Network node traffic prediction method, apparatus, device, and storage medium

Abstract

The application discloses a network node traffic prediction method and device, equipment and a storage medium, and the method comprises the following steps: obtaining a preset node data set through a target multi-modal network, inputting the preset node data set into a target modal graph embedding model to obtain target network transmission information; inputting the target network transmission information into a preset feature extraction network to obtain a network transmission feature vector; obtaining a preset network information vector corresponding to the target multi-modal network, and splicing the network transmission feature vector and the preset network information vector to obtain a network combination vector; and based on the network combination vector, performing traffic prediction on the target multi-modal network through a preset deep learning model to obtain a node traffic prediction result. The network transmission feature vector and the preset network information vector are integrated into the network combination vector containing various key information, so that the preset deep learning model accurately predicts the node traffic of the target multi-modal network based on the network combination vector.

Description

Technical Field

[0001] This invention relates to the field of screen information adjustment technology, and in particular to a method, apparatus, device and storage medium for predicting network node traffic. Background Technology

[0002] Multimodal intelligent networks are a novel network architecture proposed to address challenges such as the scalability of traditional networks, reliance on single-IP networks, and the difficulty in suppressing unknown threats. Accurate prediction of network node traffic during operation significantly aids in node management and control within multimodal intelligent networks.

[0003] However, since multimodal intelligent network environments often carry and operate multiple networks of different modes, and different modes of networks have different characteristics, existing node prediction methods have problems such as insufficient analysis and low prediction accuracy when analyzing the network node traffic of multimodal intelligent networks.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a method, apparatus, device, and storage medium for predicting network node traffic, aiming to solve the technical problems of insufficient analysis and low prediction accuracy in existing network node prediction methods.

[0006] To achieve the above objectives, the present invention provides a method for predicting network node traffic, the method comprising:

[0007] A preset node dataset is obtained through a target multimodal network, and the preset node dataset is input into a target modality graph embedding model to obtain the target network transmission information;

[0008] The target network transmission information is input into a preset feature extraction network to obtain a network transmission feature vector;

[0009] Obtain the preset network information vector corresponding to the target multimodal network, and concatenate the network transmission feature vector and the preset network information vector to obtain the network combination vector;

[0010] Based on the network-linked vectors, a preset deep learning model is used to predict the traffic flow of the target multimodal network, thereby obtaining the node traffic prediction results.

[0011] Optionally, the target multimodal network includes a preset number of submodal networks; the step of obtaining a preset node dataset through the target multimodal network includes:

[0012] By randomly sampling the nodes of the target submodal network using a preset random walk algorithm, a first node dataset corresponding to the target submodal network is obtained, wherein the target submodal network is any submodal network in the target multimodal network;

[0013] Data transmission sampling is performed on the data stream of the target submodal network to obtain the second node dataset corresponding to the target submodal network;

[0014] Based on the first node dataset, the second node dataset, and the preset structure sampling ratio, a preset node dataset corresponding to the target submodal network is generated.

[0015] Optionally, the step of performing data transmission sampling on the data stream of the target submodal network to obtain the second node dataset corresponding to the target submodal network includes:

[0016] The data stream of the target submodal network is sampled according to the preset data sampling time and target routing information to obtain the transmission data sequence set corresponding to the target submodal network;

[0017] Based on the transmitted data sequence set, determine the target transmission nodes corresponding to the preset number of transmissions in the target submodal network;

[0018] The target transmission path for the preset number of transmissions is determined based on each node and the target transmission node corresponding to each node;

[0019] The second node dataset corresponding to the target submodal network is generated based on the target transmission path and the transmission data sequence set.

[0020] Optionally, the step of sampling the data stream of the target submodal network according to a preset data sampling time and target routing information to obtain the transmission data sequence set corresponding to the target submodal network further includes:

[0021] Obtain the difference in target bandwidth utilization of the target submodal network during the preset data sampling time;

[0022] When the difference in the target bandwidth utilization rate is detected to be greater than a preset threshold, the data stream of the target submodal network is sampled according to a preset time and a preset period sampling ratio to obtain the transmission data sequence set corresponding to the target submodal network.

[0023] Optionally, the step of inputting the preset node dataset into the target modal graph embedding model to obtain the target network transmission information includes:

[0024] The target modal graph embedding model is obtained from the preset graph embedding model library according to the network type of each submodal network in the target multimodal network;

[0025] The preset node dataset is input into the target modality graph embedding model to obtain the preset number of intermediate network transmission vectors;

[0026] The target network transmission information is obtained by combining the preset number of intermediate network transmission vectors.

[0027] Optionally, the step of combining the preset number of intermediate network transmission vectors to obtain the target network transmission information includes:

[0028] When it is detected that the vector dimensions of the preset number of intermediate network transmission vectors are inconsistent, the intermediate network transmission vector with the longest vector dimension is used as the filling reference, and the remaining intermediate network transmission vectors are filled in a preset manner to obtain the completed network vector.

[0029] The target network transmission information is obtained by concatenating the intermediate network transmission vector with the longest vector dimension and the supplementary network vector.

[0030] Optionally, before inputting the target network transmission information into a preset feature extraction network to obtain the network transmission feature vector, the method further includes:

[0031] The initial convolutional neural network is iteratively updated based on the transmission matrix of the training network to obtain the preset convolutional neural network.

[0032] A preset structure adjustment operation is performed on the preset convolutional neural network to obtain a preset feature extraction network.

[0033] Furthermore, to achieve the above objectives, the present invention also proposes a network node traffic prediction device, the network node traffic prediction device comprising:

[0034] The data acquisition module is used to acquire a preset node dataset through the target multimodal network and input the preset node dataset into the target modality graph embedding model to obtain the target network transmission information;

[0035] The feature extraction module is used to input the target network transmission information into a preset feature extraction network to obtain the network transmission feature vector;

[0036] The vector concatenation module is used to obtain the preset network information vector corresponding to the target multimodal network, and to concatenate the network transmission feature vector and the preset network information vector to obtain the network combination vector.

[0037] The traffic prediction module is used to predict the traffic of the target multimodal network based on the network combined vectors and through a preset deep learning model to obtain the node traffic prediction results.

[0038] Furthermore, to achieve the above objectives, the present invention also proposes a network node traffic prediction device, the device comprising: a memory, a processor, and a network node traffic prediction program stored in the memory and executable on the processor, the network node traffic prediction program being configured to implement the steps of the network node traffic prediction method as described above.

[0039] In addition, to achieve the above objectives, the present invention also proposes a storage medium storing a network node traffic prediction program, which, when executed by a processor, implements the steps of the network node traffic prediction method as described above.

