Heterogeneous link communication method and communication device for unmanned area inspection

Abstract

The application relates to the technical field of communication, and discloses a heterogeneous link communication method and device for unmanned area inspection, which comprises collecting real-time performance parameters, inputting the performance parameters into a prediction network 1, outputting an environment interference level perceived by a link type, and qualitatively indicating the link type interfered; if the interference level exceeds a threshold value, a to-be-optimized link subset is screened out according to the indicated link type; the performance parameters of the to-be-optimized link subset are input into a prediction network 2, and a normalized link reliability score of 0-1 is mapped; the link with the highest reliability score is selected as a target communication link, and if the communication delay of the target communication link exceeds a service threshold value, the modulation mode and the transmission power of the target communication link are adjusted in cooperation. The method and device improve the communication continuity and reliability, and improve the optimization efficiency when the communication quality deteriorates.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a heterogeneous link communication method and communication device for unmanned area inspection. Background Technology

[0002] In uninhabited area inspection missions, the reliability of communication links is the lifeline for ensuring the safe operation of inspection equipment and successful data transmission. The complex and ever-changing environment of uninhabited areas, with terrain obstructions and severe weather causing continuous and unpredictable interference to communication links, presents a challenge. To address this challenge, heterogeneous link communication technology has emerged. It integrates multiple communication media such as satellite, public terrestrial wireless networks, and private terrestrial wireless networks, utilizing the redundancy between links to enhance the robustness of the overall communication system.

[0003] However, existing heterogeneous link communication still has significant limitations when facing the special needs of unmanned area inspection, mainly in the following aspects:

[0004] First, it relies on a passive response mechanism based on fixed thresholds; for example, link switching is only triggered when the link's packet loss rate, latency, and other performance parameters deteriorate to a preset fault threshold. This post-event remediation strategy has obvious lag and cannot provide early warning and intervention in the early stages of a gradual decline in link stability, resulting in a persistently high risk of communication interruption and seriously threatening the continuity of inspection tasks.

[0005] Secondly, when the system detects a performance degradation, it is often difficult to determine whether it is caused by equipment failure, random fluctuations, or specific external environmental interference. Due to the lack of accurate diagnosis of the root cause and scope of the interference, optimization strategies (such as link switching and parameter adjustment) are often blind and cannot concentrate limited computing and communication resources on the subset of links that need the most optimization, resulting in low efficiency. Summary of the Invention

[0006] Therefore, the purpose of this invention is to overcome the lag and blindness of existing heterogeneous link communication methods, and to provide a heterogeneous link communication method and device for unmanned area inspection, which improves communication continuity and reliability, and improves optimization efficiency when communication quality deteriorates.

[0007] Firstly, to address the aforementioned technical problems, this invention provides a heterogeneous link communication method for unmanned area inspection, comprising:

[0008] The system acquires real-time performance parameters of multiple heterogeneous communication links, including communication latency, data packet loss rate, signal strength, link bit error rate, and heartbeat response time.

[0009] The real-time performance parameters are input into prediction network one, and the real-time performance parameters are classified into link type-aware environmental interference levels; the link type-aware environmental interference levels are used to indicate one or more communication link types affected by environmental changes.

[0010] In response to the environmental interference level exceeding a threshold level, a subset of heterogeneous links to be optimized is selected from the plurality of heterogeneous communication links according to the communication link type indicated therein;

[0011] The real-time performance parameters of the heterogeneous link subset are input into prediction network two, and the real-time performance parameters are mapped to scores that characterize the reliability of the communication link to obtain the link quality index.

[0012] The target communication link is determined based on the link quality index, and the communication parameters of the target communication link are adjusted in response to the delay of the target communication link exceeding the threshold delay.

[0013] Preferably, the prediction network comprises: an input layer that receives a normalized feature vector composed of communication delay, data packet loss rate, signal strength, link bit error rate, and heartbeat packet response time; a radial basis function kernel layer that nonlinearly maps the normalized feature vector to a high-dimensional feature space so that data corresponding to different environmental interference levels are linearly separable in the high-dimensional feature space; a weighted decision layer that constructs a classification hyperplane based on support vectors and Lagrange multipliers in the high-dimensional feature space and calculates the classification result; and an output layer that maps the classification result of the weighted decision layer to the environmental interference level.

[0014] Preferably, the output layer maps the environmental interference level as a composite semantic label, and the composite semantic label uniquely corresponds to a combination of environmental interference level and affected link type.

[0015] Preferably, the process of selecting a subset of heterogeneous links to be optimized based on the affected link types indicated by the environmental interference level includes: predefining an interference-link mapping table, wherein the mapping table defines the correspondence between different environmental interference levels and one or more communication link types to be optimized; querying the interference-link mapping table based on the environmental interference level output by the prediction network to obtain the target link type to be optimized; and selecting all communication links belonging to the target link type among the multiple heterogeneous communication links to obtain the subset of heterogeneous links.

