Training device and method and application device of low-altitude economical UAV dispatching model

Abstract

The present application provides a training device, method and application device for a low-altitude economical UAV scheduling model. The training device for the low-altitude economical UAV scheduling model includes: a preprocessing module, which is used to preprocess multiple groups of initial input quantities in the training set based on a fuzzy decision tree to obtain multiple groups of preprocessed input quantities; a feature extraction module, which is used to extract features of the multiple groups of preprocessed input quantities based on a deep belief network to extract multiple groups of feature quantities; a prediction module, which is used to fuse the first group of feature quantities based on an optimal power flow network, and output the route path of the target UAV corresponding to the first group of feature quantities, the first group of feature quantities is any group of feature quantities in the multiple groups of feature quantities, the multiple groups of feature quantities correspond to multiple route paths of the target UAV, and the UAV scheduling model is trained based on the multiple route paths and a predetermined objective function.

Description

Technical Field

[0001] The present application relates to the field of low-altitude economic technology, and in particular to a training device and method and an application device for a low-altitude economic UAV scheduling model. Background Art

[0002] The emerging low-altitude economic operation management is an economic method based on an open, large-scale, and wide-range aerial three-dimensional transportation system. It is necessary to plan the optimal path based on environmental characteristics, route support, multi-aircraft conflicts, emergency support, and minimum time constraints to achieve the goals of improving the level of safety assurance, optimizing flight paths, and reducing the consumption of manpower, material resources, and drone resources. However, there are uncertain factors such as weather changes, airspace use conflicts, and differences in drone performance in low-altitude airspace. Aerial three-dimensional transportation involves drones, ground facilities, and support personnel from different companies. Therefore, multi-source, heterogeneous, asynchronous, and uncertain fuzzy information has become the main source of information for the scheduling of low-altitude economic systems. The traditional management method based on precise data planning and closed environment scheduling cannot support the safe and efficient operation needs of low-altitude economy. Summary of the invention

[0003] The present application provides a training device and method and an application device for a low-altitude economic UAV dispatching model, which helps to solve the problem that the current precise data planning and closed environment dispatching management methods cannot support the low-altitude economic operation needs of fuzzy information. The following introduces various aspects involved in this application.

[0004] In the first aspect, the present application provides a training device for a low-altitude economical UAV scheduling model, comprising: a preprocessing module, which is used to preprocess multiple groups of initial input quantities in a training set based on a fuzzy decision tree to obtain multiple groups of preprocessed input quantities, and any group of the initial input quantities includes multiple ground support resource parameters and multiple airspace scheduling resource parameters within a preset circular range; a feature extraction module, which is used to perform feature extraction on the multiple groups of preprocessed input quantities based on a deep belief network to extract multiple groups of feature quantities; a prediction module, which is used to fuse the first group of feature quantities based on an optimal power flow network, and output the route path of the target UAV corresponding to the first group of feature quantities, wherein the first group of feature quantities is any group of feature quantities among the multiple groups of feature quantities, and the multiple groups of feature quantities correspond to multiple route paths of the target UAV, and the UAV scheduling model is trained based on the multiple route paths and a predetermined objective function.

[0005] In the second aspect, the present application provides an application device of a low-altitude economical UAV scheduling model, comprising: an acquisition module, used to obtain a first group of initial input quantities, the first group of initial input quantities including a plurality of ground support resource parameters and a plurality of airspace scheduling resource parameters within a preset circular range; a model processing module, used to input the first group of initial input quantities into a pre-trained low-altitude economical UAV scheduling model for processing, and obtain a target route path corresponding to the target UAV in a predetermined application scenario within a predetermined time; wherein the low-altitude economical UAV scheduling model is a network model trained by using the training device described in the first aspect, and the predetermined time is associated with the radius size of the preset circle.

[0006] In the third aspect, the present application provides a training method for a low-altitude economical UAV scheduling model, comprising: preprocessing multiple groups of initial input quantities in a training set based on a fuzzy decision tree to obtain multiple groups of preprocessed input quantities, any group of the initial input quantities including multiple ground support resources and multiple airspace scheduling resource data within a preset circular range; performing feature extraction on the multiple groups of preprocessed input quantities based on a deep belief network to extract multiple groups of feature quantities; performing fusion processing on the first group of feature quantities based on an optimal power flow network, and outputting the route path of the target UAV corresponding to the first group of feature quantities, the first group of feature quantities being any group of feature quantities among the multiple groups of feature quantities, and the multiple groups of feature quantities corresponding to multiple route paths of the target UAV, and training the UAV scheduling model based on the multiple route paths and a predetermined objective function.