[0040] This invention discloses a method for predicting network node traffic. The method includes: obtaining a preset node dataset through a target multimodal network; obtaining target network transmission information based on the preset node dataset and a target modal graph embedding model; inputting the target network transmission information into a preset feature extraction network to obtain a network transmission feature vector; obtaining a preset network information vector corresponding to the target multimodal network; concatenating the network transmission feature vector and the preset network information vector to obtain a network combination vector; and predicting the traffic of the target multimodal network based on the network combination vector using a preset deep learning model to obtain node traffic prediction results. This invention can obtain the Euclidean data representation of nodes on the corresponding modal network, i.e., the aforementioned target network transmission information, through the target modal graph embedding model, to express the similarity of each node in the target multimodal network structure and the importance of nodes to each other during data communication. Since the target network transmission information contains the network structure and routing features of the target multimodal network, the subsequent preset deep learning model can analyze and mine complex networks based on this target network transmission information. At the same time, this invention can integrate various key information in the multimodal network environment by using network transmission feature vectors containing network structure information and preset network information vectors containing time and node information to obtain network combination vectors. This allows the preset deep learning model to learn more features of the multimodal intelligent network environment based on the network combination vectors, thereby accurately predicting the node traffic of the target multimodal network. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the structure of the network node traffic prediction device in the hardware operating environment involved in the embodiments of the present invention;

[0042] Figure 2 This is a flowchart illustrating the first embodiment of the network node traffic prediction method of the present invention.

[0043] Figure 3 This is a schematic diagram of the first process of the second embodiment of the network node traffic prediction method of the present invention;

[0044] Figure 4 This is a schematic diagram of the second process of the second embodiment of the network node traffic prediction method of the present invention;

[0045] Figure 5 This is a flowchart illustrating the third embodiment of the network node traffic prediction method of the present invention;

[0046] Figure 6 This is a schematic diagram of the target multimodal network structure in the third embodiment of the network node traffic prediction method of the present invention;

[0047] Figure 7 This is a schematic diagram of the preset feature extraction network structure of the third embodiment of the network node traffic prediction method of the present invention;

[0048] Figure 8 This is a structural block diagram of the first embodiment of the network node traffic prediction device of the present invention.

[0049] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0051] Reference Figure 1 , Figure 1 This is a schematic diagram of the network node traffic prediction device structure in the hardware operating environment involved in the embodiments of the present invention.

[0052] like Figure 1As shown, the network node traffic prediction device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0053] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the network node traffic prediction device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0054] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and a network node traffic prediction program.

[0055] exist Figure 1 In the network node traffic prediction device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the network node traffic prediction device of the present invention can be set in the network node traffic prediction device, and the network node traffic prediction device calls the network node traffic prediction program stored in the memory 1005 through the processor 1001 and executes the network node traffic prediction method provided in the embodiment of the present invention.

[0056] This invention provides a method for predicting network node traffic, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the network node traffic prediction method of the present invention.

[0057] Understandably, different modes of networks in a multimodal intelligent network will have different characteristics, such as network topology, network protocols, routing algorithms, and traffic characteristics. However, current popular traffic prediction methods often only analyze the network characteristics of a specific network (e.g., an IP network). When multiple networks are operating in the network environment, these methods suffer from low versatility and incomplete analysis. Therefore, existing traffic prediction methods may exhibit low accuracy when predicting the traffic of network nodes in a multimodal intelligent network.

[0058] Therefore, integrating various key information within a multimodal network environment and acquiring more features of the multimodal intelligent network environment is crucial for more accurate traffic and rate prediction. To achieve this, this embodiment can combine deep learning algorithms for node traffic analysis. Specifically, this embodiment enables the deep learning model to learn multiple features of the multimodal intelligent network environment based on various key information, thereby generating a model more suitable for traffic prediction within the multimodal intelligent network environment. Therefore, in this embodiment, the network node traffic prediction method includes the following steps:

[0059] Step S10: Obtain a preset node dataset through the target multimodal network, and input the preset node dataset into the target modal graph embedding model to obtain the target network transmission information.

[0060] It should be noted that the executing entity of the method in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer or personal computer, or other electronic devices capable of performing the same or similar functions. Here, the network node traffic prediction method provided in this embodiment and the following embodiments will be specifically described using the aforementioned network node traffic prediction device (hereinafter referred to as the prediction device).

[0061] It should be understood that the aforementioned target multimodal network can be a multimodal intelligent network for traffic prediction. Various network technologies can be loaded and run on the target multimodal network in modal form. One type of network technology can correspond to one type of modal network. For example, IP network, NDN network, and geoidentification network can correspond to three types of modal networks in the target multimodal network. Therefore, the aforementioned target multimodal network can contain a preset number of submodal networks (such as the aforementioned IP network, NDN network, and geoidentification network). The preset number is the total number of submodal networks in the target multimodal network. The specific value is not limited in this embodiment.

[0062] It is understood that the aforementioned preset node dataset can be network transmission data obtained from network nodes corresponding to each sub-modal network in the target multimodal network. That is, this embodiment can obtain features reflecting the network transmission environment of the target multimodal network by analyzing the preset node dataset. The preset node dataset may contain a node set V composed of network nodes corresponding to each sub-modal network and an edge set E composed of connecting edges between each network node.

[0063] It is readily understood that in this embodiment, the aforementioned target graph embedding model corresponds to the network topology of the target multimodal network. After inputting the preset node dataset corresponding to each submodal network into the target graph embedding model, the model can train and process the preset node dataset to obtain the vector representation of each network node in each modal network. The obtained vectors can contain structural information and routing information used on each modal network. Therefore, based on this target graph embedding model, the prediction device can convert the non-Euclidean data (graph structure or network structure) contained in the preset node dataset into Euclidean data, thereby obtaining Euclidean data containing network structure information, i.e., the aforementioned target network transmission information. Traffic analysis and processing can be performed based on this target network transmission information. Since the target network transmission information contains the network structure features and routing features of the target multimodal network, more features of the target multimodal network can be mined and learned based on this information.

[0064] Step S20: Input the target network transmission information into a preset feature extraction network to obtain the network transmission feature vector.

[0065] Step S30: Obtain the preset network information vector corresponding to the target multimodal network, and concatenate the network transmission feature vector and the preset network information vector to obtain the network combination vector.

[0066] It is understood that the target network transmission information may contain different information from each mode in the target multimodal network. Therefore, before performing traffic prediction, this embodiment can integrate and extract features from the target network transmission information through a preset feature extraction network to obtain network transmission feature vectors. Specifically, in this embodiment, the preset feature extraction network can be a Convolutional Neural Network (CNN). It is easy to understand that this embodiment can also use other network structures for feature extraction based on the actual situation, such as a Recurrent Neural Network (RNN) or a Long Short-Term Memory (LSTM). The specific neural network used for feature extraction can be selected based on the actual situation, and this embodiment does not impose any restrictions on this.

