Unmanned autonomous system performance mode analysis method and device based on multilayer clustering

Abstract

The invention relates to an unmanned autonomous system performance mode analysis method and device based on multilayer clustering. The method comprises the following steps: performing initial clustering on a sample set according to a mean shift clustering technology to obtain a plurality of initial class clusters; the initial class cluster is a performance mode set of the unmanned aerial vehicle autonomous system in different scenes; clustering the initial class cluster by using a DBSCAN clustering method to generate a sub-cluster set about the boundary region; calculating two class clusters with different performance mode labels in the sub-cluster set by using a KNN algorithm to generate a scene boundary; and performing unmanned autonomous system performance mode distinguishing according to the scene boundary. By adopting the method, the performance analysis efficiency of the unmanned autonomous system can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and device for analyzing performance patterns of unmanned autonomous systems based on multi-layer clustering. Background Art

[0002] With the rapid development of unmanned autonomous systems, the problem of identifying boundary scenarios has emerged. Researchers define boundary scenarios when parameter values ​​reach or approach the performance limits of unmanned autonomous systems. Methods to solve this problem, such as relying on defining the maximum and minimum values ​​of key variables to perform boundary value analysis, find boundary scenarios close to the extreme values ​​of key input parameters, and thus test the performance of unmanned intelligent systems under extreme and marginal conditions. However, the disadvantage of this method is that when dealing with more complex and changeable systems, the difficulty of setting the size of boundary values ​​and the cost of analyzing boundary values ​​will increase significantly. In addition, the scenario generation method based on actual traffic data to generate scenarios and conduct performance testing and evaluation relies heavily on real-time scenario data, and also involves the problem of computing resource consumption.

[0003] These methods mostly focus on the boundary between safety and unsafety, but for drone clusters, it is more meaningful to focus on their different performance modes in different scenarios. The current adaptive sampling algorithm uses experimental information to gradually sample areas where key performance mode changes may occur. However, based on the sampling results, it is impossible to directly parse the boundaries of the performance mode. There is an urgent need for a new method to efficiently divide the scenes corresponding to different performance modes. Summary of the invention

[0004] Based on this, it is necessary to provide a method and device for analyzing performance patterns of unmanned autonomous systems based on multi-layer clustering, which can improve the efficiency of performance analysis of unmanned autonomous systems, in response to the above technical problems.

[0005] A method for analyzing performance patterns of an unmanned autonomous system based on multi-layer clustering, the method comprising:

[0006] Obtaining the scene space and sample set of the unmanned autonomous system application; the sample set includes a set of indicator values ​​of the unmanned autonomous system in the scene space;

[0007] The sample set is initially clustered according to the mean shift clustering technology to obtain multiple initial clusters; the initial clusters are the performance mode sets of the UAV autonomous system in different scenarios;

[0008] The DBSCAN clustering method is used to cluster the initial clusters and generate sub-cluster sets about the boundary area;

[0009] The KNN algorithm is used to calculate two clusters with different performance mode labels in the sub-cluster set to generate scene boundaries; the performance modes of the unmanned autonomous system are distinguished according to the scene boundaries.

[0010] In one embodiment, the sample set is initially clustered according to the mean shift clustering technique to obtain initial clusters, including:

[0011] The sample set is initially clustered according to the mean shift clustering technology, and the category labels are automatically assigned using the sample responses. The performance of the unmanned autonomous system is classified into different performance modes. The unmanned autonomous system exhibits different performance modes in different areas, and a set of scenarios corresponding to different performance modes of the unmanned autonomous system is obtained.

[0012] In one embodiment, the initial clusters are clustered using the DBSCAN clustering method to generate a sub-cluster set about the boundary area, including:

[0013] Mean shift clustering is used to divide the scenes of different performance modes in the initial clusters into different categories, that is, the scenes are divided into several main performance modes according to different indicators. The scene points in each performance mode are clustered according to DBSCAN. The data column mainly selects the two-dimensional coordinates of the scene points. The dense sub-clusters represent the boundary areas of the performance modes.

