Node trajectory prediction method and device for air-sky-sea-diving unmanned cluster

Abstract

The invention discloses a node trajectory prediction method and device for an air-sky-sea-diving unmanned cluster, and belongs to the technical field of mobile ad hoc networks, and the method comprises the steps: collecting data in real time, and carrying out the preprocessing of the data; constructing an unmanned node trajectory prediction model based on a gating circulation unit, a Transform and a width learning system, and inputting data into a gating circulation unit and Transform parallel cascade module; performing weighted fusion on module output results, and dynamically adjusting the weight by using a back propagation algorithm; and finally, the weighted fusion result and the dynamic environment index are input into a width learning system, the output of the width learning system outputs a final unmanned node trajectory prediction result through a full connection layer, and the width learning system can dynamically adjust the network width according to the data change. Based on the method provided by the invention, the gating circulation unit and the Transform are cascaded in parallel, the combination of real-time response and long-term prediction is realized, a wide learning system is introduced to quickly adapt to data distribution change through incremental expansion, and the environmental adaptability of a prediction model is improved.

Description

Technical Field

[0001] This invention belongs to the field of mobile ad hoc networks, specifically to a method for predicting the trajectory of dynamic nodes such as drones, unmanned ships, and unmanned submarines. The method is particularly applicable to ad hoc networks composed of dynamic maritime base stations and terminal nodes. The invention relates to trajectory prediction technology in mobile ad hoc networks, and in particular to a method and device for predicting the trajectory of nodes in unmanned aerial, space, and submarine swarms. Background Art

[0002] With the development of unmanned air, space, shore, sea and submarine systems, more and more nodes need to predict their trajectories based on the real-time environment and status. The movement trajectory of the node is affected by many factors, including natural environmental factors such as geographical location and weather, as well as the interaction between nodes. Existing trajectory prediction methods mostly use traditional machine learning and deep learning technologies such as long short-term memory networks (LSTM) and recurrent neural networks (RNN). However, these methods have problems such as poor real-time performance and large computational complexity when faced with high-dimensional, complex environments, and dynamically changing data.

[0003] Offshore nodes are affected by complex environments such as ocean currents and wind speeds, and a single model cannot simultaneously model spatiotemporal dynamics. Traditional short-term prediction methods such as Kalman filtering and single LSTM can handle sudden movements, but they lack the ability to model long-term dependencies, resulting in serious accumulation of long-term prediction errors. Long-term prediction methods based on physical models are computationally complex and have difficulty responding to sudden obstacles in real time, and cannot meet the millisecond-level requirements of sudden movements. GRU and Transformer, as powerful time series modeling tools, can be effectively combined to handle the short-term and long-term dependencies of node trajectories. However, these models often face problems with computing speed and real-time performance when processing large-scale, dynamically changing data.

[0004] Gated Recurrent Units (GRUs) are an effective model for time series modeling, capable of capturing local temporal dependencies. However, GRUs struggle to capture long-range spatiotemporal dependencies, leading to significant prediction error accumulation in the dynamic topology of mobile ad hoc networks. Trajectory prediction in mobile ad hoc networks often requires processing long-term spatial and temporal dependencies, making it difficult for a single GRU model to fully capture complex motion patterns.

[0005] Transformers excel at handling long-range dependencies and can capture global information through self-attention mechanisms. However, pure Transformer models lack sensitivity to local temporal features and have high computational complexity, making them difficult to deploy on resource-constrained mobile nodes.

[0006] Combining GRU and Transformer can achieve real-time response and long-term prediction synchronization. In the face of sudden changes in motion state, GRU responds quickly and Transformer maintains global consistency. The introduction of Broad Learning System (BLS) further enhances and optimizes feature vectors. The dynamic changes in aerospace, sea and submarine scenarios are relatively fast. BLS dynamically adjusts the network width according to the frequency of data changes to improve the environmental adaptability of the model. The combination of GRU, Transformer and BLS improves the prediction accuracy of the model and adapts it to experimental scenarios. Summary of the Invention

[0007] In view of this, an embodiment of the present invention provides a node trajectory prediction method and device for an aerospace, sea and submarine unmanned cluster to eliminate or improve one or more defects in the prior art, and solves the problems that a single model is difficult to simultaneously model short-term and long-term dependencies, the real-time performance of processing large-scale dynamic data is low, and the modeling of complex environments in aerospace, sea and submarine scenarios is difficult. The present invention realizes layered collaboration, combines the advantages of GRU and Transformer models, and realizes the combination of real-time response and long-term prediction. At the same time, BLS is introduced to dynamically adjust the network width, quickly process and learn the latest data, thereby improving the real-time performance of trajectory prediction, and quickly adapting to changes in data distribution through incremental expansion, thereby improving the environmental adaptability of the prediction model.

