Network traffic prediction method, device, equipment and storage medium

Abstract

The application discloses a network traffic prediction method and device, equipment and a storage medium, and relates to the technical field of application flow. The method comprises the following steps: acquiring historical collection data and real-time collection data; performing sample expansion on the historical collection data to obtain an expanded data set; performing feature extraction on the expanded data set by using a neural network algorithm based on molecular stability optimization to obtain a first feature vector; performing feature mapping on the first feature vector by using a liquid drop diffusion model to obtain a second feature vector; training an initial prediction model according to the second feature vector to obtain a target prediction model; and performing network traffic prediction according to the real-time collection data and the target prediction model to obtain a traffic prediction result. The application solves the problem of poor accuracy of network traffic prediction in the related art, realizes network traffic prediction by training a model first and then applying the model, and achieves the effects of improving model precision and improving traffic prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of application traffic technology, and in particular to a network traffic prediction method, apparatus, device and storage medium. Background Technology

[0002] In the process of adaptive resource adjustment to adapt to changes in user network traffic, the method of acquiring cloud service user network traffic has a significant impact on the effectiveness of resource adjustment. In a cloud environment, methods for acquiring service user network traffic include predictive methods, which forecast network traffic for a future period. This approach supports proactive adaptive resource adjustment, reduces unnecessary adjustments, and ensures the effectiveness of the adjustments.

[0003] However, the models used in current network traffic prediction methods still need improvement in terms of robustness and generalization, resulting in poor accuracy in traffic prediction. Summary of the Invention

[0004] The main objective of this application is to provide a network traffic prediction method, apparatus, device, and storage medium, aiming to solve the technical problem of poor accuracy in network traffic prediction in related technologies.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] Firstly, this application provides a network traffic prediction method, including:

[0007] Acquire historical and real-time data;

[0008] The historical data was augmented to obtain an expanded dataset.

[0009] The first feature vector is obtained by using a neural network algorithm optimized based on molecular stability to extract features from the expanded dataset.

[0010] The first feature vector is mapped using a droplet diffusion model to obtain the second feature vector;

[0011] The initial prediction model is trained based on the second feature vector to obtain the target prediction model;

[0012] Network traffic prediction is performed based on real-time collected data and target prediction models to obtain traffic prediction results.

[0013] Optionally, in the above network traffic prediction method, the step of augmenting historical data to obtain an augmented dataset includes:

[0014] For each sample data in the historical data collection, Fermat's theorem is used to determine the nearest neighbor data of the sample data.

[0015] New sample data is generated based on the sample data and neighbor data;

[0016] The initial weights of the new sample data are determined based on the weights of the sample data and the weights of the neighbor data.

[0017] The target weights of the new sample data are obtained by adjusting the weights of the new sample data using the principle of quantum fluctuation.

[0018] Generate data corresponding to the sample data based on the new sample data and the target weights;

[0019] The historical data is merged with the generated data corresponding to a preset number of sample data to obtain an expanded dataset.

[0020] Optionally, in the above network traffic prediction method, the step of using a neural network algorithm optimized based on molecular stability to extract features from the expanded dataset and obtain the first feature vector includes:

[0021] A data network is established based on the expanded dataset; the data network includes multiple data points and the connections between the data points.

[0022] Calculate the affinity between each data point and other data points in the data network; where other data points are data points in the data network that have a connection relationship with the data point.

[0023] Adjust the connection weights between data points and other data points based on affinity;

[0024] The energy of each data point in the data network is calculated based on the connection weights, and the total energy of the data network is obtained.

[0025] The data network is trained based on the total energy to obtain an optimized neural network algorithm;

[0026] The optimized neural network algorithm is used to extract features from each data point in the data network to obtain the first feature vector; the first feature vector includes the position of each data point in the data network and the relationship between the data point and other data points.

[0027] Optionally, in the above network traffic prediction method, the step of using the droplet diffusion model to perform feature mapping on the first feature vector to obtain the second feature vector includes:

[0028] A feature mapping function is defined using a droplet diffusion model and a data network;

[0029] Based on the feature mapping function, the first feature vector is connected by feature mapping to obtain the second feature vector.

[0030] Optionally, in the above network traffic prediction method, the step of training the initial prediction model based on the second feature vector to obtain the target prediction model includes:

[0031] A basic model is generated using an improved extreme learning machine algorithm based on symmetry. The basic model consists of an input layer, a hidden layer, and an output layer connected in sequence. The weights between the hidden layer and the output layer in the basic model are structurally symmetric. The output weights of the basic model are obtained by minimizing the error from the hidden layer to the output layer.

