A graph embedding-based method for testing performance of an on-board computing platform
By constructing a performance indicator pool, a test algorithm pool, and a test data pool, and using a graph embedding method to select the optimal subset of performance test indicators, the problem of the relevance and flexibility of performance testing for airborne computing platforms is solved, and efficient and accurate performance evaluation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAN AVIATION COMPUTING TECH RES INST OF AVIATION IND CORP OF CHINA
- Filing Date
- 2022-12-15
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies lack mature and standardized performance testing methods for airborne computing platforms, making it impossible to effectively measure their support capabilities for intelligent algorithms, resulting in a lack of specificity and flexibility in testing methods.
A performance indicator pool, a test algorithm pool, and a test data pool are constructed. The optimal subset of performance test indicators is selected using a graph embedding method. Combined with the service capabilities of the airborne computing platform, targeted test methods are formed. By constructing performance test methods and patent application areas, the technical challenges involved in general technical means are solved.
It has realized the application of airborne computing platform technology, and solved the problems that can be achieved by general technical means.
Smart Images
Figure CN116149994B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aviation performance testing technology, and more specifically to a performance testing method for an airborne computing platform based on graph embedding. Background Technology
[0002] With the development of artificial intelligence technology, more and more intelligent algorithms are being applied to OODA (Out-of-Home) processes such as observation, perception, decision-making, and control. Because perception and decision-making algorithms have different requirements for computing power, storage, and networking, new demands are emerging regarding the determination of the performance capabilities of airborne computing platforms, such as integration, multitasking, real-time control, and high security. However, current performance testing of airborne computing platforms mainly uses performance indicators and testing methods for general computing platforms. There is no mature, standardized, and targeted performance testing method and tool for airborne computing platforms to measure the platform's support for intelligent algorithms. Therefore, designing a method for selecting performance testing indicators and testing tools for airborne computing platforms is of great significance for platform performance evaluation, efficiency design, and improvement. Summary of the Invention
[0003] In view of this, the embodiments of this application provide a performance testing method for airborne computing platforms based on graph embedding. This method not only solves the problem that general computing platform testing methods cannot fully measure the performance of airborne computing platforms, but also flexibly and quickly forms targeted performance testing methods according to different scenario tasks, meeting the needs of airborne computing platforms for different tasks.
[0004] This application provides the following technical solution: a performance testing method for an airborne computing platform based on graph embedding, applied to an airborne computing platform, comprising:
[0005] Construct a performance indicator pool, a test algorithm pool, and a test data pool; the performance indicator pool is used to specify the test items for airborne performance, the test algorithm pool is used to specify the stress test model, and the test data pool is used to specify the data scope of the test algorithm.
[0006] By using the performance test index selection method, the optimal performance index is obtained from the performance index pool, and the optimal performance test index subset is obtained for different airborne computing platforms to form a test method.
[0007] The method for selecting performance test indicators is based on graph embedding. It uses a subset of the performance indicator pool as the independent variable and the onboard computing platform service capability as the dependent variable to obtain the optimal subset of performance test indicators.
[0008] According to one embodiment of this application, the method further includes selecting test algorithms and test data from the test algorithm pool and the test data pool based on the optimal performance test index subset to obtain a test index set and form the test method.
[0009] According to one embodiment of this application, the performance metrics in the performance metric pool include: inference latency, throughput, utilization, power consumption, energy efficiency ratio, and load balancing.
[0010] According to one embodiment of this application, the test algorithm pool includes machine learning, multi-objective optimization algorithms, deep learning, and reinforcement learning.
[0011] According to one embodiment of this application, the test data pool includes public datasets, self-built datasets, and simulation data.
[0012] According to one embodiment of this application, the publicly available dataset includes: ImageNet and CIFAR10 for image classification, COCO and MS-COCO for object detection, ADE20K for semantic segmentation, and Librispeech dev-clean for speech recognition; the self-built dataset includes data collected for different sites and different targets; the simulation data includes data for task allocation and multi-machine collaboration.
[0013] According to one embodiment of this application, the airborne computing platform includes a flight control computer, a flight management computer, and an integrated processing computer.
[0014] This invention proposes a performance testing method for airborne computing platforms based on graph embedding. It extracts and incorporates performance indicators required by current superior embedded platforms and airborne computing platforms, constructing a performance indicator pool, a test algorithm pool, and a test data pool. Performance indicators specify the test items for airborne performance, test algorithms specify the models for stress testing, and test data specifies the data scope for the test algorithms. A method for selecting performance test indicators is proposed, which maps and matches the service capabilities required by the airborne platform with the test service indicators through graph embedding to obtain performance indicators that meet the service capability requirements. This invention enables a highly flexible and targeted performance testing method for airborne computing platforms, improving the efficiency and accuracy of airborne computing platform performance testing, and possesses high scalability, effectively supporting increasingly complex airborne intelligent applications. Attached Figure Description
[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a framework diagram of an airborne computing platform performance testing method based on graph embedding according to an embodiment of the present invention;
[0017] Figure 2 This is a framework diagram of the performance test index selection method according to an embodiment of the present invention;
[0018] Figure 3 This is a performance testing tool framework according to an embodiment of the present invention. Detailed Implementation
[0019] The embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0020] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments, providing a clear and complete description of the technical solutions of the present invention. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0021] like Figures 1-3 As shown, this embodiment of the invention provides a performance testing method for an airborne computing platform based on graph embedding, applied to an airborne computing platform, including:
[0022] Construct a performance indicator pool, a test algorithm pool, and a test data pool; the performance indicator pool is used to specify the test items for airborne performance, the test algorithm pool is used to specify the stress test model, and the test data pool is used to specify the data scope of the test algorithm.