[0067] It should be understood that, in order to further improve the accuracy of traffic prediction, this embodiment can integrate the network transmission feature vector extracted by the preset feature extraction network with other information of the target multimodal network, namely the preset network information vector, to obtain the final input vector, namely the network combination vector.

[0068] It should be noted that in this embodiment, the above-mentioned network transmission feature vector can be used... This indicates that the preset network information vector will be used... The network combination vector obtained by concatenating and splicing the two is represented as v. s Specifically, this embodiment can... spliced ​​in At the end, the above network combination vector is obtained. The preset network information vector can be a network information representation vector in the target multimodal network, with each network node s as the source node. The expression of the preset network information vector can be:

[0069]

[0070] In the formula, This represents the proportion of resources allocated by node s to each submodal network; T represents the start time of the preset network information vector collection, and the collection time can be within one hour starting from T. This represents the average temperature of the processor in node s during the hour preceding T. This represents the total number of data packets received by node s in the hour preceding T; This indicates the total number of data packets sent in the hour preceding the current time T.

[0071] It's important to understand that the above data collection start time can be represented as: T = (h, d, m, holiday), where h ∈ {0, 1, 2, ..., 23} represents the hour; d ∈ {1, 2, 3, ..., 6, 7} represents Monday to Sunday; m ∈ {1, 2, 3, ..., 11, 12} represents the month; and holiday ∈ {0, 1} indicates whether the current day is a special holiday such as Spring Festival, National Day, or Mid-Autumn Festival, with 0 representing no and 1 representing yes. For example, if we assume that the preset network information vector currently being collected has T = (15, 5, 3, 0), it means that the preset network information vector was collected starting at 3 PM on a non-holiday Friday in March. The time information T is added because network traffic often exhibits a tidal effect. Adding time information allows deep learning models to better explore the correlation between time information and traffic volume, thereby obtaining the temporal characteristics of traffic.

[0072] It should be noted that the resource allocation ratio of node s to each submodal network can be expressed as:

[0073]

[0074] In the formula, M represents the total number of modal networks in the target multimodal network, and r i s This represents the proportion of resources that node s allocates to the i-th modal network.

[0075] Specifically, if we assume that the target multimodal network has a total of 6 modes in operation, and if we assume that the source node s = 6 (which represents the 6th node in the entire multimodal network environment) and that this source node s = 6 carries three modes of the network, then This can be interpreted as source node s=6 carrying the second, third, and fifth modal networks, and allocating 50%, 20%, and 30% of the resources to the second, third, and fifth modal networks, respectively.

[0076] It's easy to understand that if we assume the processor temperature of node 6 is 36℃ at the current moment, and it sent a total of 15MB of data and received a total of 25MB of data in the hour before T = (15,5,3,0), then the preset network information vector at this time... It can be represented as:

[0077] It should be noted that the above-mentioned target network transmission information Often associated with multiple pre-defined network information vectors Corresponding. Because It contains network structure information, but the network topology of the target multimodal network does not necessarily change every moment; however It contains time information and changes over time.

[0078] Step S40: Based on the network combination vector, perform traffic prediction on the target multimodal network using a preset deep learning model to obtain node traffic prediction results.

[0079] It should be noted that, in order to learn more features of the target multimodal network based on various information about the integrated network environment, this embodiment can use a deep learning network, namely the aforementioned preset deep learning model, for mining and analysis. The aforementioned preset deep learning model can be a deep learning model built on the Transformer architecture, thereby leveraging the powerful analysis and processing capabilities of the Transformer architecture to predict the data size to be output by each node in the next time period. Specifically, this embodiment can input the vectors integrated in the previous stage into the encoder structure in the Transformer, that is, the aforementioned network transmission feature vectors. With preset network information vector The network combination vector v obtained after concatenation s The input is fed into a pre-trained deep learning model to predict node traffic in a multimodal network, and the pre-trained deep learning model can output the corresponding data. in, This can represent the predicted size of the data packets that node s will output within one hour starting from T = (h, d, m, holiday); This can represent the predicted peak processing rate of node s within one hour starting from T = (h, d, m, holiday). Therefore, this embodiment can predict the traffic volume and peak processing rate of node s within one hour starting from T, thereby obtaining the node traffic prediction result. Based on the traffic prediction result, it can be determined whether a certain sub-modal network in the target multimodal network needs to add nodes, or whether the nodes in the target multimodal network are abnormal, etc.

[0080] It is important to understand that the loss function of the aforementioned preset deep learning model can be set as follows:

[0081]

[0082] Where V represents the set of nodes in the target multimodal network, T represents time, s represents a network node, and β represents the weight.

[0083] It is easy to understand that β ≥ 1 can be set in this embodiment because this embodiment aims to focus more on the peak rate of the node in the prediction. If the predicted peak rate of the node at the next moment is an outlier, it is possible to promptly check whether the node has experienced or is about to experience an anomaly, so that the node can be dealt with or prevented in a timely manner. Therefore, this embodiment can set a larger weight for the loss function related to the peak rate prediction in order to focus the prediction on the peak rate of the node.

[0084] This embodiment discloses a method for predicting network node traffic. The method includes: obtaining a preset node dataset through a target multimodal network; obtaining target network transmission information based on the preset node dataset and a target modal graph embedding model; inputting the target network transmission information into a preset feature extraction network to obtain a network transmission feature vector; obtaining a preset network information vector corresponding to the target multimodal network; concatenating the network transmission feature vector and the preset network information vector to obtain a network combination vector; and predicting the traffic of the target multimodal network using a preset deep learning model based on the network combination vector to obtain node traffic prediction results. This embodiment can obtain the Euclidean data representation of nodes on the corresponding modal network, i.e., the aforementioned target network transmission information, through the target modal graph embedding model, to express the similarity of each node in the target multimodal network structure and the importance of nodes to each other during data communication. Since the target network transmission information contains the network structure and routing features of the target multimodal network, the subsequent preset deep learning model can analyze and mine complex networks based on this target network transmission information. At the same time, this embodiment can integrate various key information in the multimodal network environment by using network transmission feature vectors containing network structure information and preset network information vectors containing time and node information to obtain network combination vectors. This allows the preset deep learning model to learn more features of the multimodal intelligent network environment based on the network combination vectors, thereby accurately predicting the node traffic of the target multimodal network.

[0085] Reference Figure 3 , Figure 3 This is a schematic diagram of the first process of the second embodiment of the network node traffic prediction method of the present invention, based on the above. Figure 2 The illustrated embodiment presents a second embodiment of the network node traffic prediction method of the present invention.

[0086] In this embodiment, the target multimodal network includes a preset number of submodal networks, and step S10 includes:

[0087] Step S11: Randomly sample the nodes of the target submodal network using a preset random walk algorithm to obtain the first node dataset corresponding to the target submodal network. The target submodal network is any submodal network in the target multimodal network.

