Investigation on situation awareness model of unmanned aerial vehicle groups communication network based on adversarial network

Abstract

With the widespread application and development of unmanned aerial vehicle (UAV) technology, ensuring the security and stability of UAV swarm communication networks has become crucial. Given the diverse forms of interference and attacks in current networks, this poses a serious threat to the normal operation of UAV swarm communication. Therefore, how to accurately identify and effectively counter these network threats has become the focus of research. This study comprehensively evaluates the core technology of UAV swarm communication network situational awareness and constructs a situational awareness model based on adversarial networks. The model utilizes adversarial network technology and combines data collection and processing to design four experiments to comprehensively evaluate the performance of the model in different scenarios. The experimental results show that as the amount of data gradually increases, the performance of the model also improves. When processing 100, 1000, and 10,000 data points, the model achieved accuracies of 0.955, 0.962, and 0.982, respectively. Furthermore, the experimental results also indicate that effective noise suppression measures can significantly improve the accuracy and stability of the situational awareness model. Additionally, it is noted that different model structures will affect training duration, accuracy, and stability. Although increasing network scale may lead to increased computational complexity and latency, its accuracy is correspondingly improved. The adversarial network-based situational awareness model proposed in this study can accurately identify and effectively counter interference and attacks in UAV swarm communication networks, thereby providing solid protection for the collaborative combat and information sharing of UAV swarms.

Introduction

With the rapid advancement and widespread application of Unmanned Aerial Vehicle (UAV) technology, the security and stability of UAV swarm communication networks have become decisive factors affecting mission success. Ensuring efficient and stable communication links is crucial for UAV swarms undertaking complex tasks such as reconnaissance, surveillance, and material transportation, while also defending against potential network interference and malicious attacks to ensure the smooth progress of missions. However, traditional network situational awareness models are primarily designed for wired and wireless networks, lacking comprehensive consideration of the unique complexity and dynamics of UAV swarm communication networks. Consequently, they fall short in addressing the security challenges posed by UAV swarm communication networks.

This study proposes an adversarial network-based model for situational awareness in UAV swarm communication networks. The model integrates cutting-edge technologies of deep learning and adversarial networks, enabling precise identification and timely response to potential threats within UAV swarm communication networks through deep data mining and analysis. Furthermore, the rapid development of big data and artificial intelligence technologies provides abundant training data and robust computational support for the model, making real-time and accurate situational awareness in UAV swarm communication networks achievable.

In reviewing existing research, Baek and Lim ¹ emphasized the importance of reliable UAV control and situational awareness, exploring the design of UAV relay tactical data links. You ² investigated the application of distributed autonomous situational awareness systems in UAV swarms from the perspective of task requirements. However, these studies primarily focus on specific aspects of UAV swarm communication networks, lacking in-depth exploration of comprehensive situational awareness.

McAree et al., ³ developed a method to improve drone awareness for long-range operations. Jeon et al., ⁴ improved on this by using a grid map to navigate environmental obstacles and improve drone situational awareness.

In the fields of deep learning and adversarial networks, Tian et al. ⁵ conducted adversarial attack analysis on UAV deep learning models and proposed corresponding defense strategies. Munir et al. ⁶ delved into adversarial threats and attacks in situational awareness systems, offering various mitigation methods. Smith and Li ⁷ comprehensively reviewed the application of deep learning in UAV communication networks, providing new insights into situational awareness in UAV swarm communication networks. The foundational principles of adversarial networks, as dissected by Goodfellow et al., ⁸ provide a solid theoretical basis for this study.

Moreover, several other studies have examined adversarial attacks and defenses in various contexts, including LESSON multi-label adversarial fake data injection attacks targeting deep learning-based location detection, adversarial attacks and defenses for deep learning-based drones, and joint adversarial examples and false data injection attacks aimed at power system state estimation.^5,9–11 These papers offer valuable insights into the vulnerabilities of deep learning models and the strategies to mitigate such attacks, which are highly relevant to proposed model for UAV swarm communication network situational awareness.

