Communication method and device, terminal equipment, computer storage medium and program product

Abstract

The invention discloses a communication method and device, terminal equipment, a computer storage medium and a program product, and the method comprises the steps: receiving network data, and determining network state information based on the network data; performing representation extraction on the network state information through a communication switching network model to obtain a feature vector; performing feature extraction on the feature vector through the communication switching network model to obtain feature information; a network switching decision is determined based on the feature information through the communication switching network model, and the network state decision is used for determining a data transmission method; therefore, the reliability, the stability and the service quality of the communication network are improved, and frequent switching and communication interruption are avoided.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and in particular to a communication method and apparatus, terminal equipment, computer storage medium and program product. Background Technology

[0002] Currently, there are many communication technology solutions suitable for different scenarios and needs. For example, network switching based on the current network signal strength involves the terminal control channel sensing module sensing the signal strength of the communication channel and controlling network switching accordingly.

[0003] Current technologies mostly perform network switching on a terminal-by-terminal basis, which can lead to interference with certain applications during the switching process. Furthermore, when network resources are scarce, a complete switchover can cause instability in terrestrial network quality, resulting in frequent switching back to satellite communication and terrestrial networks, causing repeated resource contention. Therefore, an effective communication method is urgently needed to solve this problem. Summary of the Invention

[0004] To address the aforementioned technical problems, embodiments of the present invention provide a communication method and apparatus, a terminal device, a computer storage medium, and a program product.

[0005] In a first aspect, the communication method provided in the embodiments of this application includes:

[0006] Receive network data and determine network status information based on the network data;

[0007] The network state information is characterized and feature vectors are extracted by using a communication switching network model;

[0008] Feature information is obtained by extracting features from the feature vector using the communication switching network model.

[0009] The network switching decision is determined based on the feature information through the communication switching network model, wherein the network state decision is used to determine the data transmission method.

[0010] Secondly, the communication device provided in the embodiments of this application is applied to a terminal device and includes:

[0011] The receiving unit is used to receive network data;

[0012] The determining unit is used to determine network status information based on the network data;

[0013] The processing unit is configured to characterize and extract feature vectors from the network state information using a communication switching network model; and to extract feature information from the feature vectors using the communication switching network model.

[0014] The determining unit is further configured to determine network state decisions based on the temporal information of the feature information, wherein the network state decisions are used to determine the data transmission method.

[0015] Thirdly, the terminal device provided in the embodiments of this application includes: a processor and a memory, the memory being used to store computer programs, and the processor being used to call and run the computer programs stored in the memory to execute any of the above-described communication methods.

[0016] Fourthly, the computer-readable storage medium provided in the embodiments of this application is used to store a computer program that causes a computer to execute any of the above-described communication methods.

[0017] Fifthly, the computer program product provided in the embodiments of this application includes computer program instructions that cause a computer to perform any of the communication methods described above.

[0018] In the technical solution of this application embodiment, the terminal device receives network data and determines network status information based on the network data; it then uses a communication switching network model to characterize and extract feature vectors from the network status information; it further uses the communication switching network model to extract features from the feature vectors to obtain feature information; and finally, it uses the communication switching network model to determine network switching decisions based on the feature information. The network status decision is used to determine the data transmission method. Thus, by real-time monitoring of network status, prediction of network traffic and resource requirements, and the formulation of flexible network scheduling strategies, the reliability, stability, and service quality of the communication network are improved, avoiding frequent switching and communication interruptions. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a communication method provided in an embodiment of this application;

[0020] Figure 2 This is a schematic diagram of a communication method provided in an embodiment of this application. Figure 1 ;

[0021] Figure 3 This is a schematic diagram of a communication method provided in an embodiment of this application. Figure 2 ;

[0022] Figure 4 This is a schematic diagram of a communication method provided in an embodiment of this application. Figure 3 ;

[0023] Figure 5 This is a schematic diagram of the structure of a communication method provided in an embodiment of this application;

[0024] Figure 6This is a schematic diagram of the structural composition of a communication device provided in an embodiment of this application;

[0025] Figure 7 This is a schematic structural diagram of a terminal device provided in an embodiment of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0027] It should be noted that the terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0028] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following relevant technologies are optional solutions and can be combined with the technical solutions of the embodiments of this application in any way, and they all fall within the protection scope of the embodiments of this application.

[0029] Currently, there are many communication technology solutions, each suitable for different scenarios and needs. For example, one method involves network switching based on current network signal strength. In this method, the communication terminal's control channel sensing module senses the signal strength of the 5G communication channel and controls network switching accordingly, executing the corresponding operation. When the 5G communication channel has a communication signal and its wireless signal strength is greater than a preset threshold, the system switches to the 5G communication channel; when the 5G communication channel has a communication signal and its wireless signal strength is less than the preset threshold, or when the 5G communication channel has no communication signal, the system switches to the satellite communication channel. Another example is a vehicle communication network switching method based on in-vehicle navigation applications. In this method, the user must activate the in-vehicle navigation application. Simultaneously, the software can obtain the current device terminal's location and route information in real time. Combined with the distribution of ground base stations around the route, the core network calculates the base stations along the route where there is no ground network and where the ground network has been restored, thereby enabling the in-vehicle communication terminal to switch between satellite and ground communication networks.

[0030] However, the above solutions suffer from a lack of fine-grained control over overall network handover. These solutions perform network handover on a terminal-by-terminal basis, rather than targeting specific applications, which may lead to unnecessary interference for some applications during the handover process. Furthermore, when network resources are scarce, overall handover can result in inconsistent terrestrial network quality, frequent switching back to satellite communication and terrestrial networks, causing repeated resource contention and an unstable user experience. Therefore, this application proposes the following technical solution.

[0031] To facilitate understanding of the technical solutions of the embodiments of this application, the technical solutions of this application are described in detail below through specific embodiments. The above-mentioned related technologies are optional solutions and can be arbitrarily combined with the technical solutions of the embodiments of this application, all of which fall within the protection scope of the embodiments of this application. The embodiments of this application include at least some of the following contents.