[0088] It is understandable that the aforementioned target modal graph embedding model can be run based on the Deepwalk algorithm, and the dataset required for training the traditional Deepwalk algorithm is often generated through random walks. In this embodiment, the target multimodal network may include a preset number of submodal networks. The aforementioned target submodal network is any submodal network in the target multimodal network. Correspondingly, this embodiment can also use a preset random walk algorithm to randomly sample the nodes of any submodal network to obtain the first node dataset corresponding to any submodal network. For example, [1, 3, 2] can represent the path data from node 1 to node 3 and then to node 2. The Deepwalk algorithm learns based on these node sequences from the random walk, and the resulting vector can reflect the characteristics of the network topology of each submodal network. Therefore, the target modal graph embedding model is trained and analyzed based on the first node dataset, which can reflect the network structure information of each submodal network in the target multimodal network. For example, since node 2 is adjacent to node 1, many sequences will contain both node 1 and node 2. Therefore, the vector of node 1 obtained in the end will reflect the information that "node 2 is very important to node 1".

[0089] Step S12: Perform data transmission sampling on the data stream of the target submodal network to obtain the second node dataset corresponding to the target submodal network;

[0090] It is important to understand that in a target multimodal network, due to the existence of routing protocol algorithms, some non-adjacent nodes may have a greater influence than their adjacent nodes. For example, when data transmission is performed according to the routing protocol, data packets sent by node 2 often travel through path [3, 5, 7] to node 7, and rarely transmit data based on node 2's adjacent node 1. Therefore, this embodiment aims to ensure that the target network transmission information output by the target modal graph embedding model not only expresses the network structure information of each submodal network, but also includes the routing information of each submodal network.

[0091] It is understood that this embodiment can influence the training results of the target modal graph embedding model by improving the dataset used for training, thereby improving the vector output by the target modal graph embedding model so that it also contains routing information. When sampling the data flow of each submodal network in the target multimodal network, the obtained data is path data obtained according to the routing protocol adopted by each submodal network in the target multimodal network (such as the node sequence [3, 5, 7] mentioned above). The routing algorithm corresponding to each submodal network in the target multimodal network will indicate the node that the data packet should go to next during the data transmission process.

[0092] It should be noted that, in this embodiment, the aforementioned preset node dataset can be represented as: The first node dataset, consisting of the node sequences generated by the random walk algorithm, can be represented as: The second node dataset, composed of node sequences collected according to the routing protocols of each sub-modal network, can be represented as: Where p∈{1,2,3,...,n}, represents the p-th modal network, and n is the preset number mentioned above, i.e., the total number of submodal networks. Specifically, the second node dataset... It can be constructed based on the path formed by the source and destination nodes of each data packet on the target submodal network. For example, when the target submodal network is modal network A, if there is a data packet sent from node 2 to node 6, and according to the routing algorithm used by submodal network A, the specific path of the data packet is [2, 5, 7, 6], then the transmission sequence [2, 5, 7, 6] constitutes a data in the second node dataset corresponding to submodal network A.

[0093] Furthermore, as one possible implementation method, such as Figure 4 As shown, Figure 4 This is a second flowchart illustrating a second embodiment of the network node traffic prediction method of the present invention. In this embodiment, step S12 may include:

[0094] Step S121: Sample the data stream of the target submodal network according to the preset data sampling time and target routing information to obtain the transmission data sequence set corresponding to the target submodal network;

[0095] It is understandable that the aforementioned preset data sampling time can be the total sampling duration for data stream sampling, and the aforementioned target routing information can be the routing protocol or routing algorithm corresponding to the target submodal network. It is also easy to understand that data traffic on some modal networks often exhibits characteristics of varying traffic volume over time; for example, traffic is lower in the early morning and higher during evening leisure / after-school hours; traffic is higher on specific holidays, such as New Year's Eve, than on regular weekdays. Therefore, this embodiment can sample data transmission from the target submodal network based on specific times and dates.

[0096] It is important to understand that, in order to determine whether the network traffic of the target submodal network is correlated with time changes, this embodiment can obtain the target bandwidth utilization difference of the target submodal network within a preset data sampling time before sampling the data stream of the target submodal network. That is, it monitors the bandwidth utilization fluctuation of the target submodal network within a preset data sampling time (e.g., three days, one week, or one month, depending on the specific situation; this embodiment does not impose any restrictions on this) and records the lowest bandwidth utilization B of the target submodal network during this time period. min With the highest bandwidth utilization B max When the detected difference in target bandwidth utilization is less than the preset threshold δB total(where B) total This represents the total bandwidth (where δ < 1), i.e., B. max -B min ≤δB total In this embodiment, it can be determined that the target submodal network is a network whose flow does not change with time, and the data stream of the target submodal network can be sampled using a preset uniform data transmission method; when B max -B min >δB total In this embodiment, the target submodal network can be determined to be a network whose traffic fluctuates over time. The sampling method based on a specific time and date is adopted. That is, this embodiment can sample the data stream of the target submodal network according to a preset time and a preset period sampling ratio to obtain a transmission data sequence set.

[0097] In a practical implementation, the process of pre-setting uniform data transmission sampling for the data stream of the target submodal network can be as follows: randomly select several days and perform uniform sampling within 24 hours of each day (the number of samples in each hour is consistent, and every hour...) Perform once every few minutes for a duration of The total amount of data obtained from the sampling is G, which is the data sequence set of the above transmission.

[0098] In a specific implementation, the process of sampling the data stream of the target submodal network according to a preset time and a preset period sampling ratio can be as follows: First, assuming that the total number of samples collected from the same day is G d In this embodiment, a 24-hour day can be divided into three intervals: 0:00-7:59, 8:00-15:59, and 16:00-23:59; then, the data stream is uniformly sampled through these three time periods (i.e., in each time period, every hour...). A duration is performed every minute. The sampling was performed, and the total number of samples taken in each hour within the same time period was consistent. γ1G was collected separately. d γ2G d γ3G d Data, of which And γ3 ≥ γ2 > γ1. It's easy to understand that users tend to use the network more during work / school hours or after get off work / school hours than in the early morning. Therefore, this embodiment considers the time periods of 16:00-23:59 and 8:00-15:59 to be more representative, and thus sets a larger proportion of data accordingly. Based on the above data sampling method, the total number of samples that the prediction device can collect within the preset data sampling time can be expressed as G (G > G). dAt this point, the aforementioned transmission data sequence set is not equal to the total number of samples G. Instead, the transmission data sequence set needs to be determined by sampling the preset total number of samples G according to the preset time and preset period sampling ratio. Specifically, in some modal networks, there is a significant difference in data traffic between weekdays and weekends. Therefore, this embodiment can sample λ1G and λ2G data streams from weekdays and weekends respectively as representative of the regular period, where the preset sampling ratio can be set to λ1:λ2 = 5:2. Traffic data during the Spring Festival, National Day, and May Day holidays can better represent the network traffic characteristics during holidays. Therefore, this embodiment can sample λ3G data streams from the Spring Festival, National Day, or May Day holidays as representative of holidays. Besides long holidays like the Spring Festival and National Day, short holidays such as New Year's Day, Mid-Autumn Festival, and Dragon Boat Festival, as well as holidays like Christmas and Valentine's Day, which often have special activities but are not public holidays, are also representative. This embodiment can sample λ4G data streams from these holidays as representative of these holidays, where λ3 > λ4. Furthermore, due to the large number of students, teachers, and school staff, the summer vacation from July to August is also representative. This embodiment can also sample λ5G data streams from the summer vacation as representative of the summer vacation. The above sampling ratios need to meet the following constraints. The specific values ​​of the sampling ratios λ1, λ2, λ3, λ4 and λ5 for each preset period can be set according to the actual situation, and this embodiment does not impose any restrictions on them.

