Multi-application data channel occupation method, device, equipment and storage medium

Abstract

The application relates to the technical field of communication, and provides a data channel occupation method, device and equipment for multiple applications and a storage medium. The method comprises the following steps: obtaining usage requests of multiple service applications for a data channel, and extracting demand information such as a scene type, an importance level, a priority, application content, required bandwidth and occupation duration in the usage requests; performing sequence conversion on the demand information to obtain corresponding demand sequences; inputting the demand sequences into a pre-trained data channel occupation model to obtain a sorting sequence corresponding to each service application; responding to the usage requests of the service applications according to the sorting sequence; and occupying the data channel by the service applications according to the sorting sequence. The sequence prediction of the model is used to determine the occupation sequence of the data channel by the multiple applications, the multiple applications are prevented from competing for the data channel, problems existing in manual configuration of conflict resolution rules are avoided, the processing efficiency of the competition for the data channel by the multiple applications is improved, and errors are less likely to occur.

Description

Technical Field

[0001] This application relates to the field of communication technology, specifically to a method, apparatus, device, and storage medium for occupying data channels for multiple applications. Background Technology

[0002] With the development and application of 5G technology, in the 5G era, operators can leverage VoNR / VoLTE high-definition video calling services to enhance the user's calling experience—advantages not found in OTT applications, such as no app installation required, number-based calling, and deterministic network experience guarantees. The GSMA (Global System for Mobile Communications Association) proposed EnrichedCalling technology in RCS (Rich Communication Suite), introducing enhanced capabilities into calls and defining three scenarios: Pre-Call, In-Call, and Post-Call. Before a call, the caller can send the called party information such as the call subject, importance, pictures, and location. During a call, both parties can send each other any files, send IM (Instant Message) messages, share locations, maps, doodles, and real-time video. If the called party does not answer, the caller can send a text message or voice message. Operators can upgrade VoNR / VoLTE high-definition video calling services based on these three scenarios.

[0003] The 3GPP (3rd Generation Partnership Project) has also developed the IMS (IP Multimedia Subsystem) data channel standard. Based on VoLTE / VoNR high-definition audio and video calls, and combined with WebRTC (Web Real-Time Communications) technology, it extends the data channel (IMS Data Channel) to synchronize voice and video calls. This enables screen sharing, AR overlay, and even synchronized auditory, visual, tactile, and kinematic information for a fully immersive call experience during high-definition video calls. The IMS Data Channel, based on UDP (User Datagram Protocol), provides a highly real-time single-stream or multi-stream data interaction channel, allowing simultaneous desktop sharing, whiteboard sharing, and file transfer during calls.

[0004] By introducing the Real-Time Interactive Channel (IMSData Channel), industry-specific interactive applications are added to high-definition real-time multimedia calls for vertical industries. End-to-end QoS (Quality of Service) assurance is no longer limited to the specific QoS policies and enforcement for voice and video; it also requires customized QoS policy control for different data service types, supporting negotiation of media QoS attributes based on multiple scenarios. The IMSData Channel does not concern itself with the content and format transmitted in the channel; it only requires the communicating parties to agree on the communication format. Therefore, diverse application content can be transmitted through the IMSData Channel using general methods such as scripts. In scenarios where multiple applications compete for the same number's data channel resources, the existing solution is to manually configure conflict resolution rules to address this issue. However, with the massive number of new 5G call applications, manually configuring conflict resolution rules is time-consuming, labor-intensive, inefficient, and prone to errors. Summary of the Invention

[0005] This application provides a method, apparatus, device, and storage medium for occupancy of data channels in multiple applications, in order to solve the technical problems that manual configuration of conflict resolution rules is time-consuming, labor-intensive, inefficient, and prone to errors.

[0006] In a first aspect, embodiments of this application provide a method for occupying a data channel for multiple applications, including:

[0007] The system acquires data channel usage requests from multiple business applications and extracts demand information from each application's requests. The demand information includes the application's scenario type, importance level, priority, application content, required bandwidth, and duration of use.

[0008] The demand information is sequence-transformed to obtain the demand sequence corresponding to the demand information;

[0009] The demand sequence is input into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications;

[0010] The requests for use of the business applications are responded to according to the sorting sequence; each of the business applications occupies the data channel according to the sorting sequence.

[0011] In one embodiment, inputting the demand sequence into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications includes:

[0012] The demand sequence is input into a pre-trained data channel occupancy model to encode the demand sequence and obtain the context vector corresponding to the demand sequence.

[0013] The context vector is decoded and attention aggregation is performed to obtain the sorted sequence corresponding to each of the business applications.

[0014] In one embodiment, the step of performing sequence transformation on the demand information to obtain the demand sequence corresponding to the demand information includes:

[0015] The requirement information is cleaned and transformed into a text sequence to obtain an integer sequence corresponding to the requirement information.

[0016] Based on the preset target sequence value, the target sequence with a text length less than the preset length in the integer sequence is filled to obtain the requirement sequence corresponding to the requirement information.

[0017] In one embodiment, the data channel occupancy model includes an input layer, an embedding layer, an encoding layer, a decoding layer, and an output layer, wherein:

[0018] The embedding layer is used to perform vector mapping on the demand sequence, transforming the demand sequence into a spatial vector;

[0019] The encoding layer is used to encode the spatial vector to obtain a context vector;

[0020] The decoding layer is used to decode the context vector and perform attention aggregation processing to obtain the sorted sequence corresponding to each of the business applications.