[0032] The basic model is optimized using an optimization algorithm that combines symmetry constraints and antisymmetric regularization to obtain the initial prediction model. The weights of the hidden layers in the initial prediction model are obtained by optimizing the weights of the hidden layers in the basic model using a momentum factor optimization strategy.

[0033] The second feature vector is input into the initial prediction model for model training to obtain the target prediction model.

[0034] Optionally, in the above network traffic prediction method, the step of predicting network traffic based on real-time collected data and the target prediction model to obtain the traffic prediction result includes:

[0035] A neural network algorithm optimized based on molecular stability was used to extract features from real-time acquired data to obtain extracted feature vectors.

[0036] The target feature vector is obtained by using a droplet diffusion model to perform feature mapping on the extracted feature vector;

[0037] The target feature vector is input into the target prediction model to predict network traffic and obtain the traffic prediction result.

[0038] Secondly, this application provides a network traffic prediction device, comprising:

[0039] The data acquisition module is used to acquire historical and real-time data.

[0040] The sample augmentation module is used to augment historical data to obtain an augmented dataset.

[0041] The feature extraction module is used to extract features from the expanded dataset using a neural network algorithm optimized based on molecular stability, to obtain the first feature vector.

[0042] The feature mapping module is used to perform feature mapping on the first feature vector using the droplet diffusion model to obtain the second feature vector;

[0043] The model training module is used to train the initial prediction model based on the second feature vector to obtain the target prediction model; wherein, the initial prediction model is constructed based on the improved extreme learning machine algorithm;

[0044] The traffic prediction module is used to predict network traffic based on real-time collected data and the target prediction model, and obtain the traffic prediction results.

[0045] Thirdly, this application provides a network traffic prediction device, including a processor and a memory, wherein the memory stores a network traffic prediction program, and when the network traffic prediction program is executed by the processor, it implements the network traffic prediction method as described above.

[0046] Fourthly, this application provides a computer program product, including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, implement the network traffic prediction method described above.

[0047] Fifthly, this application provides a computer-readable storage medium storing a computer program that, when executed by one or more processors, implements the network traffic prediction method described above.

[0048] The above-mentioned one or more technical solutions provided in this application may have the following advantages or at least achieve the following technical effects:

[0049] This application proposes a network traffic prediction method, apparatus, device, and storage medium. It expands historical collected data to obtain an expanded dataset, then uses a neural network algorithm optimized based on molecular stability to extract features from the expanded dataset to obtain a first feature vector. A droplet diffusion model is then used to map the first feature vector to obtain a second feature vector. The initial prediction model is then trained based on the second feature vector to obtain a target prediction model. Finally, network traffic prediction is performed based on real-time collected data and the target prediction model to obtain the traffic prediction result. This approach achieves network traffic prediction by first training the model and then applying it. In this application, expanding historical collected data and using the expanded training data to train the prediction model results in richer and more complete training data, improving model accuracy. Combining the neural network algorithm optimized based on molecular stability and the droplet diffusion model to obtain feature vectors allows for the extraction of stable features, capturing the deep structure and characteristics of the data. This makes the target prediction model trained based on these feature vectors more likely to predict real traffic behavior, improving the accuracy of traffic prediction. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these provided drawings without creative effort.

[0051] Figure 1 This is a flowchart illustrating the first embodiment of the network traffic prediction method of this application;

[0052] Figure 2 This is a schematic diagram of the hardware structure of the network traffic prediction device involved in this application;

[0053] Figure 3 This is a detailed flowchart of step S200 in the second embodiment of the network traffic prediction method of this application;

[0054] Figure 4 This is a detailed flowchart of step S300 in the second embodiment of the network traffic prediction method of this application;

[0055] Figure 5 This is a schematic diagram of the functional modules of the first embodiment of the network traffic prediction device of this application.

[0056] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] It should be noted that in this application, 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. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element. In this application, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. In this application, the use of suffixes such as "module," "component," or "unit" to denote elements is solely for illustrative purposes and has no specific meaning in itself. Therefore, "module," "component," or "unit" may be used interchangeably.

[0059] For those skilled in the art, the specific meanings of the above terms in this application can be understood according to the specific circumstances. Furthermore, the technical solutions of the various embodiments can be combined with each other; however, this is based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0060] The pay-as-you-go model emphasized in cloud computing requires cloud service systems to provide users with different levels of service. During initial cloud service deployment, the cloud service environment needs to develop an optimal virtual resource configuration scheme for specific user network traffic. However, because cloud service user network traffic frequently changes in real time, the initial resource configuration cannot ensure that the cloud service always meets the Service Level Agreement (SLA). Therefore, cloud services need to dynamically adjust resource configuration at runtime with minimal resource overhead to guarantee the SLA. Adaptive resource adjustment can more effectively cope with real-time changes in the cloud environment, thus providing more effective support for the continuous performance assurance of the cloud service system.