[0099] It is easy to understand that if a sample of data belongs to both holidays / summer vacation data and weekends or weekdays, then in this embodiment, the sample of data can be regarded as holidays / summer vacation data to ensure that the sampled data covers as many special time nodes as possible, so that the data of the second node dataset has good representativeness.

[0100] Step S122: Determine the target transmission nodes corresponding to the preset number of transmissions for each node in the target submodal network based on the transmission data sequence set;

[0101] Step S123: Determine the target transmission path for the preset number of transmissions based on each node and the target transmission node corresponding to each node;

[0102] Step S124: Generate the second node dataset corresponding to the target submodal network based on the target transmission path and the transmission data sequence set.

[0103] It can be understood that after obtaining the transmission data sequence set, this embodiment can also filter and organize the sequence data in the transmission data sequence set through a preset script code. Specifically, this embodiment can call a specific instruction (such as Tracert) through a preset script to view the actual path of the data packet reaching the destination host, so as to obtain the source address and destination address of the data stream corresponding to the captured (or sampled) data packet, as well as the data transmission volume between the source address and the destination address, thereby counting L (i.e., the above-mentioned preset transmission quantity) common paths of each network node as the source node on the target sub-modal network, that is, the above-mentioned target transmission paths, and then obtaining the second node dataset based on these target transmission path data

[0104] It should be understood that the method for determining the above-mentioned target transmission path can be: counting the different destination nodes with the top L data transmission volumes in the transmission data sequence set for each node, that is, the above-mentioned target transmission nodes, and then determining L target transmission paths based on each node and the L target transmission nodes corresponding to each node. Among them, if there is more than one path from a certain node to the same destination node, the path with the largest data transmission volume can be preferentially selected as the target transmission path corresponding to the destination node. And if the number of target transmission data paths determined for different destination nodes of a certain node in the transmission data sequence set is L less , L less < L, then this node can only include L less target transmission data paths. At the same time, this embodiment can also perform statistics on information such as addresses and data volumes at each destination node to ensure that the data information included in does not include the data information lost by the network

[0105] Step S13: Generate a preset node dataset corresponding to the target sub-modal network based on the first node dataset, the second node dataset, and a preset structure sampling ratio

[0106] It should be noted that this embodiment can add training data collected according to routing to the training dataset of the above-mentioned target modal graph embedding model by the method of collecting the second node dataset described above, so as to increase the influence proportion of the nodes on the path on the source node during the graph embedding process. Based on the first node dataset and the second node dataset The synthesized preset node dataset can include both the routing characteristics of each modal network and the randomness and network structure characteristics brought by random walks, taking into account "exploitation and exploration" In this configuration, the proportion of the first node dataset generated by random walks can be set to α, and the proportion of the second node dataset collected by routing algorithms used in each sub-modal network can be set to 1-α. That is, regarding the preset number of node datasets... This refers to the aforementioned preset structure sampling ratio. The value of α can affect whether the entire graph embedding algorithm reflects network structure or routing characteristics more. Its specific value can be set according to specific circumstances, and this embodiment does not impose any restrictions on it.

[0107] It is important to understand that, based on the characteristics of the multimodal intelligent network environment, this embodiment proposes a method to integrate the information of each modal network in the multimodal intelligent network environment, including the structural information, network routing information, traffic temporal characteristics and node processing rate of each modal network, into Euclidean data (vectors). These vectors are then used as inputs, allowing the deep learning model to learn based on this data and generate a deep learning model suitable for the multimodal intelligent network environment, thereby further improving the accuracy of node traffic prediction for the target multimodal network.

[0108] In this embodiment, the target multimodal network includes a preset number of submodal networks. A preset random walk algorithm is used to randomly sample the nodes of the target submodal networks to obtain a first node dataset corresponding to the target submodal network. The target submodal network is any submodal network within the target multimodal network. The data flow of the target submodal network is sampled according to a preset data sampling time and target routing information to obtain a transmission data sequence set corresponding to the target submodal network. A preset number of target transmission nodes are determined based on the transmission data sequence set. A preset number of target transmission paths are determined based on each node and its corresponding target transmission nodes. A second node dataset corresponding to the target submodal network is generated based on the target transmission paths and the transmission data sequence set. Based on the first node dataset, the second node dataset, and a preset structure sampling ratio, a preset node dataset corresponding to the target submodal network is generated. In this embodiment, the first node dataset can reflect the network structure information of the target modal network, and the second node dataset can reflect the routing information of the target modal network. Therefore, this embodiment can improve the training dataset of the graph embedding model of each submodal network in the target multimodal network, so that it contains routing information and analyzes and processes it, thereby further improving the accuracy of node traffic prediction of the target multimodal network.

[0109] Reference Figure 5 , Figure 5 This is a flowchart illustrating the third embodiment of the network node traffic prediction method of the present invention, based on the above. Figure 2 Alternatively, as shown in embodiment 3, a third embodiment of the network node traffic prediction method of the present invention is proposed. Figure 5 Based on Figure 1 The embodiments shown are examples of the proposed embodiments.

[0110] It's easy to understand that a target multimodal network can support multiple modes of networks operating on it. However, network nodes in the target multimodal network may support some of these modes, or they may not support any. Therefore, the network topology corresponding to each mode in the target multimodal network is usually different. For ease of understanding, we will use... Figure 6 Let's take an example to illustrate this. Figure 6 This is a schematic diagram of the target multimodal network structure in the third embodiment of the network node traffic prediction method of the present invention, as shown below. Figure 6 As shown, assuming there are ten nodes in the target multimodal network, nodes 1, 2, 4, and 6 carry the geographic identification network and transmit data using the geographic identification network protocol; nodes 1 and 3 carry the identity identification network and transmit data using the identity identification network protocol; nodes 1, 3, and 6 carry the IPv6 network and transmit data using the IPv6 protocol; while node 5 only has a physical connection with the other nodes and does not start or carry any network. Therefore, although the three modal networks all operate in a multimodal intelligent network environment, their topologies are different. Therefore, to improve the accuracy of subsequent traffic prediction, this embodiment can configure a pre-trained graph embedding model for each modal network. Furthermore, in this embodiment, step S10 also includes:

[0111] Step S14: Obtain the target modal graph embedding model from the preset graph embedding model library according to the network type of the target multimodal network. The target modal graph embedding model includes a preset number of graph embedding sub-models.