[0032] Figure 1 This is a flowchart illustrating the communication method provided in an embodiment of this application, as shown below. Figure 1 As shown, applied to a terminal device, the communication method includes:

[0033] Step 101: Receive network data and determine network status information based on the network data.

[0034] In this embodiment, the terminal device can be a mobile terminal with a satellite communication module and a cellular communication module. For example, the mobile terminal can be a vehicle-mounted terminal. The terminal device has multiple applications. By receiving data sent by each application service, the data is sent to the satellite communication and / or cellular network. Then, the network data returned by the satellite communication module and / or cellular communication module is received. Based on the performance indicators of the network data, the network status information of each application service is estimated and determined.

[0035] In some implementations, the terminal device monitors satellite communications and cellular networks in real time. By collecting, analyzing, and processing network-related data, it estimates various performance indicators of the current network to monitor its operational status and ultimately selects a communication network based on the comprehensive network status information. It should be noted that this application may perform network status estimation only on network data returned from the terrestrial cellular network.

[0036] In some implementations, network status information includes, but is not limited to, network round-trip time, network packet loss rate, packet error rate, and real-time network bandwidth.

[0037] Step 102: The network state information is represented and feature vectors are extracted by using a communication switching network model.

[0038] In this embodiment, a deep learning-based end-to-end communication switching network model is configured on the terminal device. This model receives network status information and application service information collected and processed by the terminal device from the environment. After processing, it outputs a channel switching decision for each application to use either a satellite communication channel or a cellular network channel for data stream transmission. Specifically, the communication network switching model extracts feature vectors from the network status information. The raw data is then transformed into a high-dimensional feature vector representation, providing input for subsequent enhanced feature extraction.

[0039] In some implementations, the communication network switching model may include an input processing module, which may be a Value2Vec network. The communication network switching model uses the Value2Vec network to learn representations between network states, mapping the collected network state input values ​​to a high-dimensional feature space and outputting a high-dimensional vector. Specifically, the Value2Vec network is used to extract representations from the raw data of network state information, converting the network state information into a high-dimensional feature vector representation.

[0040] Specifically, refer to Figure 2 , Figure 2 A schematic diagram of a communication method provided in an embodiment of this application. Figure 1 .like Figure 2 As shown, the communication switching network model first uses a Value2Vec network to learn representations between network states, mapping the collected network state input values ​​to a high-dimensional feature space and outputting a high-dimensional vector. The Value2Vec network is used to extract representations from the raw network state information, converting it into a high-dimensional feature vector representation. Value2Vec uses a Dense layer network to connect each data point in the input vector to each Filter node in the next layer, forming a fully connected topology. This allows for comprehensive feature combination of the input and transforms the input vector to any dimension.

[0041] Figure 3 This is a schematic diagram of a communication method provided in an embodiment of this application. Figure 2 This is a schematic diagram of the communication switching network model provided in the embodiments of this application, showing the network structure of each filter, with reference to... Figure 3 In a Value2Vec network, there are N such... Figure 3 The filter shown maps the input to an N-dimensional feature space and outputs an embedding vector of the network state. Formulas (1), (2), and (3) are as follows:

[0042]

[0043] B = [B1, B2, ..., B N ] T (3)

[0044] Among them, X t X represents the network state information at the current time t. t It is a vector that includes, but is not limited to, network round-trip time (RRT), network packet loss rate (LR), packet error rate (WR), and real-time network bandwidth (BW). For example, X... t =(RTT) t ,LR t ,WR t BW t ); m is the dimension of the network state. For example, if X... t =(RTT) t ,LR t ,WR t BW t If m = 4, then N is the number of output nodes of the Dense network. It is consistent with the dimension of the feature vector used by each token in the multi-scale temporal feature extraction module. It is a hyperparameter of the model. The larger the value, the stronger the expressive power of the model, but the higher the computation and storage cost of the model. For example, N is 128. W is the weight parameter of the Dense network, which is an N×m dimensional vector, which can be obtained through model training. B is the bias parameter of the Dense network, which is a 1×N dimensional vector, which can be obtained through model training. The output of the Dense network is a 1×N dimensional vector representing the embedding vector of the network state. For example, we have... ReLU(·) is the ReLU activation function for Dense networks.

[0045] Step 103: Extract feature information from the feature vector by switching the network model through communication.

[0046] In this embodiment of the application, the terminal device uses a deep learning-based end-to-end communication switching network model to represent and extract network state information to obtain feature vectors, and then extracts features from the feature vectors to obtain feature information.

[0047] In some implementations, the communication network handover model may include a multi-scale temporal feature extraction module, which may be a Transform-Encoder network. Specifically, refer to... Figure 2 , Figure 2 This is a schematic diagram of a communication method provided in an embodiment of this application. Figure 1 .like Figure 2As shown, the communication switching network model first converts the network state information into a high-dimensional feature vector representation through the Value2Vec network, and then uses the Transform-Encoder network to fuse network state vectors and position vectors of multiple scales to form temporal network state tokens. After learning through multiple encoder layers, it extracts complex temporal features, trend features, global correlation features, etc. between network states, and obtains more complex and comprehensive temporal enhanced feature information of network states.

[0048] Step 104: Determine network handover decisions based on feature information using a communication handover network model, wherein the network handover decisions are used to determine the data transmission channel.

[0049] In this embodiment, the terminal uses a deep learning-based end-to-end communication handover network model to characterize and extract feature vectors from network state information. Then, it extracts features from these feature vectors to obtain feature information, and determines a network handover decision based on the temporal information of this feature information. This network handover decision determines the application data transmission method, which can be either using a satellite channel or a cellular network channel for each application to transmit data. Specifically, the terminal device is a mobile terminal with both a satellite communication module and a cellular network module. Based on the network state information of each application and the communication network handover model, it can decide on the network channel for data transmission for each application. That is, the communication network handover model outputs a decision on whether each application should use a satellite communication channel or a cellular network channel to transmit data streams. The network handover decision can be a switch from cellular network to satellite communication, a switch from satellite communication to cellular network, a maintenance of cellular network communication, or a maintenance of satellite communication.