[0061] Analysis of relevant technologies reveals that the method of acquiring cloud service user network traffic significantly impacts the effectiveness of resource adjustment during adaptive resource management that adapts to changes in user network traffic. There are two main methods for acquiring service user network traffic in the cloud environment: one is real-time acquisition, using current concurrent access data to determine the current network traffic value. However, due to the constantly changing cloud environment, this method suffers from unavoidable lag and is prone to generating numerous unnecessary adjustments. The other method uses predictive methods, leveraging historical network traffic data to predict network traffic over a future period. This approach supports proactive adaptive resource adjustment, reduces unnecessary adjustments, and ensures the effectiveness of the adjustments. Therefore, network traffic prediction is a key focus in research on adaptive resource management in cloud environments.

[0062] However, current network traffic prediction methods, such as obtaining historical traffic data from a specified app and inputting the historical traffic data into a preset mathematical model for traffic prediction, still need to improve the robustness and generalization of the models used, resulting in poor traffic prediction accuracy.

[0063] In view of the technical problem of poor accuracy in network traffic prediction in related technologies, this application provides a network traffic prediction method, apparatus, device and storage medium.

[0064] The network traffic prediction method, apparatus, device, and storage medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and implementation methods.

[0065] Example 1

[0066] Reference Figure 1 This application presents a first embodiment of a network traffic prediction method, which is applied to a network traffic prediction device.

[0067] Network traffic prediction devices refer to terminal devices or network devices that can achieve network connectivity. These devices can be terminal devices such as mobile phones, computers, tablets, portable computers, and embedded industrial control computers, or network devices such as servers and cloud platforms.

[0068] like Figure 2 The diagram shown illustrates the hardware structure of a network traffic prediction device. The network traffic prediction device may include: a processor 1001, such as a CPU (Central Processing Unit), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005.

[0069] Specifically, the communication bus 1002 is used to realize the connection and communication between these components; the user interface 1003 is used to connect to the client and communicate data with the client, and the user interface 1003 may include an output unit and an input unit; the network interface 1004 is used to connect to the backend server and communicate data with the backend server, and the network interface 1004 may include an input / output interface; the memory 1005 is used to store various types of data, such as instructions for any application or method in the network traffic prediction device, as well as application-related data, and the memory 1005 may be built-in memory; optionally, the memory 1005 may also be a storage device independent of the processor 1001, and so on. Figure 2 The memory 1005 may include an operating system, a network communication module, a user interface module, and a network traffic prediction program; the processor 1001 is used to call the network traffic prediction program stored in the memory 1005 and perform the following operations:

[0070] Acquire historical and real-time data;

[0071] The historical data was augmented to obtain an expanded dataset.

[0072] The first feature vector is obtained by using a neural network algorithm optimized based on molecular stability to extract features from the expanded dataset.

[0073] The first feature vector is mapped using a droplet diffusion model to obtain the second feature vector;

[0074] The initial prediction model is trained based on the second feature vector to obtain the target prediction model;

[0075] Network traffic prediction is performed based on real-time collected data and target prediction models to obtain traffic prediction results.

[0076] Based on the network traffic prediction equipment described above, the following section combines... Figure 1 The flowchart shown illustrates the network traffic prediction method of this embodiment in detail. The method may include the following steps:

[0077] S100: Acquire historical and real-time data.

[0078] Specifically, historical data refers to network traffic data collected from historical records, which can be network traffic data for a specific time period. Historical data can consist of multiple records, each of which can include multiple attributes, such as the time the data was recorded, the number of data packets sent / received, the number of bytes sent / received, the communication protocol used, IP address information, port number, etc.

[0079] Real-time data refers to network traffic data newly collected at the current moment. It is the basic data for predicting network traffic at future moments. Before applying this real-time data, a prediction model can be determined first, and then the real-time data can be used as the input of the prediction model to predict network traffic.

[0080] In practical applications, network traffic prediction devices can obtain historical data from the data centers of large cloud service providers, or they can obtain it by aggregating real-time data collected at previous or historical times. No specific limitation is made here.

[0081] S200: Expand the historical data to obtain an expanded dataset.

[0082] Specifically, in order to enhance model training, the historical data can be expanded to obtain more and richer training data. This expanded data can then be used to train the model, achieving better training results and thus improving the model's prediction accuracy.