[0021] In one embodiment, the method further includes:

[0022] Obtain historical usage requests of the data channel from multiple business applications;

[0023] A sample dataset is constructed based on the aforementioned historical usage requests;

[0024] The preset basic channel occupancy model is iteratively trained using the sample dataset to obtain a pre-trained data channel occupancy model.

[0025] In one embodiment, the step of iteratively training a preset basic channel occupancy model using the sample dataset to obtain a pre-trained data channel occupancy model includes:

[0026] The sample dataset is input into a preset basic channel occupancy model, and vector mapping is performed on the sample dataset to transform the demand sequence in the sample dataset into a spatial vector;

[0027] The spatial vector is encoded to obtain the historical context vector;

[0028] The historical context vector is decoded and attention-aggregated to obtain the predicted sequence.

[0029] The loss value of the basic channel occupancy model is obtained based on the predicted sequence.

[0030] If the loss value is greater than a preset threshold, the gradient descent optimization algorithm is used to return and execute the step of inputting the sample dataset into the preset basic channel occupancy model. The number of iterations is increased by one until the loss value reaches the minimum value, or when the number of iterations reaches the preset value, the pre-trained data channel occupancy model is obtained.

[0031] In one embodiment, constructing the sample dataset based on the historical usage requests includes:

[0032] Obtain the tag data of the historical usage requests; the tag data is used to characterize the order in which the business application occupies the data channel;

[0033] The historical usage requests are cleaned and serialized to obtain the historical demand sequence corresponding to the historical usage requests.

[0034] Construct an access sequence set corresponding to the historical usage requests based on the tag data;

[0035] The first index length is determined based on the sequence length of the historical demand sequence, and the second index length is determined based on the longest sequence length of each access order sequence in the access order sequence set.

[0036] Based on the first index length and the second index length, the historical demand sequence and the access order sequence set are indexed to obtain a sample dataset.

[0037] Secondly, embodiments of this application provide a multi-application data channel occupancy device, comprising:

[0038] The data processing module is used to obtain multiple business applications' requests for data channel usage and extract the demand information from the usage requests of each business application; the demand information includes the scenario type, importance level, priority, application content, required bandwidth, and duration of use of the business application.

[0039] The sequence conversion module is used to convert the demand information into a sequence to obtain the demand sequence corresponding to the demand information;

[0040] The sequence prediction module is used to input the demand sequence into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications;

[0041] The channel occupancy module is used to respond to the usage requests of the business applications according to the sorting sequence; each of the business applications occupies the data channel according to the sorting sequence.

[0042] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the multi-application data channel occupancy method described in the first aspect.

[0043] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the multi-application data channel occupancy method described in the first aspect.

[0044] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the multi-application data channel occupancy method described in the first aspect.

[0045] The data channel occupancy method, apparatus, device, and storage medium provided in this application embodiment acquire data channel usage requests from multiple business applications, extract demand information such as scenario type, importance level, priority, application content, required bandwidth, and occupancy duration from each business application's usage request, and perform sequence transformation on the extracted demand information to obtain a demand sequence corresponding to the demand information. This demand sequence is input into a pre-trained data channel occupancy model to obtain a sorting sequence corresponding to each business application. The usage requests of each business application are responded to according to the sorting sequence, enabling each business application to occupy the data channel according to the sorting sequence predicted by the model. By determining the data channel occupancy order of multiple applications through model sequence prediction, the competition for data channels among multiple applications is avoided, and the problems associated with manually configuring conflict resolution rules are avoided. This improves the efficiency of handling data channel contention among multiple applications and reduces the likelihood of errors. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1This is one of the flowcharts illustrating the data channel occupancy method for multiple applications provided in this application embodiment;

[0048] Figure 2 This is a second flowchart illustrating the data channel occupancy method for multiple applications provided in this application embodiment;

[0049] Figure 3 This is a schematic diagram illustrating the working principle of the attention mechanism provided in the embodiments of this application;

[0050] Figure 4 This is the third flowchart illustrating the multi-application data channel occupancy method provided in this application embodiment;

[0051] Figure 5 This is a schematic diagram of the structure of the multi-application data channel occupancy device provided in the embodiments of this application;

[0052] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] To address the issue of multiple applications competing for the same data channel during 5G calls, and to resolve problems with manually configured conflict resolution rules, this application provides a method for multiple applications to occupy data channels. This method involves constructing a recurrent neural network and pre-training it based on historical data from the data channel to form a prediction model. The model then predicts sequences with multiple inputs or outputs, automatically sorting the data channel usage requests from multiple applications and occupying the data channel according to the sorted order, thus resolving the data channel contention problem.

[0055] Specifically, refer to Figure 1 , Figure 1 This is a flowchart illustrating a multi-application data channel occupancy method provided in an embodiment of this application. Based on Figure 1 The multi-application data channel occupancy method provided in this application embodiment may include:

[0056] Step 100: Obtain multiple business applications' requests for data channel usage and extract the demand information from each business application's usage request; the demand information includes the business application's scenario type, importance level, priority, application content, required bandwidth, and duration of use;

[0057] The system retrieves multiple data channel usage requests from various business applications. These requests carry information about the application's data channel usage requirements, including the application's scenario type, importance level, priority, application content, required bandwidth, and duration. The system then extracts the specific requirements of each business application from these requests. Different business applications correspond to different scenario types, and the importance and priority levels of the usage requests can be pre-defined based on the importance and priority of each business application.