[0008] An embodiment of the present invention provides a node trajectory prediction method for an unmanned air, space, sea and submarine swarm, comprising the following steps:

[0009] S1. Real-time collection of motion data, environmental perception data, satellite-collected meteorological and topographical data, and interaction information between nodes from each node in the unmanned aerospace, space, and submarine swarm. The motion data of the unmanned nodes includes speed, acceleration, heading angle, and position coordinates. The collected data is then preprocessed, including data cleaning, normalization, and time alignment.

[0010] S2. Build a prediction model based on the fusion of GRU gated recurrent units, Transformer, and BLS width learning system. The model uses layered temporal processing. First, the GRU and Transformer are cascaded in parallel. The pre-processed data is input into the GRU and Transformer parallel cascade module.

[0011] S3. Perform weighted fusion on the feature vectors output by GRU and Transformer. Based on the mean square error feedback between the trajectory prediction result and the true value, use the back propagation algorithm to dynamically adjust the weight of the weighted fusion of GRU output and Transformer output.

[0012] S4. The weighted fused feature vectors processed by GRU and Transformer, real-time environmental data, and dynamic environmental indicators are input into the width learning system. The width learning system can further enhance and optimize the fused features through its feature mapping and enhanced node generation. Finally, the output of the width learning system is output through the fully connected layer to output the final unmanned node trajectory prediction result, including the node's position coordinates and motion status. In addition, BLS dynamically adjusts the network width based on the changes in preprocessed data, computing load, and prediction requirements.

[0013] Furthermore, the parameter selection of the GRU and Transformer modules of the prediction model follows the principle of scenario-driven parameter selection and module collaboration.

[0014] Furthermore, node expansion or node pruning is performed according to dynamic adjustment trigger conditions, and the dynamic adjustment trigger conditions include data complexity measurement, model performance feedback and computing resource status.

[0015] The beneficial effects of the present invention are at least:

[0016] The method and device for trajectory prediction of unmanned aerospace, sea and submarine cluster nodes of the present invention achieve efficient and robust trajectory prediction and network collaborative optimization in complex and changeable aerospace, sea and submarine environments by integrating the hierarchical time series processing architecture and dynamic adaptive mechanism of GRU, Transformer and BLS.

[0017] The GRU and Transformer collaborate to improve prediction accuracy. The GRU focuses on capturing short-term dependencies in time series data, processing short-term state changes in unmanned nodes themselves, and achieving rapid response to sudden scenarios. The Transformer excels at mining the global dependencies and spatial characteristics of data, analyzing long-term relationships such as the relative positions and collaborative tasks between unmanned nodes, and maintaining long-term global consistency. The two processes in parallel and performs weighted fusion of the outputs. The fusion results provide comprehensive feature information for subsequent BLS processing. This collaborative mechanism enables the model to learn more complex mobile ad hoc network patterns, significantly improving trajectory prediction accuracy.

[0018] BLS receives the fusion results of GRU and Transformer, performs deep feature processing, and mines potential complex data patterns through feature mapping and enhanced node generation. In addition, in the changeable aerospace, sea and underwater environments, BLS automatically expands the number of nodes according to the complexity of the input data, responds to environmental mutations by generating feature nodes, and generates enhanced nodes to adjust computing resources, allowing the model to adapt to data changes and avoid overfitting or underfitting.

[0019] Unmanned aerospace, sea and submarine scenarios are diverse and uncertain, and the situations in different regions and tasks vary greatly. The combined model of GRU, Transformer and BLS can learn richer and more comprehensive feature patterns, is not prone to overfitting, has stronger generalization ability, and can maintain good performance in various complex and changeable aerospace, sea and submarine scenarios, and adapt to different mission requirements and environmental changes.