Building upon the aforementioned research background, this study’s adversarial network-based UAV swarm communication network situational awareness model, by leveraging the advantages of deep learning and adversarial networks and incorporating advanced feature extraction methods and unique network architecture designs, significantly enhances the accuracy and stability of situational awareness. The model not only enables real-time and accurate perception of potential threats in UAV swarm communication networks but also continuously improves performance through extensive data training and optimization, providing new insights and technical means for ensuring the security of UAV swarm communication networks. The findings of this study enrich the research content of UAV swarm communication network situational awareness and lay a solid foundation for the further development and application of UAV technology.

Implementation of situational awareness in UAV communication network based on adversarial networks

Key technologies for situation awareness in UAV groups communication network

With the continuous development of UAV technology, UAV groups have become indispensable and important equipment in military, civilian, and other fields. At the same time, the communication network of UAV groups is also playing an increasingly important role in information transmission, intelligence processing, command, and decision-making.^12,13 However, with the continuous expansion of UAV groups and the increase in task complexity, these networks are also facing more and more security threats and risks. Therefore, how to achieve situational awareness in UAV communication networks has become one of the important issues that need to be addressed at present.

Situation awareness in UAV communication network refers to the perception, analysis, and judgment of network status, node behavior, traffic characteristics, and other information in the UAV communication network, so as to timely identify abnormal situations and take corresponding security measures to ensure the stability and security of the network.^14,15 As shown in Figure 1, it shows the communication of the UAV groups. Specifically, the situation awareness of UAV communication network needs to address the following key technical issues.

• (1) Data collection and processing: The data generated by the UAV groups communication network mainly includes network traffic, sensor data, etc. For these data, data collection, preprocessing, and analysis are required.

Figure 1.

Interference between UAV swarms.

Data collection and preprocessing are the foundation of situational awareness in UAV communication networks. Due to the large number of nodes and high frequency of data generation in the communication network of UAV, efficient and automated data collection and preprocessing methods are needed.^16,17 Among them, packet sniffing is a commonly used method for data collection in UAV communication networks, which captures and records data packets in real-time to obtain network traffic data. Data preprocessing is the preliminary processing of collected data, such as deduplication, filtering, normalization, etc., to extract useful feature information and provide strong support for subsequent data analysis and decision-making.

• (2) Network topology analysis: There are numerous nodes in the communication network of UAV groups, and the topology structure constantly changes with the changes of task scenarios.^18,19 Therefore, it is necessary to accurately analyze the network topology structure and update the topology model in a timely manner for subsequent situation analysis.

Network topology analysis is an important part of situational awareness in UAV communication networks. By analyzing and modeling the network topology structure, it is possible to better understand the structural characteristics and key nodes of the UAV groups communication network, thus providing support for subsequent attack identification and security measures. At the same time, due to the continuous changes in the topology structure of the UAV groups communication network as the task scenario changes, it is necessary to monitor and update the topology model in real-time.

• (3) Behavior analysis and recognition: For attack behaviors in UAV communication networks, it is necessary to identify and analyze their types, characteristics, and other information. ²⁰

Behavior analysis and recognition are the core technologies for achieving situational awareness in UAV communication networks. By analyzing and judging information such as node behavior and traffic characteristics in the communication network of UAV, attack behavior can be detected in a timely manner and corresponding security measures can be taken. Behavior analysis and recognition mainly include anomaly detection, network intrusion recognition, malicious code detection, and other aspects. Regardless of the method used, it is necessary to select and optimize according to the actual situation to improve the accuracy and robustness of recognition.

• (4) Situation assessment and response strategies: After achieving situational awareness of the UAV groups communication network, it is necessary to evaluate the network status and develop corresponding response strategies.