[0058] Furthermore, the acquired usage requests include usage requests from one or more user business applications to the data channel. Any user's usage request can be initiated by one or more business applications. Multiple usage requests from different business applications can be triggered simultaneously or within the same preset time period; no specific limitations are imposed here.

[0059] Step 200: Perform sequence transformation on the demand information to obtain the demand sequence corresponding to the demand information;

[0060] The extracted demand information is sequence-transformed to obtain the demand sequence corresponding to the demand information. Specifically, the demand information of each usage request is sequence-transformed, and each usage request's demand information corresponds to a demand sequence. The demand sequences corresponding to the usage requests of multiple business applications form a sequence set.

[0061] The extracted requirement information is text data. When transforming this requirement information, it can be done by merging and integrating it according to preset rules to obtain a corresponding text sequence, which is the requirement sequence. Furthermore, the text sequence can be further transformed, for example, by mapping the text sequence, converting the words in the text sequence into corresponding vectors or feature values, obtaining a vector corresponding to the text sequence, and using this vector as the requirement sequence corresponding to the requirement information.

[0062] Step 300: Input the demand sequence into the pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications;

[0063] The demand sequence is input into a pre-trained data channel occupancy model. This model is then used to predict the demand sequence for each business application, resulting in a ranking sequence for each application. This ranking sequence represents the order in which requests occupy the data channel, thus resolving the data channel contention problem.

[0064] Understandably, different business applications have different priorities for the same user, and the same business application may have different priorities for different users. By training a data channel occupancy model according to a predetermined data channel occupancy order, and obtaining the corresponding data channel occupancy model, when multiple business applications from one or more users request the use of data channels, the usage requests of multiple business applications can be predicted and ranked, thus eliminating the need for manual configuration of conflict resolution rules.

[0065] Step 400: Respond to the usage requests of the business applications according to the sorting sequence; each business application occupies the data channel according to the sorting sequence.

[0066] The system responds to usage requests from various business applications in the order predicted by the data channel occupancy model. Based on the responses to these requests, each business application occupies the data channel according to the order, thereby avoiding contention for the data channel among multiple applications.

[0067] Preferably, in one embodiment, the multi-application data channel occupancy method is applied to a 5G new voice system. This system includes a multi-application demand extraction module, a data preprocessing module, and a data channel occupancy model. The system's processing flow for multi-application data channel occupancy is as follows: Figure 2 As shown. In Figure 2In the process, the data channel receives usage requests from multiple business applications (i.e., business application 1, business application 2, business application 3, etc.). These requests carry text data corresponding to requirement information such as scenario type, importance level, priority level, application content, required bandwidth, and usage duration. The multi-application requirement extraction module extracts this requirement information from each usage request and merges the text data corresponding to each business application's requirement information. The data preprocessing module preprocesses the text data extracted by the multi-application requirement extraction module. Preprocessing includes text cleaning and text serialization, transforming the text data corresponding to the requirement information into a requirement sequence, which is then input into a pre-trained data channel occupancy model. The data channel occupancy model predicts the requirement sequence, obtaining a corresponding predicted sequence, which is the sorting sequence for each business application. Each business application's usage request is responded to according to the predicted sequence of the data channel occupancy model. The data channel allows multiple waiting business applications to access the data channel sequentially according to their sorting order, ensuring that each business occupies the data channel in the order corresponding to its predicted sequence.

[0068] In this embodiment, by acquiring data channel usage requests from multiple business applications, demand information such as scenario type, importance level, priority, application content, required bandwidth, and usage duration is extracted from the usage requests of each business application. The extracted demand information is then sequence-transformed to obtain a corresponding demand sequence. This demand sequence is input into a pre-trained data channel occupancy model to obtain a sorting sequence corresponding to each business application. The usage requests of each business application are responded to according to the sorting sequence, ensuring that each business application occupies the data channel according to the model-predicted sorting sequence. Determining the data channel occupancy order through model sequence prediction avoids contention among multiple applications for the data channel and avoids the problems associated with manually configuring conflict resolution rules. This improves the efficiency of handling data channel contention among multiple applications and reduces the likelihood of errors.

[0069] Preferably, in step 300, the demand sequence is input into a pre-trained channel resource allocation model to obtain a sorting sequence corresponding to each of the aforementioned business applications, including:

[0070] Step 301: Input the demand sequence into the pre-trained channel resource allocation model to encode the demand sequence and obtain the context vector corresponding to the demand sequence;

[0071] Step 302: Decode and perform attention aggregation on the context vector to obtain the sorted sequence corresponding to each of the business applications.

[0072] The demand sequence is input into a pre-trained channel resource allocation model. The demand sequence is encoded to obtain a context vector corresponding to the demand sequence. The obtained context vector is then decoded and subjected to attention aggregation processing to obtain a ranking sequence corresponding to each business application. Preferably, in this embodiment, the data channel occupancy model consists of a Long Short-Term Memory (LSTM) network and an attention mechanism. The LSTM network includes an encoder and a decoder. The obtained demand sequence is input into the pre-trained data channel occupancy model. The encoder in the LSTM network encodes the demand sequence to obtain the corresponding context vector. The decoder and attention mechanism in the LSTM network decode and aggregate the context vector to obtain a ranking sequence corresponding to each business application.