[0083] In practical applications, network traffic prediction devices can employ traditional random interpolation, improved interpolation combining Fermat's theorem and quantum fluctuation principles, the SMOTE algorithm based on Fermat's theorem (a comprehensive sampling algorithm for artificially synthesized data), or the SMOTE algorithm combining Fermat's theorem and quantum fluctuation principles to expand samples; no specific limitation is made here. Among these, the improved interpolation method combining Fermat's theorem and quantum fluctuation principles, the SMOTE algorithm based on Fermat's theorem, and the SMOTE algorithm combining Fermat's theorem and quantum fluctuation principles, compared to traditional random interpolation, can better ensure the authenticity of the generated samples in network traffic prediction tasks. Furthermore, the introduction of quantum fluctuation principles can correct the sample distribution of training data, ensuring that generated samples appear more realistically where they should be, thereby improving the continuity and similarity between generated and original samples, guaranteeing the authenticity of training data, and ultimately improving model training effectiveness.

[0084] S300: The augmented dataset is used to extract features using a neural network algorithm optimized based on molecular stability, resulting in the first feature vector.

[0085] Specifically, each data point in the augmented dataset can be considered a data point, and the relationships between these data points can be viewed as interactions between atoms. Inspired by molecular stability, feature extraction is combined with molecular stability to obtain a neural network algorithm optimized for molecular stability. This algorithm enables feature extraction based on molecular stability, simulating the stability of molecules and their structure and interactions between atoms to extract features, thus achieving more stable feature learning in high-dimensional data environments.

[0086] S400: The first eigenvector is mapped using the droplet diffusion model to obtain the second eigenvector.

[0087] Specifically, the droplet diffusion model simulates the diffusion process of droplets in a medium. Considering that each data point in network traffic data is like a droplet, these "droplets" diffuse as network traffic increases, reflecting different network traffic characteristics. Based on this, the droplet diffusion model is used to define a feature mapping relationship, performing feature mapping on the first feature vector to obtain the second feature vector.

[0088] Applying the droplet diffusion model to the feature mapping of network traffic simulates the diffusion process of droplets in a medium, providing a new perspective and description method for network traffic characteristics. It can capture the deep structure and features of the data, making the model more likely to predict the actual traffic behavior.

[0089] S500: Train the initial prediction model based on the second feature vector to obtain the target prediction model.

[0090] Specifically, the initial prediction model can be constructed using the Extreme Learning Machine (ELM) algorithm or an improved Extreme Learning Machine (ELM) algorithm based on a symmetric structure. The training process involves inputting the second feature vector into the initial prediction model to obtain output data. This output data is then compared with the labeled data from historical data collection to optimize the model parameters. The training is iteratively continued until the optimal model parameters are obtained, at which point the corresponding model is used as the target prediction model. The labeled data from historical data collection can be actual traffic records for subsequent time periods after a specific time period; correspondingly, the training data can be data from a specific time period within the historical data collection.

[0091] The standard ELM algorithm is prone to overfitting. Compared to the standard ELM algorithm, the improved ELM algorithm based on symmetry introduces symmetry constraints, which can reduce overfitting and improve the model's generalization ability. It also incorporates a symmetry penalty term to enhance the model's symmetry. Training the target prediction model using the initial prediction model constructed by the improved symmetry-based ELM algorithm reduces the risk of overfitting and improves the target prediction model's prediction performance in real-world environments, thereby further improving the accuracy of traffic flow prediction.

[0092] S600: Based on real-time collected data and target prediction models, network traffic is predicted to obtain traffic prediction results.

[0093] Specifically, after obtaining the target prediction model, the network traffic prediction device can input the real-time collected data obtained in S100 into the target prediction model to perform network traffic prediction. Alternatively, it can first extract features from the real-time collected data to obtain the target feature vector, and then input the target feature vector into the target prediction model to perform network traffic prediction. The feature extraction method can refer to S300-S400, and will not be elaborated here.

[0094] The network traffic prediction method provided in this embodiment expands historical collected data to obtain an expanded dataset. Then, a neural network algorithm optimized based on molecular stability is used to extract features from the expanded dataset to obtain a first feature vector. A droplet diffusion model is used to map the first feature vector to obtain a second feature vector. The initial prediction model is then trained based on the second feature vector to obtain a target prediction model. Finally, network traffic prediction is performed based on real-time collected data and the target prediction model to obtain the traffic prediction result. This method achieves network traffic prediction by first training the model and then applying it. In this application, historical collected data is expanded, and the expanded training data is used to train the prediction model. The training data is richer and more complete, which can improve the model accuracy. Combining the neural network algorithm optimized based on molecular stability and the droplet diffusion model to obtain feature vectors allows for the extraction of stable features, capturing the deep structure and characteristics of the data. This makes the target prediction model trained based on these feature vectors more likely to predict the actual traffic behavior, improving the accuracy of traffic prediction.