[0016] Preferably, the environmental interference levels perceived by the link type include at least: a first level indicating that the satellite link is disturbed; a second level indicating that the terrestrial wireless link is disturbed; and a third level indicating that both the satellite link and the terrestrial wireless link are disturbed.

[0017] Preferably, the prediction network two is a multilayer perceptron model, and its network structure includes: an input layer for receiving a normalized feature vector composed of the real-time performance parameters of each link in the heterogeneous link subset; a feature extraction module, which is composed of a first fully connected layer, a ReLU activation function, a Dropout layer, and a second fully connected layer connected sequentially; wherein the number of neurons in the second fully connected layer is less than that in the first fully connected layer, forming a feature encoder for feature compression and abstraction; and a regression output layer for mapping the high-order features output by the feature extraction module to a normalized reliability score in the range of 0 to 1.

[0018] Preferably, the number of neurons in the first fully connected layer and the second fully connected layer are both related to the number of communication links and the number of single-link performance parameters contained in the heterogeneous link subset.

[0019] Preferably, determining the target communication link based on the link quality index includes: selecting the communication link with the highest link quality index from the heterogeneous link subset as the target communication link.

[0020] Preferably, the communication parameters include modulation scheme and transmit power; triggering the adjustment of the communication parameters of the target communication link includes: when the link quality index decreases and triggers the adjustment of communication parameters, reducing the modulation order and increasing the transmit power after reducing the modulation order; when the link quality index increases and triggers the adjustment of communication parameters, increasing the transmit power and increasing the modulation order after increasing the transmit power.

[0021] Secondly, in order to solve the above-mentioned technical problems, the present invention also provides a heterogeneous link communication device for unmanned area inspection, including: a data processing center and several inspection devices;

[0022] The inspection equipment is used to perform inspection tasks in uninhabited areas and send the data recorded during the inspection process to the data processing center.

[0023] The data processing center is used to acquire real-time performance parameters of multiple heterogeneous communication links communicating with the inspection equipment. These performance parameters include communication latency, data packet loss rate, signal strength, link bit error rate, and heartbeat response time. The acquired real-time performance parameters are categorized into link-type-aware environmental interference levels. When the environmental interference level exceeds a threshold level, a subset of heterogeneous links to be optimized is selected from the multiple heterogeneous communication links based on the communication link type indicated by the environmental interference level. The real-time performance parameters of the heterogeneous link subset are mapped to scores characterizing the reliability of the communication links. A target communication link is determined based on the reliability score. When the latency of the target communication link exceeds a threshold latency, the communication parameters of the target communication link are adjusted.

[0024] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0025] The heterogeneous link communication method and communication device for unmanned area inspection described in this invention improves communication continuity and reliability, and enhances optimization efficiency when communication quality deteriorates.

[0026] Among them, the predictive network classifies the real-time performance parameters of the communication links available to the inspection equipment into environmental interference levels. Before the micro-performance parameters of the communication links deteriorate and accumulate into macro-communication failures, it accurately identifies and characterizes the stress effect of the external environment on specific link types, thereby greatly reducing the probability of communication interruption and effectively ensuring the continuity and reliability of inspection tasks in complex uninhabited environments, achieving near-zero interruption continuous and stable communication.

[0027] Once a decline in communication quality is detected, traditional methods either blindly switch communication links or make global parameter adjustments to all links, which leads to resource waste and low optimization efficiency.

[0028] In this embodiment of the invention, prediction network one classifies the real-time performance parameters of the communication link into an environmental interference level that is perceived by the link type. That is, it identifies in advance the impact of environmental changes on the communication link and the types of communication links affected from the performance parameters of the communication link. Based on the types of communication links affected indicated by the environmental interference level, it selects a subset of heterogeneous links to be optimized, thereby determining the root cause and main scope of impact of the communication link performance degradation. Then, combined with the quantitative evaluation of prediction network two, it locks the target communication link in the heterogeneous link subset and performs targeted parameter adjustment, avoiding the inefficiency caused by traditional global optimization. Attached Figure Description

[0029] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0030] Figure 1 This is a flowchart of a heterogeneous link communication method for uninhabited area inspection in a preferred embodiment of the present invention;

[0031] Figure 2 This is a block diagram of the prediction network in a preferred embodiment of the present invention;

[0032] Figure 3 This is a flowchart of obtaining a heterogeneous link subset in a preferred embodiment of the present invention;

[0033] Figure 4 This is a block diagram of the prediction network two in a preferred embodiment of the present invention;

[0034] Figure 5This is a structural block diagram of heterogeneous link communication for unmanned area inspection in a preferred embodiment of the present invention. Detailed Implementation

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0036] The purpose of this invention is to overcome the lag and blindness of existing heterogeneous link communication methods, and to provide a heterogeneous link communication method and device for unmanned area inspection, which improves communication continuity and reliability, and improves optimization efficiency when communication quality deteriorates.

