A method and system for calculating scene complexity of unmanned system testing

Abstract

This invention relates to the field of deep learning technology, specifically disclosing a method and system for calculating the scenario complexity of unmanned system testing. The method includes selecting evaluation elements for complexity assessment and acquiring evaluation data based on simulation software; normalizing the evaluation element data and calculating the weight of each evaluation element using the entropy weight method; calculating the scenario complexity score based on the weights and feature values ​​of each evaluation element; clustering the calculated complexity score sample data using the K-means algorithm to obtain complexity level labels; establishing a dataset from complexity evaluation elements to complexity levels; and training a complexity assessment model based on a semi-supervised learning method. By inputting relevant evaluation elements, the method effectively simulates the complexity level of the current scenario, providing a scenario complexity assessment level for unmanned system testing scenarios, thus enabling testers to provide a reference for task deployment and execution in unmanned system testing scenarios.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, specifically to a method and system for calculating the scenario complexity of unmanned system testing. Background Technology

[0002] In recent years, unmanned systems (UHS) have rapidly penetrated diverse application scenarios in the military field, including reconnaissance and surveillance, target location, electronic warfare, and swarm operations, thanks to their advantages of low cost, high mobility, and low detectability. They have also been deployed and applied in civilian fields such as fire rescue, disaster search and rescue, and material delivery. As most application scenarios evolve towards intelligence and networking, the complexity of UHS mission scenarios is increasing exponentially. The evaluation method has become a core technical bottleneck affecting mission execution efficiency. Accurate complexity evaluation methods can provide positive references for mission deployment and execution. Currently, most evaluation methods relying on expert experience are heavily influenced by subjective factors, lack comprehensive coverage of multi-factor scenarios, and lack dynamic adaptability. Therefore, this invention studies a method for calculating scenario complexity in UHS testing, combining expert experience with semi-supervised learning to train and validate a complexity evaluation model. It clearly and concisely provides scenario complexity evaluation levels, offering a reference for testers in the deployment and execution of UHS missions. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for calculating the scenario complexity of unmanned system testing, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for calculating the scenario complexity of unmanned system testing, the method comprising:

[0006] For different unmanned system testing scenarios, evaluation elements for complexity assessment are selected, and evaluation data is obtained based on simulation software;

[0007] The evaluation element data is normalized, and the weight of each evaluation element is calculated using the entropy weight method. Based on the weight of each evaluation element and the feature value of the evaluation element, the complexity score of the scenario is calculated.

[0008] The K-means algorithm is used to cluster the calculated complexity score sample data to obtain complexity level labels, and the complexity level classification obtained by clustering is verified.

[0009] Establish a dataset from complexity evaluation elements to complexity levels, and train a complexity evaluation model based on a semi-supervised learning method;

[0010] The trained complexity evaluation model is validated using the test set. If the validation fails, the process of creating the dataset and training the complexity evaluation model is returned.

[0011] As a further aspect of the present invention, the unmanned system test scenarios include: fire rescue scenario, obstacle crossing scenario, coastal rescue scenario, and search and discovery scenario.

[0012] As a further aspect of the present invention, for fire rescue scenarios and obstacle crossing scenarios, the evaluation factors are: terrain complexity, vegetation coverage, distribution of rivers and lakes, development level of transportation network, complexity of building layout, weather, temperature, and light intensity.

[0013] For coastal rescue scenarios, the assessment factors include the terrain complexity of the landing area, weather, visibility index, light intensity, wave and tide intensity, and temperature.

[0014] For search and discovery scenarios, the evaluation factors are vegetation coverage, building layout complexity, weather, visibility index, light intensity, and wind force.

[0015] As a further embodiment of the present invention, the normalization formula is as follows:

[0016] or ;

[0017] Where, x ij This represents the j-th evaluation element of the i-th data.