[0014] In one embodiment, a scene boundary sample is generated by calculating using a KNN algorithm based on two clusters with different performance mode labels in a plurality of initial clusters and a sub-clustering set, including:

[0015] Two clusters with different performance mode labels in multiple initial clusters and sub-clustering sets are taken as clustered samples. The KNN algorithm is first used to find the K nearest neighbors of each clustered sample in the feature space. Among the K neighbors found, check whether there is a sample with a different label from the current sample. If so, select the sample with the closest distance and different label as the "boundary neighbor" of the sample; these nearest neighbor samples with different labels are combined into boundary pairs, and each sample of the boundary pair represents the boundary of a potential performance change area; by listing all the formed boundary pairs, the area between samples with different labels is connected, and the area is marked as a boundary scene sample.

[0016] A device for analyzing performance patterns of unmanned autonomous systems based on multi-layer clustering, the device comprising:

[0017] A data acquisition module is used to acquire the scene space and sample set of the unmanned autonomous system application; the sample set includes a set of indicator values ​​of the unmanned autonomous system in the scene space;

[0018] The mean shift clustering module is used to perform initial clustering on the sample set according to the mean shift clustering technology to obtain multiple initial clusters; the initial clusters are performance mode sets of the UAV autonomous system in different scenarios;

[0019] The DBSCAN clustering module is used to cluster the initial clusters using the DBSCAN clustering method to generate a sub-cluster set about the boundary area;

[0020] The unmanned autonomous system performance mode differentiation module is used to calculate two clusters with different performance mode labels in the sub-cluster set using the KNN algorithm to generate scene boundaries; and differentiate the performance modes of the unmanned autonomous system according to the scene boundaries.

[0021] The above-mentioned unmanned autonomous system performance mode analysis method and device based on multi-layer clustering, this application adopts a combination of three unsupervised or weakly supervised algorithms: mean shift, DBSCAN and KNN, which reduces the dependence on parameter settings and improves the adaptability to complex data sets, and is particularly suitable for pattern recognition in complex scenes. Compared with traditional methods, it has a higher degree of automation and intelligence, thereby reducing the difficulty of manual intervention. In the performance analysis of unmanned systems, the scene generation method based on real-time traffic data often requires a lot of computing resources, and the initial clustering and density discrimination can effectively reduce the sample size, making the subsequent performance analysis more efficient. This application significantly reduces the computing cost and can still maintain a high computing efficiency when the scene complexity increases. Through multi-level clustering and local label discrimination, not only the recognition accuracy of the boundary area is improved, but also a more accurate basis is provided for the transformation of the performance mode, reducing the possibility of misjudgment and missed judgment, and ensuring the accuracy of the system performance analysis. The application scenarios of unmanned systems are often in dynamic change. This application can quickly identify new performance modes and boundary changes through the combination of multiple clustering and classification algorithms, adapt to the dynamic changes of the scene, and improve the analysis efficiency of the system in an uncertain environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of a flow chart of a method for analyzing performance patterns of an unmanned autonomous system based on multi-layer clustering in one embodiment;

[0023] Figure 2 A schematic diagram of dense distribution of boundary scene areas in one embodiment;

[0024] Figure 3 A schematic diagram of the result of mean shift clustering in one embodiment;

[0025] Figure 4 A schematic diagram of DBSCAN clustering results in another embodiment;

[0026] Figure 5 A result schematic diagram of a boundary scene in an embodiment;

[0027] Figure 6 It is a structural block diagram of a device for analyzing performance patterns of an unmanned autonomous system based on multi-layer clustering in one embodiment;

[0028] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0030] In one embodiment, Figure 1 As shown, a method for analyzing performance patterns of unmanned autonomous systems based on multi-layer clustering is provided, comprising the following steps:

[0031] Step 102, obtaining the scene space and sample set of the unmanned autonomous system application; the sample set includes a set of indicator values ​​of the unmanned autonomous system in the scene space.

