Unmanned equipment intelligent evaluation method based on GAT and MLP

Abstract

The invention discloses an unmanned equipment intelligent assessment method based on GAT and MLP, and the method comprises the steps: firstly determining the intelligent capability in an unmanned equipment intelligent assessment index system, quantifying the intelligent capability as an index, constructing a task difficulty assessment model, and assessing the task difficulty; the method comprises the following steps: constructing an index weight system based on MLP, constructing an unmanned equipment intelligent evaluation model based on GAT, training the constructed unmanned equipment intelligent evaluation model, and outputting an unmanned equipment intelligent evaluation result based on the trained unmanned equipment intelligent evaluation model. According to the scheme, the intelligent ability of the unmanned equipment is evaluated by building the evaluation system and combining the task difficulty, manual scoring is replaced with a deep learning mode, an index system of the unmanned equipment and the task difficulty involved in the evaluation process are quantified, and the evaluation efficiency of the unmanned equipment is improved. According to the method, the index weight and the correlation degree of each index in evaluation are determined, the influence of the environmental parameters and task complexity on the intelligent index performance of the unmanned equipment can be considered more comprehensively, the manual dependence is reduced, and the scoring accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned equipment evaluation, and in particular relates to an intelligent evaluation method for unmanned equipment based on GAT and MLP. Background Art

[0002] With the rapid development of unmanned equipment technology, intelligent capability has become a key factor in evaluating the performance of unmanned equipment. The ability of unmanned equipment to perform tasks in complex environments directly affects its application effect and safety. Existing evaluation methods for the intelligent capability of unmanned equipment have problems such as single evaluation indicators, lack of a systematic framework, neglect of environmental factors, and insufficient evaluation of task complexity.

[0003] Traditional evaluation systems are mostly empirical, with a high degree of manual dependence in the evaluation process, relatively single indicators, and a lack of a scientific structured framework. It is difficult to integrate the decision-making, perception, and execution capabilities of unmanned equipment, and it is impossible to systematically evaluate its intelligent capabilities. In order to solve the problem of single evaluation indicators and lack of a systematic framework, the patent application number CN 115271312 A discloses a method for evaluating the autonomous ground attack capability of unmanned aerial vehicles. The method builds a model of the autonomous performance evaluation index system of unmanned aerial vehicles by analyzing the basic capabilities that unmanned aerial vehicles need to have, but the scoring needs to be manually weighted based on experience, and the efficiency of numerical calculation is not high.

[0004] In addition, environmental factors and task complexity are extremely important in practical applications and are directly related to the task completion of unmanned equipment. Most existing evaluation methods have the problem of ignoring environmental factors and insufficient assessment of task complexity. To solve this problem, the patent with application number CN 103234763 A discloses a quantitative evaluation system and evaluation method for unmanned vehicles. The method takes into account the complexity of tasks, the complexity of the environment and the degree of human intervention, but only makes a rough classification, does not fully consider the impact of environmental factors (such as weather, terrain, lighting, etc.) on the performance of unmanned equipment, and does not adequately assess the complexity of tasks, ignoring the multidimensional characteristics of task complexity, such as task type, number of executions and operation requirements.

[0005] Therefore, a new unmanned equipment intelligent capability assessment method is needed that can comprehensively consider multiple capabilities and environmental impacts, solve the problem of high manual dependence of traditional assessment methods, and more efficiently and accurately reflect the performance of unmanned equipment in actual applications. This method should have the characteristics of systematic and comprehensive evaluation indicators to meet the needs of intelligent operation of unmanned equipment in complex environments. Summary of the invention

[0006] In view of the above problems, the purpose of the present invention is to provide an unmanned equipment intelligent evaluation method based on GAT and MLP, which aims to evaluate the intelligent capabilities of unmanned equipment by building an evaluation system and combining task difficulty, replacing manual scoring with deep learning, reducing manual dependence and improving scoring accuracy.