[0023] By using the performance test index selection method, the optimal performance index is obtained from the performance index pool, and the optimal performance test index subset is obtained for different airborne computing platforms. Based on the optimal performance test index subset, test algorithms and test data are selected from the test algorithm pool and the test data pool to obtain a set of test indices and form a test method.
[0024] The method for selecting performance test indicators is based on graph embedding. It uses a subset of the performance indicator pool as the independent variable and the onboard computing platform service capability as the dependent variable to obtain the optimal subset of performance test indicators.
[0025] The graph embedding-based airborne computing platform performance testing method of this invention can be used for performance testing of various airborne computing platforms, including flight control computers, flight management computers, integrated processing computers, etc.
[0026] The performance metric pool is used to load performance metrics for general benchmark testing methods and performance metrics of interest to airborne computers. The test algorithm pool is used to verify algorithms that validate performance metrics, encompassing traditional machine learning, deep learning, reinforcement learning, etc. The test data pool is used to test the algorithm inference process.
[0027] The performance test index selection method is used to assign the optimal subset of performance test indices to different airborne computing platforms for performance evaluation.
[0028] In this embodiment, the general benchmark testing methods included in the performance indicator pool include: MLperf, AIbenchmark, Fathom, DeepSpeech, etc.; the airborne computing platforms included in the performance indicator pool include: flight management computer, flight control computer, integrated mission processor, etc.; the performance indicators of the performance indicator pool include: inference latency, throughput, utilization, power consumption, energy efficiency ratio, load balancing, etc.
[0029] The test algorithm pool includes traditional machine learning algorithms such as random forest and Bayesian algorithm, as well as multi-objective optimization algorithms such as bee colony algorithm and ant colony algorithm, and AI algorithms such as neural network (backpropagation, deep learning, graph learning, etc.), reinforcement learning, and deep reinforcement learning.
[0030] The data pool consists of: public datasets, self-built datasets, and simulation data.
[0031] The publicly available datasets in the dataset pool include: ImageNet and CIFAR10 for image classification, COCO and MS-COCO for object detection, ADE20K for semantic segmentation, and Librispeech dev-clean for speech recognition. The self-built datasets in the data pool include data collected using drones equipped with visible light and infrared cameras for different locations and targets. The simulation data in the data pool includes data used for task allocation and multi-machine collaboration.
[0032] The performance test index selection method uses a subset of performance indices as independent variables and the required airborne computing platform service capabilities as the dependent variable. A graph embedding method is used to map the subsets of performance indices to the measurement dimensions of the airborne computing platform service capabilities, forming a standardized N-dimensional vector. This vector is then compared to determine the optimal combination of performance indices for evaluating the airborne computing platform service capabilities. Assuming there are n performance indices, there are 2^n-1 subsets; and there are m types of platform service capabilities, first, each subset is represented as an n*n matrix, with the performance indices within the subset marked as 1 in their corresponding positions and 0 in the remaining positions. Then, using graph embedding, the n*n matrix is mapped to a 1*m service capability measurement vector. Finally, by comparing the service capability measurement vectors obtained from different subset mappings, the optimal combination of performance indices for evaluating the airborne computing platform service capabilities can be obtained.
[0033] The airborne computing platform performance testing method of this invention incorporates the performance indicators, testing algorithms, and data of current superior embedded platforms and airborne computing platforms. By utilizing a subset of performance indicators and the service capabilities of the airborne computing platform, it can analyze the platform service capabilities required to support specific task scenarios. It effectively combines platform service capabilities with performance indicators, testing algorithms, and testing data for performance testing, selects the optimal subset of performance indicators to measure the airborne computing platform, and conducts targeted experiments to verify the true performance of the airborne computing platform and support its design.
[0034] In one embodiment, an airborne computing platform performance testing tool is formed based on the testing method of the present invention. This testing tool integrates a performance indicator pool, a test algorithm pool, a test data pool, and a performance test indicator selection method to form a tool capable of testing the performance of an airborne computing platform.
[0035] The invention will be further described in detail using a typical air-to-ground multi-target sensing example.
[0036] 1. A comprehensive test component pool has been built, including a performance metric pool, a test algorithm pool, and a test dataset pool.