[0032] In the application of unmanned autonomous systems, the system performance in different scenarios can vary greatly. In order to analyze the system performance more accurately, we must first obtain a rich scene space and sample set. This step discretizes the application scenarios of the system so that each sample contains multiple indicator values ​​of the system in a specific scenario. This scenario sampling approach helps to fully cover various situations of system operation at the macro level, avoiding the disadvantages of traditional methods of repeated testing of a few scenarios and difficulty in comprehensively evaluating system performance.

[0033] Step 104, performing initial clustering on the sample set according to the mean shift clustering technology to obtain a plurality of initial clusters; the initial clusters are performance mode sets of the autonomous system of the UAV in different scenarios.

[0034] After obtaining the scene sample set, the mean shift clustering technology is first used for initial clustering to divide the sample set into multiple initial clusters of performance modes. The mean shift algorithm is an unsupervised clustering method that dynamically adjusts the cluster center according to the local maximum value of the sample density, making the clustering result more consistent with the characteristics of the sample distribution.

[0035] Mean shift clustering can adapt to changes in sample density and has good adaptability to the non-uniform performance of unmanned autonomous systems in different scenarios. Unlike traditional k-means clustering, which requires a preset number of clusters, this algorithm is more suitable for use in scenarios with more complex sample distributions, reducing errors caused by manually set parameters. A series of initial clusters can be obtained through mean shift clustering, which can be regarded as "performance mode sets" of unmanned autonomous systems in different scenarios. These mode sets characterize the typical performance of the system in various scenarios and provide basic data for subsequent analysis.

[0036] Step 106: cluster the initial clusters using the DBSCAN clustering method to generate a sub-cluster set about the boundary area.

[0037] The initial clusters obtained based on mean shift clustering are further refined into a set of sub-clusters in the boundary area through the DBSCAN method. DBSCAN is used to identify lower density areas as boundary areas through density discrimination. Since boundary areas often represent the critical state of the system's transition from one performance mode to another, this step helps to obtain boundary information more accurately and provide support for performance mode division. Using DBSCAN clustering based on the initial clustering can avoid density analysis of all samples one by one, reduce the amount of calculation, and improve efficiency. This method ensures the flexibility and accuracy of the system in identifying different performance modes, and helps to speed up performance mode analysis.

[0038] Step 108, using the KNN algorithm to calculate two clusters with different performance mode labels in the sub-cluster set to generate a scene boundary; and distinguishing the performance modes of the unmanned autonomous system according to the scene boundary.

[0039] The boundary area identified by DBSCAN may contain multiple sets of sub-clusters. In order to further refine the boundary area of ​​the performance mode, it is necessary to introduce the KNN algorithm to calculate the sub-clusters with different performance mode labels to generate the scene boundary. The KNN algorithm calculates the differences between adjacent sub-clusters through the distance discrimination method and uses the label difference to determine the boundary of the performance mode. This step can help identify the critical point of the boundary area, that is, the specific boundary where the system switches between different modes. The scene boundary can be calculated using the local distance between samples, and the different performance modes in the sample set can be distinguished more finely. The scene boundary generated in this way can be used as a reference in testing or actual applications, making the performance of unmanned autonomous systems in different modes more predictable.

[0040] Compared with single-layer clustering, multi-layer clustering performs preliminary clustering of sample sets through mean shift clustering to form an initial performance mode set. The clustering algorithm adjusts the cluster center according to the local density distribution so that the initial cluster can cover different performance modes to the greatest extent. After the preliminary division, DBSCAN clustering is used for further refinement, and the areas close to the boundaries in the initial clusters are separated and processed, and the sample areas with higher density are identified as boundary scenes. This process can more accurately identify the boundaries, solve the problem that single-layer clustering is difficult to capture complex boundary changes. At the same time, through the hierarchical processing of multi-layer clustering, not only can the definition of each mode be refined, but also the predictability of mode conversion can be improved through the division of boundary areas. DBSCAN's recognition of boundary areas and KNN algorithm's neighbor discrimination of label differences enable the area of ​​mode conversion to be accurately defined, increasing the prediction accuracy of system conversion behavior. This division method with clear boundaries and distinct modes makes the performance of unmanned systems in various scenarios easier to predict and manage, thereby improving the system's operating reliability and stability in complex environments.