[0037] Example 1: Refer to Figure 1 As shown, this embodiment of the invention discloses a heterogeneous link communication method for unmanned area inspection, including:

[0038] S100: Obtain real-time performance parameters of multiple heterogeneous communication links; performance parameters include communication latency, data packet loss rate, signal strength, link bit error rate, and heartbeat response time;

[0039] S200. Input the real-time performance parameters into the prediction network and classify the real-time performance parameters into link type-aware environmental interference levels; the link type-aware environmental interference levels are used to indicate one or more communication link types affected by environmental changes.

[0040] S300, in response to environmental interference levels exceeding a threshold level, selects a subset of heterogeneous links to be optimized from multiple heterogeneous communication links based on the type of communication link indicated therein;

[0041] S400: Input the real-time performance parameters of the heterogeneous link subset into the prediction network 2, map the real-time performance parameters into scores that characterize the reliability of the communication link, and obtain the link quality index.

[0042] S500: Determine the target communication link based on the link quality indicators, and in response to the target communication link's delay exceeding the threshold delay, trigger the adjustment of the target communication link's communication parameters.

[0043] In specific application scenarios, inspection equipment, including drones and ground robots, periodically reads underlying performance data from multiple heterogeneous communication interfaces such as satellite communication modules, 4G / 5G public network modules, and wireless private network modules. The real-time performance parameters collected include at least communication latency, data packet loss rate, signal strength, link error rate, and heartbeat packet response time.

[0044] Communication latency represents the end-to-end time required for a data packet to travel from the sender to the receiver, reflecting the combined impact of signal transmission path length, intermediate node queuing, and channel congestion; it is used to assess the link's real-time performance and availability in supporting instant services. Packet loss rate represents the proportion of data packets lost during transmission out of the total transmitted packets, primarily caused by channel congestion, signal attenuation, or severe jitter; it is used to assess the link's stability and continuity. Signal strength represents the power of the signal received at the receiver, directly reflecting the path loss of the wireless signal in space propagation, and is greatly affected by factors such as distance, obstruction, and weather; it is used to assess the link's physical connection foundation and potential reliability. Link bit error rate represents the ratio of erroneous bits to the total number of bits after physical layer transmission and modulation / demodulation, directly reflecting the severity of the wireless channel's impact from noise, interference, and multipath effects; it is used to assess the link's original transmission quality and channel cleanliness. Heartbeat response time represents the round-trip delay of small data packets (heartbeat packets) periodically sent by the device to the control center; it is used to assess the link's connection survivability and basic reachability. By acquiring communication latency, data packet loss rate, signal strength, link error rate, and heartbeat response time, a comprehensive and multi-dimensional real-time perception of the communication link status can be achieved.

[0045] After cleaning and normalizing the acquired performance parameters, they are input into the pre-trained prediction network 1. The prediction network 1 is trained on a support vector machine model based on historical data. It can identify the performance parameter degradation patterns caused by specific environmental interference (such as rain attenuation and terrain occlusion) and output the link type-aware environmental interference level. This level is not excellent, good, or poor, but directly contains a composite label of the affected link type. For example: satellite link - severe interference; terrestrial wireless link - mild interference.

[0046] The environmental interference level output in step S200 is compared with a preset threshold level (e.g., moderate interference or above). If the threshold is exceeded, affected communication links are selected from all available links based on the affected link type indicated in the environmental interference level, forming a subset of heterogeneous links to be optimized. For example, if the environmental interference level is "satellite links are disturbed," all satellite links are included in the subset, while ground links are not included in this optimization. This step avoids initiating unnecessary and resource-intensive deep optimization processes when communication links are in good condition. At the same time, it narrows the optimization target from all links to the problematic link set, greatly improving optimization efficiency and fundamentally solving the problem of blind optimization.

[0047] The real-time performance parameters of a subset of heterogeneous links are input into a pre-trained prediction network II. This network II performs deep fusion and abstraction of the real-time performance parameters through multi-layer nonlinear transformations, ultimately outputting a normalized score between 0 and 1 as the link quality indicator for each communication link. This allows the reliability of satellite links and 4G links to be compared on the same level, providing an objective and quantitative basis for selecting the optimal communication path and avoiding the risks of subjective or biased decision-making.

[0048] Based on the quantification score in step S400, the link with the highest link quality index is selected as the target communication link for the current highest priority business data, and the communication latency of the target link is continuously monitored. If it exceeds the service tolerance threshold latency, parameter adjustment is triggered.

[0049] The heterogeneous link communication method for unmanned area inspection of the present invention improves communication continuity and reliability, and enhances optimization efficiency when communication quality deteriorates.