Situation assessment and response strategies are the last step in the situational awareness of UAV communication networks. By evaluating the status of the UAV groups communication network, abnormal situations can be detected in a timely manner and corresponding measures can be taken to ensure the stability and security of the network. The response strategy needs to be selected and formulated based on specific attack types and scenarios. At present, commonly used coping strategies include defense, interference, and countermeasures. Among them, defense strategies mainly protect the network from attacks through encryption, access control, and other means; The interference strategy is to prevent the attacker’s attack behavior by interfering with their communication signals.

Design of situation awareness model for UAV groups communication network based on adversarial networks

The situational awareness model of UAV communication network based on adversarial networks mainly includes three layers: data acquisition and preprocessing layer, deep learning layer, and decision layer. Figure 2 shows the basic structure of the perception model.

Figure 2.

Basic structure of perception model.

Data collection and preprocessing layer

In the data collection stage, it is necessary to collect real UAV groups communication network data, and virtual data samples can also be generated through simulation and other means. The data preprocessing is required to reduce data noise and increase data diversity.

The data preprocessing formula is as follows:

$${\mathrm{X}}_{\mathrm{t}+1}=\mathrm{f}\left({\mathrm{X}}_{\mathrm{t}}\right)+{\in }_{\mathrm{t}}$$

(1)

Formula 1 indicates that in data preprocessing, some processing is required on the original data ${\mathrm{X}}_{\mathrm{t}}$ to obtain new data ${\mathrm{X}}_{\mathrm{t}+1}$ . $\mathrm{f}$ here is a function used to convert ${\mathrm{X}}_{\mathrm{t}}$ ; ${\in }_{\mathrm{t}}$ is a noise term used to increase the diversity of data.

For UAV communication networks, it is necessary to consider factors such as communication legality, data timeliness, and data reliability, and perform data preprocessing based on these factors. The specific operations include data cleaning, data conversion, and data enhancement.

Deep learning layer

In the deep learning layer, Generate Adversarial Networks (GAN) is used to construct a situational awareness model for UAV communication networks. GAN consists of a generator network and a discriminator network, which are responsible for generating false data and determining the authenticity of the data, respectively. The generator network utilizes input random noise to generate false data. By iteratively optimizing, the generated false data becomes closer to the real data; the discriminator network evaluates the authenticity of input data. By iterative optimization, the discriminator’s ability to distinguish between real and false data is improved.

The loss functions of generator network and discriminator network are as follows:

$${\mathrm{L}}_{\mathrm{G}}=-\frac{1}{\mathrm{m}}\sum _{\mathrm{i}=1}^{\mathrm{m}}\mathrm{logD}\left(\mathrm{G}\left({\mathrm{Z}}^{\left(\mathrm{i}\right)}\right)\right)$$

(2)

$${\mathrm{L}}_{\mathrm{D}}=-\frac{1}{\mathrm{m}}\sum _{\mathrm{i}=1}^{\mathrm{m}}[\mathrm{logD}\left({\mathrm{X}}^{\left(\mathrm{i}\right)}\right)+\log (1-\mathrm{D}\left(\mathrm{G}\left({\mathrm{Z}}^{\left(\mathrm{i}\right)}\right)\right))]$$

(3)

Among them, ${\mathrm{Z}}^{\left(\mathrm{i}\right)}$ represents input random noise; ${\mathrm{X}}^{\left(\mathrm{i}\right)}$ is a real data sample; $\mathrm{m}$ is the number of samples; $\mathrm{G}\left(\mathrm{Z}\right)$ is false data generated by the generator network; $\mathrm{D}\left(\mathrm{X}\right)$ is a discriminator network that evaluates the authenticity of input data X.

In the situational awareness model of UAV communication network, the generator network needs to generate false data similar to real communication data to increase the number and diversity of data samples. Discriminator networks need to identify real and false data in order to improve the robustness and generalization ability of the model. Through this approach, GAN can effectively improve the accuracy, stability, and generalization performance of the model.