[0007] The specific technical solution for achieving the purpose of the present invention is:

[0008] An intelligent evaluation method for unmanned equipment based on GAT and MLP includes the following steps:

[0009] Step 1: Determine the intelligent capabilities in the unmanned equipment intelligent evaluation index system and quantify the intelligent capabilities into indicators;

[0010] Step 2: Construct a task difficulty assessment model to assess the task difficulty;

[0011] Step 3: Construct an indicator weight system based on MLP and determine the reference weight coefficient of each indicator;

[0012] Step 4: Build an intelligent evaluation model for unmanned equipment based on GAT;

[0013] Step 5: Train the constructed unmanned equipment intelligent evaluation model, and output the unmanned equipment intelligent evaluation result based on the trained unmanned equipment intelligent evaluation model.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] The solution of the present invention quantifies the index system of unmanned equipment and the difficulty of tasks involved in the evaluation process, and determines the weights of indicators in the evaluation and the correlation between each indicator through a graph attention network (GAT) and a multi-layer perceptron (MLP). It can more comprehensively consider the impact of environmental parameters and task complexity on the performance of intelligent indicators of unmanned equipment, improve the accuracy and comprehensiveness of the evaluation, have a high degree of automation, reduce the subjectivity of manual scoring, and improve the evaluation efficiency.

[0016] The present invention is further described below in conjunction with specific implementation modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a flow chart of the intelligent evaluation method for unmanned equipment based on GAT and MLP of the present invention.

[0018] Figure 2 Schematic diagram of the overall architecture of the unmanned equipment intelligent evaluation method based on GAT and MLP of the present invention.

[0019] Figure 3The present invention is a schematic diagram of determining the index weights and the evaluation results based on MLP and GAT in the intelligent evaluation method for unmanned equipment based on GAT and MLP. DETAILED DESCRIPTION

[0020] Example

[0021] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an kind" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0023] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to those of ordinary skill in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be considered as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so that once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0024] Combination Figures 1 to 3 ,A method for intelligent evaluation of unmanned equipment based on GAT and MLP includes the following steps:

[0025] Step 1: Determine the intelligent capabilities in the unmanned equipment intelligent evaluation index system and quantify the intelligent capabilities into indicators;

[0026] The intelligent capabilities are selected according to actual needs, and the intelligent capabilities are quantified as indicators;

[0027] According to the selected intelligent capabilities and their quantitative indicators, actual or experimental data is obtained, and after processing the data, the unmanned equipment intelligent evaluation index vector X is obtained. scores =[s1,s2,...,s N ],s i For each indicator performance, i∈[1,N], N represents the number of indicators.

[0028] Taking some anti-terrorist drones as an example, the intelligent capabilities include the first-level intelligent capabilities of 1 perception and monitoring, 2 decision-making and planning, and 3 execution of actions;

[0029] The 1 perception and monitoring intelligent capability includes the secondary intelligent capabilities of 1.1 environmental perception and 1.2 situational awareness; the 1 decision-making and planning intelligent capability includes the secondary intelligent capabilities of 2.1 trajectory planning and 2.2 mission decision-making; the 3 execution action intelligent capability includes the secondary intelligent capabilities of 3.1 mobility and 3.2 instant matching precision strike capability;

[0030] Among them, the 1.1 environmental perception intelligent capability includes three levels of intelligent capabilities: 1.1.1 automatic weather condition detection and prediction, 1.1.2 three-dimensional map generation and update, 1.1.3 high-precision positioning capability and navigation capability, and 1.1.4 dynamic environment adaptability;

[0031] 1.2 Situational awareness intelligent capabilities include three levels of intelligent capabilities: 1.2.1 Identification of friend or foe and situation analysis, 1.2.2 Target identification, 1.2.3 Judgment, and 1.2.44 Tracking and prioritization;

[0032] 2.1 Intelligent trajectory planning capabilities include 2.1.1 three-level intelligent capabilities of autonomous navigation and path planning;

[0033] 2.2 Mission decision-making intelligent capabilities include 2.2.1 attack decision-making capability, 2.2.2 defense decision-making capability, and 2.2.3 collaborative operations and communications.

[0034] 3.1 Maneuverability Intelligent capabilities include three levels of intelligent capabilities: 3.1.1 Maneuverability obstacle avoidance and 3.1.2 Adaptive routing;

[0035] 3.2 Real-time matching precision strike intelligent capabilities include three levels of intelligent capabilities: 3.2.1 ammunition matching, 3.2.2 command reception and execution, and 3.2.3 resource release and synchronous update.

