A low-altitude logistics route planning method, system, device, medium and product

Abstract

The application discloses a low-altitude logistics route planning method, system, device, medium and product, relates to the technical field of unmanned aerial vehicle logistics, and comprises the following steps: performing data processing on historical demand tensor data and node topological structure to obtain input data; constructing a demand prediction model based on a space-time graph neural network and in combination with a Transformer attention mechanism; inputting the input data into the demand prediction model to obtain demand tensor prediction data in multiple time periods; based on low-altitude airspace constraints, constructing a route optimization model with the target of minimizing energy consumption, and then obtaining a route planning result based on the demand tensor prediction data in multiple time periods; and adopting an end-to-end collaborative training mechanism, reversely propagating the route planning result to the demand prediction model in the form of the node topological structure, and combining the route optimization model to realize real-time updating of the demand prediction model. The application can make low-altitude logistics demand prediction directly serve the improvement of route quality in a large-scale dynamic demand scenario.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) logistics technology, and in particular to a low-altitude logistics route planning method, system, equipment, medium, and product. Background Technology

[0002] Drone logistics has important applications in urban delivery, emergency medical supplies transportation in mountainous areas, and island resupply. However, existing low-altitude logistics route planning technologies for drones (such as the two-stage method of prediction followed by optimization and dynamic programming methods) are insufficient to meet the high-efficiency operation requirements of large-scale dynamic demand scenarios. Specifically, existing low-altitude logistics route planning technologies for drones mainly suffer from the following problems: (1) The two-stage method of prediction followed by optimization has limitations.

[0003] The existing two-stage approach of forecasting first and then optimizing follows the process of "demand forecasting → route planning." It uses forecasting models such as ARIMA to predict demand, which is then input into a mixed-integer programming model to generate routes. In this type of method, the forecasting model aims to minimize the mean squared error, failing to consider the sensitivity of downstream planning to key demand points, leading to increased logistics and distribution costs.

[0004] (2) Bottlenecks in dynamic programming methods.

[0005] Existing dynamic programming methods directly output routes by introducing reinforcement learning (RL), but they do not explicitly model the demand distribution and are prone to getting trapped in local optima (such as over-concentrating high-density areas while ignoring peripheral demands).

[0006] Therefore, there is an urgent need in this field to provide a low-altitude logistics route planning method that enables low-altitude logistics demand forecasting to directly serve the improvement of route quality and is applicable to large-scale dynamic demand scenarios. Summary of the Invention

[0007] The purpose of this application is to provide a method, system, equipment, medium, and product for low-altitude logistics route planning, which can enable low-altitude logistics demand forecasting to directly serve the improvement of route quality in large-scale dynamic demand scenarios, thereby improving the accuracy of low-altitude logistics route planning.

[0008] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a low-altitude logistics route planning method, including: Acquire historical demand tensor data and node topology; the node topology is formed with historical demand points as nodes and connecting routes between historical demand points as line segments. Data processing is performed on the historical demand tensor data and the node topology to obtain input data; A demand prediction model is constructed based on a spatiotemporal graph neural network and combined with the Transformer attention mechanism. The input data is fed into the demand forecasting model to obtain demand tensor forecast data for multiple time periods. Based on low-altitude airspace constraints, a route optimization model is constructed with the goal of minimizing energy consumption. The route optimization model is used to obtain route planning results based on demand tensor prediction data from multiple time periods; An end-to-end collaborative training mechanism is adopted to backpropagate the route planning results to the demand prediction model in the form of a node topology structure. Combined with the route optimization model, the demand prediction model is updated in real time. Then, the process returns to the step of inputting the input data into the demand prediction model to obtain demand tensor prediction data for multiple time periods.

[0009] Optionally, the historical demand tensor data and the node topology are processed to obtain input data, including... The historical demand tensor data is subjected to data standardization processing to obtain standardized data; the historical demand tensor data includes: the time, location, and load data of historical low-altitude logistics demand points, as well as environmental information associated with the historical low-altitude logistics demand points. Based on the geographical distance between the required points in the node topology, a geographical adjacency matrix is ​​constructed; The standardized data and the geographic adjacency matrix are used as the input data.

[0010] Optionally, the demand prediction model includes a spatial convolution module, a temporal attention module, a fusion gating module, and a demand decoding layer; The spatial convolution module is used to acquire the spatial features of the input data; the temporal attention module is used to acquire the temporal features of the input data; the fusion gating module is used to perform spatiotemporal feature fusion on the spatial features and the temporal features to obtain spatiotemporal features; the demand decoding layer is used to map the spatiotemporal features to the demand prediction values ​​of all demand points in each time step within a future set time period to obtain multi-time period demand tensor prediction data.