[0018] As a further aspect of the present invention, the step of calculating the weight of each evaluation element using the entropy weight method specifically includes:

[0019] Calculate the indicator weight of the i-th data under the j-th evaluation element: ;

[0020] Calculate the entropy value of the j-th evaluation element: ,in .

[0021] Calculate the entropy weights for each evaluation element: .

[0022] As a further aspect of the present invention, the formula for calculating the scene complexity score is as follows:

[0023] .

[0024] As a further aspect of the present invention, the K-means clustering process includes:

[0025] Determine the number of clusters and calculate the distance from each sample point to the centroid of its cluster;

[0026] Update the centroid of each cluster, and repeat the above steps of calculating distance and updating centroid until the distance from each sample point to the centroid of its cluster is less than the set threshold.

[0027] The present invention also provides a system for calculating the scenario complexity of unmanned system testing, the system comprising:

[0028] The data acquisition module is used to select evaluation elements for complexity assessment for different unmanned system test scenarios and acquire evaluation data based on simulation software.

[0029] The normalization module is used to normalize the evaluation element data and calculate the weight of each evaluation element using the entropy weight method. Based on the weight of each evaluation element and the feature value of the evaluation element, the complexity score of the scenario is calculated.

[0030] The clustering module is used to cluster the calculated complexity score sample data using the K-means algorithm to obtain complexity level labels and to verify the complexity level classification obtained by clustering.

[0031] The training module is used to build a dataset from complexity evaluation elements to complexity levels, and to train a complexity evaluation model based on a semi-supervised learning method.

[0032] The verification module is used to verify the trained complexity evaluation model using the test set. If the verification fails, the operation of establishing the dataset and training the complexity evaluation model is returned.

[0033] Compared with the prior art, the beneficial effects of the present invention are: by inputting relevant evaluation elements, it can effectively simulate the complexity level of the current scenario, and provide a clear and concise scenario complexity evaluation level in unmanned system test scenarios, so as to provide testers with a reference for task deployment and execution in unmanned system test scenarios. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0035] Figure 1 A flowchart illustrating a method for calculating the scenario complexity of unmanned system testing, provided in an embodiment of the present invention.

[0036] Figure 2 This is an analysis diagram showing the selection of scene complexity evaluation elements in an embodiment of the present invention. Detailed Implementation

[0037] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0038] Figure 1 Here is a flowchart illustrating a method for calculating the scenario complexity in unmanned system testing, as shown below. Figure 1 and Figure 2 As shown in the embodiment of the present invention, a method for calculating the scenario complexity of unmanned system testing includes:

[0039] For different unmanned system testing scenarios, evaluation elements for complexity assessment are selected, and evaluation data is obtained based on simulation software;

[0040] The evaluation element data is normalized, and the weight of each evaluation element is calculated using the entropy weight method. Based on the weight of each evaluation element and the feature value of the evaluation element, the complexity score of the scenario is calculated.

[0041] The K-means algorithm is used to cluster the calculated complexity score sample data to obtain complexity level labels, and the complexity level classification obtained by clustering is verified.

[0042] Establish a dataset from complexity evaluation elements to complexity levels, and train a complexity evaluation model based on a semi-supervised learning method;

[0043] The trained complexity evaluation model is validated using the test set. If the validation fails, the process of establishing the dataset and training the complexity evaluation model is returned. The model validation uses 5-fold cross-validation to debug and obtain the scene complexity evaluation model.

[0044] In this embodiment, multi-sensor fusion unmanned system test scenario simulation software will be used to obtain evaluation data. In order to obtain complexity level labels and more intuitively represent the complexity, the complexity score sample data will be clustered using the K-means algorithm. The complexity level classification will be verified using expert classification data. If the verification fails, the process of normalizing the evaluation element data and calculating the weight of each evaluation element using the entropy weight method will be returned.

[0045] By establishing a dataset that maps complexity evaluation elements to complexity levels and dividing it into training and testing sets, a complexity evaluation model is trained using a semi-supervised learning method.