[0112] It is understood that the aforementioned preset graph embedding model library can store graph embedding sub-models pre-configured according to each modal network, so that the target network transmission information output by the target modal graph embedding model can also include the topology information and routing information of the target multimodal network. Specifically, this embodiment can obtain the graph embedding sub-models corresponding to each modal network in the target multimodal network from the preset graph embedding model library based on the network type of the target multimodal network, that is, the type of modal network specifically included in the target multimodal network. The aforementioned preset number is the number of types of each modal network in the target multimodal network, and the aforementioned target modal graph embedding model can be composed of the preset number of graph embedding sub-models.

[0113] Step S15: Input the preset node dataset into the target modal graph embedding model to obtain the preset number of intermediate network transmission vectors;

[0114] Step S16: Combine the preset number of intermediate network transmission vectors to obtain the target network transmission information.

[0115] It is important to understand that, based on the preset node dataset When inputting the target modal graph embedding model to obtain intermediate network transmission vectors, network nodes can be used as input units. Specifically, assuming that the target multimodal network in this embodiment has a total of 6 modal networks, the target modal graph embedding model can contain 6 graph embedding sub-models. After inputting the preset node dataset of each submodal network into the corresponding graph embedding sub-model, the vector representation of each node s in the target multimodal network can be obtained. Where p∈{1,2,3,4,5,6}, it represents different modal networks. It can include the structural information and routing information of each submodal network. That is, each network node s can obtain six intermediate network transmission vectors in six modes through the target modal graph embedding model, which can be represented as follows:

[0116] It is understandable that, for ease of data management and analysis, this embodiment can combine intermediate network transmission vectors belonging to the same node to obtain unified target network transmission information, which can be represented as a matrix. However, as the above analysis shows, in a multimodal intelligent network environment, not all nodes carry all network modes, and the network modes carried by each node are different. Therefore, in this embodiment, the topology and number of nodes of different modes of networks in the target multimodal network will also be different, resulting in the possibility that the dimension of the intermediate transmission vectors output by the graph embedding sub-model corresponding to each mode of network may be different. Therefore, before concatenating these intermediate network transmission vectors into the target network transmission information in matrix form, it is necessary to first check the dimension of the intermediate network transmission vectors. Therefore, further, as an implementation method, in this embodiment, step S16 may include:

[0117] Step S161: When it is detected that the vector dimensions of the preset number of intermediate network transmission vectors are inconsistent, the intermediate network transmission vector with the longest vector dimension is used as the filling reference, and the remaining intermediate network transmission vectors are filled in a preset manner to obtain the completed network vector.

[0118] Step S162: Concatenate the intermediate network transmission vector with the longest vector dimension and the padded network vector to obtain the target network transmission information.

[0119] It should be noted that in this embodiment, when the vector dimensions of the intermediate network transmission vectors are found to be inconsistent, the embodiment can use the intermediate network transmission vector with the longest vector dimension as the padding reference, and pad the remaining smaller vectors to the same dimension as the largest vector. The aforementioned pre-padding method can be to zero-padded the remaining smaller intermediate network transmission vectors at the ends. After ensuring that the graph embedding results of each modal network, i.e., the dimensions of each intermediate network transmission vector, are consistent through pre-padding, this embodiment can concatenate them into a matrix form of target network transmission information, i.e., the target network transmission matrix. For some nodes, if the script determines that they do not carry any modal networks, these nodes will be directly represented by a 0 vector under the corresponding modal network. For example, Figure 6 Node 5 in the network does not support modal network 3, so Specifically, this embodiment can transmit the intermediate network vector of node s in six modes. spliced ​​into the target network transmission matrix M s The target network transmission matrix M s It can be represented as:

[0120]

[0121] Furthermore, as one possible implementation, in this embodiment, before step S20, the following steps are also included:

[0122] Step S17: Iteratively update the initial convolutional neural network based on the transmission matrix of the training network to obtain the preset convolutional neural network;

[0123] Step S18: Perform a preset structure adjustment operation on the preset convolutional neural network to obtain a preset feature extraction network.

[0124] It is understood that the aforementioned preset feature extraction network can be a pre-trained and iterated convolutional neural network. In practical applications, this embodiment can directly extract network transmission feature vectors based on this preset feature extraction network. During the pre-training process, the training network transmission matrix and the target network transmission matrix are obtained in the same way. Since the training network transmission matrix M corresponds to different topologies... S The difference is that if the amount of data (i.e., the training network transfer matrix and its corresponding labels) of the preset convolutional neural network is small during the training process, more training data can be obtained by changing the network topology, thereby improving the generalization ability of the preset feature extraction network.

[0125] It should be noted that the above-mentioned preset structure adjustment operation can remove the fully connected layers of a pre-trained preset convolutional neural network model, retaining only the feature extraction part, thereby obtaining a preset feature extraction network. This also reduces the model complexity of the preset feature extraction network and improves the computation speed. For ease of understanding, let's take... Figure 7 Let's take an example to illustrate this. Figure 7 This is a schematic diagram of the preset feature extraction network structure of the third embodiment of the network node traffic prediction method of the present invention. In this diagram, CONV1-3 represent convolutional layers, all of which can use 3x3 convolutional kernels and have a stride of 1. Among them, only CONV1 uses zero-padding. Pooling represents pooling layers, which can use average pooling. Flatten is an unfolding layer, which can transform the results obtained from the convolutional layers and pooling layers into one-dimensional data, namely the network transmission feature vector mentioned above.

[0126] This embodiment obtains a target modal graph embedding model from a preset graph embedding model library based on the network type of the target multimodal network. The target modal graph embedding model includes a preset number of graph embedding sub-models. A preset node dataset is input into the target modal graph embedding model to obtain a preset number of intermediate network transmission vectors. When the vector dimensions of the preset number of intermediate network transmission vectors are found to be inconsistent, the intermediate network transmission vector with the longest vector dimension is used as the padding reference, and the remaining intermediate network transmission vectors are padded in a preset manner to obtain a padded network vector. The intermediate network transmission vector with the longest vector dimension and the padded network transmission vector are concatenated to obtain the target network transmission information. The initial convolutional neural network is iteratively updated based on the training network transmission matrix to obtain a preset convolutional neural network. A preset structure adjustment operation is performed on the preset convolutional neural network to obtain a preset feature extraction network. This embodiment can configure graph embedding sub-models for each modality of the target multimodal network to determine the target modal graph embedding model, so that the target network transmission information output by the target modal graph embedding model can also include the topology and routing information of the target multimodal network, thereby further improving the accuracy of subsequent node traffic prediction.