[0011] Optionally, the mapping relationship used by the demand decoding layer is represented as follows: ; In the formula, Spatiotemporal characteristics, For learnable weight matrix, For bias terms, Let be the predicted demand values ​​for all demand points in the t-th prediction time step.

[0012] Optionally, an end-to-end collaborative training mechanism is adopted to backpropagate the route planning results to the demand prediction model in the form of a node topology, and to combine the route optimization model to achieve real-time updates of the demand prediction model, including: The route planning results are backpropagated to the demand forecasting model in the form of a node topology to obtain the initial demand forecast value. Using the aforementioned route optimization model, the first objective function value is obtained based on the initial demand prediction value; Obtain actual flight route data; Using the aforementioned route optimization model, the second objective function value is obtained based on the actual route data; Determine whether the difference between the first objective function value and the second objective function value meets the accuracy requirement, and obtain the determination result; When the judgment result is yes, the prediction parameters of the demand prediction model are not adjusted. When the judgment result is negative, the difference between the first objective function value and the second objective function value is used as the loss function of the demand forecasting model, and backpropagation is performed on the demand forecasting model to adjust the prediction parameters of the demand forecasting model and realize the real-time update of the demand forecasting model.

[0013] Optionally, the low-altitude airspace constraints include: path continuity constraints, demand satisfaction constraints, and range constraints.

[0014] Secondly, this application provides a low-altitude logistics route planning system, comprising: The dynamic demand forecasting module is used to acquire historical demand tensor data and node topology, process the historical demand tensor data and node topology to obtain input data, construct a demand forecasting model based on a spatiotemporal graph neural network and combined with the Transformer attention mechanism, and input the input data into the demand forecasting model to obtain demand tensor forecasting data for multiple time periods; the node topology is formed by using historical demand points as nodes and connecting routes between historical demand points as line segments. The route planning and optimization module is used to construct a route optimization model based on low-altitude airspace constraints with the goal of minimizing energy consumption, and to obtain route planning results based on multi-time period demand tensor prediction data using the route optimization model. The end-to-end collaborative training module is used to backpropagate the route planning results to the demand prediction model in the form of a node topology structure using an end-to-end collaborative training mechanism. It combines the route optimization model to realize the real-time update of the demand prediction model and returns to the step of inputting the input data into the demand prediction model in the demand dynamic prediction module to obtain demand tensor prediction data for multiple time periods.

[0015] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the low-altitude logistics route planning method provided above.

[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the low-altitude logistics route planning method provided above.

[0017] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the low-altitude logistics route planning method described above.

[0018] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, system, equipment, medium, and product for low-altitude logistics route planning. In large-scale dynamic demand scenarios, route planning results are obtained by using a route optimization model based on demand tensor prediction data obtained from a demand forecasting model, enabling low-altitude logistics demand forecasting to directly improve route quality. Furthermore, by employing an end-to-end collaborative training mechanism and combining it with the route optimization model, the demand forecasting model can be calibrated in real time during the low-altitude logistics route planning process, significantly reducing systematic errors and improving the accuracy of low-altitude logistics route planning. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a low-altitude logistics route planning method provided in an embodiment of this application; Figure 2 A schematic diagram of the implementation architecture of a low-altitude logistics route planning method provided in an embodiment of this application; Figure 3 This is a schematic diagram of the demand prediction model training process provided in one embodiment of this application; Figure 4 A functional module diagram of a low-altitude logistics route planning system provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] In one exemplary embodiment, this application provides a low-altitude logistics route planning method. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is described using a server as an example. Figure 1 As shown, the method includes: Step 100: Obtain historical demand tensor data and node topology. The node topology is formed by using historical demand points as nodes and connecting routes between historical demand points as line segments.

[0024] Step 101: Process the historical demand tensor data and node topology to obtain the input data.

[0025] Step 102: Construct a demand prediction model based on a spatiotemporal graph neural network and combined with the Transformer attention mechanism.

[0026] Step 103: Input the input data into the demand forecasting model to obtain multi-period demand tensor forecast data. The purpose of this step is mainly to predict the distribution and intensity of demand points in the target area across multiple time periods based on historical order data and environmental factors (such as weather, holidays, airspace control, etc.) in the historical demand tensor data.