[0046] like Figure 1 and Figure 2As shown, in a preferred embodiment of the present invention, the unmanned system test scenarios include: fire rescue scenario, obstacle crossing scenario, coastal rescue scenario, and search and discovery scenario.

[0047] like Figure 1 and Figure 2 As shown, in a preferred embodiment of the present invention, for fire rescue scenarios and obstacle crossing scenarios, the evaluation factors are terrain complexity, vegetation coverage, river and lake distribution, transportation network development level, building layout complexity, weather, temperature, and light intensity.

[0048] For coastal rescue scenarios, the assessment factors include the terrain complexity of the landing area, weather, visibility index, light intensity, wave and tide intensity, and temperature.

[0049] For search and discovery scenarios, the evaluation factors are vegetation coverage, building layout complexity, weather, visibility index, light intensity, and wind force.

[0050] In this embodiment, the elements for assessing the complexity of fire rescue scenarios include the complexity of the scenario's terrain (low, medium, high), vegetation coverage (low, medium, high), distribution of rivers and lakes (few and scattered, many and scattered, few and concentrated, many and concentrated), development level of the transportation network (low, medium, high), complexity of building layout (few and scattered, many and scattered, few and concentrated, many and concentrated), weather (sunny, cloudy, rainy / snowy, foggy), temperature (actual ambient temperature), light intensity, and visibility (low, medium, high).

[0051] As a preferred embodiment of the present invention, the normalization formula is:

[0052] or ;

[0053] Where, x ij This represents the j-th evaluation element of the i-th data.

[0054] In this embodiment, simulation software for unmanned system test scenarios using multi-sensor fusion is employed to acquire simulation data of evaluation elements and assign values ​​to the feature values ​​of these elements. Assuming the scenario to be evaluated contains m data points and n complexity evaluation elements, evaluation index data for the dataset to be evaluated are collected, and an evaluation index matrix is ​​constructed:

[0055] ;

[0056] Where i = 1, 2, ..., m; j = 1, 2, ..., n.

[0057] In decision-making problems involving multiple indicators, considering the differences in magnitude or meaning between the indicators, it is necessary to normalize the original indicator data matrix to obtain a matrix. Normalization can be expressed by the following formula:

[0058] or ;

[0059] For benefit-type indicators, the former formula is used; for cost-type indicators, the latter formula is used.

[0060] As a preferred embodiment of the present invention, the step of calculating the weight of each evaluation element using the entropy weight method specifically includes:

[0061] Calculate the indicator weight of the i-th data under the j-th evaluation element: ;

[0062] Calculate the entropy value of the j-th evaluation element: ,in .

[0063] Calculate the entropy weights for each evaluation element: .

[0064] As a preferred embodiment of the present invention, the formula for calculating the scene complexity score is as follows:

[0065] ;

[0066] Where, r ij This represents the feature value of the j-th evaluation element of the i-th data in the scenario.

[0067] As a preferred embodiment of the present invention, the K-means clustering process includes:

[0068] Determine the number of clusters and calculate the distance from each sample point to the centroid of its cluster;

[0069] Update the centroid of each cluster, and repeat the above steps of calculating distance and updating centroid until the distance from each sample point to the centroid of its cluster is less than the set threshold.

[0070] In this embodiment, the sample data is clustered using the K-means algorithm to form a dataset with ground-truth labels, and the scene complexity level is verified by expert classification data.

[0071] A semi-supervised learning approach was used to train a scene complexity evaluation model. During training, 80% of the labeled dataset was used as the training set and 20% as the test set, employing 5-fold cross-validation. After repeated adjustments, the model parameters were selected, and the weights of the obtained parameters and regressed data were used to evaluate the complexity of new scenes.

[0072] This invention also provides a system for calculating the scenario complexity of unmanned system testing, the system comprising:

[0073] The data acquisition module is used to select evaluation elements for complexity assessment for different unmanned system test scenarios and acquire evaluation data based on simulation software.

