate sequence reflecting instantaneous current fluctuations. This sequence is used for further analysis of motor load change trends. In another embodiment, assuming a sliding window length of 50 points and a sliding step size of 10 points, sliding window extraction is performed on a current sequence with a total length of 6000 points, yielding approximately 596 current segments and corresponding current rate sequences. If the original current value within a certain window is [4.1, 4.3, 4.6, 4.9, 5.2, …], then the differential current rate sequence is [+0.2, +0.3, +0.3, +0.3, …], showing that the current continuously increases within that segment.

[0058] Step S324: Perform sliding window integration on the current rate sequence, calculate the cumulative current change within each window, and generate an electrical work conversion sequence; In one embodiment, the total change in current within each current rate sequence is obtained through integration, which is considered as the relative conversion rate of electrical work. Using a cumulative summation method, while maintaining a consistent time window, the sum of the current rates for each segment is recorded as a single electrical work conversion data point. Based on this, a complete time series of electrical work conversion is constructed, reflecting the dynamic changes in the energy transfer process. In another embodiment, assuming the current rate sequence comprises 500 segments, the average electrical power conversion per segment after integration is 1.2 (unit normalized), with a maximum value of 2.7 and a minimum value of 0.4. During high-maneuverability flight, the electrical power conversion exceeds 2.0 for five consecutive segments, demonstrating the high energy consumption characteristic resulting from the motor's rapid response.

[0059] Step S325: Locate the high-load output period based on the electrical power conversion sequence, collect the propeller climb speed during the high-load output period; record the propulsion energy consumption of the propeller climb speed; In one embodiment, by setting a threshold for electrical power conversion (e.g., mean + 1.5 × standard deviation), periods of high load output are identified from the electrical power conversion sequence. Then, synchronously recorded propeller speed changes are extracted within these high load intervals, and the speed change value per unit time (i.e., climb rate) is calculated. Combined with voltage and current data, the energy consumption of the propulsion process is evaluated, and a propulsion energy consumption statistical report is generated.

[0060] In another embodiment, assuming the electrical power conversion threshold is set to 2.2, eight high-load periods were detected, with an average duration of approximately 0.6 seconds. During one of these periods, the propeller speed increased from 6500 rpm to 7900 rpm, a climb rate of 1400 rpm / s; the voltage was 11.7V, and the average current was 7.8A. The estimated propulsion energy consumption for this period was 91.3 W·s. The propulsion energy consumption values ​​for multiple high-load sections ranged from 80 to 110 W·s, which can be used to identify critical load nodes and structural design bottlenecks.

[0061] Preferably, step S33 includes the following steps: Step S331: Extract the attitude angle parameters of the flight state; In one embodiment, a six-axis inertial measurement unit (IMU) is used to acquire the aircraft's attitude data at a frequency of 100 Hz, including pitch, roll, and yaw. A quaternion filtering algorithm is used to fuse and smooth the raw gyroscope and accelerometer data, extracting a stable sequence of attitude angle parameters. Specifically for the pitch parameter, a sliding window (1.5 seconds long) is established to filter out short-term fluctuation noise, retaining only the trend change data of the aircraft over a long period. This sequence serves as the basic input for subsequent trend tracking and fluctuation identification.

[0062] Step S332: Track the pitch angle fluctuation trend based on attitude angle parameters, and decompose the pitch angle fluctuation trend into a continuous rising segment and a continuous falling segment; In one embodiment, a first-order derivative model is constructed to analyze the slope value of the pitch angle changing over time. When the slope of consecutive frames remains positive and exceeds a minimum change threshold (e.g., ≥0.3° per frame), it is considered a "continuous ascent segment"; when the slope remains negative and the absolute value exceeds the threshold, it is judged as a "continuous descent segment". The system uses a threshold filtering mechanism (minimum segment length is 20 frames) to remove misjudgments of minor fluctuations. Within each flight cycle, ascent and descent segments alternate, marked as key fluctuation segments of the flight dynamic cycle.

[0063] In another embodiment, it is assumed that among the extracted 6000 frames of pitch angles, 32 effective ascent segments and 33 descent segments were identified. The longest ascent segment was 162 frames, corresponding to a time of approximately 1.62 seconds; the shortest ascent segment was 21 frames. The maximum pitch angle increase was 28°, and the minimum increase was 4°. Some ascent segments correspond to rapid ascent and climb maneuvers of the aircraft, while the descent segments mostly correspond to dives or attitude adjustments.