[0073] LSTM networks are a special type of recurrent neural network. The encoder-decoder neural network in LSTM networks is an organized recurrent neural network, mainly used to solve sequence prediction problems with multiple inputs or multiple outputs. It includes an encoder and a decoder. The encoder is responsible for encoding the input sequence word by word, generating a fixed-length vector, namely the context vector. The decoder is responsible for reading the context vector output by the encoder and generating the output sequence.

[0074] The attention mechanism addresses the limitations of the encoder-decoder structure. First, it provides the decoder with richer contextual information from the encoder, allowing the encoder to pass more information to the decoder. In contrast to traditional models where the encoder only passes the last hidden state from the encoding stage, the attention mechanism model passes all hidden states to the decoder. Simultaneously, attention provides a learning mechanism where, when predicting the output sequence at each time step, the decoder can learn where to focus within a richer context. The attention network assigns an attention weight to each input; the more relevant the input is to the current operation, the closer the attention weight is to 1, and vice versa. These attention weights are recalculated at each output step.

[0075] Reference Figure 3 The diagram shown illustrates the working principle of the attention mechanism. Figure 3 In the middle, T x To input the number of time steps, T y To output the number of time steps, attention i c represents the attention weight at output time step i. iThis is the context vector at output time step i. An RNN (Recurrent Neural Network) is a recurrent neural network; in this embodiment, the RNN network can be considered as an LSTM. The attention weights are calculated according to Formula 1 below, with a weight length of T. x The sum of all weights is 1:

[0076] attention i =softmax(dense(x,y) i-1 (1)

[0077] The context vector is obtained by summing the products of the attention weights and the input sequence according to Formula 2:

[0078]

[0079] In Equation 2, m represents the number of input sequences. The obtained context vector is input into the LSTM network to obtain the predicted sequence.

[0080] y i =LSTM(c i (3)

[0081] LSTM is a special type of recurrent neural network, where the same neural network is reused repeatedly. LSTM can learn long-term dependent information; by controlling how long the values ​​in the cache are stored, it can remember long-term information, making it suitable for learning long sequences. Each neuron has four inputs and one output, and each neuron contains a cell unit to store the remembered values. Each LSTM neuron contains three gates: a forget gate, an input gate, and an output gate, demonstrating good learning performance for long sequences.

[0082] Furthermore, the data channel occupancy model includes an input layer, an embedding layer, an encoding layer, a decoding layer, and an output layer. The embedding layer is used to perform vector mapping on the demand sequence, transforming the demand sequence into a spatial vector. The encoding layer is used to encode the spatial vector to obtain a context vector. The decoding layer is used to decode the context vector and perform attention aggregation processing to obtain the sorting sequence corresponding to each business application.

[0083] Preferably, step 200, which involves performing sequence transformation on the extracted demand information to obtain the demand sequence corresponding to the demand information, may further include:

[0084] Step 201: Perform text cleaning and text sequence transformation on the demand information to obtain the integer sequence corresponding to the demand information;

[0085] Step 202: Based on the preset target sequence value, fill the target sequence in the integer sequence whose text length is less than the preset length to obtain the requirement sequence corresponding to the requirement information.

[0086] First, the requirement information is cleaned and transformed into a text sequence to obtain the integer sequence corresponding to the requirement information. The text data is then converted into a vector. Then, based on the preset target sequence value, the target sequences with text lengths less than the preset length in the obtained integer sequence are padded to ensure that all sequence lengths are consistent, thus obtaining the requirement sequence corresponding to the requirement information.

[0087] The target sequence value is, for example, 0. Text data is cleaned and serialized into an integer sequence of length L. Sequences shorter than L are padded with zeros to obtain the required sequence. This sequence is then input into an attention-based encoder-decoder neural network. The encoder extracts features from the simultaneously triggered multi-application access request sequence and encodes them separately into a fixed-length context vector h. This context vector h is input into the decoder, which aggregates the learned features to generate a sorted sequence of multiple applications using data channel resources that simultaneously meet the access requirements of multiple applications. This sorted sequence is then imported into the Data Channel, which sequentially allows each application to access the waiting applications according to its sorted order.

[0088] Preferably, the multi-application data channel occupancy method provided in this application embodiment mainly includes stages such as data preprocessing, model building and training, and model application. Based on this, the multi-application data channel occupancy method provided in this embodiment may further include:

[0089] Step 001: Obtain historical usage requests of the data channel from multiple business applications;

[0090] Step 002: Construct a sample dataset based on the historical usage requests;

[0091] Step 003: Use the sample dataset to iteratively train the preset basic channel occupancy model to obtain the pre-trained data channel occupancy model.

[0092] In the data preprocessing and model building and training phases, the historical usage requests of multiple business applications for data channels are first obtained. In the data preprocessing phase, the obtained historical usage requests are preprocessed to build a sample dataset. In the model building and training phase, the built sample dataset is input into the preset basic channel occupancy model, and the preset basic channel occupancy model is iteratively trained to obtain the pre-trained data channel occupancy model.

[0093] Further, in step 002, a sample dataset is constructed based on the acquired historical usage requests, specifically including:

[0094] Step 0021: Obtain the tag data of the historical usage requests; the tag data is used to characterize the order in which the business application occupies the data channel;

[0095] Step 0022: Perform text cleaning and text serialization on the historical usage requests to obtain the historical demand sequence corresponding to the historical usage requests;

[0096] Step 0023: Construct an access sequence set corresponding to the historical usage requests based on the tag data;

[0097] Step 0024: Determine the first index length based on the sequence length of the historical demand sequence, and determine the second index length based on the longest sequence length of each access order sequence in the access order sequence set;

[0098] Step 0025: Based on the first index length and the second index length, index the historical demand sequence and the access order sequence set to obtain a sample dataset.