[0027] Step 104: Based on low-altitude airspace constraints, construct a route optimization model with the goal of minimizing energy consumption.

[0028] Step 105: Using the route optimization model, the route planning results are obtained based on the demand tensor prediction data of multiple time periods.

[0029] Step 106: Employing an end-to-end collaborative training mechanism, the route planning results are backpropagated to the demand forecasting model in the form of a node topology. This, combined with the route optimization model, enables real-time updates to the demand forecasting model, and then returns to step 103. This step utilizes a differentiable optimization layer to backpropagate the route planning results to the demand forecasting model, using the decision-making effect to guide the updating of forecast parameters, thereby avoiding the misleading influence of forecast errors on path planning found in traditional two-stage methods.

[0030] By implementing steps 100 to 106 above, this application, through the adoption of an end-to-end integrated prediction-planning link, can make better decisions even in the presence of noisy data or inaccurate demand forecasts, thereby enhancing the robustness and accuracy of low-altitude logistics route planning.

[0031] In an exemplary embodiment of this application, in order to eliminate the impact of historical demand tensor data and node topology, the data processing method of step 101 provided above can be: Step 1): Standardize the historical demand tensor data to obtain standardized data. For example, use Z-score normalization with a mean of 0 and a standard deviation of 1. The historical demand tensor data includes the time, location, and load data of historical low-altitude logistics demand points, as well as environmental information associated with these points (such as weather, airspace control, etc.).

[0032] Step 2): Construct a geographic adjacency matrix based on the geographic distances between demand points in the node topology. Therefore, in practical applications, the geographic adjacency matrix is ​​a weighted adjacency matrix, primarily calculated based on the geographic coordinates of the demand points and road network data. The geographic distances between the demand points are used to determine the adjacency matrix. The constructed geographic adjacency matrix is ​​represented as follows: .

[0033] In the formula, The width of the Gaussian kernel. The value is 0.1. The neighborhood threshold, The value is 0.5. Indicates based on demand points i and demand points j Geographical distance between The constructed geographical adjacency matrix.

[0034] Based on the processing in steps 101-1 and 101-2 above, the standardized data and the geographic adjacency matrix are finally used as input data.

[0035] In one exemplary embodiment of this application, such as Figure 2 As shown, the demand prediction model constructed in step 102 of this application may include: a spatial convolution module, a temporal attention module, a fusion gating module, and a demand decoding layer, in order to capture the spatial correlation and temporal dynamics of demand points, and to output the demand amount of demand points in the next T hours (i.e., the future node demand prediction result).

[0036] In this embodiment, the input data for the demand forecasting model is assumed to include: 1) standardized historical demand tensor data. ;2) A geographic adjacency matrix representing the spatial relationships between demand points (obtained through distance calculation between demand points). .in, It indicates the number of historical time steps (such as the past 24 hours or a longer period of time). The total number of demand points, The feature dimensions (corresponding to load data, weather, holiday tags, etc. at low-altitude logistics demand points) are as follows. The data represents the number field. The output data of the demand forecasting model is multi-period demand tensor forecast data. That is, each time step This corresponds to the predicted demand values ​​for all demand points. Based on this, the specific data processing methods for each module included in the demand forecasting model can be described as follows: (1) The spatial convolution module is used to obtain the spatial features of the input data.

[0037] The spatial convolution module uses Chebyshev polynomial unrolled graph convolution (ChebNet), which includes: .

[0038] In the formula, For the normalized Laplace matrix, , It is the identity matrix. Let be the order of the Chebyshev polynomial. For learnable parameters, For the first Hidden state, For the first Hidden layer state.

[0039] (2) The time attention module is used to obtain the time characteristics of the input data.

[0040] The time position encoding added in the time attention module is as follows : , .

[0041] In the formula, Encode the sinusoidal component for time position. Encode the cosine component for time position. For demand point index, i.e. demand point . Indexed by feature dimensions, For time steps.

[0042] An attention mechanism is introduced into the time attention module, enabling the demand forecasting model to focus on key time features during the learning phase, thereby enhancing its ability to effectively grasp key information. Regarding the application of the attention mechanism, self-attention... The formula can be expressed as: .

[0043] In the formula, These represent the query, key, and value, respectively, and softmax() is the activation function. It is the dimension of the key. This indicates transpose. The calculation formula is as follows: .

[0044] In the formula, For input hidden layer, , and All are learning matrices.