[0064] Step S333: Monitor the number of power surges during the continuous rise phase. If the number of power surges exceeds the preset power surge threshold, mark it as propeller speed instability and obtain propeller instability data; use the propeller instability data to track the offset of the command input, identify blade structure loosening, and obtain blade loss data; record the speed sequence corresponding to the blade loss data. In one embodiment, for each ascent segment, the fluctuation of its internal current sequence is analyzed, and a local extremum detection method is used to statistically analyze power surge events (defined as instantaneous fluctuations above 0.8A). If the number of surges in a certain ascent segment is greater than 3, and the duration of the surges exceeds 500ms, it is determined that the propeller exhibits nonlinear response behavior and is marked as "propeller instability". The delay deviation between the flight control command and the actual current response in this segment is further compared. If the deviation is consistently greater than a set value (e.g., the delay between the command and the current response is greater than 150ms), it is further determined as "propeller structure loosening". All propeller speeds in this segment are recorded as a speed sequence.

[0065] In another embodiment, it is assumed that among the monitored rising segments, 8 segments exhibited ≥4 electrical abrupt changes, with an average abrupt change amplitude of 1.2A and a maximum of 2.4A; among these, 5 segments also showed significant control response delays (average offset of 180ms). Based on comprehensive system analysis, it was determined that there was a potential for blade structure loosening. The corresponding rotational speed sequences for these 5 segments were recorded, totaling approximately 4200 rotational speed data points, with a rotational speed range of 6200–7800 RPM.

[0066] Step S334: Monitor the number of voltage jumps during the continuous rising phase, and record the voltage jump intervals where the number of voltage jumps exceeds the voltage jump threshold; within the voltage jump interval, calculate the horizontal velocity change rate; calculate the electric driving energy corresponding to the horizontal velocity change rate. In one embodiment, a voltage jump threshold of 0.4V (the drop between adjacent frames) is set. The voltage sequence in a continuously rising segment is analyzed, and jump events in each segment are counted. If there are more than three consecutive drops, the segment is defined as a "voltage jump interval". Subsequently, the horizontal velocity change rate (in m / s) of this segment is calculated using GPS and IMU data. 2 The driving energy is estimated by combining the current and voltage per unit time. The electric driving energy is recorded as an energy consumption index for this segment and used for further energy efficiency analysis and anomaly detection.

[0067] In another embodiment, it is assumed that seven distinct voltage abrupt changes are found within the 12 rising segments, with the voltage dropping as low as 10.2V (system rated 11.1V), and the number of abrupt changes ranging from four to nine. Within these segments, the average velocity change rate is 1.8 m / s². 2 The maximum speed reached 3.2 m / s 2The calculated electric drive energy range is 18.6J to 41.3J, with an average consumption of about 27.5J per segment.

[0068] Step S335: Determine the flight speed energy consumption based on the rotational speed sequence and electric drive energy.

[0069] In one embodiment, the recorded propeller rotation speed sequence is time-aligned with the corresponding electric drive energy, and divided into several equal-length energy consumption windows (e.g., 1 second per window). The average flight speed within each window is calculated, and the electric drive energy required per unit speed is defined as "speed energy consumption". If the speed energy consumption value is consistently higher than the normal threshold for a certain period of time (e.g., exceeding the average value by 30%), it may reflect blade wear or abnormal flight load.

[0070] In another embodiment, it is assumed that the extracted flight rotation speed sequence covers a total time of 30 seconds, divided into 30 1-second windows. The calculated average flight speed ranges from 3.1 m / s to 6.7 m / s; the average electric drive energy per second ranges from 15 J to 44 J. The corresponding speed energy consumption (electrical energy divided by speed) ranges from 3.8 J / (m / s) to 9.5 J / (m / s), with 4-second windows having speed energy consumption exceeding 8 J / (m / s), which are marked by the system as abnormal energy consumption segments.

[0071] Preferably, step 335 includes: Perform sliding window processing on the speed sequence, calculate the difference between the average speed point and the speed point within each window; calculate the average of the absolute values ​​of the differences to form the average difference sequence of speeds; In one embodiment, the system collects the rotational speed data of the flight motors at a frequency of 50Hz, generating a 60-second rotational speed sequence with approximately 3000 rotational speed points. The rotational speed sequence is processed into a sliding window with a fixed window width of 100 points and a sliding step size of 20 points. For each sliding window, the average rotational speed value of all rotational speed points within the window is first calculated, and then the difference between each point and the average value is calculated. The absolute values ​​of these differences are taken and averaged to form the average rotational speed difference for that window, ultimately generating a rotational speed average difference sequence of length 150.