[0036] Furthermore, the indicators for quantifying the automatic weather condition detection and prediction capability in 1.1.1 are: weather condition prediction accuracy;

[0037] The quantitative indicators of 1.1.2 3D map generation and updating are: 3D map accuracy, landform feature recognition accuracy;

[0038] The quantitative indicators of high-precision positioning capability and navigation capability in 1.1.3 are: mean square error of positioning error and mean square error of route error;

[0039] The quantitative indicators of the dynamic environment adaptability in 1.1.4 are: the success rate of risk avoidance through fault tolerance and emergency mechanisms, the proportion of stable sensor data quality time, the real-time adjustment frequency of the sensor, and the accuracy of the real-time adjustment frequency of the sensor;

[0040] The quantitative indicators of 1.2.1 IFF and situation analysis are: identification distance, precision, accuracy, and average time of enemy equipment and friendly forces’ positions and status;

[0041] The quantitative indicators of target recognition, determination, tracking and priority sorting in 1.2.2 are: image denoising, structural similarity index, compression rate; target classification accuracy, target recognition distance measurement accuracy, target recognition distance measurement time;

[0042] The quantitative indicators of autonomous navigation and path planning in 2.1.1 are: number of backup planned routes, frequency of real-time path planning updates

[0043] The quantitative indicators of attack decision-making capability in 2.2.1 are: firepower allocation combat cost-effectiveness ratio, strike accuracy, and damage estimation accuracy;

[0044] The quantitative indicators of defense decision-making capability in 2.2.2 are: threat detection success rate, average time consumption of threat response strategy;

[0045] The quantitative indicators of 2.2.3 Cooperative Operations and Communications are: accuracy of information transmission, maximum distance of information transmission, transmission speed, and maximum transmission delay;

[0046] The quantitative indicators of 3.1.1 obstacle avoidance maneuver are: obstacle avoidance maneuver success rate;

[0047] The indicators of 3.1.2 adaptive route quantification are: the overlap rate between the actual movement route and the planned route;

[0048] The quantitative indicators of ammunition matching in 3.2.1 are: weapon station ammunition selection accuracy;

[0049] The quantitative indicators of instruction reception and execution in 3.2.2 are: the time to complete the attack task;

[0050] The quantitative indicator of resource release and synchronous update in 3.2.3 is: the frequency of real-time release and update of combat resources.

[0051] The unmanned equipment can obtain index data by completing the test task, and perform simple linear processing or classification on the data to obtain the score vector X including all evaluation items. scores =[s1,s2,...,s N ], and the score of each evaluation item is a scalar.

[0052] Step 2: Construct a task difficulty assessment model to assess the task difficulty;

[0053] Among them, task difficulty includes task complexity and environmental difficulty;

[0054] Task difficulty is a key factor affecting the performance of intelligent indicators. Therefore, it is necessary to build a task difficulty assessment module to assess the difficulty of each indicator under different tasks and use it as an auxiliary indicator weight generation;

[0055] The task complexity is evaluated based on the action difficulty coefficient and the number of executions, and the environmental difficulty is evaluated based on the climate, terrain and electromagnetic interference;

[0056] Generally speaking, the difficulty of the test task can be divided into task complexity and environmental difficulty. The task complexity can be evaluated from the action difficulty coefficient and the number of executions, and the environmental difficulty can be evaluated from the aspects of climate, terrain and electromagnetic interference.

[0057] For the complexity of the task, the difficulty coefficient of the action can be determined when setting the action. Taking the drone as an example, common actions include take-off and vertical landing, hovering, one rotation, vertical rectangle, inverted vertical triangle, horizontal circling with the nose facing inward, horizontal figure-8 flight, and horizontal straight flight at an altitude of 10 meters. For example, if the difficulty coefficient of one rotation is 7, when the number of executions is 3, it means that the action of one rotation is repeated 3 times, that is, 3 rotations.

[0058] Regarding environmental difficulty, the climate can be expressed by data such as precipitation, wind speed, temperature, humidity, visibility, and illumination; the terrain can be divided into forests, bushes, grasslands, wetlands, farmlands, towns, deserts, etc.; the electromagnetic interference environment can be divided into no interference, electromagnetic confrontation, electromagnetic shielding, and electromagnetic camouflage, etc.