[0045] (3) The fusion gating module is used to perform spatiotemporal feature fusion of spatial features and temporal features to obtain spatiotemporal features.

[0046] The fusion gating module can use gated recurrent units (GRUs) to combine spatial and temporal features for spatiotemporal feature fusion, as shown in the following formula: .

[0047] .

[0048] .

[0049] .

[0050] In the formula, The spatial features output by the spatial convolution module. ; The temporal features output by the temporal attention module. ; To update the gate (control how much historical information is retained). ; To reset the gate (control how much historical information is forgotten). ; This is a candidate hidden state (a candidate for new information). ; To obtain the spatiotemporal features through fusion, . , and These are the weight matrices for the update gate, reset gate, and candidate state, respectively.

[0051] (4) The demand decoding layer is used to decode spatiotemporal features ( ) mapped to a future time period (e.g. Forecast values ​​of demand for all demand points at each time step within (hours) This yields demand tensor prediction data for multiple time periods. The mapping relationship used in the demand decoding layer is represented as follows: .

[0052] In the formula, For learnable weight matrix, . For bias terms, . For the first t Demand forecasts for all demand points within each time step. .

[0053] In another exemplary embodiment of this application, in order to achieve accurate planning of low-altitude logistics routes, the low-altitude airspace constraints used in step 104 above include: (1) Path continuity constraint.

[0054] Flow balancing constraints at intermediate demand points, for any intermediate demand point The inflow and outflow of drones are balanced, and we have: .

[0055] In the formula, Indicates drone Not at height From the demand point Fly to the point of demand , Indicates drone At height From the demand point Fly to the point of demand , Indicates a collection of drones. This indicates a set of altitudes (where drones are restricted to flying at certain altitudes). Indicates drone Not at height From the demand point Fly to the point of demand , Indicates drone At height From the demand point Fly to the point of demand .

[0056] Origin and destination constraints: For any UAV, it must start from a starting point and return to a destination, which means: . For drones The starting point of the flight For drones At height From the starting point Fly to the point of demand .

[0057] . For drones The end of the flight, For drones At height From the demand point Fly to the finish line .

[0058] (2) Demand satisfaction constraints.

[0059] The needs of each requirement point should be met: .

[0060] In the formula, Indicates drone service node The demand, Indicate demand points The actual demand for goods.

[0061] The drone's payload needs to be dynamically updated after it reaches the service demand point, including: .

[0062] .

[0063] In the formula, and All are integer variables, representing drones. Reaching the demand point and demand points Load at that time. Indicate demand points The actual demand for goods. It represents a very large constant.

[0064] The payload of a drone at any node cannot exceed its carrying capacity, including: .

[0065] In the formula, This indicates the drone's payload capacity, i.e., the drone's maximum load.

[0066] (3) Driving range constraints.

[0067] .

[0068] In the formula, Indicates drone At height From the demand point Flight to demand Required distance. This indicates the maximum range of the drone.

[0069] Based on the constraints given above, the route optimization model constructed in step 104 of this application can be expressed as: .

[0070] In the formula, Indicates energy consumption. This represents the flight payload energy consumption coefficient, which is set to 0.5. This represents the flight time energy consumption coefficient, which is set to 0.8. Indicates the altitude reached by the drone upon takeoff. The time required. Indicates the drone from altitude Time required for landing. This indicates the speed at which the drone flies horizontally at a constant speed.

[0071] In an exemplary embodiment of this application, in order to achieve joint optimization of demand forecasting and route planning through an integrated prediction-optimization framework, particularly suitable for scenarios with dynamically changing demand in low-altitude logistics delivery, step 106 of this application employs an end-to-end collaborative training mechanism. The route planning results are backpropagated to the demand forecasting model in the form of a node topology structure, and combined with the route optimization model to achieve real-time updates of the demand forecasting model, including: Step 1: Backpropagate the route planning results to the demand forecasting model in the form of a node topology to obtain the initial demand forecast.

[0072] Step 2: Using the route optimization model, the first objective function value is obtained based on the initial demand forecast.

[0073] Step 3: Obtain actual flight route data.

[0074] Step 4: Using a route optimization model, obtain the second objective function value based on actual route data.

[0075] Step 5: Determine whether the difference between the first and second objective function values ​​meets the accuracy requirements, and obtain the determination result. If the determination result is yes, the prediction parameters of the demand forecasting model are not adjusted. If the determination result is no, the difference between the first and second objective function values ​​is used as the loss function of the demand forecasting model, and backpropagation is performed on the demand forecasting model to adjust the prediction parameters of the demand forecasting model, thereby achieving real-time updates of the demand forecasting model.