[0007] In a fourth aspect, the present application provides an application method of a low-altitude economical UAV scheduling model, comprising: obtaining a first set of initial input quantities, the first set of initial input quantities including a plurality of ground support resource parameters and a plurality of airspace scheduling resource parameters within a preset circular range; inputting the first set of initial input quantities into a pre-trained low-altitude economical UAV scheduling model for processing, and obtaining a target route path corresponding to the target UAV in a predetermined application scenario within a predetermined time; wherein the low-altitude economical UAV scheduling model is a network model trained using the training method described in the third aspect, and the predetermined time is associated with the radius size of the preset circle.

[0008] In a fifth aspect, the present application provides an electronic device, comprising: a memory for storing code; and a processor connected to the memory, for executing the code stored in the memory, so that the electronic device executes the method described in the third aspect or the fourth aspect.

[0009] In a sixth aspect, the present application provides a non-volatile computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can implement the method described in the third aspect or the fourth aspect.

[0010] In a seventh aspect, the present application provides a computer program product, which, when executed on an electronic device, enables the electronic device to implement the method described in the third aspect or the fourth aspect.

[0011] In the embodiment of the present application, a fuzzy decision tree is used to preprocess the data, and a deep belief network is constructed to extract features from the preprocessed data. The feature quantity extracted by the deep belief network is used as the input of the optimal power flow network to optimize the target route path of the target drone. Through fuzzy decision trees and fuzzy logic, it is possible to reasonably handle the uncertainty factors in the scheduling process, handle the ambiguity in the constraints and optimization targets of the low-altitude economic scheduling system, so that the system can make more reasonable predictions, route planning and scheduling when facing uncertainty and ambiguity, and improve the robustness of large-scale system scheduling. Through the scheduling system based on the optimal power flow network, under the premise of meeting the system constraints, it is possible to optimize specific goals such as minimizing idle resources or ensuring the highest safety in low-altitude airspace for a certain route. The embodiment of the present application helps to adapt to the low-altitude economy of multi-source, heterogeneous, asynchronous, and uncertain fuzzy information sources, find the optimal solution that meets the real-time status of the drone and the route stage path, avoid the idleness of ground resources in the low-altitude economy, and achieve the purpose of reducing airspace route conflicts and improving overall operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0013] Figure 1 It is a schematic diagram of a training device for a low-altitude economical UAV scheduling model provided in an embodiment of the present application.

[0014] Figure 2 It is a schematic diagram of fuzzy classification and fuzzy weighting based on a fuzzy decision tree provided in an embodiment of the present application.

[0015] Figure 3 It is a schematic diagram of an application device of a low-altitude economical UAV scheduling model provided in an embodiment of the present application.

[0016] Figure 4 It is a flowchart of a training method for a low-altitude economical UAV scheduling model provided in an embodiment of the present application.

[0017] Figure 5It is a flow chart of the application method of the low-altitude economical UAV scheduling model provided in the embodiment of the present application.

[0018] Figure 6 It is a schematic diagram of a component unit / partial component unit of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The same or similar reference numerals are used in the drawings to represent the same or similar modules. It should be understood that the drawings are only schematic, and the protection scope of the present application is not limited thereto.

[0020] The low-altitude economy with drones at its core has evolved from a niche track to a new outlet for competition in various places, and has developed into a national strategic emerging industry. The research on drone technology has evolved from single drone applications, multiple drone applications to large-scale and wide-range drone applications, and has established a new low-altitude economic business model characterized by safe, automated, and cost-effective passenger and cargo aerial three-dimensional transportation.

[0021] The main constraints of low-altitude economy are: 1) Climate and restricted areas lead to safety level adjustments. The higher the safety level, the lower the route interest rate (reducing the use of drones). 2) County-level airspace is a continuous and uninterrupted three-dimensional space. The lower the safety level, the more layers (generally 15 meters per layer, 4 layers), the more drones can fly, and it is relatively slow. The higher the safety level, the fewer layers (maybe 60 meters per layer, 1 layer), the fewer drones can fly, and it is relatively slow. In other words, after the safety level of the route is determined, the low-altitude economy calculates the maximum passing efficiency (more overall, faster single units).

[0022] For example, from point A to point B, it is assumed that there are 6 airspace routes, namely A1, A2, A3, B1, B2 and B3, forming route AB. The safety level of route A1 is very high. In the airspace of 60-120 meters, it is assumed that it can only be divided into 1 layer, so you have to fly slowly and less. The safety level of route A2 is high. In the airspace of 60-120 meters, it is assumed that it can only be divided into 2 layers, so you can fly a little faster and a little more. By analogy, the safety level of route B3 is very low. In the airspace of 60-120 meters, it is assumed that it can only be divided into 4 layers, so you can fly very fast and a lot.

[0023] Currently, drone operating companies usually plan the optimal path and ground support on their own based on precise information such as planned fixed routes and regular aerial intelligence, forming a closed or semi-closed working environment.