Decision makers

At the decision-making level, probability graph models can be used to predict the situation of UAV communication networks. The probability graph model is a probability model based on graph theory, which can describe the dependency relationship between variables, and infer the conditional probability between variables by combining the probability density function.

The conditional probability distribution of the probability graph model is as follows:

$$\mathrm{P}\left(\mathrm{X}|\mathrm{\theta }\right)=\frac{1}{\mathrm{Z}\left(\mathrm{\theta }\right)}\underset{\mathrm{i}=1}{\overset{\mathrm{n}}{\Pi }}{\mathrm{f}}_{\mathrm{i}}({\mathrm{X}}_{\mathrm{i}},{\mathrm{P}}_{\mathrm{a}}\left({\mathrm{X}}_{\mathrm{i}}\right),{\mathrm{\theta }}_{\mathrm{i}})$$

(4)

Among them, $\mathrm{X}$ is the state variable of the node; $\mathrm{\theta }$ is a parameter; $\mathrm{Z}\left(\mathrm{\theta }\right)$ is the normalized constant; ${\mathrm{f}}_{\mathrm{i}}$ is a local probability distribution function.

In the situational awareness model of UAV communication network, the probability graph model can describe the states, connection relationships, and behavior patterns of different nodes in the UAV communication network. Through the establishment of directed acyclic graph and Bayesian network and other models, the real-time movement track, communication quality, node fault, and other information of UAV groups can be predicted, so as to realize real-time monitoring and state awareness of UAV groups communication network. The joint probability density function of Bayesian network is as follows:

$$\mathrm{P}({\mathrm{X}}_1,{\mathrm{X}}_2,\cdots ,{\mathrm{X}}_{\mathrm{n}})=\underset{\mathrm{i}=1}{\overset{\mathrm{n}}{\Pi }}\mathrm{P}\left({\mathrm{X}}_{\mathrm{i}}|{\mathrm{P}}_{\mathrm{a}}\left({\mathrm{X}}_{\mathrm{i}}\right)\right)$$

(5)

Among them, ${\mathrm{X}}_1,{\mathrm{X}}_2,\cdots ,{\mathrm{X}}_{\mathrm{n}}$ represents the state variables of different nodes, and ${\mathrm{P}}_{\mathrm{a}}\left({\mathrm{X}}_{\mathrm{i}}\right)$ represents the parent node of ${\mathrm{X}}_{\mathrm{i}}$ .

In summary, the situational awareness model of UAV communication network based on adversarial networks mainly includes three layers: data acquisition and preprocessing layer, deep learning layer, and decision layer. This three-layer structure can enhance the robustness and generalization ability of the model, thereby more effectively ensuring the security and stability of the UAV communication network.

Experimental evaluation of situational awareness model for UAV communication network

The dataset used in this experiment originates from two main sources: firstly, real data collected from actual unmanned aerial vehicle (UAV) communication networks, which include key indicators such as communication messages exchanged between UAVs, UAV positions, velocities, angles, among others, across different scenarios and time periods; secondly, data generated by high-precision UAV communication network simulators. These simulated data encompass various flight and communication conditions to ensure the diversity and richness of the dataset.

Data collection strictly adhered to principles of research ethics and privacy protection, with all data undergoing anonymization and encryption prior to experimentation. The preprocessing stage involved data cleaning to eliminate anomalies, missing values, or errors arising from equipment malfunctions or transmission issues, followed by data standardization to ensure uniformity in analysis across different sources and formats.

During the data preparation phase, the collected dataset was partitioned into training and testing sets. To comprehensively assess the model’s performance across different data volumes, three levels of data volume were set: 100, 1000, and 10,000, representing small, medium, and large-scale datasets, respectively. This setup aimed to simulate various data volume scenarios encountered in real-world applications, thus enabling a more comprehensive evaluation of the model’s robustness and generalization capabilities.