[0020] Additional advantages, objects, and features of the present invention will be set forth in part in the following description and will become apparent to those skilled in the art upon examination of the following or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained by the structures particularly pointed out in the description and drawings.

[0021] Those skilled in the art will understand that the purposes and advantages that can be achieved by the present invention are not limited to the above specific descriptions, and the above and other purposes that can be achieved by the present invention will be more clearly understood based on the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention. In the drawings:

[0023] Figure 1 Flowchart of the node trajectory prediction method and device for an unmanned aerospace, sea and submarine swarm provided in an embodiment of the present invention.

[0024] Figure 2 A schematic diagram of the overall structure of the unmanned node trajectory prediction model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the implementation methods and drawings. Here, the schematic implementation methods of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0026] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, the accompanying drawings only show structures and / or processing steps closely related to the solutions according to the present invention, while other details that are not closely related to the present invention are omitted.

[0027] It should be emphasized that the term "include / comprises" when used herein refers to the existence of features, elements, steps or components, but does not exclude the existence or addition of one or more other features, elements, steps or components.

[0028] It should also be noted that, unless otherwise specified, the term "connection" herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0029] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings, in which the same reference numerals represent the same or similar components, or the same or similar steps.

[0030] The aerospace, sea and submarine network in the present invention includes space-based, air-based, sea-based and submarine-based networks. The space-based network is composed of high-orbit, high-throughput satellites. The air-based network includes medium-altitude, long-endurance UAVs and high-altitude, long-endurance UAVs. The sea-based network includes unmanned ships, civilian ships equipped with high-throughput satellite terminals, hydroacoustic receiving terminals and ship-borne base stations. The submarine-based network is unmanned submarines. Medium-altitude, long-endurance UAVs, unmanned ships and unmanned submarines access civilian ships through unmanned ad hoc access to build a maritime local area network, and form a collaborative neighbor discovery algorithm based on mobile trajectory prediction.

[0031] The present invention provides a node trajectory prediction method for an air, space, sea and submarine unmanned cluster. Figure 1 As shown, the method includes the following steps S1 to S4:

[0032] S1. Real-time collection of motion data, environmental perception data, satellite-collected meteorological and topographical data, and interaction information between nodes from each node in the unmanned aerospace, space, and submarine swarm. The motion data of the unmanned nodes includes speed, acceleration, heading angle, and position coordinates. The collected data is then preprocessed, including data cleaning, normalization, and time alignment.

[0033] S2. Build a prediction model based on the fusion of GRU gated recurrent units, Transformer, and BLS width learning system. The model uses layered temporal processing. First, the GRU and Transformer are cascaded in parallel. The pre-processed data is input into the GRU and Transformer parallel cascade module.

[0034] S3. Perform weighted fusion on the feature vectors output by GRU and Transformer. Based on the mean square error feedback between the trajectory prediction result and the true value, use the back propagation algorithm to dynamically adjust the weight of the weighted fusion of GRU output and Transformer output.

[0035] S4. The weighted fused feature vectors processed by GRU and Transformer, real-time environmental data, and dynamic environmental indicators are input into the width learning system. The width learning system can further enhance and optimize the fused features through its feature mapping and enhanced node generation. Finally, the output of the width learning system is output through the fully connected layer to output the final unmanned node trajectory prediction result, including the node's position coordinates and motion status. In addition, BLS dynamically adjusts the network width based on the changes in preprocessed data, computing load, and prediction requirements.

[0036] By predicting the movement trajectories of unmanned nodes, we can provide prior information for neighbor discovery in ad hoc networks, thereby optimizing network topology maintenance, reducing communication overhead, and improving network stability in dynamic environments.

[0037] The above steps are described in detail below.

[0038] In step S1, data is collected in real time from each node in the unmanned air, space, shore, and sea-based swarm. Unmanned nodes include drones, unmanned ships, and unmanned submarines. Data includes the node's position, velocity, acceleration, and heading angle. In addition to local node data, the system can also collect information from other nodes, including relative positions and velocities, environmental perception data, and meteorological and topographic data collected by satellites.