[0024] The emerging low-altitude economic operation management is an economic method based on an open (operated by different enterprises), large-scale (≥1,000 drones), and large-scale (county-level airspace or above) three-dimensional air transportation system. It is necessary to plan, organize, coordinate and control the procurement, storage, distribution and use of environmental characteristics, planned routes, signal channels, sensing facilities, take-off and landing and relay stations, support resources, and support teams. It is necessary to plan the optimal path based on constraints such as environmental characteristics, route support, multi-aircraft conflicts, emergency support, and minimum time to achieve the goal of improving the level of safety and security, optimizing flight paths, and reducing the consumption of manpower, material resources, and drone resources. However, there are uncertain factors such as weather changes, airspace use conflicts, and differences in drone performance in low-altitude airspace. Three-dimensional air transportation involves drones, ground facilities, and support personnel from different companies.

[0025] In summary, the speed of aerial three-dimensional traffic in the low-altitude economy varies more than that of ground traffic, and the constraints are more difficult. The time and space of drones relative to the ground and drones relative to drones change dramatically, causing dynamic data such as environment, intelligence, orders, and guarantees to change from definite input to fuzzy input. Therefore, multi-source, heterogeneous, asynchronous, and uncertain fuzzy information has become the main source of information for the scheduling of the low-altitude economic system. The traditional management method based on precise data planning and closed environment scheduling cannot support the safe and efficient operation needs of the low-altitude economy.

[0026] Therefore, it is necessary to design a technical solution for a low-altitude economical UAV dispatching system that is adaptable to multi-source, heterogeneous, and uncertain fuzzy information input sources.

[0027] Based on this, the present application embodiment proposes a training device for a low-altitude economical drone scheduling model. The drone scheduling model is a deep learning model, such as Figure 1 As shown, the training device 100 for the low-altitude economical UAV scheduling model may include a preprocessing module 110 , a feature extraction module 120 , and a dual-channel prediction module 130 .

[0028] The preprocessing module 110 is used to preprocess multiple groups of initial input quantities in the training set based on a fuzzy decision tree to obtain multiple groups of preprocessed input quantities. Any group of initial input quantities includes multiple ground support resource parameters and multiple airspace scheduling resource parameters within a preset circular range.

[0029] The preprocessing module 110 can use fuzzy logic to preprocess multiple groups of initial input quantities, including processing missing values, outliers, and feature scaling to ensure data quality. Through fuzzy logic, uncertain factors in the scheduling process can be reasonably handled to improve the robustness of large-scale system scheduling.

[0030] In the open (operated by different companies), large-scale (≥1,000 drones), and wide-range (county-level airspace or above) low-altitude economic dispatch, it is necessary to solve the problem in an extremely short time (such as within 1 minute) and make judgments and decisions on many uncertain factors (ground support resources, air dispatch resources, security resources), such as weather changes, airspace use conflicts, drone performance differences, airspace perception information, etc., when they cannot be transmitted back in real time.

[0031] The radius of the preset circle is positively correlated with the rated flight speed of the drone in the low-altitude economy. The higher the rated flight speed of the drone, the larger the radius of the preset circle. If data is sent back too frequently, the communication pressure will be high and the system may not have time to process it; if the time interval between data returns is large, the timeliness of the response will be affected. Taking the operation of urban drones as an example, the speed of common multi-rotor drones is generally 500 meters / min. In order to ensure the timeliness of information feedback, data can be sent back once / min or twice / min. The radius of the preset circle can be 500 meters or 600 meters.

[0032] It is necessary to regularly allocate support resources for the low-altitude economy, regularly adjust scheduling resources, and regularly synchronize safety assurance resources. In some implementations, multiple ground support resource parameters within a preset circular range may at least include: parameters of dynamic data such as micro-meteorological station parameters (meteorological data), spectrum sensing parameters (spectrum data), optoelectronic sensing parameters (video data), millimeter wave radar parameters (azimuth, speed data) within a preset circular range, and communication base station parameters (communication capabilities, such as 4 / 5G), Beidou differential station parameters (positioning capabilities), take-off and landing site parameters (cabin specifications and quantity), emergency rescue site parameters (types of equipment and facilities for rescue) and other static data, that is, dynamic data of 4 types of IoT perception and static data parameters of 4 types of site capabilities. For example, the data of multiple ground support resource parameters of relevant sites within a range of 500 meters are calculated once a minute and allocated to the sites of the drone route path according to the results of the fuzzy decision tree. Multiple airspace scheduling resource parameters may at least include: route plan (time, longitude and latitude, altitude, dynamic data), route capacity (number of drone flights within the route, dynamic data), and some or all parameters in restricted areas. Based on the results of the fuzzy decision tree, the calculation is sent to the drone every half a minute, and the drone adjusts the latest path or executes the existing route.

[0033] In some implementations, any set of initial inputs may also include multiple security resource parameters. Multiple security resource parameters may include at least route status parameters within a preset circular range (such as 500 meters), such as the number, specifications, load, purpose and other dynamic data of drones. The dynamic data of the route status within a preset circular range (such as 500 meters) is synchronized to the emergency support unit (social organization, personnel are dynamic data) of the route grid (geographic unit, static data). When a drone fails, the grid where the route is located will execute the corresponding protection plan, report it step by step according to the grid-village / community-town-district-county-city, and implement the corresponding protection plan.