Performance of situation awareness model for UAV groups communication network based on adversarial networks under different data volumes

This experiment aims to verify the feasibility of a situational awareness model for UAV communication networks based on adversarial networks, and explore the impact of different data volumes on model training.

Specific steps: Firstly, it is necessary to collect data from the UAV communication network. Each data includes communication information between UAV, UAV position, speed, angle, and other indicators; the collected data is divided into training and testing sets, and proportions are set based on the amount of data (100, 1000, and 10,000); the data from the training set is subjected to data cleaning and preprocessing operations, such as removing missing values, standardizing data, etc.; next, the situational awareness model of UAV communication network based on adversarial networks is used for training; after the training is completed, the model is tested on the test set and various performance indicators are calculated, such as the model’s recall rate, accuracy, F1 value, etc. In order to ensure the accuracy of the experiment, this article conducted eight tests under different data volumes.

Required materials: UAV communication network data, which can be collected through simulation or actual scenarios; software tools include Python, Tensorflow, etc.

Figure 3 shows the performance of the UAV groups communication network situational awareness model for adversarial networks under different data volumes. Figure 3(a) shows 100 data volumes. Figure 3(b) shows 1000 data volumes, and Figure 3(c) shows 10,000 data volumes.

Figure 3.

Performance of UAV communication network situational awareness model in adversarial networks under different data volumes: (a) shows 100 pieces of data, (b) shows 1000 pieces of data, and (c) shows 10,000 pieces of data.

In Figure 3, the meanings of Accuracy, F1 score, and Recall rate are as follows:

• (1) Accuracy: This measures the degree of consistency between model predictions and actual outcomes. In classification tasks, accuracy is the ratio of correctly classified samples to the total number of samples. For example, if a model correctly classifies 90 out of 100 samples, its accuracy is 90%.

• (2) F1 score: This is a metric that combines precision and recall to evaluate the performance of a classification model. The F1 score is the harmonic mean of precision and recall. It measures the model’s performance in balancing precision and recall. Precision refers to the proportion of true positive samples among the samples predicted as positive by the model, while recall refers to the proportion of true positive samples among all actual positive samples. A higher F1 score indicates better performance in both precision and recall.

• (3) Recall rate: Recall rate measures the proportion of correctly predicted positive examples among all true positive examples. It can also be understood as the proportion of recognized positive examples out of all true positive examples. In a binary classification model, where there are P true positive samples and N negative samples, if the model incorrectly predicts some positive samples as negative and correctly predicts others as positive, with TP representing true positives and FN representing false negatives, the recall rate can be defined as: Recall = TP/(TP + FN). Recall rate evaluates the model’s ability to identify true positive examples.

Among them, the UAV groups communication network situational awareness model based on adversarial networks had an accuracy of 0.955, an F1 value of 0.940, and a recall rate of 0.937 for 100 data volumes; at a data volume of 1000, its accuracy was 0.962, F1 value was 0.960, and recall rate was 0.949; at a data volume of 10,000, its accuracy was 0.982, F1 value was 0.981, and recall rate was 0.975. It could be seen that as the amount of data increased, the accuracy, F1 value, and recall rate of the model all improved. This meant that in practical applications, collecting more data could significantly improve the accuracy of the UAV groups communication network situational awareness model.

Performance of situation awareness model for UAV groups communication network based on adversarial networks in different noise environments

The purpose of this experiment is to investigate the robustness and accuracy of a situation awareness model for UAV communication networks based on adversarial networks in noisy environments.

Specific steps: Firstly, in different noise environments such as electromagnetic interference and signal interference, a communication network composed of UAV is used, and data is collected; the collected data is cleaned and preprocessed, such as removing missing values, standardizing data, etc.; the data is divided into training and testing sets, and the same training set is used to train the model; on a dataset in a noisy environment, noise reduction operations are carried out, such as filtering, denoising, enhancing vibration, etc.; Next, the UAV groups communication network situational awareness model based on adversarial networks is used for training and tested on the test set; the performance indicators such as error rate, accuracy, and robustness of the model in various noise environments are calculated. In order to ensure the accuracy of the experiment, this article conducted eight tests under different signal interferences.