[0050] In some implementations, the communication switching network model may include a prediction module and an intelligent decision-making module. The prediction module may be a Temporal Predictive Network (T-Predict), and the intelligent decision-making module may be a Reciprocal Link Network (RLN). Specifically, refer to... Figure 2 , Figure 2 A schematic diagram of the communication handover network model provided in the embodiments of this application. Figure 1 .like Figure 2As shown, the T-Predict network consists of a temporal convolutional layer (T-Conv), a pooling layer (Pool), and two fully connected layers (FC). The RLN decision network includes a Q-network, a target network, and an action decision network. The T-Predict network performs scale compression and feature compression on the temporal augmentation feature representation output by the Transform-Encoder to generate network state predictions for future time steps, thus completing the temporal prediction task. The RLN decision network receives the current network state token output by the Transform-Encoder network, makes scheduling and decisions for network switching among various service applications, and can learn and update the logic processing and decision-making strategies for network switching online through real-world feedback.

[0051] As described above, the communication method provided in this application involves a terminal device receiving network data and determining network status information based on the network data. A network status information is then characterized and extracted using a communication switching network model to obtain a feature vector. Feature information is extracted from the feature vector, and a network switching decision is made based on this feature information. The network status decision is used to determine the data transmission channel. This approach, compared to existing technologies, does not rely entirely on data acquisition and processing from third-party applications or external data sources. It enables more accurate network switching decisions by combining models with real-time data, thereby improving the system's adaptability and response speed to changes in the network environment. Furthermore, by monitoring the network status information of each application and dynamically adjusting the communication network of each application, intelligent network management is achieved, ensuring that network resources are always allocated and used in the best way, thereby maximizing network resource utilization efficiency, avoiding resource waste, and improving overall communication efficiency. Finally, by dynamically selecting the application's communication network based on the current network status and the network resources occupied by the application through an end-to-end communication switching network model decision, more flexible and intelligent network switching decisions are provided, optimizing resource allocation and improving overall network performance and user experience.

[0052] In some implementations, step 101 above may include: receiving network data and determining the data packets received from the network data, the number of lost data packets, and the number of erroneous data packets.

[0053] Here, the terminal device receives data information sent by various application services, processes the data information and sends it to the network device. After receiving the data information sent by the terminal device, the network device processes the data and generates network data composed of multiple data packets, and returns the network data to the mobile device.

[0054] In some implementations, the terminal device determines the time information of the network status based on data packets.

[0055] Specifically, the round-trip time (RTT) of a network refers to the time it takes for data to travel from the sender to the receiver and back. The sender can be a terminal device, and the receiver can be a network device. Specifically, when a terminal device sends a data packet to a network device, it adds a timestamp T to the packet. A1 When a network device receives a data packet, it records the data packet reception timestamp T. B1 and T A1 T B1 and the current timestamp T B2 The data is packaged into a data packet and sent back to the terminal device. The terminal device records a timestamp T upon receiving the data packet. A2 and take out T A1 T B1 T B2 The data can be used to calculate the RTT delay as shown in the following formula (4):

[0056] RTT=T A2 -T A1 -(T B2 -T B1 (4)

[0057] In some implementations, the terminal device determines the network packet loss rate based on the number of lost data packets.

[0058] Specifically, network packet loss rate (LR) refers to the ratio of the number of data packets lost during data transmission to the total number of data packets sent, usually expressed as a percentage. Specifically, it is calculated by dividing the number of lost data packets by the expected number of data packets to be delivered within a given period. The formulas for calculating the network packet loss rate (LR) are as follows: (5) and (6):

[0059] Exp = S max -S min +1(5)

[0060]

[0061] Where LR is the packet loss rate within one period; Exp is the expected number of packets received; S max S is the sequence number of the largest data packet received within this period. min The sequence number of the smallest data packet received within this period; N is the actual number of data packets received.

[0062] In some implementations, the terminal device determines the packet error rate of the network status based on the number of erroneous packets.

[0063] Here, the network packet error rate usually refers to the proportion of packets that have errors during network data transmission. A Cyclic Redundancy Check (CRC) is added to each packet. CRC is an error detection mechanism widely used in digital networks and storage devices. It generates a checksum by performing mathematical operations on data blocks so that the receiving end can detect errors that may occur during transmission. CRC is widely used in communication protocols, file transfer, storage devices, and other fields to ensure data integrity. The data receiving end can obtain data error packets through verification. The formula for calculating the packet error rate is as follows: (7)

[0064]

[0065] Where WR is the packet error rate within one period; N w N represents the number of data packets that failed verification within this period. t This represents the total number of data packets received during this period.

[0066] In some implementations, the terminal device determines network status information based on time information, network packet loss rate, and packet error rate. It should be noted that the terminal device can determine network status information based on one or more of these factors, including time information, network packet loss rate, and packet error rate; it can also determine network status information based on other network data.

[0067] As can be seen from the above, the communication method provided in this application, by receiving network data, determines the data packets received, the number of lost data packets, and the number of erroneous data packets; based on the data packets, determines the time information of the network status; based on the number of lost data packets, determines the network packet loss rate of the network status; based on the number of erroneous data packets, determines the data packet error rate of the network status; and based on the time information, network packet loss rate, and data packet error rate, determines the network status information. In this way, it can determine the communication mode selected by each application based on the real-time bandwidth usage of each application and the current packet loss rate of the ground network, avoiding frequent network switching of the entire terminal due to the needs of a single application, reducing the switching frequency, reducing resource consumption and communication interruptions during the switching process, improving network stability, and reducing the negative impact of switching. At the same time, network problems in one application will not affect the normal operation of other applications, ensuring the performance and stability of application operation.

[0068] Figure 4 This is a schematic diagram of a communication method provided in an embodiment of this application. Figure 3 This is a schematic diagram of the communication handover network model provided in the embodiments of this application, as shown below. Figure 4 The diagram shown illustrates the structure of the Transform-Encoder network. (Refer to...) Figure 4 Before extracting feature information from the feature vector, the following steps are also included:

[0069] Obtain historical network status information within a preset period.