[0034] The output results in the fuzzy decision tree mainly have two forms: discrete decision tree and continuous decision tree. Discrete decision tree mainly processes discrete independent variables and dependent variables, while continuous decision tree can process continuous independent variables and dependent variables. The output result form of the embodiment of the present application is mainly a continuous decision tree, which can meet the needs of continuous independent variables in the low-altitude economy.

[0035] In a fuzzy decision tree, fuzzy partitioning can be used to deal with uncertainty and ambiguity in data. There are currently three main fuzzy partitioning methods, namely grid partitioning, tree partitioning, and scattered partitioning. Among them, grid partitioning is used to divide the data space into regular grids, which is suitable for problems that require regular grid partitioning, such as image processing, spatial data analysis, etc. Tree partitioning is suitable for hierarchical data processing and is used for classification and regression problems. Common fuzzy partitioning functions include triangle membership function, Gaussian membership function, trapezoidal membership function, and Sigmoid membership function. Among them, the triangle membership function has been widely used in fuzzy logic and fuzzy control systems because of its simple and intuitive characteristics. The embodiment of the present application uses the triangle membership function for tree and grid fuzzy partitioning.

[0036] The preprocessing module 110 utilizes the fuzzy decision tree to target the uncertainty and fuzziness of multiple groups of initial input quantities, and fuzzifies the discrete and continuous values ​​of the most recently collected ground resource parameters (four types of IoT perception, four types of site capabilities) and scheduling resource parameters (route plans, capacity, restricted areas), and outputs continuous values ​​(regression tree).

[0037] The feature extraction module 120 is used to perform feature extraction on multiple groups of preprocessed input quantities based on a deep belief network to extract multiple groups of feature quantities.

[0038] Construct a deep belief net (DBN), use DBN to learn features of multiple sets of preprocessed inputs, and extract the representation of each set of inputs layer by layer. DBN can initialize the network weights through layer-by-layer pre-training and unsupervised learning, and then fine-tune the network parameters through supervised learning.

[0039] Classification is the core problem of deep learning. Combining fuzzy theory with DBN classifier can handle uncertain problems. In some implementations, the input data of low-altitude economy (any set of initial input quantities) can be fuzzy weighted to handle the fuzziness in system constraints (such as meteorological influences, restricted areas) and optimization goals (such as safety, efficiency, and economy), so as to find the optimal solution that is more in line with the actual situation.

[0040] When building a classification model, in order to better approximate the local details of the target model being built, the "fuzzy partitioning" approach can be used to divide the input space into multiple fuzzy regions or fuzzy subspaces. Then, a classification sub-model is constructed in each subspace, and finally the results of each sub-model are integrated and output.

[0041] In some implementations, a fuzzy decision tree (FDT) algorithm is used to perform fuzzy partitioning and fuzzy weighting on a data set to construct and optimize a DBN network structure. The preprocessing module 110 is used to divide multiple groups of initial input quantities into multiple subsets through a fuzzy decision tree, and the feature extraction module 120 is used to construct multiple deep belief network sub-models with different structures in multiple subspaces, train multiple deep belief network sub-models in parallel on multiple subsets, perform fuzzy weighting on the results of multiple deep belief network sub-models, and extract multiple groups of feature quantities. This helps to improve the degree of approximation of local details of the constructed target model.

[0042] Figure 2 is a schematic diagram of fuzzy classification and fuzzy weighting based on a fuzzy decision tree provided in an embodiment of the present application. Figure 2 As shown, the main process of building and optimizing the DBN network structure is as follows:

[0043] First, the preprocessing module 110 can divide the training data into multiple subsets through the fuzzy decision tree algorithm, for example, multiple groups of initial input quantities are divided into subset 1, subset 2, ... subset k. Then, the feature extraction module 120 can construct classification DBN sub-models with different structures in each subspace, and train multiple DBN sub-models in parallel on each subset; referencing the idea of ​​fuzzy set theory, the results of each DBN sub-model are fuzzy weighted to extract multiple groups of feature quantities.

[0044] Using fuzzy decision trees to fuzzily divide and fuzzily weight the data set, build and optimize the DBN network structure, which helps to better explore the expressive power of the DBN deep network. In view of the unclear category boundaries or noisy data in the low-altitude economic data set, the accuracy of DBN is further improved in system scheduling and the training of DBN is accelerated. It can effectively and quickly solve the classification problem of large sample data, overcome the shortcomings of a single DBN model such as long data classification time and high complexity, avoid overfitting problems, and have high classification accuracy.