Required materials: UAV communication network data, which can be collected through simulation or actual scenarios; experimental equipment such as noise generators and filters; software tools such as Python, Tensorflow, etc.

Figure 4 shows the performance of the UAV groups communication network situational awareness model in different noise environments for adversarial networks. Figure 4(a) shows electromagnetic interference, and Figure 4(b) shows signal interference.

Figure 4.

Performance of UAV groups communication network situational awareness model in different noise environments for adversarial networks: (a) shows electromagnetic interference and (b) shows signal interference.

The situation awareness model of UAV communication network based on adversarial network had an error rate of 2.5% under electromagnetic interference. The accuracy was 97.5%, and the robustness was 92.3%; the error rate was 3.1% under signal interference. The accuracy was 96.9%, and the robustness was 89.9%. It could be seen that there were certain differences in the performance of the model under different noise environments. The impact of electromagnetic interference on the model was smaller than signal interference, and the accuracy and robustness of the model were improved. In practical applications, effective noise suppression could improve the accuracy and stability of the situational awareness model for UAV communication networks.

Performance of situation awareness model for UAV groups communication network based on adversarial networks under different model structures

The purpose of this experiment is to investigate the impact of different model structures on the situational awareness model of UAV communication network based on adversarial networks.

Specific steps: The situational awareness model for UAV communication network based on adversarial networks is designed. Two model structures, convolutional neural network and recurrent neural network, were designed separately; the same dataset was used to train these two models and tested on the test set; the performance indicators such as training time, accuracy, and stability of different models are compared. In order to ensure the accuracy of the experiment, eight tests were conducted under different model structures in this article.

Required materials: UAV communication network data, which can be collected through simulation or actual scenarios; software tools such as Python, Tensorflow, etc.

Figure 5 shows the performance of the UAV groups communication network situational awareness model in different model structures for adversarial networks. Figure 5(a) shows a convolutional neural network, and Figure 5(b) shows a recurrent neural network.

Figure 5.

Performance of UAV groups communication network situational awareness model in different model structures for adversarial networks: (a) shows a convolutional neural network and (b) shows a recurrent neural network.

The situation awareness model of UAV groups communication network based on adversarial network was trained with convolutional neural network model, and its training time was 14.8 s. The accuracy was 96.3%, and the stability was 90.2%; under the training of the recurrent neural network model, the training time was 16.9 s. The accuracy was 97.3%, and the stability was 92.1%. It could be seen that there were certain differences in the training time, accuracy, and stability of models under different model structures. Recurrent neural networks had higher accuracy. However, training time was longer and stability was better. In practical applications, selecting a reasonable model structure could improve the performance of the situational awareness model for UAV communication networks.

Performance of situation awareness model for UAV groups communication network based on adversarial networks at different network sizes

This experiment aims to investigate the impact of different scale networks on the situational awareness model of UAV communication networks based on adversarial networks.

Specific steps: A situational awareness model for UAV communication network based on adversarial networks is designed and applied to networks of different sizes, including 10 nodes, 100 nodes, and 1000 nodes; the same dataset is used to train these networks and test them on the test set; the performance indicators such as computational complexity, accuracy, and latency of the model were calculated, and the performance of networks of different scales was compared. In order to ensure the accuracy of the experiment, this article conducted eight tests on different network scales.

Required materials: UAV communication network data, which can be collected through simulation or actual scenarios; software tools such as Python, Tensorflow, etc.

Table 1 shows the performance of the UAV communication network situational awareness model for adversarial networks with a network size of 10 nodes. Table 2 shows the performance of the UAV groups communication network situational awareness model for adversarial networks with a network size of 100 nodes. Table 3 shows the performance of the UAV communication network situational awareness model for adversarial networks with a network size of 1000 nodes.