[0070] Here, the input to the Transform-Encoder network is a combination of historical network state vectors and position vectors at multiple scales, forming temporal network state tokens. Therefore, the input to the Transform-Encoder network is based on historical network state tokens within a preset period. For example, it includes historical network state information at three scales: network state information tokens from the previous n time periods, network state information tokens from the same time point of the previous week, and network state information tokens from the same time point of the previous month.

[0071] Accordingly, step 103 above includes: performing global temporal feature extraction on the feature vector based on historical network state information to obtain at least one feature information.

[0072] Here, the Transform-Encoder network is responsible for extracting global temporal features from the historical network state input token across multiple scales. It generates temporal features through a multi-layer Self-Attention mechanism and a Forward network, capturing multi-level temporal features and global correlation features of the network state. The input to the Transform-Encoder network is the historical network state token, and the output is a high-dimensional feature representation of the temporal network state, divided into two parts. Specifically, the first part of the Transform-Encoder network output is used by the communication switching network model for intelligent decision-making regarding network switching. This first part can be the output at position 0, representing the high-dimensional features of the current network state fused with multi-scale global temporal information. The second part of the output is used by the communication switching network model to predict the network state at the next time step. This second part can be a multi-dimensional temporal output.

[0073] For example, the Transform-Encoder network uses 6 encoder layers, each using dual-head attention. The feature vector embedding of the network state information output in step 102 is fused with the position embedding and input into the encoder module of the Transformer. After passing through multiple encoder blocks, the high-dimensional feature representation sequence of the network state is output as follows: formulas (8), (9), (10), and (11):

[0074]

[0075] H = Encode(Z) (9)

[0076] Encode iMHA (Z i =LayerNorm(Z) i +MultiHeadAttention(Z i (10)

[0077] Encode i (Z i = layerNorm(Encode) iMHA (Z i )+FeedForward(Encode iMHA (Z i (11)

[0078] in, This is a high-dimensional representation of the network state across multiple time scales, representing the network state token. `pos` is the position embedding, representing the temporal position of the historical network state. Positions 0 to n are mapped to [0, 1) using trigonometric functions. Each position is a 1×N dimensional vector; even-numbered dimensions are generated using a sine function, and odd-numbered dimensions using a cosine function. `Z` is the input token of the Encoder module, which incorporates the sequence position encoding information `pos` from the historical time step. In the vector, is an n×N dimensional vector, whose dimension remains constant throughout the entire Encoder layer; H is the output of the entire Encoder module, representing the high-dimensional feature representation sequence of the network state, which is also an n×N dimensional vector; Add(·) adds two vectors; Encode(·) is the operation of the entire Encoder module, which is obtained by stacking multiple Encoder blocks; Encodei(Z i Z represents the operation of the i-th Encoder block. i This is the input to the i-th Encoder block; MultiHeadAttention(Z) i `)` represents the multi-head attention operation of the i-th Encoder block, which combines multiple self-attention modules. The outputs of these self-attention modules are concatenated through a function and passed through a linear layer. For example, two self-attention modules are used; `LayerNorm(·)` normalizes the feature vector; `Encode`... iMHA (Z i ) represents the output of the multi-head attention module of the i-th Encoder block; FeedForward(EncodeiMHA (Z i )) is the feedforward network output of the i-th Encoder block, which consists of two fully connected layers. The activation function of the first fully connected layer is ReLU, and no activation function is used in the second layer.

[0079] In some implementations, before performing step 104, the method further includes: acquiring historical network decision information of network data within a preset period. Here, the terminal statistically analyzes the decision actions of each application service regarding whether to perform network switching, and compiles these into a set of historical network decision information.

[0080] Accordingly, step 104 includes: determining the network state based on at least one feature; determining the network handover action corresponding to the network state based on historical network handover decision information; determining the reward value corresponding to the network handover action based on the current network simulation environment; and determining the network state decision based on the reward value, wherein the network state decision is used to determine the data transmission channel.

[0081] In some implementations, network status decisions are made based on reward values. These network status decisions are used to determine data transmission channels and include: obtaining the historical operating status and historical reward values ​​of the application service within a preset period; determining a first cumulative value based on the historical operating status; determining a first reward value based on the historical reward value; determining a network status decision based on the first reward value and the first cumulative value; and determining the data transmission channel corresponding to the application service based on the network status decision.

[0082] Here, the communication handover network model includes an RLN decision network, which specifically comprises a Q-network, a Target network, and an action decision network. It is responsible for making decisions and scheduling network handover, generating decision actions for each application service regarding whether to perform a network handover. Specifically, the RLN decision network receives at least one feature information token value from the Transform-Encoder network output through a deep Q-learning network. The RLN decision network selects only the token value H0 from the most recent position of the at least one feature information as the state variable S of the Q-network. The set of decision actions for each application service regarding whether to perform a network handover is A. The Target network evaluates the value of each decision action, and the action selection network selects the action with the highest value evaluation score for execution, automatically learning the optimal action strategy.

[0083] Specifically, assume the state at time t is s. t , where s t Let H0|t be the network state token output by the multi-scale temporal feature extraction module at time t. t The input is fed into a deep neural Q-network Q(s, a, w), and the action a with the highest value evaluation is selected from the network's output.t (a t ∈A) as the current decision, where a t This describes the application service's network switching action at time t. After the system executes this decision, the environment will return a reward value R. t This value can be positive or negative, and its range is [-1, 1]. It is calculated as the ratio of the sum of reward values ​​that each application service can obtain after executing this decision to the sum of the maximum reward values ​​that the current application service can obtain.