[0041] In the above-mentioned unmanned autonomous system performance mode analysis method based on multi-layer clustering, this application adopts a combination of three unsupervised or weakly supervised algorithms: mean shift, DBSCAN and KNN, which reduces the dependence on parameter settings and improves the adaptability to complex data sets, and is particularly suitable for pattern recognition in complex scenarios. Compared with traditional methods, it has a higher degree of automation and intelligence, thereby reducing the difficulty of manual intervention. In the performance analysis of unmanned systems, scene generation methods based on real-time traffic data often require a lot of computing resources, while initial clustering and density discrimination can effectively reduce the sample size, making subsequent performance analysis more efficient. This application significantly reduces the computing cost and can still maintain a high computing efficiency when the scene complexity increases. Through multi-level clustering and local label discrimination, not only the recognition accuracy of the boundary area is improved, but also a more accurate basis is provided for the transformation of the performance mode, reducing the possibility of misjudgment and missed judgment, and ensuring the accuracy of the system performance analysis. The application scenarios of unmanned systems are often in dynamic change. This application can quickly identify new performance patterns and boundary changes through a combination of multiple clustering and classification algorithms, adapt to the dynamic changes of the scene, and improve the analysis efficiency of the system in an uncertain environment.

[0042] In one embodiment, the sample set is initially clustered according to the mean shift clustering technique to obtain initial clusters, including:

[0043] The sample set is initially clustered according to the mean shift clustering technology, and the category labels are automatically assigned using the sample responses. The performance of the unmanned autonomous system is classified into different performance modes. The unmanned autonomous system exhibits different performance modes in different areas, and a set of scenarios corresponding to different performance modes of the unmanned autonomous system is obtained.

[0044] In one embodiment, the initial clusters are clustered using the DBSCAN clustering method to generate a sub-cluster set about the boundary area, including:

[0045] Mean shift clustering is used to divide the scenes of different performance modes in the initial clusters into different categories, that is, the scenes are divided into several main performance modes according to different indicators. The scene points in each performance mode are clustered according to DBSCAN. The data column mainly selects the two-dimensional coordinates of the scene points. The dense sub-clusters represent the boundary areas of the performance modes.

[0046] In a specific embodiment, mean shift clustering is used to divide scenes of different performance modes into different categories, that is, the scenes are divided into several main performance modes, such as "high performance mode" and "low performance mode" according to indicators such as task completion time and drone swarm performance. On this basis, DBSCAN only clusters the scene points within each performance mode, and the data column mainly selects the two-dimensional coordinates of the scene points. The purpose of doing this is to use DBSCAN to identify areas with uneven position density in each performance mode. Because sampling is mainly carried out on areas where the performance mode has changed significantly, that is, the boundary areas. Then through DBSCAN, dense clustering indicates that this is the boundary area of ​​the performance mode, which is significantly affected by environmental factors (obstacle location). DBSCAN can identify sparse and dense areas, thereby helping to find non-boundary areas and boundary areas in the scene.

[0047] In one embodiment, a scene boundary sample is generated by calculating using a KNN algorithm based on two clusters with different performance mode labels in a plurality of initial clusters and a sub-clustering set, including:

[0048] Two clusters with different performance mode labels in multiple initial clusters and sub-clustering sets are taken as clustered samples. The KNN algorithm is first used to find the K nearest neighbors of each clustered sample in the feature space. Among the K neighbors found, check whether there is a sample with a different label from the current sample. If so, select the sample with the closest distance and different label as the "boundary neighbor" of the sample; these nearest neighbor samples with different labels are combined into boundary pairs, and each sample of the boundary pair represents the boundary of a potential performance change area; by listing all the formed boundary pairs, the area between samples with different labels is connected, and the area is marked as a boundary scene sample.