[0039] The collected data is then detected and processed for outliers, corrected or deleted, and noise removed using filtering algorithms. Data of different dimensions is normalized to the same scale. For location coordinate data, a maximum-minimum normalization method is used to map longitude and latitude values ​​to the [0, 1] range. For speed data, the same maximum-minimum normalization method is used to map speed values ​​to the [0, 1] range. The system aligns data from all nodes based on their timestamps, ensuring that data from different nodes can be processed within the same time window.

[0040] In the above step S2, the prediction model based on the fusion of GRU gated recurrent unit, Transformer and BLS width learning system is referenced Figure 2 As shown in the figure, the GRU and Transformer in the prediction model are combined in a parallel cascade manner, and the weighted fused feature vector processed by GRU and Transformer, real-time environmental data and dynamic environmental indicators are input into the width learning system.

[0041] GRU and Transformer are combined in a parallel cascade manner, and the results are weightedly fused to achieve an efficient combination of local and global feature extraction. GRU achieves millisecond-level response and alleviates the impact of signal transmission delay. Transformer is suitable for processing long-term cross-media navigation, maintaining global path consistency, and performing global modeling and denoising on complex interfered data. GRU processes the short-term state changes of each unmanned node, quickly responds to sudden movements, updates short-term predictions, and helps nodes adjust their neighbor lists in a timely manner; Transformer analyzes long-term relationships such as relative positions and collaborative tasks between nodes, predicts future node aggregation areas, and helps optimize the global neighbor discovery strategy.

[0042] Input to the GRU module is the motion time series feature, which represents the three-dimensional linear velocity of the unmanned node in the corresponding time series length , angular velocity and acceleration , one-dimensional heading angle and three-dimensional position coordinates , the output is the encoded hidden motion state and short-term trajectory feature vector ; Input of the Transformer module It is a multivariate spatiotemporal feature, representing the three-dimensional linear velocity, angular velocity and acceleration of the unmanned node in the corresponding time series length, the one-dimensional heading angle, the three-dimensional coordinates of the position, and the wind speed. , meteorological data consisting of wave height and precipitation intensity, ocean current speed Ocean data composed of water depth, terrain data, including the relative positions between multiple nodes , relative speed Multi-node interactive data with cluster topology, meteorological remote sensing data and terrain enhancement data collected by satellites, and output including long-term dependent feature vectors and the cross attention matrix, which represents the collaborative relationship between nodes.

[0043] GRU controls the flow of information through update gates and reset gates, effectively capturing short-term dependencies in time series. GRU contains two gates: the update gate and the reset gate. Their calculations involve input data, the previous hidden state, and the corresponding weight matrix and bias terms. Then, the candidate hidden state is calculated, combined with the information from the reset gate, and finally the current hidden state is updated. Its core calculation steps are as follows:

[0044] Reset gate, control the degree of information retention of the hidden state at the previous moment, reset gate output vector ,in, is the sigmoid activation function, and are the weight matrix and bias term of the reset gate respectively, is the hidden state at the previous moment, is the current input vector;

[0045] Update gate, which determines the degree of update of the hidden state, and the update gate output vector ;

[0046] Combine the reset gate and the current input to generate candidate states, candidate hidden states , where ⊙ represents element-by-element multiplication and tanh represents the hyperbolic tangent activation function;

[0047] The new and old states are fused through the update gate, and the final hidden state .

[0048] The Transformer is primarily composed of a self-attention mechanism and a feed-forward network. The self-attention mechanism aggregates information by calculating the weights of the query, key, and value. Multi-head attention parallelizes this process multiple times to capture information from different subspaces. Furthermore, positional encoding is used to inject positional information into the sequence, as the Transformer itself does not have the ability to process sequence order. The specific components are as follows:

[0049] The self-attention mechanism captures the global dependency within the sequence through query, key, and value matrices. First, the Q, K, and V matrices are generated. , , , where X represents the input sequence, 、 and is a learnable projection matrix, scaled dot product attention, ,in Represents the dimension of the key vector, the scaling factor to prevent the gradient from disappearing;

[0050] The multi-head attention mechanism is to use multiple attention heads in parallel to capture the features of different subspaces. , where each head is calculated as ,in 、 and are the independent projection matrices for each head, is the output projection matrix;

[0051] Feedforward neural network performs nonlinear transformation on attention output ,in and is the weight matrix, and is the bias term;

[0052] Inject sequence position information for position encoding, , , where pos is the sequence position and i is the dimension index.