[0059] Get the task difficulty vector based on the evaluation results: X task =[X comp , X env ], where X comp represents the complexity of the task, X env Indicates the difficulty of the environment.

[0060] In this embodiment, the complexity of the unmanned equipment test task is represented by a 1×2 vector X comp It includes the action difficulty coefficient and the number of executions. For example, the task complexity of 3 rotations can be expressed as X comp=[7, 3] means that the vector can indicate that the action with a difficulty coefficient of 7 in the experimental task was performed 3 times.

[0061] The environment when the unmanned equipment is performing the test mission is represented by vector X env Representation, including climate, terrain and electromagnetic interference. Climate can be represented by a 1×7 vector, in which the data include: precipitation (mm), wind speed (m / s), temperature (℃), humidity (%rh), visibility (km) and illuminance (Lux), such as [0.4,2.2,25,80,6,800] means that the precipitation during the test is 0.4mm, the wind speed is 2.2m / s, the temperature is 25℃, the humidity is 80%rh, the visibility is 6km, and the illuminance is 800Lux. Terrain can be represented by one-hot coding, for example, common terrains include forest (0), bush (1), grassland (2), wetland (3), farmland (4), town (5), and desert (6). If the terrain of the test task is grassland, it can be represented by the vector [0,0,1,0,0,0,0]. Electromagnetic interference can also be represented by one-hot coding, for example, common electromagnetic interference situations are divided into: electromagnetic countermeasure (0), electromagnetic shielding (1), and electromagnetic camouflage (2). If there is no interference, it can be represented by [0,0,0]. If there is electromagnetic confrontation in the test, it can be represented by [1,0,0]. Combining climate, terrain and electromagnetic interference conditions can obtain the environmental difficulty vector, such as X env =[0.4,2.2,25,80,6,800,0,0,1,0,0,0,0,0,0,0].

[0062] Combine the above task complexity and environment difficulty into a feature vector X task =[X comp , X env ], X task Indicates the difficulty of the test task.

[0063] Step 3: Construct an indicator weight system based on MLP and determine the reference weight coefficient of each indicator:

[0064] Multilayer Perceptron (MLP) is used to calculate the reference weight coefficient (difficulty) of each indicator. The indicator weight evaluation model is constructed based on MLP, including input layer, hidden layer and output layer. The input layer receives the feature vector X input =[X task , x i ], i∈[1,N], where x i is the i-th indicator item to be evaluated, and outputs the reference weight coefficient ζ i :

[0065] X input =[X task , x i ]

[0066] The first hidden layer is a fully connected layer using the ReLU activation function.

[0067] g1 = ReLU(W1× input +b1)

[0068] The second hidden layer is also a fully connected layer, using the ReLU activation function:

[0069] g2=ReLU(W2g1+b2)

[0070] The output layer is a fully connected layer that directly outputs the reference weight coefficient

[0071] ζ i =W3g2+b3

[0072] Among them, W1, W2, W3 are the weight matrices of each layer, b1, b2, b3 are bias vectors;

[0073] After initializing the parameters in the indicator weight evaluation model, each evaluation item in the unmanned equipment intelligent evaluation index is sequentially compared with the task difficulty vector X task After splicing, the index weight evaluation model is input, and the reference weight coefficient ζ of each index is calculated through forward propagation i .

[0074] Step 4: Build an intelligent evaluation model for unmanned equipment based on the graph attention network GAT:

[0075] Graph Attention Network (GAT) is a model that introduces attention mechanism into graph neural network, which is used to calculate the attention weight between nodes and their neighbor nodes. When aggregating neighbor information, GAT takes into account the importance of neighbors and gives greater weight to important neighbor nodes. Based on the above-mentioned unmanned equipment intelligent evaluation index system, each level of capability is defined as a node. Graph Attention Network (GAT) can be used to process the relationship between nodes and combine the impact of task difficulty on the performance of various indicators. Build a GAT model, whose structure includes input layer, attention mechanism and process of updating node features.