[0045] The prediction module 130 is used to perform fusion processing on the first set of feature quantities based on the optimal power flow network, and output the route path of the target UAV corresponding to the first set of feature quantities. The first set of feature quantities is any set of feature quantities among multiple sets of feature quantities, and the multiple sets of feature quantities correspond to multiple route paths of the target UAV. The UAV scheduling model is trained based on the multiple route paths and the predetermined objective function, and the predetermined objective function corresponds to the predetermined application scenario.

[0046] Optimal power flow (OPF) refers to the power flow distribution that can meet all operating constraints and achieve the optimal value of a certain performance indicator of the system by adjusting the available control variables when the structural parameters and load conditions of the system are given. With the increasing expansion of the low-altitude economy, the safety and economic factors that need to be considered are becoming more and more complex.

[0047] It is understandable that the target drone may be multiple drones, and the target drone may be all drones managed by the low-altitude economic integrated operation platform. The multiple flight paths are multiple preferred flight paths corresponding to the multiple drones.

[0048] The multiple sets of feature quantities extracted by DBN are used as inputs of the optimal power flow network to optimize the operating status of the low-altitude economic dispatch system. The optimal power flow network adjusts the allocation of route resources, support capabilities, and guarantee capabilities through the low-altitude economic dispatch system. Under the premise of meeting the system constraints, it optimizes the dispatch for specific goals such as minimizing idle resources or ensuring the highest safety in low-altitude airspace for a certain route.

[0049] For the processing of multiple objectives, the current optimal power flow network usually adopts the "weighted summation" modeling method, which compromises two conflicting objective functions into a single objective function through weight factors. The main difficulty of this method is how to select appropriate weight factors, and it cannot handle multiple objective functions of different dimensions. Another shortcoming is the processing of constraints. All constraints are hard constraints and cannot be violated in the slightest, which reduces the feasible domain. In actual operation, in order to obtain a more satisfactory state, some constraints are allowed to exceed the limit slightly (with scalability). Especially in low-altitude economy, there are a large number of scalability constraints.

[0050] The embodiment of the present application aims at the multi-objective optimal power flow problem in the low-altitude economy, applies fuzzy set theory to transform the multi-objective optimal power flow problem with scalable constraints into a single-objective nonlinear programming problem, and adopts the nonlinear primal-dual path tracking interior point method to solve it.

[0051] For example, scheduling optimization is performed for a specific goal (i.e., a predetermined objective function) such as minimizing idle resources on a certain route or ensuring the highest safety in low-altitude airspace. Fuzzy modeling of multi-objective optimal power flow problems, with the safety level of low-altitude airspace and the total number of drones passing as predetermined objective functions, the nonlinear programming model (P1) of the multi-objective optimal power flow problem with hard constraints and scalable constraints can be expressed as follows:

[0052] Minimize the objective function f(x)=[f1(x),f2(x)] T , where f1(x) and f2(x) represent the low-altitude airspace security level function and the total number of drones passing function, respectively.

[0053] Equality constraint: stg(x)=0, where g(x)=[g1(x),g2(x),…,g k (x)] T Represents the tidal flow equation, which ensures the balance of passing efficiency of low-altitude airspace routes, and k is an integer greater than 1.

[0054] Inequality constraints include hard constraint variables and scalable constraint variables. a Satisfaction: X a_min ≤X a ≤X a _ max , which indicates the constraints that must be strictly followed. The scalable constraint variable X b Satisfaction: X b_min ≤X b ≤X b_max , indicating constraints with certain flexibility and scalability.

[0055] Unconstrained variables represent variables without constraints. Fuzzy relations are used to represent variables that are as large as possible without exceeding them by too much, which usually involves fuzzy logic and the determination of membership functions.

[0056] The nonlinear programming model (P1) takes into account multiple constraints and objectives in the low-altitude economic system, aiming to find an optimal solution that maximizes the total number of drones passing through the route corresponding to the safety level of the low-altitude airspace, while satisfying the hard constraints and scalability constraints of the system.

[0057] In the process of generating route paths for low-altitude economy, most of the data involved in ground resource parameters, scheduling resource parameters and safety assurance parameters are dynamic data, resulting in uncertainty in scheduling and decision-making. The solution of the drone scheduling model based on the optimal power flow network can use optimization algorithms such as DBN to deal with the nonlinearity and complexity.

[0058] During the model training process, the validation set or test set is used to evaluate the performance of the fusion model, including indicators such as accuracy and recall. Based on the evaluation results, the model can be optimized, such as adjusting the number of DBN layers, the number of neurons, or adjusting the parameters of the OPF algorithm.

[0059] For different types of low-altitude economic drones and drone routes, and different predetermined application scenarios, the predetermined objective function may be different. The predetermined objective function may be an objective function that uses safety indicators, efficiency indicators, and economic indicators (cost indicators) as primary (core) indicators.