[0050] Among them, the predictive network classifies the real-time performance parameters of the communication links available to the inspection equipment into environmental interference levels. Before the micro-performance parameters of the communication links deteriorate and accumulate into macro-communication failures, it accurately identifies and characterizes the stress effect of the external environment on specific link types, thereby greatly reducing the probability of communication interruption and effectively ensuring the continuity and reliability of inspection tasks in complex uninhabited environments, achieving near-zero interruption continuous and stable communication.

[0051] Once a decline in communication quality is detected, traditional methods either blindly switch communication links or make global parameter adjustments to all links, which leads to resource waste and low optimization efficiency.

[0052] In this embodiment of the invention, prediction network one classifies the real-time performance parameters of the communication link into an environmental interference level that is perceived by the link type. That is, it identifies in advance the impact of environmental changes on the communication link and the types of communication links affected from the performance parameters of the communication link. Based on the types of communication links affected indicated by the environmental interference level, it selects a subset of heterogeneous links to be optimized, thereby determining the root cause and main scope of impact of the communication link performance degradation. Then, combined with the quantitative evaluation of prediction network two, it locks the target communication link in the heterogeneous link subset and performs targeted parameter adjustment, avoiding the inefficiency caused by traditional global optimization.

[0053] Specifically, refer to Figure 2As shown, the prediction network includes: an input layer that receives a normalized feature vector composed of communication delay, data packet loss rate, signal strength, link bit error rate, and heartbeat packet response time; a radial basis kernel function layer that nonlinearly maps the normalized feature vector to a high-dimensional feature space so that data corresponding to different environmental interference levels are linearly separable in this high-dimensional feature space; a weighted decision layer that constructs a classification hyperplane based on support vectors and Lagrange multipliers in the high-dimensional feature space and calculates the classification result; and an output layer that maps the classification result of the weighted decision layer to the environmental interference level. The environmental interference level mapped by the output layer is a composite semantic label, and each composite semantic label uniquely corresponds to a combination of environmental interference level and affected link type.

[0054] In specific application scenarios, the input layer is configured to receive a five-dimensional input vector, which is composed of five performance parameters: communication latency, data packet loss rate, signal strength, link bit error rate, and heartbeat packet response time, after being normalized and preprocessed using the Z-Score normalization method.

[0055] The radial basis function kernel layer uses radial basis functions as kernel functions, and its mathematical expression is:

[0056] X is the input five-dimensional feature vector; i represents the index of the support vector; X i It is the i-th support vector learned by the model during training; exp represents the exponential function, that is, the exponential operation with the natural constant e (approximately equal to 2.71828) as the base; It is a hyperparameter that controls the radial range of the function; This represents the Euclidean distance between the input vector and the support vectors. This kernel function nonlinearly maps the originally linearly inseparable five-dimensional feature data to a higher-dimensional feature space by calculating the similarity between the input vector and all support vectors in a high-dimensional space. The radial basis function (RBF) kernel has powerful nonlinear mapping capabilities, enabling it to capture the complex, nonlinear interactions between the five performance parameters: in this high-dimensional feature space, data points corresponding to different levels of environmental interference, which were originally intertwined in the five-dimensional space, become linearly separable.

[0057] The weighted decision layer integrates the weighted contributions of all support vectors, constructing an optimal classification hyperplane in the high-dimensional feature space. It performs weighted voting on the typical interference models represented by all support vectors, making the model's decision robust, insensitive to atypical or isolated noisy data, and able to output stable classification results based on the most dominant feature patterns. The output layer maps the classification results of the weighted decision layer to environmental interference level labels through a multi-classification strategy.

[0058] Through the collaboration of the above four layers, the predictive network 1 aggregates five low-level, heterogeneous performance parameters into a high-level, qualitative, and type-information intelligent judgment on the interference in the communication link's environment, thereby achieving early, accurate, and root-cause prediction of the decline trend in link stability.

[0059] The training objective of Prediction Network 1 is to enable it to learn from historical data the interference patterns left on performance parameters by different environmental disturbances, thereby establishing a reliable classification model. Its training process includes the following core steps:

[0060] Building the training dataset: A large number of data samples are systematically collected from historical inspection tasks. Each sample is a data pair, including input features and output labels. The input features are a vector of performance parameters collected at a specific moment, consisting of communication latency, data packet loss rate, signal strength, link error rate, and heartbeat packet response time. The output label is the manually labeled environmental interference level corresponding to that moment. Label sources include: Multi-source information fusion labeling: Cross-validation and analysis are performed using task logs (such as geographical location and equipment attitude), publicly available meteorological data, and geographic information system data at that moment, followed by post-processing labeling by domain experts or automated rules. Sensor-assisted labeling: Automatic labeling is performed in conjunction with data from other sensors on the equipment (such as high-definition cameras for identifying rain and snow, and inertial measurement units for identifying body occlusion). Example labels: Labels are link-type-aware composite semantic labels. For example: ① Weak signal strength and increased bit error rate occur simultaneously, corresponding to physical satellite link obstruction interference; ② Slowly decreasing signal strength accompanied by a significant increase in latency, corresponding to strong convective weather interference on satellite links; ③ Severe fluctuations in signal strength and intermittent spikes in packet loss rate, corresponding to multipath fading interference on terrestrial wireless links; ④ A surge in heartbeat packet response time but relatively stable signal strength and bit error rate, corresponding to network congestion interference on terrestrial wireless links; ⑤ A significant and synchronous increase in bit error rate across all links, corresponding to broadband electromagnetic interference on all links; ⑥ Normal signal strength but persistently high bit error rate, corresponding to external co-channel interference on satellite links; ⑦ A slow and synchronous increase in heartbeat packet response time and communication latency, corresponding to exhaustion of processing resources of all link devices.