[0099] When constructing the sample dataset, the first step is to obtain the label data of historical usage requests. This label data is generated by annotating business applications and is used to characterize the order in which business applications occupy the data channel, i.e., the priority of occupation and / or the importance level. The historical usage requests are then cleaned and serialized to obtain the historical demand sequence corresponding to the historical usage requests. The specific text cleaning and serialization process is the same as the model application stage described above and will not be repeated here.

[0100] Based on the tag data of historical usage requests, an access sequence set corresponding to the usage requests of each business application is constructed. The first index length is determined according to the sequence length of the demand sequence, and the second index length is determined according to the longest sequence length of each access sequence in the access sequence set. Based on the determined first and second index lengths, the historical demand sequence and access sequence set are indexed to obtain the sample dataset.

[0101] It should be noted that for multiple business applications, the number of business applications may differ, and / or the different business applications may correspond to different access orders. Furthermore, for different users, different numbers of business applications, and / or the same number of different business applications, and / or the same multiple business applications may also correspond to different access orders. Therefore, different access order sequences can be obtained, and multiple different access order sequences form an access order sequence set. When indexing the access order sequence set, specifically from the user dimension, for the same user's multiple application access needs, the access order sequence is indexed using the business application as the index item and the access order sequence as the index content. The access order sequence is indexed based on the business application. The business applications can be mapped to a set of application identifiers or a set of feature values ​​corresponding to the application identifiers using application identifiers.

[0102] Further, in step 003, based on the constructed sample dataset, the preset basic channel occupancy model is iteratively trained to obtain a pre-trained data channel occupancy model, which may also include:

[0103] Step 0031: Input the sample dataset into the preset basic channel occupancy model, perform vector mapping on the sample dataset, and transform the demand sequence in the sample dataset into a spatial vector;

[0104] Step 0032: Encode the spatial vector to obtain the historical context vector;

[0105] Step 0033: Decode and perform attention aggregation on the historical context vector to obtain the predicted sequence.

[0106] Step 0034: Obtain the loss value of the basic channel occupancy model based on the predicted sequence.

[0107] Step 0035: If the loss value is greater than a preset threshold, the gradient descent optimization algorithm is used to return and execute the step of inputting the sample dataset into the preset basic channel occupancy model. The number of iterations is increased by one until the loss value reaches the minimum value, or when the number of iterations reaches the preset value, the pre-trained data channel occupancy model is obtained.

[0108] The constructed sample dataset is input into a pre-defined basic channel occupancy model. Vector mapping is performed on the sample dataset. In the data preprocessing stage, the constructed sample dataset is an indexed demand sequence. After inputting the sample dataset into the pre-defined basic channel occupancy model, the model performs vector mapping on the sample dataset, transforming each historical demand sequence in the sample dataset into a spatial vector. The spatial vector is encoded to obtain the corresponding historical context vector. This historical context vector is then decoded and attention-aggregated to obtain the predicted sequence. The loss value of the basic channel occupancy model is obtained based on this predicted sequence. Specifically, this can be achieved by comparing the predicted sequence with access sequence sequences in a pre-generated set of access sequence sequences. Specifically, based on multiple business applications corresponding to the predicted sequence, the access sequence set is indexed to determine the access sequence sequences corresponding to these multiple business applications. The model's predicted sequence is compared with the indexed access sequence sequences to determine the matching degree, and the model's loss value is determined based on this matching degree. If the loss value is greater than the preset threshold, the gradient descent optimization algorithm is used to train the model. The model parameters are adjusted and the sample dataset is re-input into the basic channel occupancy model for retraining. The number of iterations is increased by one until the loss value reaches the minimum value, or until the number of iterations reaches the preset value, to obtain the pre-trained data channel occupancy model.

[0109] The following combination Figure 4 The diagram illustrating the multi-application data channel occupancy process describes in detail the multi-application data channel occupancy method provided in this application embodiment. Based on Figure 4 In this embodiment, the multi-application data channel occupancy method includes data preprocessing, model training, and model application stages. In the data preprocessing stage, historical multi-application simultaneous access demand text data and the corresponding manually labeled and sorted multi-application access order are obtained from the Data Channel as the total dataset. This dataset undergoes text cleaning and text serialization. In the model training stage, the basic channel occupancy model is a pre-built encoder-decoder neural network based on long short-term memory neurons. The encoder extracts features from the simultaneously triggered multi-application access demand sequence, encoding it separately into a fixed-length context vector h. This context vector h is input to the decoder, which aggregates the learned features through an attention decoder to generate a sequential sequence of multi-application data channel resource usage that simultaneously satisfies multi-application access demands. This sequence is then compared with the labeled multi-application data channel resource usage ranking result to calculate the objective function. The gradient descent optimization algorithm is used iteratively to determine the model parameters that minimize the objective function; these model parameters are the weight values. The following is based on... Figure 4 The document elaborates on each stage, including data preprocessing, model training, and model application.