[0074] The normalization module is used to normalize the evaluation element data and calculate the weight of each evaluation element using the entropy weight method. Based on the weight of each evaluation element and the feature value of the evaluation element, the complexity score of the scenario is calculated.

[0075] The clustering module is used to cluster the calculated complexity score sample data using the K-means algorithm to obtain complexity level labels and to verify the complexity level classification obtained by clustering.

[0076] The training module is used to build a dataset from complexity evaluation elements to complexity levels, and to train a complexity evaluation model based on a semi-supervised learning method.

[0077] The verification module is used to verify the trained complexity evaluation model using the test set. If the verification fails, the operation of establishing the dataset and training the complexity evaluation model is returned.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of calculating scenario complexity of unmanned system testing, characterized in that, The method includes: For different unmanned system testing scenarios, evaluation elements for complexity assessment are selected, and evaluation data is obtained based on simulation software; The evaluation element data is normalized, and the weight of each evaluation element is calculated using the entropy weight method. Based on the weight of each evaluation element and the feature value of the evaluation element, the complexity score of the scenario is calculated. The K-means algorithm is used to cluster the calculated complexity score sample data to obtain complexity level labels, and the complexity level classification obtained by clustering is verified. Establish a dataset from complexity evaluation elements to complexity levels, and train a complexity evaluation model based on a semi-supervised learning method; The trained complexity evaluation model is validated using the test set. If the validation fails, the process of creating the dataset and training the complexity evaluation model is returned.

2. The method of claim 1, wherein, The unmanned system test scenarios include: fire rescue scenario, obstacle crossing scenario, coastal rescue scenario, and search and discovery scenario.

3. The method of claim 2, wherein, For fire rescue scenarios and obstacle crossing scenarios, the evaluation factors are: terrain complexity, vegetation coverage, distribution of rivers and lakes, development level of transportation network, complexity of building layout, weather, temperature, and light intensity. For coastal rescue scenarios, the assessment factors include the terrain complexity of the landing area, weather, visibility index, light intensity, wave and tide intensity, and temperature. For search and discovery scenarios, the evaluation factors are vegetation coverage, building layout complexity, weather, visibility index, light intensity, and wind force.

4. The method of claim 1, wherein, The normalization formula is: or ; wherein x ij represents the jth evaluation element of the ith data.

5. The method of claim 4, wherein, The specific steps for calculating the weight of each evaluation element using the entropy weight method include: calculating the index weight of the i-th data under the j-th evaluation element: ; The entropy value of the jth evaluation element is calculated as: wherein ; Calculate the entropy weights for each evaluation element: .

6. The method for calculating the scenario complexity of an unmanned system test according to claim 5, characterized in that, The formula for calculating the scenario complexity score is as follows: 。 7. The method for calculating the scenario complexity of an unmanned system test according to claim 1, characterized in that, The process of K-means clustering includes: Determine the number of clusters and calculate the distance from each sample point to the centroid of its cluster; Update the centroid of each cluster, and repeat the above steps of calculating distance and updating centroid until the distance from each sample point to the centroid of its cluster is less than the set threshold.

8. A system for calculating the scenario complexity of unmanned system testing, used to implement the method for calculating the scenario complexity of unmanned system testing as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to select evaluation elements for complexity assessment for different unmanned system test scenarios and acquire evaluation data based on simulation software. The normalization module is used to normalize the evaluation element data and calculate the weight of each evaluation element using the entropy weight method. Based on the weight of each evaluation element and the feature value of the evaluation element, the complexity score of the scenario is calculated. The clustering module is used to cluster the calculated complexity score sample data using the K-means algorithm to obtain complexity level labels and to verify the complexity level classification obtained by clustering. The training module is used to build a dataset from complexity evaluation elements to complexity levels, and to train a complexity evaluation model based on a semi-supervised learning method. The verification module is used to verify the trained complexity evaluation model using the test set. If the verification fails, the operation of establishing the dataset and training the complexity evaluation model is returned.

Images