[0061] Model selection and hyperparameter setting: Support vector machine is selected as the base model, and its kernel function is determined to be the radial basis function kernel. Key hyperparameters for model training are defined, including the penalty coefficient C and the radial basis function parameters.

[0062] Model training and parameter optimization: The preprocessed training dataset is input into the support vector machine algorithm. The algorithm solves a convex quadratic programming problem to find the optimal classification hyperplane in the feature space that can separate data points of different classes with the maximum margin. In this process, the algorithm automatically determines: support vectors: those most critical training samples that are located on the classification margin or misclassified; Lagrange multipliers: the weights corresponding to each support vector, representing the importance of the vector to the decision boundary.

[0063] A strategy combining K-fold cross-validation and grid search is used to optimize the hyperparameters (C, γ): the training data is randomly divided into K parts, one part is used as the validation set, and the remaining K-1 parts are used as the training set. Training and validation are carried out under different combinations of (C, γ) parameters. Finally, the hyperparameters with the highest average classification accuracy in K validations are selected as the configuration of the final model.

[0064] Model Validation and Deployment: A reserved test dataset, unused during training and hyperparameter optimization, is used to evaluate the performance of the final model. Evaluation metrics include overall classification accuracy, precision, and recall for each category, to ensure the model has good generalization ability. The trained and validated model parameters are then archived into a model file, integrated, and deployed into the communication management system of the inspection equipment for real-time inference.

[0065] Based on the above embodiments, the environmental interference levels for link type awareness include at least: a first level indicating that the satellite link is disturbed; a second level indicating that the terrestrial wireless link is disturbed; and a third level indicating that both the satellite link and the terrestrial wireless link are disturbed.

[0066] In specific application scenarios, the first level is the prediction network that identifies performance parameters exhibiting typical satellite link disturbance patterns, for example:

[0067] Mode A (physical obstruction): Signal strength and bit error rate deteriorate drastically at the same time, while ground link parameters remain relatively stable.

[0068] Mode B (Weather Impact): Signal strength exhibits a slow and continuous decline, accompanied by a significant increase in communication latency (due to error correction and retransmission), with no significant change in the ground link.

[0069] The second level is to predict when the network identifies performance parameters exhibiting typical terrestrial wireless link disruption patterns, such as:

[0070] Mode C (Multipath Fading): Signal strength fluctuates drastically and rapidly, while data packet loss rate spikes intermittently, and satellite link parameters remain stable.

[0071] Mode D (Network Congestion): Heartbeat response time and communication delay increase significantly, but signal strength and bit error rate are not abnormal. This mode is different from physical channel impairment.

[0072] The third level is when the prediction network identifies a globally perturbed pattern in the performance parameters, for example:

[0073] Mode E (Strong Electromagnetic Interference): The bit error rate increases significantly and synchronously across all link types, and the signal strength may be normal or affected inconsistently.

[0074] Mode F (Extreme Weather): Under conditions such as torrential rain and sandstorms, multiple indicators such as signal strength, bit error rate, and latency of satellite and ground wireless links (especially private networks for line-of-sight transmission) deteriorate simultaneously and comprehensively.

[0075] Mode G (Device Failure): Such as overload of the main processor, causing abnormal synchronization of heartbeat packet response time across all links.

[0076] Based on the above embodiments, referring to Figure 3 As shown, the process involves filtering out a subset of heterogeneous links to be optimized based on the affected link types indicated by the environmental interference level. This includes: a predefined interference-link mapping table, which defines the correspondence between different environmental interference levels and one or more communication link types to be optimized; querying the interference-link mapping table based on the predicted environmental interference level output by the network to obtain the target link type to be optimized; and filtering out all communication links belonging to the target link type from multiple heterogeneous communication links to obtain a subset of heterogeneous links.

[0077] In specific application scenarios, during system initialization, a key-value pair mapping table is created and stored in memory. This table uses the environmental interference level as the key and the set of target link types to be optimized as the value. The mapping relationship is defined as follows: based on knowledge in the field of communication, the precise mapping logic is predefined, for example:

[0078] Key: Satellite link disrupted → Value: Inmarsat, Tiantong satellite;

[0079] Key: Terrestrial wireless link disrupted → Value: 4G, 5G, private wireless network;

[0080] Key: Comprehensive harsh environment → Value: Maritime satellite, Tiantong satellite, 4G, 5G, private wireless network.