[0076] An unmanned equipment intelligent index evaluation model is constructed based on the graph attention network GAT. The input layer of the unmanned equipment intelligent index evaluation model receives the evaluation item score X scores =[s1,s2,...,s N ] and the weight coefficients ζ of each indicator determined based on MLP i , define each level of capability as a node, given N node features h = [h1,h2,...,h N ],h i ∈R 2, using the bucketing method to i and i Upgrade to vector s i and i , each node feature h i =s i ·ζ i ,i∈[1,N];

[0077] Determine the attention coefficient e between each pair of nodes ij To measure the correlation between them, and update the node feature representation of each node in the process;

[0078] e ij =LeakyReLU(a T [W4h i ||W4h j ])

[0079] Normalized attention coefficient α ij The calculation formula is:

[0080]

[0081] Through this formula, the attention coefficients of each node i can be normalized to ensure that their sum is 1;

[0082] The calculation formula for updating node features is:

[0083]

[0084] Among them, e ij represents the attention coefficient between nodes i and j, a is a learnable attention parameter vector, W4 is a learnable weight matrix, || represents the concatenation operation, represents the set of neighbor nodes of node i, σ is the activation function, which can be ReLU or Sigmoid;

[0085] The output layer of the unmanned equipment intelligent index evaluation model receives the node feature representation of all nodes and outputs the required intelligent capability index score and comprehensive index score:

[0086] t = ReLU(W5h′+b5)

[0087] y=W6t+b6

[0088] Among them, W5, W6 are weight matrices, b5, b6 are bias vectors, and y is the final output vector, including the scores of the required abilities and the total score of the assessment;

[0089] Initialize the output layer model, define the input dimension and output dimension; input the node feature representation output by GAT into the output layer; calculate the final score through forward propagation. Specifically in this embodiment, the output vector is used to generate the scores of three first-level capabilities (perception monitoring, decision planning, and execution action) and the comprehensive score

[0090] Step 5: After training the constructed unmanned equipment intelligent evaluation model, the unmanned equipment intelligent evaluation result is output based on the trained unmanned equipment intelligent evaluation model;

[0091] In this embodiment, the process of training the model includes: ① Initializing model parameters, defining loss functions and optimizers. ② Inputting data into MLP to calculate weight coefficients of all indicators. ③ Applying the reference weight coefficients calculated by MLP to GAT to calculate node feature representations. ④ Inputting the node feature representation output by GAT into the output layer to generate the final score. ⑤ Calculate the loss function and update the model parameters through back propagation. Specifically, the mean square error (MSE) is used as the loss function, assuming that we have some known unmanned equipment intelligent capability scores as labels. The optimizer uses the Adam optimizer to minimize the loss function. Minimize the loss function through the back propagation algorithm and adjust the model parameters. ⑥ Repeat the above steps until the stopping condition is met (such as reaching the maximum number of iterations or the loss function converges).

[0092] When evaluating model performance, we use methods such as cross-validation to evaluate. According to the evaluation results, we adjust hyperparameters such as learning rate, number of layers, number of hidden units, etc. Finally, we input various indicator performance and task difficulty parameters into the trained model to obtain the scores of the three first-level capabilities (perception monitoring, decision planning, and execution action) and the comprehensive score.

[0093] It can be seen that this scheme can quantify the index system of unmanned equipment and the difficulty of tasks involved in the evaluation process according to actual conditions, and automatically determine the weights of indicators in the evaluation and the correlation between each indicator. It can more comprehensively consider the impact of environmental parameters and task complexity on the performance of unmanned equipment intelligent indicators, improve the accuracy and comprehensiveness of the evaluation, have a high degree of automation, reduce the subjectivity of manual scoring, and improve evaluation efficiency.

[0094] The present solution also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0095] Step 1: Determine the intelligent capabilities in the unmanned equipment intelligent evaluation index system and quantify the intelligent capabilities into indicators;

[0096] Step 2: Construct a task difficulty assessment model to assess the task difficulty;

[0097] Step 3: Construct an indicator weight system based on MLP and determine the reference weight coefficient of each indicator;

[0098] Step 4: Build an intelligent evaluation model for unmanned equipment based on GAT;

[0099] Step 5: After training the constructed unmanned equipment intelligent evaluation model, the unmanned equipment intelligent evaluation result is output based on the trained unmanned equipment intelligent evaluation model.