[0084] For example, after executing the action with the highest value as the current decision, before the next decision cycle arrives, the operational status of each application service is collected, and the calculated reward value can be determined according to the following rules:

[0085] The first condition is set as follows: Within N historical decision periods, there have been A or more consecutive network handovers, which is taken as the first preset value. If the first condition is not met, the service times out, and the service is running on a satellite network, which is taken as the first preset value. If the first condition is not met, the service times out, and the service is running on a cellular network, which is taken as the second preset value. If the first condition is not met, the service runs normally, and the service is running on a satellite network, which is taken as the third preset value. If the first condition is not met, the service runs normally, and the service is running on a cellular network, which is taken as the fourth preset value. For example, it can be set as follows: Within 5 historical decision periods, there have been 2 or more consecutive network handovers, which is taken as -2. If the first condition is not met, the service times out, and the service is running on a satellite network, which is taken as -2. If the first condition is not met, the service times out, and the service is running on a cellular network, which is taken as -1. If the first condition is not met, the service runs normally, and the service is running on a satellite network, which is taken as 1. If the first condition is not met, the service runs normally, and the service is running on a cellular network, which is taken as 2.

[0086] Simultaneously, the environment changes, and the first network state transitions to the next state, namely the second network state s. t+1 At this time, the Target network has state s t+1 There is an action value estimate Q(s) t+1 a t+1 w t Thus, through iterative updates via model training, the maximum value (max) among all action value estimates within a preset period can be obtained. a Q(s t+1 ,a,w t In each decision-making cycle, the action corresponding to the highest value evaluation score is selected as the decision action a|max. a Q(s t+1 ,a,w tBy maximizing the total reward, the optimal action strategy is automatically learned. Specifically, the following formulas (12), (13), (14), and (15) are used:

[0087] target=R t +γQ(s t+1 ,a t+1 ,w t (12)

[0088]

[0089] Among them, s t The network state token output by the multi-scale temporal feature extraction module at time t is H0|t; t The application service switching network action at time t; target is the target value of the decision action at time t; Q(s) t a t w t Q(s) represents the predicted value of the evaluation corresponding to the decision action at time t; t+1 a t+1 w t ) represents the predicted value of the action at time t+1; max a Q(s t+1 ,a,w t ) represents the largest value among all action value estimates; w represents the weight parameters of the neural network, which are iteratively updated through model training. t Let be the network parameters at time t; γ be the discount factor for future value; R t This represents the action a of executing a decision. t The actual reward value after environmental feedback; R max The sum of the maximum reward values ​​that all application services can obtain; r i To implement this decision, and before the next decision cycle arrives, the reward value is calculated by collecting the operational status of each application service.

[0090] Thus, in evaluating the value of an action, Q(s) t a t w t The value R in the current state is included. t and the cumulative discounted value γQ(s) of the estimated reward for future states t+1 a t+1 w tWhen γ is 0, the deep learning network learns only based on the short-term reward of the current state. When γ is 1, the reward of the current state and the reward of the future state are proportional, taking into account both short-term and long-term rewards. In each decision cycle, the action corresponding to the highest value score is selected as the decision action a|max. a Q(s t+1 ,a,w t It automatically learns the optimal action strategy by maximizing total reward.

[0091] In some implementations, before the communication switching network model determines network state decisions based on the time-series information of feature information, it further includes: performing time-series prediction on at least one feature information to obtain the time-series information of at least one feature information.

[0092] Here, the communication switching network model first uses the T-Predict temporal prediction network to achieve scale compression and feature compression, generating network state prediction values ​​for future time moments, thus completing the temporal prediction task. Specifically, the T-Predict network is responsible for feature aggregation and feature transformation of the high-dimensional temporal features of the network state, generating network state prediction values ​​for future time moments. T-Predict receives at least one feature information from the output of the Transformer-Encoder module, which can be all features except for the token value H0 at the most recent time moment. First, a T-Conv convolution merges the data in the time dimension, and then, through regression, outputs the predicted values ​​of the network state as shown in formulas (16) and (17):

[0093] V = ReLu(H*) T W1 T (16)

[0094]

[0095] Among them, W1 T These are the weights of T-Conv, an N×n×1 dimensional vector; T It is a temporal convolution operator; Max(0, H) represents a max pooling layer, which outputs a 1×N dimensional vector; B1 is the weight of the first fully connected layer, which is an N×64-dimensional vector. The output of the first fully connected layer is a 1×64-dimensional vector. B2 is the bias of the first fully connected layer, which is a 1×64-dimensional vector. B1 is the weight of the second fully connected layer, which is a 64×m dimensional vector. The output of the second fully connected layer is a 1×m dimensional vector. For example, m=4. B2 is the bias of the second fully connected layer, which is a 1×m dimensional vector.

[0096] In some implementations, the communication handover network model is a trained communication handover network model, specifically, it also includes:

[0097] Acquire historical network status information and historical network status prediction values ​​within a preset period, as well as historical network handover decision information within a preset period.

[0098] Iteratively execute the following steps until the loss value of the communication switching network model to be trained meets the preset convergence condition:

[0099] Historical network state information, historical network state predictions, and historical network handover decision information are input into the communication handover network model to be trained to obtain the prediction results of the network state decisions. The loss value of the prediction results is calculated, and the network parameters of the communication handover network model to be trained are adjusted based on the loss value. The communication handover network model to be trained that meets the preset convergence conditions is used as the trained communication handover network model.

[0100] Here, a training set for training the communication handover network model is formed by collecting historical network status information from multiple terminal devices at various times and a set of network handover actions based on certain rules. Specifically, the terminal devices acquire historical network status information of network data within a preset period, as well as historical network handover decision information of network handover decisions within the preset period. The preset period can be one month, one week, or one day, etc., and this application does not impose any limitations on it.

[0101] In this application, the time interval for the terminal device to obtain historical network status information and historical network switching decision information of each application within a preset period is not limited in any way.

[0102] The historical network status information includes, but is not limited to, historical network round-trip time, historical network packet loss rate, historical data packet error rate, and historical network bandwidth. This application does not make any specific limitations on the specific reference of the historical network status information.

[0103] The historical network handover decision information includes, but is not limited to, network handover decisions made by the application within a preset period. These network handover decisions may be a switch from cellular network to satellite communication, a switch from satellite communication to cellular network, a switch to maintaining cellular network communication, or a switch to maintaining satellite communication. This application does not limit the specific reference to historical network handover decisions.