[0081] By modifying this table, different inspection equipment (which may have different types of communication modules) or new communication technologies can be adapted without changing the core algorithm, thus improving the flexibility and versatility of the solution.

[0082] The system receives a specific environmental interference level (e.g., the string "terrestrial wireless link is disturbed") from the prediction network. Using this environmental interference level as the query key, the system performs a lookup operation in a predefined interference-link mapping table to obtain and return the set of target link types (e.g., 4G, 5G, private wireless network) corresponding to the key.

[0083] The system traverses all available and active heterogeneous communication links, compares the actual type of each link with the target link type set obtained in the previous step, filters out all links that match the type, and uses these filtered links to form a subset of heterogeneous links to be optimized. This subset is then passed as input to the subsequent prediction network 2 for in-depth evaluation.

[0084] Based on the above embodiments, referring to Figure 4 As shown, prediction network 2 is a multilayer perceptron model. Its network structure includes: an input layer, which receives a normalized feature vector composed of the real-time performance parameters of each link in the heterogeneous link subset; a feature extraction module, which consists of a first fully connected layer, a ReLU activation function, a Dropout layer, and a second fully connected layer connected sequentially; wherein, the second fully connected layer has fewer neurons than the first fully connected layer, forming a feature encoder for feature compression and abstraction; and a regression output layer, which maps the high-order features output by the feature extraction module to a normalized reliability score in the range of 0 to 1.

[0085] In specific application scenarios, the first fully connected layer extensively captures the pairwise interactions and complex nonlinear relationships that may exist between all input performance parameters. The ReLU activation function is applied to the output of the first fully connected layer, introducing a nonlinear transformation that enables the network to learn and fit complex mapping functions. The dropout rate of the Dropout layer is set to 0.2. During training, the Dropout layer randomly discards 20% of the neurons in the output of the first fully connected layer as a regularization to prevent excessive co-adaptation between neurons. The second fully connected layer has fewer neurons than the first fully connected layer, forming a bottleneck structure together with the previous layer. As a narrow layer, it compresses and refines the high-dimensional features of the previous layer's output, forcing the network to retain the core information most relevant to reliability assessment and filtering out redundant noise. The regression output layer consists of a fully connected layer with an output dimension of 1 and a cascaded Sigmoid activation function. The Sigmoid activation function compresses and maps any scalar value of the previous layer's output to the interval (0, 1).

[0086] Furthermore, the number of neurons in both the first and second fully connected layers is related to the number of communication links and the number of single-link performance parameters contained in the heterogeneous link subset.

[0087] In specific application scenarios, the width of the first fully connected layer and the second fully connected layer should be matched with the complexity of the input data and the scale of the problem to be analyzed. The complexity of the input data is determined by the number of communication links contained in the heterogeneous link subset and the number of single link performance parameters.

[0088] Specifically, the number of neurons N1 in the first fully connected layer is determined based on the following method: N1 represents the number of neurons in the first fully connected layer; N0 represents the number of basic neurons, with a value of 64 or 120. represents the scaling factor, which controls the rate at which the number of neurons grows with the problem size; in one scheme, it can be 16; P represents the number of single-link performance parameters; K represents the number of communication links contained in the heterogeneous link subset. This represents the floor function.

[0089] The number of neurons N2 in the second fully connected layer is determined based on the following method: ;

[0090] The number of neurons in the second fully connected layer is directly determined by the number of neurons in the first fully connected layer, and a fixed compression ratio is maintained to obtain a stable feature abstraction and dimensionality reduction ratio.

[0091] Based on the above embodiments, the target communication link is determined according to the link quality index, including: selecting the communication link with the highest link quality index from the heterogeneous link subset as the target communication link. Since the link quality index is a deep evaluation result of the comprehensive reliability of the two-way link prediction network (comprehensively considering multiple dimensions such as latency, packet loss, strength, and bit error rate), selecting the highest score means selecting the most stable, robust, and least likely to be interrupted connection, guiding communication traffic to the most reliable path with the highest efficiency and determinism. This directly maximizes the probability of successful transmission on the first attempt, providing the highest level of communication guarantee for the inspection task.

[0092] Based on the above embodiments, the communication parameters include modulation scheme and transmission power; the communication parameters that trigger the adjustment of the target communication link include: when the link quality index decreases and triggers the adjustment of communication parameters, reducing the modulation order and increasing the transmission power after reducing the modulation order; when the link quality index increases and triggers the adjustment of communication parameters, increasing the transmission power and increasing the modulation order after increasing the transmission power.