[0076] Based on the descriptions of steps 1-5 above, in practical applications, the end-to-end prediction-optimization collaborative training process is as follows: Figure 3 As shown, it includes: (1) Prepare historical logistics demand, environmental datasets, and low-altitude logistics delivery route datasets. The historical logistics demand and environmental datasets are obtained through actual data collection. The low-altitude logistics delivery route dataset is obtained by inputting the actual demand tensor data into the aforementioned demand prediction model and solving it using a solver. The prepared historical logistics demand, environmental datasets, and low-altitude logistics delivery route datasets are divided into training and testing sets, for example, with the training set accounting for 70% and the testing set accounting for 30%.

[0077] (2) Using the data in the training set, perform demand tensor prediction based on the demand prediction model constructed in step 102 above. Input the predicted demand tensor prediction data into the route optimization model established in step 104 above, and call the solver to solve the problem, generate the initial route set and calculate the optimal solution objective function value z1 (i.e. the first objective function value).

[0078] .

[0079] In the formula, The objective function is denoted as , which corresponds to the route optimization model. For demand points Forecasted demand for goods, For demand points The actual demand for goods.

[0080] (3) Input the actual route data into the route optimization model established in step 104 to calculate the actual optimal solution objective function value z2 (i.e. the second objective function value).

[0081] .

[0082] (4) Determine whether the deviation between the two objective function values ​​meets the accuracy requirements. If not, use this deviation as the loss function of the demand forecasting model and perform backpropagation to adjust the forecasting parameters. The loss function is expressed as: .

[0083] In the formula, This represents the loss function value.

[0084] (5) If the condition is met, training ends. After completing parameter training based on the training set, the test set data is used to calculate and evaluate the quality of the predicted solution in the demand prediction model.

[0085] In summary, compared with traditional methods, this application has at least the following advantages: 1) This application can improve the accuracy of low-altitude logistics route planning. Traditional methods usually adopt a sequential process of prediction followed by optimization. Demand prediction deviations can easily mislead UAV route decisions (such as missing capacity in high-demand areas). However, the end-to-end training adopted in this application directly calibrates the prediction model through decision feedback, which significantly reduces systematic errors.

[0086] (2) Enhance the robustness of the planning scheme. This application adopts an end-to-end forecasting-planning integrated link, which can still make better decisions even in the presence of noisy data or inaccurate demand forecasts.

[0087] Based on the same inventive concept, this application also provides a low-altitude logistics route planning system for implementing the low-altitude logistics route planning method described above. The solution provided by this system is similar to the solution described in the above method; therefore, the specific limitations of one or more low-altitude logistics route planning system embodiments provided below can be found in the limitations of the low-altitude logistics route planning method described above, and will not be repeated here.

[0088] In one exemplary embodiment, such as Figure 4 As shown, a low-altitude logistics route planning system is provided, including: a demand dynamic forecasting module, a route planning optimization module, and an end-to-end collaborative training module.

[0089] The dynamic demand forecasting module acquires historical demand tensor data and node topology; processes this data to obtain input data; constructs a demand forecasting model based on a spatiotemporal graph neural network and the Transformer attention mechanism; and inputs the input data into the demand forecasting model to obtain multi-period demand tensor forecast data. The node topology is formed by using historical demand points as nodes and connecting routes between these points as line segments.

[0090] The route planning and optimization module is used to construct a route optimization model based on low-altitude airspace constraints with the goal of minimizing energy consumption, and to obtain route planning results based on demand tensor prediction data from multiple time periods using the route optimization model.

[0091] The end-to-end collaborative training module is used to backpropagate the route planning results to the demand forecasting model in the form of a node topology structure using an end-to-end collaborative training mechanism. Combined with the route optimization model, it realizes the real-time update of the demand forecasting model and returns to the dynamic demand forecasting module to input the input data into the demand forecasting model to obtain demand tensor forecasting data for multiple time periods.

[0092] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 5 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores low-altitude logistics route planning data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a low-altitude logistics route planning method.

[0093] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0094] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0095] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0096] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0098] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (RRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0099] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc.