[0104] In some implementations, before inputting historical network state information and historical network handover decision information into the communication handover network model to be trained to obtain the prediction result of the network state decision, the following steps are also included:

[0105] Historical network state information and historical network switching decision information are collected to obtain training information; the training information is then processed to obtain network parameters.

[0106] Accordingly, historical network state information and historical network handover decision information are input into the communication handover network model to be trained to obtain the prediction results of network state decision, including: inputting network parameters into the communication handover network model to be trained to obtain the prediction results of network state decision.

[0107] Specifically, the actual network state values ​​of multiple terminals at various times are collected, and a set of network switching actions based on certain rules is formed to create training information. The gradient descent method is used to process the model information and update the network model parameters. The calculation formulas are as follows: (18), (19) and (20):

[0108]

[0109] Among them, Loss t |predict represents the error in predicting the network state at time t, which can be achieved using the Mean Absolute Error (MAE) loss function; Loss t |policy is the deviation between the estimated value of the deep Q(s, a, w) network at time t and the target value, which can be the mean square error (MSE) loss function; α is the learning rate, which can be a constant (empirical value) or can vary with the number of training iterations, depending on the convergence of the training. Take the partial derivative of the loss function with respect to the weight parameter w; w t+1 These are the updated network parameters at time t+1.

[0110] For example, network status and network switching action sets for M months were collected offline, with a collection period of T seconds. Assuming the current time is the 10th moment of the day, the network status corresponding to the historical network status information can be obtained as follows: the status X = (X9, X8, ..., X0, X...) at the same moment in the previous 10 moments, the same moment in the previous week, and the same moment in the previous month. 10 |last_week,X 10 The last_month value is input into the end-to-end communication handover network model, and the T-Predict network in the communication handover network model outputs the predicted value. Will Compared with the actual collected X 10 Compare and calculate the loss. t |predict, and update the parameters w of the input processing module, multi-scale temporal feature extraction module, and prediction module in the end-to-end communication switching network model using gradient descent.10 |predict to w 11 |predict, thus enabling iterative training.

[0111] After the input processing module, multi-scale temporal feature extraction module, and prediction module are trained, the intelligent decision-making model is trained offline. The output H|t=10 of the multi-scale temporal feature extraction module at the current time is extracted, and the feature vector H|t=10 closest to the current time is selected as the state s of the Q(s, a, w) network. 10 The environmental feedback Q(s) 10 a 10 R 10 s 11 Input the data into Q(s, a, w), and retrieve the data corresponding to action a. 10 The model prediction value Q(s) 10 a 10 w 10 ), representing state s 10 Next, execute action a 10 The estimated value of the model at this time, when the network state transitions to s 11 In the actual environment, action a is performed. 10 Receive reward R 10 For s 11 There is an estimated value m A axQ(s 11 ,A|t=10,w 10 w is updated using formula (19) and gradient descent. 10 Parameters to w 11 Through this iterative training, the parameters of the network model are trained, making the network model's estimation of the action value function increasingly accurate.

[0112] Thus, in the offline training of the intelligent decision-making model, what can be learned are switching rules formulated according to certain empirical patterns and rules learned from previously encountered operating scenarios. However, the actual operating environment is ever-changing, and a model trained only offline cannot cover and be applicable to all applicable scenarios. Based on the basic model parameters completed in offline training, the intelligent decision-making module can adaptively learn and update the model parameters online according to the actual environment and operating status. The update method is consistent with that in offline training. Since each iteration update in online learning includes the real network situation and service operating status under real-time decision-making, the neural network model's estimation of the value function of decision actions becomes closer and closer to the real situation, realizing intelligent learning and updating of action decisions and improving the system's adaptability to complex environments.

[0113] As can be seen from the above, the communication method provided in this application provides a unified neural network architecture through an end-to-end communication switching network model, thereby realizing the integration of prediction, scheduling and decision-making. In actual operation, it dynamically learns logic processing and network switching decisions online. This dynamic adjustment mechanism not only optimizes the utilization efficiency of network resources, but also significantly improves the reliability of data communication, ensuring that the application can continue to operate stably in various complex network environments, thereby enhancing the stability and reliability of the overall communication service.

[0114] Figure 5 This is a schematic diagram of the structure of a communication method provided in an embodiment of this application, such as... Figure 5 As shown, taking a vehicle-mounted terminal as an example, the mobile terminal monitors the status of satellite communication and cellular networks in real time, including network latency, packet loss rate, and data packet error rate. Through a deep learning-based end-to-end communication switching network model, it performs representation learning, temporal enhancement feature extraction, temporal prediction, and network switching decisions on multi-scale, multi-temporal network states. Inputting network state information and application service information collected from the environment, it directly outputs the decision of whether each application should use a satellite communication channel or a cellular network channel to transmit data streams. The switching decision is sent to the mobile terminal, which then executes the corresponding application's communication channel switching action. Thus, the end-to-end communication switching network model of this application provides a unified neural network architecture, integrating prediction, scheduling, and decision-making. During actual operation, it dynamically learns logical processing and network switching decisions online. This dynamic adjustment mechanism not only optimizes the utilization efficiency of network resources but also significantly improves the reliability of data communication, ensuring that applications can operate continuously and stably in various complex network environments, thereby enhancing the overall stability and reliability of communication services. The communication method provided in this application integrates satellite communication and terrestrial communication resources to achieve all-round, all-weather communication coverage, thereby meeting the communication needs in different scenarios. Through the mutual backup and complementarity of multiple communication methods, the reliability and stability of the communication system are improved, and problems such as communication failures caused by insufficient terrestrial base station coverage are avoided.