[0093] In specific application scenarios, when link quality indicators decline, the system switches to a connection-preserving mode:

[0094] The first step is to switch the modulation scheme to a lower-order one of the currently available schemes (e.g., downgrading from 64-QAM to 16-QAM, or from 16-QAM to QPSK). This action is the first response action triggered after the link quality index deteriorates, and it does not need to wait for other conditions, aiming to be executed as quickly as possible.

[0095] The second step is to increase the transmit power of the target link after the modulation order is reduced and executed. The trigger for increasing the transmit power is not immediate, but requires a stability condition to be met. For example, within a preset short time window (such as 100 milliseconds) after the modulation order is reduced, the link quality indicators stop decreasing and tend to stabilize.

[0096] Lowering the modulation order is equivalent to lowering the minimum signal-to-noise ratio threshold required for correct decoding by the demodulator, instantly widening the security margin of communication and resolving response lag issues before communication interruptions occur. Increasing transmit power is equivalent to directly increasing signal energy to overwhelm noise and interference. In poor channel conditions, blindly increasing transmit power to maintain high-order modulation is inefficient. This solution first ensures connection availability by lowering the modulation order, and then increases transmit power to solidify this more robust connection, achieving efficient use of power resources.

[0097] When link quality metrics rise and trigger communication parameter adjustments to efficiency-enhancing mode:

[0098] Step 1: Increase the transmit power of the target link. This is the first response action triggered after the link quality index rises.

[0099] Step 2: After the transmit power is increased and executed, the modulation method is switched to the currently available scheme, which is a higher-order modulation method. The trigger for the modulation order upgrade needs to meet a high-performance condition: after the transmit power is increased, the link quality index can not only be stabilized at a high level, but also continuously exceed the upgrade threshold required by the next-order modulation method for a preset time.

[0100] Increasing the transmit power improves the signal-to-noise ratio margin under the current configuration, while increasing the modulation order transmits more data per unit time. Increasing the transmit power first, and then increasing the modulation order, significantly reduces the risk of immediate failure due to instantaneous fluctuations after modulation upgrade.

[0101] Example 2: Refer to Figure 5 As shown, this embodiment of the invention discloses a heterogeneous link communication device for unmanned area inspection, including: a data processing center and several inspection devices;

[0102] The inspection equipment is used to perform inspection tasks in uninhabited areas and send the data recorded during the inspection process to the data processing center.

[0103] The data processing center acquires real-time performance parameters of multiple heterogeneous communication links communicating with the inspection equipment. These parameters include communication latency, data packet loss rate, signal strength, link bit error rate, and heartbeat response time. The acquired real-time performance parameters are categorized into link-type-aware environmental interference levels. When an environmental interference level exceeds a threshold, a subset of heterogeneous links to be optimized is selected from the multiple heterogeneous communication links based on the communication link type indicated by the environmental interference level. The real-time performance parameters of this subset are mapped to scores characterizing the reliability of the communication links. The target communication link is determined based on the reliability score. When the latency of the target communication link exceeds a threshold latency, the communication parameters of the target communication link are adjusted.

[0104] The embodiments of the present invention are used to execute the heterogeneous link communication method for uninhabited area inspection in Embodiment 1. The two belong to the same inventive concept and have the same technical effect, which will not be repeated here.

[0105] In summary, the heterogeneous link communication method and communication device for unmanned area inspection described in this invention improves communication continuity and reliability, and enhances optimization efficiency when communication quality deteriorates.

[0106] Among them, the predictive network classifies the real-time performance parameters of the communication links available to the inspection equipment into environmental interference levels. Before the micro-performance parameters of the communication links deteriorate and accumulate into macro-communication failures, it accurately identifies and characterizes the stress effect of the external environment on specific link types, thereby greatly reducing the probability of communication interruption and effectively ensuring the continuity and reliability of inspection tasks in complex uninhabited environments, achieving near-zero interruption continuous and stable communication.

[0107] Once a decline in communication quality is detected, traditional methods either blindly switch communication links or make global parameter adjustments to all links, which leads to resource waste and low optimization efficiency.

[0108] In this embodiment of the invention, prediction network one classifies the real-time performance parameters of the communication link into an environmental interference level that is perceived by the link type. That is, it identifies in advance the impact of environmental changes on the communication link and the types of communication links affected from the performance parameters of the communication link. Based on the types of communication links affected indicated by the environmental interference level, it selects a subset of heterogeneous links to be optimized, thereby determining the root cause and main scope of impact of the communication link performance degradation. Then, combined with the quantitative evaluation of prediction network two, it locks the target communication link in the heterogeneous link subset and performs targeted parameter adjustment, avoiding the inefficiency caused by traditional global optimization.