[0072] In another embodiment, assuming a total of 2400 rotational speed sampling points are obtained, a sliding window with a width of 80 points and a sliding step size of 10 points is used, resulting in 232 windows. Within each window, the average rotational speed is assumed to be approximately 3300 rpm, with deviations between points ranging from ±300 rpm. After processing, the resulting average difference sequence shows a minimum value of 15 rpm and a maximum value of 112 rpm, exhibiting a periodic fluctuation.

[0073] Identify the time periods where the average difference is below the stable threshold, extract the voltage and current data for those time periods, and then multiply the voltage and current data point by point and sum them up to obtain the electrical work conversion. In one embodiment, a stability threshold of 30 rpm is set. The average speed difference sequence obtained in step S1 is iterated through, and consecutive time periods below 30 rpm are filtered out to identify three stable segments: seconds 12-18, seconds 26-31, and seconds 45-51. The system synchronously collects voltage and current data during these periods (sampling frequency also 50 Hz). For each sampling point, the product of voltage and current is calculated and accumulated point-by-point within the stable segments to ultimately determine the electrical power conversion.

[0074] Extract the initial and final speeds of the electric drive energy and calculate the speed increment; divide the electrical work conversion by the speed increment to determine the energy consumption at flight speed.

[0075] In one embodiment, speed data corresponding to the aforementioned stable time periods are extracted from the rotational speed sequence. The rotational speed values ​​at the start and end of each stable segment are selected and converted into linear velocity using motor parameters. Taking stable segment one as an example, the starting speed is 5.2 m / s, the ending speed is 7.8 m / s, and the speed increment is 2.6 m / s. Combining this with the electrical work conversion of this segment (e.g., 1250 joules), the energy consumed per unit speed increase is calculated to be approximately 481 J / (m / s), which is the energy consumption index for flight speed.

[0076] In another embodiment, it is assumed that three stable acceleration processes were evaluated, with the start and end speeds for each segment being (4.0→6.5 m / s), (6.2→9.0 m / s), and (5.5→7.5 m / s), corresponding to speed increments of 2.5 m / s, 2.8 m / s, and 2.0 m / s, respectively. The corresponding electrical work is 1080 J, 1560 J, and 940 J, respectively. Thus, the speed energy consumption is calculated to be 432, 557, and 470 J / (m / s), respectively. The average energy consumption is approximately 486 J / (m / s), which can be used as a reference for flight efficiency evaluation in flight missions.

[0077] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0078] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A digital twin decision-making method for city-level low-altitude economy, characterized in that, Includes the following steps: Step S1: Obtain urban 3D geographic information; detect the 3D outlines of buildings in the urban 3D geographic information; Determine the landable area using the three-dimensional outline of the building; Step S2: Optimize the landing points of the UAV in the landable area using a pre-trained digital twin model to generate landing point data; obtain UAV delivery mission information; determine the first delivery route based on the UAV delivery mission information and landing point data; Step S3: Continuously monitor the flight status of the drone in the first delivery route; detect the drone's energy consumption based on the flight status, and if the drone's energy consumption is greater than the preset energy consumption threshold, it is judged as an abnormal flight status. Step S4: Collect wind parameters for abnormal flight conditions; based on the wind parameters, filter out the strong wind reverse flight sections of the first delivery route to obtain the second delivery route.

2. The digital twin decision-making method for city-level low-altitude economy according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain three-dimensional geographic information of the city; Step S12: Establish a raster structure, import the original point cloud of the city's 3D geographic information into the raster structure, and output the city's vertical point cloud set; construct a vertical slice sequence of the city's vertical point cloud set along the Z-axis direction; Step S13: Calculate the normal vector of the vertical slice sequence; identify the surface structure based on the vertical slice sequence; if the normal vector points to the surface structure, record it as a potential building area; Step S14: Identify the boundary points of potential building areas, connect the boundary points in the outline order to generate the three-dimensional outline of the building; Step S15: Read the 3D outline of the building and construct the projection plane; obtain the deployment site data; map the deployment site data onto the projection plane to determine the landing area.

3. The digital twin decision-making method for city-level low-altitude economy according to claim 1, characterized in that, Step S2 involves using a pre-trained digital twin model to optimize the UAV landing point in the landable area, including: A pre-trained digital twin model is used to optimize UAV landing points in landable areas; The digital twin model includes an input layer, multiple initial layers, a fusion layer, and a decision layer. The initial layer is used to extract multi-scale landing area features, while the fusion layer is used to upsample and weightedly fuse the landing area features to obtain a fused feature map. In the decision layer, the fused feature map is comprehensively scored to select the optimal landing point.