[0115] Compared to traditional communication methods, the communication method provided in this application offers several advantages. First, unlike traditional communication methods which are often limited by the deployment range of ground base stations, this method achieves wide coverage by utilizing satellite resources to provide global communication coverage, including land, sea, and air. Stable communication connections are maintained in cities, remote areas, and even on airplanes. Second, this application utilizes multiple complementary communication methods to ensure communication continuity even in disaster-prone or complex terrain environments, avoiding communication interruptions caused by natural disasters or terrain factors, as is common in traditional communication. Third, unlike traditional communication methods which are easily affected by weather and terrain, the communication method provided in this application utilizes satellite communication technology to avoid the influence of ground obstacles, natural and human interference, thus ensuring communication stability.

[0116] Specifically, the communication method proposed in this application achieves independence and autonomy, allowing it to operate largely independently without entirely relying on data acquisition and processing from third-party in-vehicle navigation applications. It does not depend on external data sources and, through its own algorithms and models combined with real-time data, can make more accurate network switching decisions. This comprehensive data processing capability improves the system's adaptability and responsiveness to changes in the network environment. The communication method proposed in this application achieves fine-grained control, deciding the communication mode selected by each application based on its real-time bandwidth usage and the current packet loss rate of the ground network. This avoids frequent network switching for the entire terminal due to the needs of a single application, reducing switching frequency, minimizing resource consumption and communication interruptions during switching, improving network stability, and reducing the negative impact of switching. Simultaneously, network problems in one application will not affect the normal operation of other applications, ensuring application performance and stability. The communication method proposed in this application optimizes resource utilization by dynamically adjusting the communication network of each application through real-time monitoring of its network requirements and current network status, achieving intelligent network management and ensuring that network resources are always allocated and used optimally. This maximizes network resource utilization efficiency, avoids resource waste, and improves overall communication efficiency. This application proposes a communication method that incorporates an end-to-end communication switching network model. Based on the current network status and the network resources used by the application, the method dynamically selects the application's communication network through the end-to-end communication switching network model. This provides more flexible and intelligent network switching decisions, optimizes resource allocation, and improves overall network performance and user experience.

[0117] Figure 6 This is a schematic diagram of the structural composition of the communication device provided in the embodiments of this application. Figure 1 Applied to terminal devices, such as Figure 6 As shown, the communication device 600 includes:

[0118] The receiving unit 601 is used to receive network data.

[0119] The determining unit 602 is used to determine network status information based on the network data.

[0120] The processing unit 603 is used to characterize and extract feature vectors from the network state information through a communication switching network model; and to extract feature information from the feature vectors through the communication switching network model.

[0121] The determining unit 602 is further configured to determine network state decisions based on the temporal information of the feature information, wherein the network state decisions are used to determine the data transmission method.

[0122] In some embodiments, the receiving unit 601 receives the network data, and the determining unit 602 is further configured to determine the data packets received from the network data, the number of lost data packets, and the number of erroneous data packets; determine the time information of the network status based on the data packets; determine the network packet loss rate of the network status based on the number of lost data packets; determine the data packet error rate of the network status based on the number of erroneous data packets; and determine network status information based on the time information, the network packet loss rate, and the data packet error rate.

[0123] In some embodiments, the processing unit 603 is further configured to acquire historical network status information of network data within a preset period; and to perform global temporal feature extraction on the feature vector based on the historical network status information to obtain at least one feature information.

[0124] In some embodiments, the processing unit 603 is further configured to acquire network decision information of network data within a preset period.

[0125] Accordingly, the determining unit 602 is further configured to: determine the network state based on the at least one feature information; determine a network switching action corresponding to the network state based on the network switching decision information; determine a first reward value corresponding to the network switching action based on the current network simulation environment; and determine the network state decision based on the first reward value, wherein the network state decision is used to determine the data transmission channel.

[0126] In some implementations, the processing unit 603 is also used to obtain the running status and historical reward value of the application service within a preset period.

[0127] Accordingly, the determining unit 602 is further configured to determine a first cumulative value based on the historical operating status, determine a first reward value based on the historical reward value, determine a network status decision based on the first reward value and the first cumulative value, and determine the data transmission channel corresponding to the application service based on the network status decision.

[0128] In some embodiments, the processing unit 603 is further configured to perform time-series prediction on the at least one feature information to obtain time-series information of the at least one feature information; and to perform feature processing on the at least one feature information based on the time-series information of the at least one feature information to obtain a network state prediction value.

[0129] In some embodiments, the communication handover network model is a trained communication handover network model. The processing unit 603 is further configured to acquire historical network state information, historical network state prediction values, and historical network handover decision information of the network data within a preset period.

[0130] Iteratively execute the following steps until the loss value of the communication switching network model to be trained meets the preset convergence condition:

[0131] The historical network state information, the predicted historical network state values, and the historical network handover decision information are input into the communication handover network model to be trained to obtain the prediction results of the network state decision.

[0132] Calculate the loss value of the prediction result, and adjust the network parameters of the communication switching network model to be trained based on the loss value.

[0133] The communication switching network model to be trained that meets the preset convergence conditions is used as the trained communication switching network model.

[0134] In some embodiments, the processing unit 603 is further configured to collect information from the historical network status information and the historical network switching decision information to obtain training information; and to process the training information to obtain network parameters.

[0135] Accordingly, in some embodiments, the processing unit 603 is further configured to input the network parameters into the communication switching network model to be trained, and obtain the prediction result of the network state decision.

[0136] Those skilled in the art should understand that Figure 6 The functions of each unit in the communication device shown can be understood by referring to the relevant description of the aforementioned method. Figure 6 The functions of each unit in the communication device shown can be implemented by a program running on a processor or by specific logic circuits.

[0137] Figure 7 This is a schematic structural diagram of a terminal device 700 provided in an embodiment of this application. Figure 7The terminal device 700 shown includes a processor 710, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0138] Optionally, such as Figure 7 As shown, the terminal device 700 may further include a memory 720. The processor 710 can retrieve and run computer programs from the memory 720 to implement the methods described in this embodiment.

[0139] The memory 720 can be a separate device independent of the processor 710, or it can be integrated into the processor 710.