[0053] Next, we will explain the parameter selection of the GRU and Transformer modules of the prediction model:

[0054] The parameter selection of the GRU and Transformer modules of the prediction model follows the principle of scenario-driven parameter selection and module collaboration.

[0055] Dynamically adjust based on motion patterns, environmental complexity, and computing resource constraints. The required delays for drones, unmanned ships, and unmanned submarines vary depending on their mission scenarios:

[0056] Drone swarms require millisecond-level response and high-precision prediction. GRUs require short time steps and a reduced number of layers. The number of hidden units is dynamically adjusted based on real-time computing resources, increasing it when resources are sufficient to improve accuracy and reducing it when resources are limited to reduce latency. Transformers use a moderate number of heads and layers, adapting the feedforward network dimensions to the sensor data dimensions, prioritizing local attention to reduce computational overhead, and updating the network parameter weights within GRUs and Transformers through online learning to adapt to sudden changes in motion states.

[0057] Long-term cruise and steady-state operation of unmanned vessel swarms require robustness against environmental interference. GRUs use a medium time step and number of layers. The number of hidden units balances short-term response and noise immunity. Transformers adapt the number of heads and layers to environmental data. The feedforward network dimension is positively correlated with the number of interacting nodes. Positional encoding is used to enhance long-sequence modeling capabilities, and the attention window covers a typical cruise cycle.

[0058] Unmanned submarine clusters require low power consumption and long-term prediction. GRU extends the timing window to capture slowly changing motion patterns. The number of hidden units and layers is dynamically compressed according to the power budget. Transformer reduces the number of attention heads, increases the layer depth and feedforward network dimension to improve representation capabilities, and uses sparse attention to reduce computational load.

[0059] Considering the inter-module synergy from a system perspective, the GRU and Transformer are time-aligned. The GRU time step is T, and the Transformer attention window is set to kT. k is adjusted according to the task stage. In the emergency situation, k is smaller, and in the steady-state stage, k is larger. The outputs of GRU and Transformer are aligned to the same dimension through linear projection to ensure the consistency of the fusion module input.

[0060] In the above step S3, GRU and Transformer generate short-term trajectory feature vectors respectively and the long-term dependent eigenvector , their output results are fused by weighted summation, and weights are assigned to the outputs of GRU and Transformer respectively, so the output after weighted summation is It can be expressed as ,in and are learnable weights.

[0061] Dynamically adjusting the GRU and Transformer weighted fusion weights through the back-propagation algorithm includes the following steps:

[0062] Define the forward propagation structure of the model, including two parallel prediction modules, GRU and Transformer, and a weighted sum layer to fuse the outputs of the two modules;

[0063] The mean square error is selected as the loss function, and the loss function L is expressed as: ;

[0064] Use the defined loss function to calculate the loss between the weighted fusion output and the true label; starting from the output layer, reversely calculate the gradient of the loss function for each parameter, and the loss function for The gradient of , calculate the pair by the chain rule and The gradients are and ;

[0065] Update weights using gradient descent;

[0066] Repeat the above steps of calculating loss, calculating gradients, and updating weights until the performance of the model on the validation set no longer improves significantly or the preset number of iterations is reached.

[0067] In the above step S4, the input of BLS includes the weighted fused feature vector after GRU and Transformer processing, real-time environmental data and dynamic environmental indicators. Dynamic environmental indicators include data change rate and resource status. It quickly integrates multi-source data including satellite positioning and environmental factors, provides real-time updated prediction results, and supports dynamic adjustment of neighbor discovery parameters. The width learning system generates high-dimensional feature representations through feature nodes and enhancement nodes, maps them to the target dimension through a fully connected layer, and outputs the trajectory prediction results of unmanned nodes in the future, including the node's position coordinates and motion status. In addition, in the case of changeable air, space, sea and underwater environments, BLS automatically expands the number of nodes according to the complexity of the input data, responds to environmental mutations by generating feature nodes, and generates enhancement nodes to adjust computing resources, so that the model can adapt to data changes and avoid overfitting or underfitting.