[0049] In a specific embodiment, for each clustered sample, two samples with the closest distance and different labels are retained, and the scene boundary is defined by the area between these scenes with different labels. For each sample, the KNN algorithm is first used to find its K nearest neighbors in the feature space. Among the K neighbors found, check whether there is a sample with a different label from the current sample. If so, select the sample with the closest distance and different label as the "boundary neighbor" of the sample; these nearest neighbor samples with different labels are combined into boundary pairs. Each sample of the boundary pair represents the boundary of a potential performance change area; by listing all the formed boundary pairs, the area between samples with different labels is connected, and the area is marked as a boundary scene. The boundary area formed in this way can effectively represent the transition area between different performance modes.

[0050] like Figure 2 As shown in the figure, the gray sample points represent the center coordinates of the obstacles, indicating that simulation experiments were conducted at these locations. The spatial distribution of the samples intuitively illustrates the relationship between the sample locations and the significant fluctuations in the simulation duration. The sample distribution is significantly denser in areas of the scene space where the simulation duration changes significantly. This sampling strategy effectively achieves the purpose of capturing key scenes.

[0051] In order to identify the scenarios constituted by performance boundaries, this application designs a clustering step as shown in Algorithm 1.

[0052]

[0053] First, due to the continuity of sample responses, a mean-shift clustering technique is used, which automatically assigns class labels based on sample responses, thereby classifying the performance of the drone swarm into different performance modes. Figure 3 As shown, cluster 1 can be defined as a set of scenarios with longer mission times and poor drone swarm performance; on the contrary, cluster 2 represents the exact opposite performance pattern. Secondly, in view of the problem of uneven sample density in the scene space, this application proposes to use sample locations as features and adopt a density-based spatial clustering algorithm (DBSCAN) to identify areas with higher density. The core concept of the DBSCAN clustering method is to determine the core points by the number of samples within a specific radius (ε), thereby gradually expanding to high-density areas to form clusters. As Figure 4As shown in the figure, gray dots and gray triangles represent scenes close to potential boundary scenes, while gray squares represent scenes far from the boundary. This method effectively reduces the number of non-boundary scene samples, optimizes the clustering process, and improves the clustering quality. Ultimately, the first two clustering applications lay the foundation for the subsequent identification of boundary scenes based on the concept of K nearest neighbors (KNN). Specifically, for each clustered sample, the two most recent samples with different labels are retained, and the scene boundary is defined by the area between these samples with different labels. Figure 5 As shown, the samples marked with “×” symbol represent boundary scenes, accounting for 45% of the sample space. This shows that the proposed sampling method effectively targets boundary scenes.

[0054] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0055] In one embodiment, Figure 6 As shown, an unmanned autonomous system performance mode analysis device based on multi-layer clustering is provided, including: a data acquisition module 602, a mean shift clustering module 604, a DBSCAN clustering module 606 and an unmanned autonomous system performance mode differentiation module 608, wherein:

[0056] The data acquisition module 602 is used to acquire the scene space and sample set of the unmanned autonomous system application; the sample set includes the indicator value set of the unmanned autonomous system in the scene space;

[0057] The mean shift clustering module 604 is used to perform initial clustering on the sample set according to the mean shift clustering technology to obtain multiple initial clusters; the initial clusters are performance mode sets of the autonomous system of the UAV in different scenarios;

[0058] A DBSCAN clustering module 606 is used to cluster the initial clusters using the DBSCAN clustering method to generate a sub-cluster set about the boundary area;

[0059] The unmanned autonomous system performance mode differentiation module 608 is used to calculate two clusters with different performance mode labels in the sub-cluster set using the KNN algorithm to generate a scene boundary; and differentiate the unmanned autonomous system performance modes according to the scene boundary.