[0100] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0101] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A low-altitude logistics route planning method, characterized in that, include: Obtain historical demand tensor data and node topology; The node topology is formed with historical demand points as nodes and connecting routes between historical demand points as line segments. Data processing is performed on the historical demand tensor data and the node topology to obtain input data, including: standardizing the historical demand tensor data to obtain standardized data; the historical demand tensor data includes: time, location, and load data of historical low-altitude logistics demand points, as well as environmental information associated with the historical low-altitude logistics demand points; constructing a geographic adjacency matrix based on the geographic distance between demand points in the node topology; and using the standardized data and the geographic adjacency matrix as the input data. A demand prediction model is constructed based on a spatiotemporal graph neural network and combined with the Transformer attention mechanism. The input data is fed into the demand forecasting model to obtain demand tensor forecast data for multiple time periods. Based on low-altitude airspace constraints, a route optimization model is constructed with the goal of minimizing energy consumption. The route optimization model is used to obtain route planning results based on demand tensor prediction data from multiple time periods; An end-to-end collaborative training mechanism is adopted to backpropagate the route planning results to the demand prediction model in the form of a node topology structure. The route optimization model is combined to realize the real-time update of the demand prediction model, and the process returns to the step of inputting the input data into the demand prediction model to obtain demand tensor prediction data for multiple time periods. The method employs an end-to-end collaborative training mechanism, backpropagating the route planning results to the demand prediction model in the form of a node topology structure. This, combined with the route optimization model, enables real-time updates to the demand prediction model, including: The route planning results are backpropagated to the demand forecasting model in the form of a node topology to obtain the initial demand forecast value. Using the aforementioned route optimization model, the first objective function value is obtained based on the initial demand prediction value; Obtain actual flight route data; Using the aforementioned route optimization model, the second objective function value is obtained based on the actual route data; Determine whether the difference between the first objective function value and the second objective function value meets the accuracy requirement, and obtain the determination result; When the judgment result is yes, the prediction parameters of the demand prediction model are not adjusted. When the judgment result is negative, the difference between the first objective function value and the second objective function value is used as the loss function of the demand forecasting model, and backpropagation is performed on the demand forecasting model to adjust the prediction parameters of the demand forecasting model and realize the real-time update of the demand forecasting model.

2. The low-altitude logistics route planning method according to claim 1, characterized in that, The demand prediction model includes a spatial convolution module, a temporal attention module, a fusion gating module, and a demand decoding layer; The spatial convolution module is used to acquire the spatial features of the input data; the temporal attention module is used to acquire the temporal features of the input data; the fusion gating module is used to perform spatiotemporal feature fusion on the spatial features and the temporal features to obtain spatiotemporal features. The demand decoding layer is used to map the spatiotemporal features to the predicted demand values ​​of all demand points in each time step within a future set time period, thereby obtaining multi-time period demand tensor prediction data.

3. The low-altitude logistics route planning method according to claim 2, characterized in that, The mapping relationship used by the demand decoding layer is represented as follows: ； In the formula, Spatiotemporal characteristics, For learnable weight matrix, For bias terms, Let be the predicted demand values ​​for all demand points in the t-th prediction time step.

4. The low-altitude logistics route planning method according to claim 1, characterized in that, The low-altitude airspace constraints include: path continuity constraints, demand satisfaction constraints, and range constraints.

5. A low-altitude logistics route planning system, characterized in that, Used to implement the low-altitude logistics route planning method as described in any one of claims 1-4; The low-altitude logistics route planning system includes: The dynamic demand forecasting module is used to acquire historical demand tensor data and node topology, process the historical demand tensor data and node topology to obtain input data, construct a demand forecasting model based on a spatiotemporal graph neural network and combined with the Transformer attention mechanism, and input the input data into the demand forecasting model to obtain demand tensor forecasting data for multiple time periods; the node topology is formed by using historical demand points as nodes and connecting routes between historical demand points as line segments. The route planning and optimization module is used to construct a route optimization model based on low-altitude airspace constraints with the goal of minimizing energy consumption, and to obtain route planning results based on multi-time period demand tensor prediction data using the route optimization model. The end-to-end collaborative training module is used to backpropagate the route planning results to the demand prediction model in the form of a node topology structure using an end-to-end collaborative training mechanism. It combines the route optimization model to realize the real-time update of the demand prediction model and returns to the step of inputting the input data into the demand prediction model in the demand dynamic prediction module to obtain demand tensor prediction data for multiple time periods.

6. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the low-altitude logistics route planning method according to any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the low-altitude logistics route planning method as described in any one of claims 1-4.

8. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the low-altitude logistics route planning method as described in any one of claims 1-4.

Images