[0109] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A heterogeneous link communication method for inspection in uninhabited areas, characterized in that, include: Obtain real-time performance parameters of multiple heterogeneous communication links; The performance parameters include communication latency, data packet loss rate, signal strength, link bit error rate, and heartbeat response time. The real-time performance parameters are input into prediction network one, and the real-time performance parameters are classified into link type-aware environmental interference levels; the link type-aware environmental interference levels are used to indicate one or more communication link types affected by environmental changes. In response to the environmental interference level exceeding a threshold level, a subset of heterogeneous links to be optimized is selected from the plurality of heterogeneous communication links according to the communication link type indicated therein; The real-time performance parameters of the heterogeneous link subset are input into prediction network two, and the real-time performance parameters are mapped to scores that characterize the reliability of the communication link to obtain the link quality index. The target communication link is determined based on the link quality index, and the communication parameters of the target communication link are adjusted in response to the delay of the target communication link exceeding the threshold delay. The prediction network comprises: an input layer that receives a normalized feature vector composed of communication delay, data packet loss rate, signal strength, link bit error rate, and heartbeat packet response time; a radial basis function kernel layer that nonlinearly maps the normalized feature vector to a high-dimensional feature space so that data corresponding to different environmental interference levels are linearly separable in the high-dimensional feature space; a weighted decision layer that constructs a classification hyperplane based on support vectors and Lagrange multipliers in the high-dimensional feature space and calculates the classification result; and an output layer that maps the classification result of the weighted decision layer to the environmental interference level. The environmental interference level mapped by the output layer is a composite semantic label, which uniquely corresponds to a combination of environmental interference degree and affected link type.

2. The heterogeneous link communication method for unmanned area inspection according to claim 1, characterized in that, Based on the affected link types indicated by the environmental interference level, a subset of heterogeneous links to be optimized is selected, including: A predefined interference-link mapping table defines the correspondence between different environmental interference levels and one or more communication link types to be optimized; Based on the environmental interference level output by the prediction network, the interference-link mapping table is queried to obtain the target link type to be optimized; Filter all communication links belonging to the target link type from the multiple heterogeneous communication links to obtain the heterogeneous link subset.

3. The heterogeneous link communication method for unmanned area inspection according to claim 1 or 2, characterized in that, The environmental interference levels sensed by the link type include at least: The first level of indication of satellite link disruption; The second level indicates that the terrestrial radio link is being disrupted; Level 3 indicates that both satellite links and terrestrial wireless links are disrupted.

4. The heterogeneous link communication method for unmanned area inspection according to claim 1, characterized in that, The second prediction network is a multilayer perceptron model, and its network structure includes: The input layer is used to receive a normalized feature vector composed of the real-time performance parameters of each link in the heterogeneous link subset; The feature extraction module consists of a first fully connected layer, a ReLU activation function, a Dropout layer, and a second fully connected layer connected sequentially; wherein the second fully connected layer has fewer neurons than the first fully connected layer, forming a feature encoder for feature compression and abstraction. The regression output layer is used to map the high-order features output by the feature extraction module to a normalized reliability score in the range of 0 to 1.

5. The heterogeneous link communication method for unmanned area inspection according to claim 4, characterized in that, The number of neurons in the first and second fully connected layers is related to the number of communication links and the number of single-link performance parameters contained in the heterogeneous link subset.

6. The heterogeneous link communication method for unmanned area inspection according to claim 1, characterized in that, Determining the target communication link based on the link quality index includes: selecting the communication link with the highest link quality index from the heterogeneous link subset as the target communication link.

7. The heterogeneous link communication method for unmanned area inspection according to claim 1 or 6, characterized in that, The communication parameters include modulation scheme and transmit power; the communication parameters that trigger adjustment of the target communication link include: When the link quality index decreases and triggers communication parameter adjustment, the modulation order is reduced, and the transmit power is increased after the modulation order is reduced. When the link quality index rises and triggers communication parameter adjustment, the transmit power is increased, and the modulation order is increased after the transmit power is increased.

8. A heterogeneous link communication device for uninhabited area inspection, used to execute the heterogeneous link communication method for uninhabited area inspection as described in any one of claims 1-7, characterized in that, The communication device includes: a data processing center and several inspection devices; The inspection equipment is used to perform inspection tasks in uninhabited areas and send the data recorded during the inspection process to the data processing center. The data processing center is used to acquire real-time performance parameters of multiple heterogeneous communication links communicating with the inspection equipment. These performance parameters include communication latency, data packet loss rate, signal strength, link bit error rate, and heartbeat response time. The acquired real-time performance parameters are categorized into link-type-aware environmental interference levels. When the environmental interference level exceeds a threshold level, a subset of heterogeneous links to be optimized is selected from the multiple heterogeneous communication links based on the communication link type indicated by the environmental interference level. The real-time performance parameters of the heterogeneous link subset are mapped to scores characterizing the reliability of the communication links. A target communication link is determined based on the reliability score. When the latency of the target communication link exceeds a threshold latency, the communication parameters of the target communication link are adjusted.

Images