[0068] BLS fuses multi-source data through the feature mapping layer of BLS using a randomly generated weight matrix Map the fused input data to the high-dimensional feature space to obtain the feature node Z, which is expressed as: , where X is the fused input data, is the activation function, is the bias vector.

[0069] Based on the feature nodes, enhanced nodes are generated through nonlinear transformation, using random matrices Transform the feature nodes to generate enhanced nodes Y, which can be expressed as: ,in is the bias vector.

[0070] The output of BLS is input into the fully connected layer and the final node trajectory prediction result is obtained through weight matrix calculation. The calculation formula is: ,in is the weight matrix of the fully connected layer, is the bias vector, is the output of BLS; in this way, the BLS module plays a key role in the unmanned node trajectory prediction model. It not only integrates the advantages of GRU and Transformer, but also improves the adaptability and prediction performance of the model through dynamic adjustment and optimization mechanisms.

[0071] BLS automatically adjusts network width and optimizes computing resource usage based on the system's real-time forecast demand and computing resource status. When real-time forecast demand is high and computing resources are sufficient, the number of nodes in the feature mapping layer and the enhancement node layer is increased to enhance the model's expressiveness. When forecast demand is low or computing resources are limited, the number of nodes is reduced to avoid resource waste. The specific steps are as follows:

[0072] Define dynamic adjustment trigger conditions, including data complexity measurement, model performance feedback, and computing resource status;

[0073] Data complexity metrics include entropy monitoring and gradient change rate, and calculating the Shannon entropy of the input data window: When the entropy increase exceeds 20% due to a sudden change in environmental data, node expansion is triggered, and the mean absolute value of the gradient in the time window is calculated for the motion state data, environmental perception data, and node interaction data: ,like , triggering adjustments;

[0074] The indicator of model performance feedback is the model prediction residual. The mean square error (MSE) between the predicted trajectory and the actual trajectory is calculated in real time. If the residual for k consecutive steps exceeds the threshold, an adjustment is triggered.

[0075] The computing resource status is divided into hardware resource utilization rate. Idle hardware resources allow node expansion, while overload triggers pruning.

[0076] Adjusting the network width includes node expansion and node pruning;

[0077] When it is detected that the data complexity increases or the model performance decreases, and there are sufficient resources, the nodes are expanded. The nodes include feature nodes and enhanced nodes. Perform feature screening and select features that are highly correlated with the current mutation, i.e. The features of the network are randomly generated, the weights and biases of new feature nodes are randomly generated, the new nodes are integrated into the network, and nonlinear transformations are performed on the newly added feature nodes to generate enhanced nodes.

[0078] Node pruning is performed when data complexity decreases or resources are tight. The importance of each node is evaluated by calculating the L1 norm of the node weight, and nodes with lower importance are removed from the network.

[0079] It should be understood by those skilled in the art that the various exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether to implement the system in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via a data signal carried in a carrier wave.

[0080] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0081] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments.

[0082] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations to the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A node trajectory prediction method for an air, space, sea and submarine unmanned swarm, characterized by: The aerospace, sea and submarine network includes space-based, air-based, sea-based and submarine-based networks. The space-based network is composed of high-orbit high-throughput satellites, the air-based network is composed of drones, the sea-based network includes unmanned ships, civilian ships equipped with high-throughput satellite terminals, hydroacoustic receiving terminals and ship-borne base stations, and the submarine-based network is composed of unmanned submarines. UAVs, unmanned ships and unmanned submarines access civilian ships through unmanned random access to build a maritime local area network. Based on mobile trajectory prediction, a collaborative neighbor discovery algorithm is formed. Trajectory prediction includes the following steps: S1. Real-time collection of motion data, environmental perception data, satellite-collected meteorological and topographical data, and interaction information between unmanned nodes in the air, space, sea, and submarine unmanned cluster, and pre-processing of the collected data. S2. Build a prediction model based on the fusion of gated recurrent units, transformers, and a width learning system. The model uses layered temporal processing. First, the gated recurrent unit and transformer are cascaded in parallel. After preprocessing, the real-time collected data is input into the gated recurrent unit and transformer parallel cascade module. S3, perform weighted fusion on the feature vectors output by the gated recurrent unit and the Transformer, and dynamically adjust the weight of the weighted fusion using the back-propagation algorithm based on the mean square error feedback between the trajectory prediction result and the true value; S4. The weighted fused feature vectors processed by the gated recurrent unit and the Transformer, along with real-time environmental data and dynamic environmental indicators, are input into the width learning system. The width learning system can further enhance and optimize the fused features through its feature mapping and enhanced node generation. Finally, the output of the width learning system is output through the fully connected layer to produce the final unmanned node trajectory prediction result. In addition, the width learning system dynamically adjusts the network width based on changes in preprocessed data, computational load, and prediction requirements. By predicting the movement trajectories of unmanned nodes, we can provide prior information for neighbor discovery in ad hoc networks, thereby optimizing network topology maintenance, reducing communication overhead, and improving network stability in dynamic environments.