[0100] The present invention also provides a computer storable medium having a computer program stored thereon, and when the computer program is executed by a processor, the following steps are implemented:

[0101] Step 1: Determine the intelligent capabilities in the unmanned equipment intelligent evaluation index system and quantify the intelligent capabilities into indicators;

[0102] Step 2: Construct a task difficulty assessment model to assess the task difficulty;

[0103] Step 3: Construct an indicator weight system based on MLP and determine the reference weight coefficient of each indicator;

[0104] Step 4: Build an intelligent evaluation model for unmanned equipment based on GAT;

[0105] Step 5: After training the constructed unmanned equipment intelligent evaluation model, the unmanned equipment intelligent evaluation result is output based on the trained unmanned equipment intelligent evaluation model.

[0106] 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. An intelligent evaluation method for unmanned equipment based on GAT and MLP, characterized in that: The following steps are involved: Step 1: Determine the intelligent capabilities in the unmanned equipment intelligent evaluation index system and quantify the intelligent capabilities into indicators; Step 2: Construct a task difficulty assessment model to assess the task difficulty; Step 3: Construct an indicator weight system based on MLP and determine the reference weight coefficient of each indicator; Step 4: Build an intelligent evaluation model for unmanned equipment based on GAT; Step 5: Train the constructed unmanned equipment intelligent evaluation model, and output the unmanned equipment intelligent evaluation result based on the trained unmanned equipment intelligent evaluation model.

2. The unmanned equipment intelligent evaluation method based on GAT and MLP according to claim 1 is characterized in that: The intelligent capabilities are selected according to actual needs, and the intelligent capabilities are quantified as indicators; According to the selected intelligent capabilities and their quantitative indicators, actual or experimental data is obtained, and after processing the data, the unmanned equipment intelligent evaluation index vector X is obtained. scores =[s1,s2,...,s N ],s i For each indicator performance, i∈[1,N], N represents the number of indicators.

3. The unmanned equipment intelligent evaluation method based on GAT and MLP according to claim 2 is characterized in that: The intelligent capabilities include the first-level intelligent capabilities of perception, monitoring, decision-making, planning and action execution; The intelligent capabilities of perception and monitoring include the secondary intelligent capabilities of environmental perception and situational awareness, the intelligent capabilities of decision-making and planning include the secondary intelligent capabilities of trajectory planning and mission decision-making, and the intelligent capabilities of action execution include the secondary intelligent capabilities of mobility and instant matching and precision strike capability; The environmental perception intelligent capabilities include three levels of intelligent capabilities: automatic weather condition detection and prediction, three-dimensional map generation and updating, high-precision positioning and navigation capabilities, and dynamic environmental adaptation capabilities; The intelligent capabilities of situational awareness include three levels of intelligent capabilities: friend-or-foe identification and situation analysis, target identification, determination, tracking and priority sorting; The intelligent capabilities of trajectory planning include three levels of intelligent capabilities: autonomous navigation and path planning; The mission decision-making intelligent capability includes three levels of intelligent capabilities: attack decision-making capability, defense decision-making capability, and coordinated operations and communications; The intelligent capabilities of mobility include three levels of intelligent capabilities such as obstacle avoidance and adaptive routing; The instant matching precision strike intelligent capability includes three levels of intelligent capabilities: ammunition matching, command reception and execution, and resource release and synchronous update.

4. The unmanned equipment intelligent evaluation method based on GAT and MLP according to claim 3 is characterized in that: The quantitative indicators of the automatic weather condition detection and prediction capability are: weather condition prediction accuracy; The quantitative indicators of the three-dimensional map generation and update are: three-dimensional map accuracy and landform feature recognition accuracy; The indicators for quantifying the high-precision positioning capability and navigation capability are: mean square error of positioning error and mean square error of route error; The quantitative indicators of the dynamic environment adaptability are: the success rate of risk avoidance through fault tolerance and emergency mechanisms, the proportion of stable time of sensor data quality, the real-time adjustment frequency of the sensor, and the accuracy of the real-time adjustment frequency of the sensor; The quantitative indicators of the enemy identification and situation analysis are: the identification distance, precision, accuracy, and average time of the enemy equipment and friendly forces' positions and status; The quantitative indicators of target recognition, determination, tracking and priority sorting are: image denoising, structural similarity index, and compression rate; Target classification accuracy, target identification distance measurement accuracy, target identification distance measurement time; The quantitative indicators of autonomous navigation and path planning are: the number of backup planned routes, the frequency of real-time path planning updates The quantitative indicators of the attack decision-making capability are: firepower allocation combat cost-effectiveness ratio, strike accuracy, and damage estimation accuracy; The quantitative indicators of the defense decision-making capability are: threat detection success rate, average time consumption of threat response strategy; The quantitative indicators of the collaborative operations and communications are: accuracy of information transmission, maximum distance of information transmission, transmission speed, and maximum transmission delay; The quantitative indicators of the maneuver obstacle avoidance are: maneuver obstacle avoidance success rate; The indicators of the adaptive route quantification are: the overlap rate between the actual movement route and the planned route; The quantitative indicators of ammunition matching are: weapon station ammunition selection accuracy; The quantitative indicators of command reception and execution are: time to complete the attack task; The quantitative indicator of resource release and synchronous update is: the real-time release and update frequency of combat resources.