[0060] In some implementations, the number of target drones in the low-altitude economy is greater than or equal to 1000. The predetermined objective function corresponds to a predetermined application scenario, and the predetermined application scenario may be a route area with a dense population and many hazardous chemicals, and the predetermined objective function is an objective function with safety indicators as the primary indicator. When using airspace scheduling resources, ground support resources, and security resources, the drone can reduce efficiency and economic indicators to match a more reliable and safe operation plan for the route section, ensuring the safety of the airspace section. Alternatively, the predetermined application scenario may be a route area with a sparse population and sufficient support, and the predetermined objective function is an objective function with cost indicators as the primary indicator. When using airspace scheduling resources, ground support resources, and security resources, the drone can reduce efficiency and safety indicators to match the most profitable operation plan for the route section, thereby increasing the economic benefits of the airspace section. Alternatively, the predetermined application scenario may be a route area with a complex environment and limited support, and the predetermined objective function is an objective function with efficiency indicators as the primary indicator. When using airspace scheduling resources, ground support resources, and security resources, drones can reduce economic and safety indicators, match the most efficient operation plan for that section of the route, and improve the traffic efficiency of that section of the airspace.

[0061] In the embodiment of the present application, a fuzzy decision tree is used to preprocess the data, and a deep belief network is constructed to extract features from the preprocessed data. The feature quantity extracted by the deep belief network is used as the input of the optimal power flow network to optimize specific targets (such as the route path of the target drone). Through fuzzy decision trees and fuzzy logic, it is possible to reasonably handle the uncertainty factors in the scheduling process, handle the ambiguity in the constraints and optimization targets of the low-altitude economic scheduling system, so that the system can make more reasonable predictions, route planning and scheduling when facing uncertainty and ambiguity, and improve the robustness of large-scale system scheduling. Through the scheduling system based on the optimal power flow network, under the premise of meeting the system constraints, it is possible to optimize specific targets such as minimizing idle resources or ensuring the highest safety in low-altitude airspace for a certain route. The embodiment of the present application helps to adapt to the low-altitude economic system with multi-source, heterogeneous, asynchronous and uncertain fuzzy information sources, find the optimal solution that meets the real-time status of the drone and the route stage path, avoid the idleness of low-altitude economic ground resources, and achieve the purpose of reducing airspace route conflicts and improving overall operating efficiency.

[0062] The implementation example of this application adjusts the allocation of route resources, support capabilities, and guarantee capabilities through the drone dispatch system, while meeting the constraints of low-altitude airspace resources facilities, laws and regulations. The implementation of this application is suitable for multi-level collaboration and complex environments in low-altitude economy, and solves the scenario where accurate data is difficult to obtain and decision makers need to dispatch quickly. The implementation of this application can be used to optimize the flight route, flight altitude, flight speed and other parameters of the drone to ensure the safe and efficient operation of the drone in a complex and changeable low-altitude environment, and improve the operating efficiency and safety of the drone.

[0063] Figure 3 Schematic diagram of an application device of a low-altitude economical UAV dispatching model provided in an embodiment of the present application. Figure 3 As shown, the application device 300 of the low-altitude economic drone scheduling model may include an acquisition module 310 and a model processing module 320.

[0064] The acquisition module 310 is used to acquire a first group of initial input quantities. The first group of initial input quantities includes a plurality of ground support resource parameters and a plurality of airspace scheduling resource parameters within a preset circular range.

[0065] The model processing module 320 is used to input the first set of initial input quantities into a pre-trained low-altitude economical UAV scheduling model for processing, and obtain the target route path corresponding to the target UAV in a predetermined application scenario within a predetermined time.

[0066] The low-altitude economic UAV dispatching model is a network model trained by using any of the training devices described above. The scheduled time is associated with the radius of the preset circle, and the scheduled time can be, for example, 0.5 minutes or 1 minute.

[0067] Combination of the above Figure 1-Figure 3 The training device and application device embodiments of the present application are described in detail. Figures 4 to 5 The method embodiment of the present application is described in detail. It should be understood that the description of the method embodiment corresponds to the description of the device embodiment, so the parts not described in detail can refer to the previous device embodiment.

[0068] Figure 4 1 is a flow chart of a training method for a low-altitude economical UAV dispatching model provided in an embodiment of the present application. Figure 4 As shown, the training method of the low-altitude economical drone scheduling model in the embodiment of the present application may mainly include steps S410 to S430, and these steps are described in detail below.

[0069] It should be pointed out that the size of the serial numbers of the steps in the embodiments of the present application does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0070] In step S410, multiple groups of initial input quantities in the training set are preprocessed based on the fuzzy decision tree to obtain multiple groups of preprocessed input quantities. Any group of initial input quantities includes multiple ground support resources and multiple airspace scheduling resource data within a preset circular range.

[0071] In step S420, feature extraction is performed on the preprocessed multiple groups of input quantities based on the deep belief network to extract multiple groups of feature quantities.

[0072] In step S430, the first set of feature quantities is fused based on the optimal power flow network, and the route path of the target UAV corresponding to the first set of feature quantities is output. The first set of feature quantities is any one of multiple sets of feature quantities, and the multiple sets of feature quantities correspond to multiple route paths of the target UAV. The UAV scheduling model is trained based on the multiple route paths and the predetermined objective function.