2. The node trajectory prediction method of the air, space, sea and submarine unmanned swarm according to claim 1 is characterized in that: The functions of the gated recurrent unit, Transformer, and width learning system modules in the prediction model in step S2 and their inputs and outputs are: The gated recurrent unit and Transformer are combined in a parallel cascade manner, and the results are weightedly fused to achieve efficient integration of local and global feature extraction. The gated recurrent unit achieves millisecond-level response, alleviating the impact of signal transmission delay. The Transformer is suitable for handling long-term cross-medium navigation, maintaining global path consistency, and performing global modeling and denoising on complex data subject to interference. The gated recurrent unit processes short-term state changes of each unmanned node, quickly responds to sudden movements, updates short-term predictions, and helps nodes adjust their neighbor lists in a timely manner. Transformer analyzes long-term relationships between nodes, such as relative positions and collaborative tasks, to predict future node aggregation areas and help optimize the global neighbor discovery strategy. Input to the Gated Recurrent Unit module is the motion time series feature, which represents the three-dimensional linear velocity of the unmanned node in the corresponding time series length , angular velocity and acceleration , one-dimensional heading angle and three-dimensional position coordinates , the output is the encoded hidden motion state and short-term trajectory feature vector ; Input of the Transformer module It is a multivariate spatiotemporal feature, representing the three-dimensional linear velocity, angular velocity and acceleration of the unmanned node in the corresponding time series length, the one-dimensional heading angle, the three-dimensional coordinates of the position, and the wind speed. , meteorological data consisting of wave height and precipitation intensity, ocean current speed Ocean data composed of water depth, terrain data, including the relative positions between multiple nodes , relative speed Multi-node interactive data with cluster topology, meteorological remote sensing data and terrain enhancement data collected by satellites, and output including long-term dependent feature vectors and the cross attention matrix, which represents the collaborative relationship between nodes; The input of the width learning system includes the weighted fused feature vectors after gated recurrent units and Transformer processing, real-time environmental data and dynamic environmental indicators. Dynamic environmental indicators include data change rate and resource status. It quickly integrates multi-source data, provides real-time updated prediction results, and supports dynamic adjustment of neighbor discovery parameters. The width learning system generates high-dimensional feature representations through feature nodes and enhancement nodes, maps them to the target dimension through a fully connected layer, and outputs the trajectory prediction results of unmanned nodes in the future, including the node's position coordinates and motion status. In addition, in the case of changeable aerospace, sea and underwater environments, BLS automatically expands the number of nodes according to the complexity of the input data, responds to environmental mutations by generating feature nodes, and generates enhancement nodes to adjust computing resources, so that the model can adapt to data changes and avoid overfitting or underfitting.