[0127] Furthermore, this embodiment of the invention also proposes a storage medium storing a network node traffic prediction program, which, when executed by a processor, implements the steps of the network node traffic prediction method described above.

[0128] refer to Figure 8 , Figure 8 This is a structural block diagram of the first embodiment of the network node traffic prediction device of the present invention.

[0129] like Figure 8 As shown, the network node traffic prediction device proposed in this embodiment of the invention includes:

[0130] Data acquisition module 801 is used to acquire a preset node dataset through the target multimodal network and input the preset node dataset into the target modality graph embedding model to obtain the target network transmission information;

[0131] Feature extraction module 802 is used to input the target network transmission information into a preset feature extraction network to obtain network transmission feature vector;

[0132] The vector concatenation module 803 is used to obtain the preset network information vector corresponding to the target multimodal network, and to concatenate the network transmission feature vector and the preset network information vector to obtain the network combination vector.

[0133] The traffic prediction module 804 is used to predict the traffic of the target multimodal network based on the network combination vector and through a preset deep learning model to obtain the node traffic prediction result.

[0134] This embodiment obtains a preset node dataset through a target multimodal network, and obtains target network transmission information based on the preset node dataset and the target modality graph embedding model. The target network transmission information is input into a preset feature extraction network to obtain network transmission feature vectors. A preset network information vector corresponding to the target multimodal network is obtained, and the network transmission feature vector and the preset network information vector are concatenated to obtain a network combination vector. Based on the network combination vector, a preset deep learning model is used to predict the traffic of the target multimodal network, obtaining node traffic prediction results. This embodiment can obtain the Euclidean data representation of nodes on the corresponding modality network, i.e., the aforementioned target network transmission information, through the target modality graph embedding model, to express the similarity of each node in the target modality network structure and the importance of nodes to each other during data communication. Since the target network transmission information contains the network structure and routing features of the target multimodal network, the subsequent preset deep learning model can analyze and mine complex networks based on this target network transmission information. At the same time, this embodiment can integrate various key information in the multimodal network environment by using network transmission feature vectors containing network structure information and preset network information vectors containing time and node information to obtain network combination vectors. This allows the preset deep learning model to learn more features of the multimodal intelligent network environment based on the network combination vectors, thereby accurately predicting the node traffic of the target multimodal network.

[0135] Based on the first embodiment of the network node traffic prediction device of the present invention, a second embodiment of the network node traffic prediction device of the present invention is proposed.

[0136] In this embodiment, the data acquisition module 801 is further configured to perform random walk sampling on the nodes of the target submodal network using a preset random walk algorithm to obtain a first node dataset corresponding to the target submodal network, wherein the target submodal network is any submodal network in the target multimodal network;

[0137] The data acquisition module 801 is also used to perform data transmission sampling on the data stream of the target submodal network to obtain the second node dataset corresponding to the target submodal network;

[0138] The data acquisition module 801 is also used to generate a preset node dataset corresponding to the target submodal network based on the first node dataset, the second node dataset, and the preset structure sampling ratio.

[0139] Furthermore, as one implementation, the data acquisition module 801 is also used to sample the data stream of the target submodal network according to the preset data sampling time and target routing information to obtain the transmission data sequence set corresponding to the target submodal network;

[0140] The data acquisition module 801 is also used to determine the target transmission nodes corresponding to the preset transmission quantity of each node in the target submodal network based on the transmission data sequence set.

[0141] The data acquisition module 801 is further configured to determine the target transmission path for the preset transmission quantity based on each node and the target transmission node corresponding to each node;

[0142] The data acquisition module 801 is further configured to generate a second node dataset corresponding to the target submodal network based on the target transmission path and the transmission data sequence set.

[0143] Furthermore, as one implementation, the data acquisition module 801 is also used to obtain the target modal graph embedding model from the preset graph embedding model library according to the network type of each submodal network in the target multimodal network;

[0144] The data acquisition module 801 is also used to input the preset node dataset into the target modal graph embedding model to obtain the preset number of intermediate network transmission vectors;

[0145] The data acquisition module 801 is also used to perform vector combination on the preset number of intermediate network transmission vectors to obtain target network transmission information.

[0146] Furthermore, as one possible implementation, the data acquisition module 801 is also used to, when it is detected that the vector dimensions of the preset number of intermediate network transmission vectors are inconsistent, take the intermediate network transmission vector with the longest vector dimension as the filling reference, and perform preset filling on the remaining intermediate network transmission vectors to obtain the completed network vector.

[0147] The data acquisition module 801 is also used to concatenate the intermediate network transmission vector with the longest vector dimension and the supplementary network vector to obtain the target network transmission information.

[0148] Furthermore, as one possible implementation, the data acquisition module 801 is also used to iteratively update the initial convolutional neural network based on the training network transfer matrix to obtain a preset convolutional neural network.

[0149] The data acquisition module 801 is also used to perform a preset structure adjustment operation on the preset convolutional neural network to obtain a preset feature extraction network.

[0150] This embodiment uses a preset random walk algorithm to randomly sample nodes of the target submodal network to obtain a first node dataset corresponding to the target submodal network, where the target submodal network is any submodal network in the target multimodal network. The data stream of the target submodal network is sampled according to a preset data sampling time and target routing information to obtain a transmission data sequence set corresponding to the target submodal network. Based on the transmission data sequence set, a preset number of target transmission nodes are determined for each node in the target submodal network. A preset number of target transmission paths are determined based on each node and its corresponding target transmission nodes. A second node dataset corresponding to the target submodal network is generated based on the target transmission paths and the transmission data sequence set. Finally, a preset node dataset corresponding to the target submodal network is generated based on the first node dataset, the second node dataset, and a preset structure sampling ratio. In this embodiment, the first node dataset reflects the network structure information of the target modal network, and the second node dataset reflects the routing information of the target modal network. Therefore, this embodiment can improve the training set of the graph embedding model of each submodal network in the target multimodal network to include routing information and analyze and process it, thereby further improving the accuracy of node traffic prediction of the target multimodal network. In addition, this embodiment obtains the target modal graph embedding model from the preset graph embedding model library according to the network type of the target multimodal network. The target modal graph embedding model includes a preset number of graph embedding sub-models. The preset node dataset is input into the target modal graph embedding model to obtain a preset number of intermediate network transmission vectors. When the vector dimensions of the preset number of intermediate network transmission vectors are found to be inconsistent, the intermediate network transmission vector with the longest vector dimension is used as the padding reference, and the remaining intermediate network transmission vectors are pre-padded to obtain the padded network vector. The intermediate network transmission vector with the longest vector dimension and the padded network transmission vector are concatenated to obtain the target network transmission information. The initial convolutional neural network is iteratively updated based on the training network transfer matrix to obtain a preset convolutional neural network; a preset structure adjustment operation is then performed on the preset convolutional neural network to obtain a preset feature extraction network. This embodiment can configure a graph embedding sub-model for each modality of the target multimodal network to determine the target modal graph embedding model. This allows the target network transmission information output by the target modal graph embedding model to also include the topology and routing information of the target multimodal network, thereby further improving the accuracy of subsequent node traffic prediction.