4. The digital twin decision-making method for city-level low-altitude economy according to claim 3, characterized in that, Upsampling and weighted fusion of multi-scale features include: Receive the landable region feature map from the initial layer and calculate the resolution of the landable region features; use the resolution to filter low-resolution feature maps; Each pixel of the low-resolution feature map is copied to a larger region of surrounding pixels to obtain an upsampled feature map. Assign weight coefficients to the upsampled feature map, and multiply the pixel values ​​of the upsampled feature map by the corresponding weight coefficients to obtain the fused feature map.

5. The digital twin decision-making method for city-level low-altitude economy according to claim 2, characterized in that, Step S2, which determines the first delivery route based on the drone delivery mission information and landing point data, includes: Based on the analysis of drone delivery mission information, the takeoff point coordinates and delivery mission sequence are determined. The lateral difference is obtained by subtracting the lateral coordinates of the takeoff point from the lateral coordinates of the landing point. The longitudinal difference is obtained by subtracting the longitudinal coordinates of the takeoff point from the longitudinal coordinates of the landing point. Subtract the altitude of the takeoff point from the altitude of the landing point to obtain the altitude difference; Square the lateral difference, longitudinal difference, and height difference respectively, and take the squared result; add the squared results and solve for the positive square root, which is the straight-line flight distance; An ordered flight path sequence is generated using the delivery task order and straight-line flight distance to determine the first delivery path.

6. The digital twin decision-making method for city-level low-altitude economy according to claim 5, characterized in that, The generation of an ordered flight path sequence based on delivery task order and straight-line flight distance includes: Extract the location information of the delivery task sequence and arrange them according to the delivery task sequence to construct an initial task point sequence; In the initial task point sequence, the relative flight cost between task points is calculated using the straight-line flight distance to obtain flight cost data; the flight cost data is then used to sequentially connect the initial task point sequence to generate an ordered flight path sequence to determine the first delivery path.

7. The digital twin decision-making method for city-level low-altitude economy according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Continuously monitor the flight status of the drone in the first delivery route; Step S32: Determine propulsion energy consumption based on flight status; Step S33: Determine flight speed energy consumption based on flight status; Step S34: Integrate propulsion energy consumption and flight speed energy consumption into UAV energy consumption; Step S35: If the drone's energy consumption is greater than the preset energy consumption threshold, it is judged as an abnormal flight state.

8. The digital twin decision-making method for city-level low-altitude economy according to claim 7, characterized in that, Step S32 includes the following steps: Step S321: Extract flight control data of flight status; Step S322: Extract attitude adjustment commands from flight control data and record the current, voltage, and propeller speed parameters driven by the attitude adjustment commands to form a drive response data sequence aligned with timestamps; Step S323: Set a fixed time window in the drive response data sequence, extract the current signal segment, calculate the first-order difference value of the current signal segment, and record it as a current rate sequence; Step S324: Perform sliding window integration on the current rate sequence, calculate the cumulative current change within each window, and generate an electrical work conversion sequence; Step S325: Locate the high-load output period based on the electrical power conversion sequence, collect the propeller climb speed during the high-load output period, and record the propulsion energy consumption of the propeller climb speed.

9. The digital twin decision-making method for city-level low-altitude economy according to claim 7, characterized in that, Step S33 includes the following steps: Step S331: Extract attitude angle parameters of the flight state; Step S332: Track the pitch angle fluctuation trend based on attitude angle parameters, and decompose the pitch angle fluctuation trend into a continuous rising segment and a continuous falling segment; Step S333: Monitor the number of power surges during the continuous rise phase. If the number of power surges exceeds the preset power surge threshold, mark it as propeller speed instability and obtain propeller instability data; use the propeller instability data to track the offset of the command input, identify blade structure loosening, and obtain blade loss data; record the speed sequence corresponding to the blade loss data. Step S334: Monitor the number of voltage jumps during the continuous rising phase, and record the voltage jump intervals where the number of voltage jumps exceeds the voltage jump threshold; within the voltage jump interval, calculate the horizontal velocity change rate; calculate the electric driving energy corresponding to the horizontal velocity change rate. Step S335: Determine the flight speed energy consumption based on the rotational speed sequence and electric drive energy.

10. The digital twin decision-making method for city-level low-altitude economy according to claim 9, characterized in that, Step 335 includes: Perform sliding window processing on the speed sequence, calculate the difference between the average speed point and the speed point within each window; calculate the average of the absolute values ​​of the differences to form the average difference sequence of speeds; Identify the time periods where the average difference is below the stable threshold, extract the voltage and current data for those time periods, and then multiply the voltage and current data point by point and sum them up to obtain the electrical work conversion. Extract the initial and final speeds of the electric drive energy and calculate the speed increment; divide the electrical work conversion by the speed increment to determine the energy consumption at flight speed.

Images