[0037] 2. Task Scenario
[0038] In the context of intelligent computing tasks for multi-target perception and attack decision-making, this project focuses on target detection, recognition, tracking, and localization. It acquires ground images through visual sensors, analyzes and labels the types and locations of targets in the images, and verifies the ability of the intelligent computing platform to perform target recognition.
[0039] 3. Platform service capability refinement
[0040] Based on a comprehensive analysis of the mission, the airborne computing platform should possess service capabilities in three aspects: high performance, real-time performance, and high reliability.
[0041] 4. Selection of performance test indicators
[0042] (1) Assume there are 9 performance indicators, including: accuracy, computing power, storage, inference latency, recognition rate, throughput, false alarm rate, image size limit and power consumption.
[0043] (2) Assume there are three types of service capabilities: high performance, real-time performance, and high reliability.
[0044] (3) For the 9 indicators, there are 2^9-1 subsets, and each subset is represented as a 9*9 matrix.
[0045] (4) A matrix with a service capacity of 1*3.
[0046] (5) Using the graph embedding method, the above 9*9 matrix is mapped to 1*3 service capabilities.
[0047] (6) By comparing, we can obtain the measurement strength of different subsets for different service capabilities, and thus obtain the set of performance indicators for measuring different service capabilities.
[0048] 5. Selected testing algorithm
[0049] PVA-Net network was chosen as the basic network model for the multi-target perception algorithm.
[0050] 6. Selected test dataset
[0051] (1) Public datasets
[0052] (2) Data set collected using a drone image acquisition system
[0053] 7. Formulation of test methods
[0054] The above performance indicators, intelligent algorithms, and datasets are used to form a performance testing method for airborne computing platforms that support air-to-ground multi-target perception applications.
[0055] 8. Results Analysis
[0056] Based on the experimental results, a comprehensive analysis of the performance of the airborne computing platform was conducted, providing effective evidence for whether the platform can meet the requirements of air-to-ground multi-target intelligent perception applications and for subsequent redesign.
[0057] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A performance testing method for an airborne computing platform based on graph embedding, applied to an airborne computing platform, characterized in that, include: Build a performance metric pool, a test algorithm pool, and a test data pool; The performance index pool is used to specify the test items for airborne performance, the test algorithm pool is used to specify the stress test model, and the test data pool is used to specify the data scope of the test algorithm. By using the performance test index selection method, the optimal performance index is obtained from the performance index pool, and the optimal performance test index subset is obtained for different airborne computing platforms to form a test method. The method for selecting performance test indicators is based on graph embedding, using a subset of the performance indicator pool as the independent variable and the service capability of the airborne computing platform as the dependent variable to obtain the optimal subset of performance test indicators. It also includes selecting test algorithms and test data from the test algorithm pool and test data pool based on the optimal performance test index subset to obtain a test index set and form the test method; The performance test index selection method uses a subset of the performance index pool as the independent variable and the required airborne computing platform service capability as the dependent variable. It utilizes graph embedding to map the subsets of the performance index pool to the measurement dimensions of the airborne computing platform service capability, forming a standardized N-dimensional vector. This vector is then compared to find the optimal combination of performance test indices for evaluating the airborne computing platform service capability. Specifically, if there are n performance test indices in the performance index pool, then there are 2^n-1 subsets. Since there are m types of airborne computing platform service capabilities, each subset is first represented as n... In an n-ary matrix, the performance test metrics contained in the subset are marked as 1 in the corresponding positions of the matrix, and the rest are set to 0; then, using a graph embedding method, the n-ary matrix is... The matrix mapping of n is 1 The service capability metric vector of the airborne computing platform is m; finally, by comparing the service capability metric vectors of the airborne computing platform obtained by different subset mappings, the optimal combination of performance test indicators for evaluating the service capability of the airborne computing platform is obtained.
2. The performance testing method for an airborne computing platform based on graph embedding according to claim 1, characterized in that, The performance metrics in the performance metric pool include: inference latency, throughput, utilization, power consumption, energy efficiency ratio, and load balancing.
3. The performance testing method for an airborne computing platform based on graph embedding according to claim 1, characterized in that, The test algorithm pool includes machine learning, multi-objective optimization algorithms, deep learning, and reinforcement learning.
4. The performance testing method for an airborne computing platform based on graph embedding according to claim 1, characterized in that, The test data pool includes public datasets, self-built datasets, and simulation data.
5. The performance testing method for an airborne computing platform based on graph embedding according to claim 4, characterized in that, The publicly available datasets include: ImageNet and CIFAR10 for image classification, COCO and MS-COCO for object detection, ADE20K for semantic segmentation, and Librispeech dev-clean for speech recognition; the self-built datasets include data collected for different sites and different targets; the simulation data includes data for task allocation and multi-machine collaboration.
6. The performance testing method for an airborne computing platform based on graph embedding according to claim 1, characterized in that, The airborne computing platform includes a flight control computer, a flight management computer, and an integrated processing computer.