Table 1.

Performance of the UAV communication network situational awareness model for adversarial networks with a network size of 10 nodes.

Number of tests	Calculation complexity (s)	Accuracy	Delay (s)
1	2.3	0.944	2
2	2	0.947	1.8
3	2.2	0.949	2
4	2.1	0.943	1.9
5	2.5	0.946	2.1
6	2.5	0.945	2.4
7	2.3	0.943	2.3
8	2	0.948	1.8
Average	2.2	0.946	2.0

Table 2.

Performance of the UAV communication network situational awareness model for adversarial networks with a network size of 100 nodes.

Number of tests	Calculation complexity (s)	Accuracy	Delay (s)
1	13	0.965	8.6
2	13.3	0.965	8.5
3	13.4	0.965	8.8
4	13	0.966	8.7
5	13.1	0.962	8.5
6	12.8	0.965	8.4
7	12.8	0.963	8.4
8	13.3	0.962	8.8
Average	13.1	0.964	8.6

Table 3.

Performance of UAV communication network situational awareness model in adversarial networks with a network size of 1000 nodes.

Number of tests	Calculation complexity (s)	Accuracy	Delay (s)
1	78.8	0.984	45
2	78.8	0.982	45.1
3	78.7	0.98	44.7
4	78.6	0.983	45.1
5	79.1	0.982	45.2
6	79.1	0.979	45.1
7	79	0.982	45
8	79.1	0.984	45
Average	78.9	0.982	45.0

The situation awareness model of UAV communication network based on adversarial network had a computational complexity of 2.2 s, an accuracy of 0.946, and a delay of 2.0 s with a network size of 10 nodes; under a network size of 100 nodes, its computational complexity was 13.1 s, accuracy was 0.964, and latency was 8.6 s; at a network size of 1000 nodes, its computational complexity was 78.9 s, accuracy was 0.982, and latency was 45.0 s. It could be seen that as the network size increased, the computational complexity and latency of the model increased, while the accuracy improved. This meant that in practical applications, high-performance computing devices and optimization algorithms could improve the performance of UAV communication network situational awareness models.

Conclusions

The situation awareness of UAV groups communication network is an important technology for UAV groups cooperative operations, which can effectively improve the efficiency and accuracy of UAV cooperative operations. The situational awareness model of UAV communication network based on adversarial networks is a new method. By training the generator and discriminator, the generator can generate more realistic network data, thereby improving the accuracy of the model. This article explored the performance of the model under different data volumes, noise environments, model structure, and network scale through experimental design. The experimental results showed that as the amount of data increases, the accuracy, F1 value, and recall rate of the model all improved and tended to stabilize. This also conforms to the rule of “data is king” in deep learning. More data can help the model learn and predict better. However, it should be noted that an increase in data volume also means an increase in computational complexity, requiring the use of high-performance computing devices. The impact of noise environment on the model varies, and the impact of electromagnetic interference is smaller than that of signal interference. This is also consistent with the actual situation. In wireless communication, electromagnetic interference can be offset by increasing antenna gain and other methods. Signal interference is more likely to cause communication noise, thereby affecting communication quality. The recurrent neural network model has higher accuracy, but longer training time and better stability. This indicates that recurrent neural networks have certain advantages in processing sequence data. However, in practical applications, choices need to be made based on specific circumstances and needs. If time is tight or the amount of data is small, other types of neural networks can be considered. As the network size increases, the computational complexity and latency of the model both increase, while the accuracy improves. This also conforms to the relationship between “layers and accuracy” in deep learning. Increasing the number of network layers can enable the model to obtain more feature information, thereby improving the accuracy of the model. However, it also increases computational complexity and latency. In practical applications, effective noise suppression, reasonable selection of model structure, use of high-performance computing equipment, and optimization algorithms can all improve the performance of UAV communication network situational awareness models.