[0110] First, in the data preprocessing stage, text data of historical multi-application simultaneous access requests and corresponding manually labeled, sorted multi-application access order sequences are obtained from the Data Channel as the total dataset. This dataset is then cleaned and serialized. The multi-application simultaneous access request information includes text data carried in the application usage requests, such as the application's scenario type, importance level, priority, application content, required bandwidth, and duration. The multi-application access request of the i-th user can be represented as {S1}. i S2 i S3 i S L i}, with length L. In the sorted multi-application access sorting sequence, the multi-application access sorting sequence of the i-th user can be represented as {R1}. i R2 i R3 i ..., R M i}, with a length of M.

[0111] Text data containing historical data on the simultaneous access requirements of multiple applications to the data channel, along with the corresponding sorted sequence of access orders for multiple applications, undergoes text cleaning and serialization. Specifically, all punctuation marks are retained; if the text is Chinese, it is segmented into words; if the text is English, the English letters are converted to lowercase; each word is tokenized, transforming each segment of text into an index number; and zeros are padded to sequences that do not reach the maximum text length.

[0112] Furthermore, the longest length L of the text data containing the demand information for multiple applications to simultaneously access the data channel is taken as its index sequence length, and its dictionary size is multiapp_vocab_size. The longest sequence length M in the corresponding sorted multiapp access sequence set is taken as its index sequence length, and its dictionary size is odered_multiapp_vocab_size. The obtained text-based sequence dataset is indexed to obtain the sample dataset. The sample dataset is then divided into a training set and a test set; for example, 80% of the sample dataset is allocated to the training set, and 20% to the test set. The training set is used to train the model, and the test set is used to test the model.

[0113] During model training, the encoder can extract features from the multi-application access request sequences triggered simultaneously, encoding them individually into a fixed-length context vector h. Specifically, the first layer of the model is the input layer, which takes the indexed multi-application access request sequences as input, with each sequence having a length of L. The output data of this layer is (None, L). The second layer is the embedding layer, which uses word embedding to convert each word in the request sequence into a vector. The input data dimension is multiapp_vocab_size, which is the length of the first index corresponding to the request sequence. The output is set to convert the words into 128-dimensional spatial vectors. Since the input sequence length is L, the shape of the output data of this layer is (None, L, 128). The function of this layer is to perform vector mapping on the words in the input request sequence, converting the index of each word into a fixed-dimensional vector of 128 dimensions. The third layer is the LSTM encoding layer, which contains multiple parallel LSTM layers, each with 128 LSTM neurons. The activation function is set to "ReLU". The output data of this layer is in the form of (None, L, 128), encoded into three fixed-length context vectors.

[0114] Furthermore, the decoder aggregates the learned features through an attention mechanism to generate a sequential sequence of data channel resources used by multiple applications, simultaneously satisfying the access requirements of multiple applications. Specifically, the fourth layer of the model is an attention-based LSTM decoding layer, containing 256 LSTM neurons with the activation function set to "ReLU". The output data of this attention-based LSTM decoding layer is in the form (None, L, 256). The fifth layer of the model is a fully connected (Dense) layer, which is also the output layer of the model. It contains Dense fully connected neurons of the size "odered_multiapp_vocab_size" with the activation function set to "softmax". The output of softmax is fed into the multi-class cross-entropy loss function. The output data of this layer is in the form (None, "odered_multiapp_vocab_size"), generating a sequential sequence of data channel resources used by multiple applications, simultaneously satisfying the access requirements of multiple applications.

[0115] During model training, the number of iterations was set to 1000 (epochs = 1000), and the batch size was set to 100 (batch_size = 100). Categorical cross-entropy was chosen as the loss function, also known as the objective function (loss = 'categorical_crossentropy'). Gradient descent was employed, and the Adam optimizer was selected to improve the learning speed of traditional gradient descent (optimizer = 'adam'). During iteration, the predicted sequence of the model obtained in each iteration was compared with the pre-labeled ranking results of data channel resources used by multiple applications to calculate the objective function. Gradient descent was used to gradually find the weight values ​​that minimized the objective function, thus obtaining the model training result. Alternatively, when the model reached the pre-set number of iterations, the weight value that minimized the objective function during the iterations was used as the model training result. The trained model weights were used as a data channel occupancy model to predict the subsequent data channel occupancy order for multiple applications.

[0116] Furthermore, in the model application phase, the usage requests of multiple applications for the data channel are first obtained from the data channel. Following the processing method in the data preprocessing phase, the usage requests are preprocessed to obtain the corresponding demand sequence. Then, the trained data channel occupancy model is used to predict the demand sequence to obtain the corresponding sorting sequence of multiple applications. The data channel usage requests of multiple applications are responded to according to this sorting sequence, so that multiple applications occupy the data channel in the order of responding to the requests, avoiding contention and conflict for the data channel.

[0117] In this embodiment, historical textual data of multiple applications simultaneously accessing the data channel, along with the corresponding manually labeled and sorted access order of the multiple applications, are obtained from the data channel as the total dataset. This dataset is then cleaned and serialized to construct a corresponding sample dataset. A pre-defined encoder-decoder neural network based on long short-term memory neurons is trained. The encoder extracts features from the sequence of multiple applications accessing the data channel simultaneously and encodes it into a fixed-length context vector h. This context vector h is input into the decoder, which uses attention to aggregate the learned features, generating a sequence of multiple applications using data channel resources that simultaneously meet the access requirements. This sequence is then compared with the sorted results of the labeled multiple applications using data channel resources to calculate the objective function. During model training, a gradient descent optimization algorithm is used to determine the weight values ​​that minimize the objective function, completing the model training. The trained model is then used for online processing of data channel contention among multiple applications, thereby achieving orderly use of the data channel by multiple applications, avoiding data channel contention conflicts, and improving the user experience.