[0151] Other embodiments or specific implementations of the network node traffic prediction device of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0152] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0153] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0154] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0155] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for predicting network node traffic, characterized in that, The network node traffic prediction method includes: A preset node dataset is obtained through a target multimodal network, and the preset node dataset is input into a target modal graph embedding model to obtain target network transmission information. The target multimodal network includes a preset number of submodal networks. The target network transmission information is input into a preset feature extraction network to obtain a network transmission feature vector; Obtain the preset network information vector corresponding to the target multimodal network, and concatenate the network transmission feature vector and the preset network information vector to obtain the network combination vector; Based on the network-linked vectors, a preset deep learning model is used to predict the traffic of the target multimodal network and obtain the node traffic prediction results. The step of obtaining the preset node dataset through the target multimodal network includes: By randomly sampling the nodes of the target submodal network using a preset random walk algorithm, a first node dataset corresponding to the target submodal network is obtained, wherein the target submodal network is any submodal network in the target multimodal network; Data transmission sampling is performed on the data stream of the target submodal network to obtain the second node dataset corresponding to the target submodal network; Based on the first node dataset, the second node dataset, and the preset structure sampling ratio, a preset node dataset corresponding to the target submodal network is generated. The step of performing data transmission sampling on the data stream of the target submodal network to obtain the second node dataset corresponding to the target submodal network includes: The data stream of the target submodal network is sampled according to the preset data sampling time and target routing information to obtain the transmission data sequence set corresponding to the target submodal network; Based on the transmitted data sequence set, determine the target transmission nodes corresponding to the preset number of transmissions in the target submodal network; The target transmission path for the preset number of transmissions is determined based on each node and the target transmission node corresponding to each node; Generate the second node dataset corresponding to the target submodal network based on the target transmission path and the transmission data sequence set; The step of inputting the preset node dataset into the target modal graph embedding model to obtain the target network transmission information includes: The target modal graph embedding model is obtained from the preset graph embedding model library according to the network type of each submodal network in the target multimodal network; The preset node dataset is input into the target modality graph embedding model to obtain the preset number of intermediate network transmission vectors; The target network transmission information is obtained by combining the preset number of intermediate network transmission vectors. The step of combining the preset number of intermediate network transmission vectors to obtain the target network transmission information includes: When the vector dimensions of the preset number of intermediate network transmission vectors are found to be inconsistent, the intermediate network transmission vector with the longest vector dimension is used as the filling reference, and the remaining intermediate network transmission vectors are filled in a preset manner to obtain the completed network vector. The target network transmission information is obtained by concatenating the intermediate network transmission vector with the longest vector dimension and the supplementary network vector.

2. The network node traffic prediction method as described in claim 1, characterized in that, The step of sampling the data stream of the target submodal network according to a preset data sampling time and target routing information to obtain the transmission data sequence set corresponding to the target submodal network further includes: Obtain the difference in target bandwidth utilization of the target submodal network during the preset data sampling time; When the difference in the target bandwidth utilization rate is detected to be greater than a preset threshold, the data stream of the target submodal network is sampled according to a preset time and a preset period sampling ratio to obtain the transmission data sequence set corresponding to the target submodal network.

3. The network node traffic prediction method as described in claim 1, characterized in that, Before inputting the target network transmission information into a preset feature extraction network to obtain the network transmission feature vector, the method further includes: The initial convolutional neural network is iteratively updated based on the transmission matrix of the training network to obtain the preset convolutional neural network. A preset structure adjustment operation is performed on the preset convolutional neural network to obtain a preset feature extraction network.

4. A network node traffic prediction device, characterized in that, The network node traffic prediction device includes: The data acquisition module is used to acquire a preset node dataset through the target multimodal network and input the preset node dataset into the target modality graph embedding model to obtain the target network transmission information. The target multimodal network includes a preset number of submodal networks. The feature extraction module is used to input the target network transmission information into a preset feature extraction network to obtain the network transmission feature vector; The vector concatenation module is used to obtain the preset network information vector corresponding to the target multimodal network, and to concatenate the network transmission feature vector and the preset network information vector to obtain the network combination vector. The traffic prediction module is used to predict the traffic of the target multimodal network based on the network combination vector and through a preset deep learning model to obtain the node traffic prediction result. The data acquisition module is further configured to perform random walk sampling on the nodes of the target submodal network using a preset random walk algorithm to obtain a first node dataset corresponding to the target submodal network, wherein the target submodal network is any submodal network in the target multimodal network; perform data transmission sampling on the data stream of the target submodal network to obtain a second node dataset corresponding to the target submodal network; and generate a preset node dataset corresponding to the target submodal network based on the first node dataset, the second node dataset, and a preset structure sampling ratio. The data acquisition module is further configured to sample the data stream of the target submodal network according to a preset data sampling time and target routing information to obtain a transmission data sequence set corresponding to the target submodal network; determine a preset number of target transmission nodes corresponding to each node in the target submodal network according to the transmission data sequence set; determine a target transmission path for the preset number of transmissions according to each node and the target transmission nodes corresponding to each node; and generate a second node dataset corresponding to the target submodal network according to the target transmission path and the transmission data sequence set. The data acquisition module is further configured to obtain a target modal graph embedding model from a preset graph embedding model library according to the network type of each submodal network in the target multimodal network; input the preset node dataset into the target modal graph embedding model to obtain the preset number of intermediate network transmission vectors; and perform vector combination on the preset number of intermediate network transmission vectors to obtain target network transmission information. The data acquisition module is further configured to, when detecting inconsistencies in the vector dimensions of the preset number of intermediate network transmission vectors, use the intermediate network transmission vector with the longest vector dimension as the filling reference and perform preset filling on the remaining intermediate network transmission vectors to obtain a completed network vector; and concatenate the intermediate network transmission vector with the longest vector dimension and the completed network vector to obtain the target network transmission information.

5. A network node traffic prediction device, characterized in that, The device includes: a memory, a processor, and a network node traffic prediction program stored in the memory and executable on the processor, the network node traffic prediction program being configured to implement the steps of the network node traffic prediction method as described in any one of claims 1 to 3.

6. A storage medium, characterized in that, The storage medium stores a network node traffic prediction program, which, when executed by a processor, implements the steps of the network node traffic prediction method as described in any one of claims 1 to 3.

Images