[0060] For the specific definition of the unmanned autonomous system performance pattern analysis device based on multi-layer clustering, please refer to the definition of the unmanned autonomous system performance pattern analysis method based on multi-layer clustering above, which will not be repeated here. Each module in the above-mentioned unmanned autonomous system performance pattern analysis device based on multi-layer clustering can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0061] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for analyzing the performance mode of an unmanned autonomous system based on multi-layer clustering is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0062] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0063] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0064] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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.

[0065] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for analyzing performance patterns of unmanned autonomous systems based on multi-layer clustering, characterized in that: The method comprises: Acquire a scene space and a sample set for application of the unmanned autonomous system; the sample set includes a set of indicator values ​​of the unmanned autonomous system in the scene space; Performing initial clustering on the sample set according to the mean shift clustering technology to obtain a plurality of initial clusters; the initial clusters are performance mode sets of the autonomous system of the UAV in different scenarios; Clustering the initial clusters using the DBSCAN clustering method to generate a sub-cluster set about the boundary area; The KNN algorithm is used to calculate two clusters with different performance mode labels in the sub-cluster set to generate a scene boundary; and the performance modes of the unmanned autonomous system are distinguished according to the scene boundary.

2. The method according to claim 1, characterized in that The sample set is initially clustered according to the mean shift clustering technique to obtain initial clusters, including: The sample set is initially clustered according to the mean shift clustering technology, and category labels are automatically assigned using sample responses. The performance of the unmanned autonomous system is classified into different performance modes. The unmanned autonomous system exhibits different performance modes in different areas, and a scene set corresponding to different performance modes of the unmanned autonomous system is obtained.

3. The method according to claim 1, characterized in that The initial clusters are clustered using the DBSCAN clustering method to generate a sub-cluster set about the boundary area, including: Mean shift clustering is used to divide the scenes of different performance modes in the initial clusters into different categories, that is, the scenes are divided into several main performance modes according to different indicators. The scene points in each performance mode are clustered according to DBSCAN. The data column mainly selects the two-dimensional coordinates of the scene points. The dense sub-clusters represent the boundary areas of the performance modes.

4. The method according to claim 1, characterized in that: The method of generating scene boundary samples by calculating based on the two clusters with different performance mode labels in the plurality of initial clusters and the sub-clustering set using a KNN algorithm includes: Taking multiple initial clusters and two clusters with different performance mode labels in the sub-clustering set as clustered samples, the KNN algorithm is first used to find the K nearest neighbors of each clustered sample in the feature space. Among the found K neighbors, check whether there is a sample with a different label from the current sample. If so, select the sample with the closest distance and different label as the "boundary neighbor" of the sample; combine these nearest neighbor samples with different labels into boundary pairs, and each sample of the boundary pair represents the boundary of a potential performance change area; by listing all the formed boundary pairs, connect the area between samples with different labels, and mark the area as a boundary scene sample.

5. A device for analyzing performance patterns of unmanned autonomous systems based on multi-layer clustering, characterized in that: The device comprises: A data acquisition module, used to acquire the scene space and sample set of the unmanned autonomous system application; the sample set includes a set of indicator values ​​of the unmanned autonomous system in the scene space; A mean shift clustering module is used to perform initial clustering on the sample set according to the mean shift clustering technology to obtain a plurality of initial clusters; the initial clusters are performance mode sets of the autonomous system of the UAV in different scenarios; A DBSCAN clustering module is used to cluster the initial clusters using the DBSCAN clustering method to generate a sub-cluster set about the boundary area; The unmanned autonomous system performance mode differentiation module is used to calculate two clusters with different performance mode labels in the sub-cluster set by using the KNN algorithm to generate a scene boundary; and differentiate the unmanned autonomous system performance mode according to the scene boundary.

Images