[0140] Optionally, such as Figure 7 As shown, the terminal device 700 may also include a transceiver 730, and the processor 710 may control the transceiver 730 to communicate with other devices. Specifically, it may send information or data to other devices or receive information or data sent by other devices.

[0141] The transceiver 730 may include a transmitter and a receiver. The transceiver 730 may further include antennas, and the number of antennas may be one or more.

[0142] Optionally, the terminal device 700 may specifically be a mobile terminal / terminal device according to the embodiments of this application, and the terminal device 700 may implement the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0143] It should be understood that the processor in the embodiments of this application may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0144] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0145] It should be understood that the above-described memory is exemplary and not a limiting description. For example, the memory in the embodiments of this application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM), etc. That is to say, the memory in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0146] This application also provides a computer-readable storage medium for storing computer programs.

[0147] Optionally, the computer-readable storage medium can be applied to the network device in the embodiments of this application, and the computer program causes the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0148] Optionally, the computer-readable storage medium can be applied to the mobile terminal / terminal device in the embodiments of this application, and the computer program causes the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0149] This application also provides a computer program product, including computer program instructions.

[0150] Optionally, the computer program product can be applied to the network device in the embodiments of this application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of this application. For the sake of brevity, they will not be described in detail here.

[0151] Optionally, the computer program product can be applied to the mobile terminal / terminal device in the embodiments of this application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, they will not be described in detail here.

[0152] This application also provides a computer program.

[0153] Optionally, the computer program can be applied to the network device in the embodiments of this application. When the computer program is run on the computer, it causes the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0154] Optionally, the computer program can be applied to the mobile terminal / terminal device in the embodiments of this application. When the computer program is run on a computer, it causes the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0155] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0156] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0157] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0158] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0159] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0160] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0161] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A communication method characterized by comprising: The method comprises: receiving network data, determining network state information based on the network data; characteristic vector extraction of the network state information is performed by a communication switching network model; feature extraction of the characteristic vector is performed by the communication switching network model to obtain feature information; determining a network switching decision based on the feature information by the communication switching network model, wherein the network state decision is used to determine a data transmission method.

2. The method of claim 1, wherein, The receiving network data, determining network state information based on the network data, comprises: receiving the network data, determining the number of received data packets, lost data packets and error data packets; determining the time information of the network state based on the data packets; determining the network packet loss rate of the network state based on the number of lost data packets; determining the data packet error rate of the network state based on the number of error data packets; determining network state information based on the time information, the network packet loss rate and the data packet error rate.

3. The method of claim 1, wherein, Before the feature extraction of the characteristic vector to obtain the feature information, the method further comprises: obtaining historical network state information of network data within a preset period; Correspondingly, the feature extraction of the characteristic vector to obtain the feature information comprises: global time domain feature extraction of the characteristic vector based on the historical network state information to obtain at least one feature information.

4. The method of claim 3, wherein, Before the network state decision based on the feature information by the communication switching network model, wherein the network state decision is used to determine the data transmission channel, the method further comprises: obtaining network decision information of network data within a preset period; Correspondingly, the network state decision based on the feature information by the communication switching network model, wherein the network state decision is used to determine the data transmission channel, comprises: determining the network state based on the at least one feature information; determining the network switching action corresponding to the network state based on the network switching decision information; determining a first reward value corresponding to the network switching action based on the current network simulation environment; determining the network state decision based on the first reward value, wherein the network state decision is used to determine the data transmission channel.

5. The method of claim 4, wherein, The determination of the network state decision based on the first reward value, wherein the network state decision is used to determine the data transmission channel, comprises: obtaining the running state and historical reward value of the application service within a preset period; determining a first cumulative value based on the historical running state and the first reward value based on the historical reward value; determining the network state decision based on the first reward value and the first cumulative value, and determining the data transmission channel corresponding to the application service based on the network state decision.

6. The method of claim 3, wherein, The method further comprises: time series prediction of the at least one feature information to obtain time series information of the at least one feature information; feature processing of the at least one feature information based on the time series information of the at least one feature information to obtain a network state prediction value.

7. The method of any one of claims 1-6, wherein the communication switching network model is a trained communication switching network model; and the method further comprises: obtaining historical network state information, historical network state prediction values of the network data in a preset period, and historical network switching decision information of network switching decisions in the preset period; iteratively performing the following steps until a loss value of a to-be-trained communication switching network model meets a preset convergence condition: inputting the historical network state information, the historical network state prediction values, and the historical network switching decision information into the to-be-trained communication switching network model to obtain a prediction result of the network state decision; calculating a loss value of the prediction result and adjusting network parameters of the to-be-trained communication switching network model based on the loss value; training the to-be-trained communication switching network model that meets the preset convergence condition as the trained communication switching network model.

8. The method of claim 7, wherein, Before the inputting the historical network state information and the historical network switching decision information into the to-be-trained communication switching network model to obtain the prediction result of the network state decision, the method further comprises: performing information collection on the historical network state information and the historical network switching decision information to obtain training information; and performing processing on the training information to obtain network parameters. Correspondingly, the inputting the historical network state information and the historical network switching decision information into the to-be-trained communication switching network model to obtain the prediction result of the network state decision comprises: inputting the network parameters into the to-be-trained communication switching network model to obtain the prediction result of the network state decision.

9. A communications device, characterized by The apparatus applied to a terminal device comprises: a receiving unit configured to receive network data; a determining unit configured to determine network state information based on the network data; a processing unit configured to perform feature extraction on the network state information through a communication switching network model to obtain a feature vector, and perform feature extraction on the feature vector through the communication switching network model to obtain feature information; the determining unit is further configured to determine a network state decision based on time sequence information of the feature information, wherein the network state decision is used to determine a data transmission method.

10. A terminal device, comprising: comprise: a processor and a memory configured to store a computer program, the processor being configured to invoke and run the computer program stored in the memory to execute the method of any one of claims 1-8.

11. A computer readable storage medium, characterized in that, a computer program configured to cause a computer to execute the method of any one of claims 1-8.

12. A computer program product, characterised in that, computer program instructions configured to cause a computer to execute the method of any one of claims 1-8.

Images