[0118] The following describes the multi-application data channel occupancy device provided in the embodiments of this application. The multi-application data channel occupancy device described below and the multi-application data channel occupancy method described above can be referred to in correspondence with each other.

[0119] Reference Figure 5 The multi-application data channel occupancy device provided in this application embodiment includes:

[0120] The data processing module 10 is used to acquire multiple business applications’ requests to use the data channel and extract the demand information from the requests of each business application; the demand information includes the scenario type, importance level, priority, application content, required bandwidth and duration of the business application.

[0121] Sequence conversion module 20 is used to perform sequence conversion on the demand information to obtain the demand sequence corresponding to the demand information;

[0122] The sequence prediction module 30 is used to input the demand sequence into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications;

[0123] The channel occupancy module 40 is used to respond to the usage requests of the business applications according to the sorting sequence; each of the business applications occupies the data channel according to the sorting sequence.

[0124] In one embodiment, the sequence prediction module 30 is further configured to:

[0125] The demand sequence is input into a pre-trained data channel occupancy model to encode the demand sequence and obtain the context vector corresponding to the demand sequence.

[0126] The context vector is decoded and attention aggregation is performed to obtain the sorted sequence corresponding to each of the business applications.

[0127] In one embodiment, the sequence conversion module 20 is further configured to:

[0128] The requirement information is cleaned and transformed into a text sequence to obtain an integer sequence corresponding to the requirement information.

[0129] Based on the preset target sequence value, the target sequence with a text length less than the preset length in the integer sequence is filled to obtain the requirement sequence corresponding to the requirement information.

[0130] In one embodiment, the data channel occupancy model includes an input layer, an embedding layer, an encoding layer, a decoding layer, and an output layer, wherein:

[0131] The embedding layer is used to perform vector mapping on the demand sequence, transforming the demand sequence into a spatial vector;

[0132] The encoding layer is used to encode the spatial vector to obtain a context vector;

[0133] The decoding layer is used to decode the context vector and perform attention aggregation processing to obtain the sorted sequence corresponding to each of the business applications.

[0134] In one embodiment, the multi-application data channel occupancy device further includes a pre-training module, used for:

[0135] Obtain historical usage requests of the data channel from multiple business applications;

[0136] A sample dataset is constructed based on the aforementioned historical usage requests;

[0137] The preset basic channel occupancy model is iteratively trained using the sample dataset to obtain a pre-trained data channel occupancy model.

[0138] In one embodiment, the pre-training module is further configured to:

[0139] The sample dataset is input into a preset basic channel occupancy model, and vector mapping is performed on the sample dataset to transform the demand sequence in the sample dataset into a spatial vector;

[0140] The spatial vector is encoded to obtain the historical context vector;

[0141] The historical context vector is decoded and attention-aggregated to obtain the predicted sequence.

[0142] The loss value of the basic channel occupancy model is obtained based on the predicted sequence.

[0143] If the loss value is greater than a preset threshold, the gradient descent optimization algorithm is used to return and execute the step of inputting the sample dataset into the preset basic channel occupancy model. The number of iterations is increased by one until the loss value reaches the minimum value, or when the number of iterations reaches the preset value, the pre-trained data channel occupancy model is obtained.

[0144] In one embodiment, the pre-training module is further configured to:

[0145] Obtain the tag data of the historical usage requests; the tag data is used to characterize the order in which the business application occupies the data channel;

[0146] The historical usage requests are cleaned and serialized to obtain the historical demand sequence corresponding to the historical usage requests.

[0147] Construct an access sequence set corresponding to the historical usage requests based on the tag data;

[0148] The first index length is determined based on the sequence length of the historical demand sequence, and the second index length is determined based on the longest sequence length of each access order sequence in the access order sequence set.

[0149] Based on the first index length and the second index length, the historical demand sequence and the access order sequence set are indexed to obtain a sample dataset.

[0150] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 can call a computer program in the memory 630 to execute steps of a multi-application data channel occupancy method, such as including:

[0151] The system acquires data channel usage requests from multiple business applications and extracts demand information from each application's requests. The demand information includes the application's scenario type, importance level, priority, application content, required bandwidth, and duration of use.

[0152] The demand information is sequence-transformed to obtain the demand sequence corresponding to the demand information;

[0153] The demand sequence is input into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications;

[0154] The requests for use of the business applications are responded to according to the sorting sequence; each of the business applications occupies the data channel according to the sorting sequence.

[0155] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, 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.

[0156] On the other hand, embodiments of this application also provide a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the multi-application data channel occupancy method provided in the above embodiments, such as including:

[0157] The system acquires data channel usage requests from multiple business applications and extracts demand information from each application's requests. The demand information includes the application's scenario type, importance level, priority, application content, required bandwidth, and duration of use.

[0158] The demand information is sequence-transformed to obtain the demand sequence corresponding to the demand information;

[0159] The demand sequence is input into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications;

[0160] The requests for use of the business applications are responded to according to the sorting sequence; each of the business applications occupies the data channel according to the sorting sequence.

[0161] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing a processor to execute the steps of the multi-application data channel occupancy method provided in the above embodiments, such as including:

[0162] The system acquires data channel usage requests from multiple business applications and extracts demand information from each application's requests. The demand information includes the application's scenario type, importance level, priority, application content, required bandwidth, and duration of use.