5. The unmanned equipment intelligent evaluation method based on GAT and MLP according to claim 2 is characterized in that: The task difficulty in step 2 includes task complexity and environment difficulty; The task complexity is evaluated based on the action difficulty coefficient and the number of executions, and the environmental difficulty is evaluated based on the climate, terrain and electromagnetic interference; Get the task difficulty vector based on the evaluation results: X task =[X comp , X env ], where X comp represents the complexity of the task, X env Indicates the difficulty of the environment.

6. The unmanned equipment intelligent evaluation method based on GAT and MLP according to claim 5 is characterized in that: The step 3 constructs an indicator weight evaluation model based on MLP and determines the reference weight coefficient of each indicator, specifically: An indicator weight evaluation model is constructed based on MLP, including an input layer, a hidden layer and an output layer. The input layer receives the feature vector X input =[X task , x i ], i∈[1,N], where x i is the i-th indicator item to be evaluated, and outputs the reference weight coefficient ζ i : X input =[X task ,x i ] g1=ReLU(W1X input +b1) g2=ReLU(W2g1+b2) g i =W3g2+b3 Among them, W1, W2, W3 are the weight matrices of each layer, b1, b2, b3 are bias vectors, After initializing the parameters in the indicator weight evaluation model, each evaluation item in the unmanned equipment intelligent evaluation index is sequentially compared with the task difficulty vector X task After splicing, the index weight evaluation model is input, and the reference weight coefficient ζ of each index is calculated through forward propagation i .

7. The unmanned equipment intelligent evaluation method based on GAT and MLP according to claim 5 is characterized in that: The step 4 of building an intelligent evaluation model for unmanned equipment based on GAT is specifically as follows: An unmanned equipment intelligent index evaluation model is constructed based on the graph attention network GAT. The input layer of the unmanned equipment intelligent index evaluation model receives the evaluation item score X scores =[s1,s2,...,s N ] and the weight coefficients ζ of each indicator determined based on MLP i , define each level of capability as a node, given N node features h = [h1,h2,...,h N ],h i ∈R 2 , using the bucketing method to i and i Upgrade to vector s i and i , each node feature h i =s i ·ζ i ,i∈[1,N]; Determine the attention coefficient e between each pair of nodes ij To measure the correlation between them, and update the node feature representation of each node in the process; e ij =LeakyReLU(a T [W4h i ||W4h j ]) Node feature h i The update formula is: Among them, e ij represents the attention coefficient between nodes i and j, a is a learnable attention parameter vector, W4 is a learnable weight matrix, || represents the concatenation operation, represents the set of neighbor nodes of node i, σ is the activation function; The output layer of the unmanned equipment intelligent index evaluation model receives the node feature representation of all nodes and outputs the required intelligent capability index score and comprehensive index score: t = ReLU(W5h′+b5) y=W6t+b6 Among them, W5, W6 are weight matrices, b5, b6 are bias vectors, and y is the final output vector, including the scores of the required abilities and the total score of the evaluation.

8. The unmanned equipment intelligent evaluation method based on GAT and MLP according to claim 7 is characterized in that: The unmanned equipment intelligent evaluation model adjusts model parameters through a loss function and an optimizer during the training process until the loss function converges or reaches the number of iterations set for the training.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer storable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images