3. The node trajectory prediction method of the air, space, sea and submarine unmanned swarm according to claim 1 is characterized in that: The selection of the gated recurrent unit and Transformer parameters of the prediction model in step S2 is: Scenario-driven parameter selection, dynamically adjusted based on motion patterns, environmental complexity, and computing resource constraints. UAVs, unmanned ships, and unmanned submarines have different mission scenarios and require different delays. Drone swarms require millisecond-level response and high-precision prediction. Gated recurrent units require short time steps and a reduced number of layers. The number of hidden units is dynamically adjusted based on real-time computing resources, increasing it to improve accuracy when resources are sufficient and reducing it to reduce latency when resources are limited. Transformers use a moderate number of heads and layers, with the feedforward network dimensions adapted to the sensor data dimensions. Local attention is prioritized to reduce computational overhead, and network parameter weights within gated recurrent units and Transformers are updated through online learning to adapt to sudden changes in motion states. Unmanned vessel swarms cruise for extended periods of time and operate in a steady state, requiring robustness against environmental interference. Gated recurrent units are designed with a medium time step and number of layers. The number of hidden units balances short-term response with noise immunity. The number of Transformer heads and layers adapts to environmental data. The feedforward network dimension is positively correlated with the number of interacting nodes. Positional encoding is used to enhance long-sequence modeling capabilities, and the attention window covers a typical cruise cycle. Unmanned submarine swarms require low power consumption and long-term prediction. Gated recurrent units extend the timing window to capture slowly changing motion patterns. The number of hidden units and layers is dynamically compressed based on the power budget. Transformers reduce the number of attention heads, increase the layer depth and feedforward network dimensions to improve representation capabilities, and use sparse attention to reduce computational load. Considering the synergistic effect between modules from a system perspective, the gated recurrent unit is aligned with the Transformer timing. The time step of the gated recurrent unit is T, and the Transformer attention window is set to kT. k is adjusted according to the task stage. In the emergency situation, k is smaller, and in the steady-state stage, k is larger. The outputs of the gated recurrent unit and the Transformer are aligned to the same dimension through linear projection to ensure the consistency of the fusion module input.

4. The node trajectory prediction method of the air, space, sea and submarine unmanned swarm according to claim 1 is characterized in that: In step S3, the steps of dynamically adjusting the weights of the gated recurrent unit and the Transformer weighted fusion through the back-propagation algorithm are as follows: S31. Define the forward propagation structure of the model, including two parallel prediction modules: the gated recurrent unit and the Transformer, and a weighted sum layer to fuse the outputs of the two modules, assigning weights to the outputs of the gated recurrent unit and the Transformer respectively. and , the output of the gated recurrent unit is , the output of Transformer is , then the output after weighted summation is It can be expressed as: ; S32. Select mean square error as the loss function. The loss function L is expressed as: , use the defined loss function to calculate the loss between the weighted fusion output and the true label; S33. Starting from the output layer, reversely calculate the gradient of the loss function for each parameter. The loss function is The gradient of , calculate the pair by the chain rule and The gradients are and , update the weights using gradient descent; S34. Repeat the above steps of calculating loss, calculating gradients, and updating weights until the performance of the model no longer improves significantly or reaches the preset number of iterations.

5. The node trajectory prediction method of the air, space, sea and submarine unmanned swarm according to claim 1 is characterized in that: In step S4, the width learning system dynamically adjusts the network width according to the system's real-time forecast demand and computing resource status as follows: S41. Define dynamic adjustment trigger conditions: data complexity measurement, model performance feedback, and computing resource status; Data complexity metrics include entropy monitoring and gradient change rate, and calculating the Shannon entropy of the input data window: When the entropy increases by more than 20% due to a sudden change in environmental data, node expansion is triggered. The mean absolute value of the gradient in the time window is calculated for the motion state data, environmental perception data, and node interaction data: ,like , triggering adjustments; The indicator of model performance feedback is the model prediction residual. The mean square error (MSE) between the predicted trajectory and the actual trajectory is calculated in real time. If the residual for k consecutive steps exceeds the threshold, an adjustment is triggered. The computing resource status is divided into hardware resource utilization rate. Idle hardware resources allow node expansion, while overload triggers pruning. S42. Dynamically adjust the specific process: node expansion, node pruning; When it is detected that the data complexity increases or the model performance decreases, and there are sufficient resources, the nodes are expanded. The nodes include feature nodes and enhanced nodes. Perform feature screening and select features that are highly correlated with the current mutation, i.e. The features of the network are randomly generated, the weights and biases of new feature nodes are randomly generated, the new nodes are integrated into the network, and nonlinear transformations are performed on the newly added feature nodes to generate enhanced nodes. Node pruning is performed when data complexity decreases or resources are tight. The importance of each node is evaluated by calculating the L1 norm of the node weight, and nodes with lower importance are removed from the network.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images