[0073] Optionally, preprocessing the multiple groups of initial input quantities in the training set based on the fuzzy decision tree in step S410 may include: dividing the multiple groups of initial input quantities into multiple subsets through the fuzzy decision tree. Extracting features from the preprocessed multiple groups of input quantities based on the deep belief network in step S420 to extract multiple groups of feature quantities may include: constructing multiple deep belief network sub-models with different structures in multiple subspaces, training multiple deep belief network sub-models in parallel on multiple subsets, and fuzzy weighting the results of the multiple deep belief network sub-models to extract multiple groups of feature quantities.

[0074] Optionally, the multiple ground support resource parameters within the preset circumference include at least: micro-meteorological station parameters, spectrum sensing parameters, photoelectric sensing parameters, millimeter wave radar parameters, communication base station parameters, Beidou differential station parameters, take-off and landing site parameters, and emergency rescue site parameters within the preset circumference. The multiple airspace scheduling resource parameters include at least: route plan, route capacity, and some or all parameters in restricted areas. The radius of the preset circle is positively correlated with the rated flight speed of the drone in the low-altitude economy.

[0075] Optionally, the number of target drones in the low-altitude economy is greater than or equal to 1000. The predetermined objective function corresponds to a predetermined application scenario, and the predetermined application scenario may be a route area with a dense population and many hazardous chemicals, and the predetermined objective function is an objective function with safety indicators as the primary indicator. Alternatively, the predetermined application scenario may be a route area with a sparse population and sufficient support, and the predetermined objective function is an objective function with cost indicators as the primary indicator. Alternatively, the predetermined application scenario may be a route area with a complex environment and limited support, and the predetermined objective function is an objective function with efficiency indicators as the primary indicator. When using airspace scheduling resources, ground support resources, and security resources, drones can reduce economic and safety indicators to match the most efficient operation plan for this route section, thereby improving the traffic efficiency of this airspace section.

[0076] Figure 5 1 is a flow chart of an application method of a low-altitude economical UAV dispatching model provided in an embodiment of the present application. Figure 5 As shown, the application method of the low-altitude economical drone scheduling model in the embodiment of the present application may mainly include steps S510 to S520, and these steps are described in detail below.

[0077] In step S510, a first set of initial input quantities is obtained, which includes a plurality of ground support resource parameters and a plurality of airspace scheduling resource parameters within a preset circular range.

[0078] In step S520, the first set of initial input quantities is input into a pre-trained low-altitude economical UAV scheduling model for processing, and a target route path corresponding to the target UAV in a predetermined application scenario is obtained within a predetermined time.

[0079] The low-altitude economical UAV dispatching model is a network model trained by applying any of the training methods described above. The scheduled time is associated with the radius size of the preset circle.

[0080] Figure 6 Schematic diagram of component units / partial component units of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device 600 may include: a memory 610 and at least one processor 620 .

[0081] The memory 610 is used to store codes or computer programs.

[0082] The processor 620 is connected to the memory 610 and is used to execute the code or computer program stored in the memory 610 to control the electronic device 600 to execute any of the processing methods described above.

[0083] Exemplarily, the computer program may be divided into one or more modules / units, one or more modules / units are stored in the memory 610 and executed by the processor 620 to complete the present application.

[0084] Those skilled in the art will understand that Figure 6 The electronic device 600 is merely an example and does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure, or combine some components, or different components.

[0085] The processor 620 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, etc.

[0086] The electronic device 600 provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0087] An embodiment of the present application further provides a non-volatile computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0088] An embodiment of the present application provides a computer program product. When the computer program product runs on an electronic device, the electronic device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0089] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above method embodiments, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments can be implemented. Among them, the computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera / electronic device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), read-only compact disc (CD-ROM), magnetic tape, floppy disk and optical data storage device. The computer-readable storage medium mentioned in the present application can be a non-volatile storage medium, in other words, it can be a non-transient storage medium.