[0163] The demand information is sequence-transformed to obtain the demand sequence corresponding to the demand information;

[0164] The demand sequence is input into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications;

[0165] The requests for use of the business applications are responded to according to the sorting sequence; each of the business applications occupies the data channel according to the sorting sequence.

[0166] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0167] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0168] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for occupying a data channel for multiple applications, characterized in that, include: Obtain data channel usage requests from multiple business applications and extract the demand information from the usage requests of each business application; The requirement information includes the scenario type, importance level, priority, application content, required bandwidth, and duration of the business application. The demand information is sequence-transformed to obtain the demand sequence corresponding to the demand information; The demand sequence is input into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications; The requests for the use of the aforementioned business application are responded to according to the sorting sequence; Each of the aforementioned business applications occupies the data channel according to the aforementioned sorting sequence; The method further includes: Obtain historical usage requests of the data channel from multiple business applications; A sample dataset is constructed based on the aforementioned historical usage requests; The preset basic channel occupancy model is iteratively trained using the sample dataset to obtain a pre-trained data channel occupancy model. The construction of the sample dataset based on the historical usage requests includes: Obtain the tag data of the historical usage requests; the tag data is used to characterize the order in which the business application occupies the data channel; The historical usage requests are cleaned and serialized to obtain the historical demand sequence corresponding to the historical usage requests. Construct an access sequence set corresponding to the historical usage requests based on the tag data; The first index length is determined based on the sequence length of the historical demand sequence, and the second index length is determined based on the longest sequence length of each access order sequence in the access order sequence set. Based on the first index length and the second index length, the historical demand sequence and the access order sequence set are indexed to obtain a sample dataset.

2. The multi-application data channel occupancy method according to claim 1, characterized in that, The step of inputting the demand sequence into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications includes: The demand sequence is input into a pre-trained data channel occupancy model to encode the demand sequence and obtain the context vector corresponding to the demand sequence. The context vector is decoded and attention aggregation is performed to obtain the sorted sequence corresponding to each of the business applications.

3. The multi-application data channel occupancy method according to claim 1, characterized in that, The step of performing sequence transformation on the demand information to obtain the demand sequence corresponding to the demand information includes: The requirement information is cleaned and transformed into a text sequence to obtain an integer sequence corresponding to the requirement information. Based on the preset target sequence value, the target sequence with a text length less than the preset length in the integer sequence is filled to obtain the requirement sequence corresponding to the requirement information.

4. The multi-application data channel occupancy method according to claim 1, characterized in that, The data channel occupancy model includes an input layer, an embedding layer, an encoding layer, a decoding layer, and an output layer, wherein: The embedding layer is used to perform vector mapping on the demand sequence, transforming the demand sequence into a spatial vector; The encoding layer is used to encode the spatial vector to obtain a context vector; The decoding layer is used to decode the context vector and perform attention aggregation processing to obtain the sorted sequence corresponding to each of the business applications.

5. The multi-application data channel occupancy method according to claim 1, characterized in that, The step of iteratively training a pre-defined basic channel occupancy model using the sample dataset to obtain a pre-trained data channel occupancy model includes: The sample dataset is input into a preset basic channel occupancy model, and vector mapping is performed on the sample dataset to transform the demand sequence in the sample dataset into a spatial vector; The spatial vector is encoded to obtain the historical context vector; The historical context vector is decoded and attention aggregation is performed to obtain the predicted order sequence; The loss value of the basic channel occupancy model is obtained based on the predicted sequence. If the loss value is greater than a preset threshold, the gradient descent optimization algorithm is used to return and execute the step of inputting the sample dataset into the preset basic channel occupancy model. The number of iterations is increased by one until the loss value reaches the minimum value, or when the number of iterations reaches the preset value, the pre-trained data channel occupancy model is obtained.

6. A multi-application data channel occupancy device, characterized in that, include: The data processing module is used to obtain multiple business applications' requests for data channel usage and extract the demand information from the usage requests of each business application. The requirement information includes the scenario type, importance level, priority, application content, required bandwidth, and duration of the business application. The sequence conversion module is used to convert the demand information into a sequence to obtain the demand sequence corresponding to the demand information; The sequence prediction module is used to input the demand sequence into a pre-trained data channel occupancy model to obtain the sorting sequence corresponding to each of the business applications; The channel occupancy module is used to respond to the usage requests of the business applications according to the sorting sequence; each of the business applications occupies the data channel according to the sorting sequence; The multi-application data channel occupancy device further includes a pre-training module, which is used for: Obtain historical usage requests of the data channel from multiple business applications; A sample dataset is constructed based on the aforementioned historical usage requests; The preset basic channel occupancy model is iteratively trained using the sample dataset to obtain a pre-trained data channel occupancy model. The pre-training module is also used for: Obtain the tag data of the historical usage requests; the tag data is used to characterize the order in which the business application occupies the data channel; The historical usage requests are cleaned and serialized to obtain the historical demand sequence corresponding to the historical usage requests. Construct an access sequence set corresponding to the historical usage requests based on the tag data; The first index length is determined based on the sequence length of the historical demand sequence, and the second index length is determined based on the longest sequence length of each access order sequence in the access order sequence set. Based on the first index length and the second index length, the historical demand sequence and the access order sequence set are indexed to obtain a sample dataset.

7. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the multi-application data channel occupancy method according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-application data channel occupancy method according to any one of claims 1 to 5.

Images