[0090] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0091] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0092] In the embodiments provided in the present application, it should be understood that the disclosed devices / equipment and methods can be implemented in other ways. For example, the device / equipment embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0093] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0094] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0095] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0096] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A training device for a low-altitude economical UAV dispatching model, characterized in that: include: A preprocessing module, used for preprocessing multiple groups of initial input quantities in a training set based on a fuzzy decision tree, dividing the multiple groups of initial input quantities into multiple subsets, wherein any group of the initial input quantities includes multiple ground support resource parameters and multiple airspace scheduling resource parameters within a preset circular range; A feature extraction module, used to construct multiple deep belief network sub-models with different structures, train the multiple deep belief network sub-models on the multiple subsets respectively, perform fuzzy weighting on the results of the multiple deep belief network sub-models, and extract multiple groups of feature quantities; A prediction module, configured to perform fusion processing on a first set of feature quantities based on an optimal power flow network, output a route path of a target UAV corresponding to the first set of feature quantities, wherein the first set of feature quantities is any one of the multiple sets of feature quantities, and the multiple sets of feature quantities correspond to multiple route paths of the target UAV, and train the UAV scheduling model based on the multiple route paths and a predetermined objective function; Among them, the radius of the preset circle is positively correlated with the rated flight speed of the UAV in the low-altitude economy, and the multiple ground support resource parameters within the preset circle include at least: micro-meteorological station parameters, spectrum sensing parameters, optoelectronic sensing parameters, millimeter wave radar parameters, communication base station parameters, Beidou differential station parameters, take-off and landing site parameters, and emergency rescue site parameters within the preset circle; the multiple airspace scheduling resource parameters include at least: route plan, route capacity, and some or all parameters in restricted areas.

2. The training device according to claim 1, characterized in that The predetermined objective function corresponds to a predetermined application scenario, the predetermined application scenario is a route area with a dense population and many hazardous chemicals, and the predetermined objective function is an objective function with a safety index as the primary index; or The predetermined application scenario is a route area with a sparse population and sufficient support, and the predetermined objective function is an objective function with a cost index as a primary index; or, The predetermined application scenario is a route area with a complex environment and limited support, and the predetermined objective function is an objective function with efficiency index as the primary index.

3. An application device of a low-altitude economical UAV dispatching model, characterized in that: include: An acquisition module, configured to acquire a first set of initial input quantities, wherein the first set of initial input quantities includes a plurality of ground support resource parameters and a plurality of airspace scheduling resource parameters within a preset circular range; A model processing module, used for inputting the first group of initial inputs into the pre-trained low-altitude economic UAV scheduling model for processing, and obtaining a target route path corresponding to the target UAV in a predetermined application scenario within a predetermined time; Among them, the low-altitude economical drone scheduling model is a network model trained using the training device described in any one of claims 1-2, and the predetermined time is associated with the radius size of the preset circle.

4. A training method for a low-altitude economical UAV dispatching model, characterized in that: include: Preprocessing multiple groups of initial input quantities in a training set based on a fuzzy decision tree, dividing the multiple groups of initial input quantities into multiple subsets, any group of the initial input quantities including multiple ground support resource parameters and multiple airspace scheduling resource parameters within a preset circular range; Constructing multiple deep belief network sub-models of different structures, respectively training the multiple deep belief network sub-models on the multiple subsets, performing fuzzy weighting on the results of the multiple deep belief network sub-models, and extracting multiple groups of feature quantities; Based on the optimal power flow network, a first group of feature quantities is fused and processed, and a route path of the target UAV corresponding to the first group of feature quantities is output, where the first group of feature quantities is any one of the multiple groups of feature quantities, and the multiple groups of feature quantities correspond to multiple route paths of the target UAV, and the UAV scheduling model is trained based on the multiple route paths and a predetermined objective function; Among them, the multiple ground support resource parameters within the preset circular range include at least: micro-meteorological station parameters, spectrum sensing parameters, optoelectronic sensing parameters, millimeter wave radar parameters, communication base station parameters, Beidou differential station parameters, take-off and landing site parameters, and emergency rescue site parameters within the preset circular range. Part or all of the parameters; the multiple airspace scheduling resource parameters include at least: route plan, route capacity, and part or all of the parameters in restricted areas; the radius of the preset circle is positively correlated with the rated flight speed of the UAV in the low-altitude economy.

5. The training method according to claim 4, characterized in that: The predetermined objective function corresponds to a predetermined application scenario, the predetermined application scenario is a route area with a dense population and many hazardous chemicals, and the predetermined objective function is an objective function with a safety index as the primary index; or The predetermined application scenario is a route area with a sparse population and sufficient support, and the predetermined objective function is an objective function with a cost index as a primary index; or, The predetermined application scenario is a route area with a complex environment and limited support, and the predetermined objective function is an objective function with efficiency index as the primary index.

6. An application method of a low-altitude economical UAV dispatching model, characterized in that: include: Acquire a first set of initial input quantities, wherein the first set of initial input quantities includes a plurality of ground support resource parameters and a plurality of airspace scheduling resource parameters within a preset circular range; Inputting the first group of initial inputs into the pre-trained low-altitude economical UAV scheduling model for processing, and obtaining the target route path corresponding to the target UAV in the predetermined application scenario within a predetermined time; Among them, the low-altitude economical drone scheduling model is a network model trained using the training method described in any one of claims 4-5, and the predetermined time is associated with the radius size of the preset circle.

7. An electronic device, characterized in that: include: A memory for storing instructions; A processor is used to execute instructions stored in the memory so that the electronic device executes the training method described in any one of claims 4-5 or the application method described in claim 6.

8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, it is used to implement the training method described in any one of claims 4-5 or the application method described in claim 6.

Images