[0095] Example 2

[0096] Based on the same technical concept, referring to Figures 3 to 4 The second embodiment of the network traffic prediction method of this application is proposed, which is applied to a network traffic prediction device.

[0097] The following is combined Figure 3 The flowchart shown illustrates the network traffic prediction method of this embodiment in detail. The method may include the following steps:

[0098] S100: Acquire historical and real-time data.

[0099] Specifically, the sources of historical and real-time data can be the data centers of large cloud service providers. These data centers process a large number of network requests every day and collect network traffic data for each specific time period using unstructured JSON format.

[0100] In this embodiment, all collected data is gathered using unstructured JSON format, where each record represents network traffic data for a specific time period. For each data record, its attributes include: the time the data was recorded (timestamp t), and the number of data packets sent from the source IP address to the destination IP address (number of data packets P). s The number of data packets sent from the destination IP address to the source IP address, i.e., the number of data packets received, P. r The number of bytes sent from the source IP address to the destination IP address, i.e., the number of bytes sent (B). s The number of bytes sent from the destination IP address to the source IP address, i.e., the number of bytes received (B). r The protocol used for data transmission (such as HTTP, HTTPS, etc.), i.e., the communication protocol; and the starting IP address for data transmission, i.e., the source IP address. s The target IP address for data transmission is the destination IP address. d The starting port for data transmission, i.e., the source port number. s The target port for data transmission is the target port number. d .

[0101] For example, in a specific example, the content of one piece of data is as follows:

[0102] {"Timestamp": "2023-09-09 08:00:00"}

[0103] Number of data packets sent: 12345

[0104] Number of data packets received: 54321

[0105] "Number of bytes sent": 56789

[0106] "Number of bytes received": 98765

[0107] Communication protocol: HTTP

[0108] Source IP address: 192.168.1.1

[0109] Target IP address: 10.0.0.1

[0110] "Source port number": 80,

[0111] "Target port number": 8080

[0112] The data indicates that at 08:00:00 on 2023-09-09, 2000 data packets were sent from port 80 of IP address 192.168.1.100 to port 8080 of IP address 10.0.0.50, and this IP address responded with 2500 data packets.

[0113] In practice, in order to accurately predict future network traffic, such as the traffic in the next hour, the actual traffic in the subsequent time period is used as the label of the current data point. The current data point is a data point in the historical data of a certain time period before the subsequent time period.

[0114] In the example above, the data label is network traffic within a future time period (e.g., network traffic in the next hour). For each data point, the traffic data from the hour following that point is used as the label. The labeling process can be represented as:

[0115] L(t)=P r (t+1)

[0116] Where L(t) is the label of time t, representing the number of data packets received within one hour after time t.

[0117] For the above data point, if the network traffic at 09:00:00 on 2023-09-09 is 3000 data packets, then the label of this data point is 3000. That is, the traffic data one hour after the current time point t is used as the label to obtain the predicted traffic label L(t).

[0118] In this embodiment, sample data is given as an example, presented in tabular form, as shown in Table 1:

[0119] Table 1

[0120]

[0121]

[0122] S200: Expand the historical data to obtain an expanded dataset.

[0123] In optional implementations of this embodiment, such as Figure 3 The detailed process diagram shown below, S200 may include:

[0124] S210: For each sample data in the historical data collection, use Fermat's theorem to determine the nearest neighbor data of the sample data;

[0125] S220: Generate new sample data based on sample data and neighbor data;

[0126] S230: Determine the initial weight of the new sample data based on the weight of the sample data and the weight of the neighbor data;

[0127] S240: The weights of the new sample data are adjusted using the principle of quantum fluctuation to obtain the target weights of the new sample data;

[0128] S250: Generate data corresponding to the sample data based on the new sample data and the target weight;

[0129] S260: Merge the historical collected data with the generated data corresponding to a preset number of sample data to obtain an expanded dataset.

[0130] Imagine that each data point exists in a high-dimensional space and can find its nearest neighbor data point through the "shortest path".

[0131] The first step is to set the initialization parameters.

[0132] Initialization parameters can include the minimum number of neighbors, the maximum number of neighbors, and the number of new samples generated. Given sample data s in vector format, it can be represented as:

[0133] s = [t, P] s P r B s B r , π, IP s IP d Port s Port d ].

[0134] The second step is to perform a neighbor search.

[0135] For each sample in the dataset composed of historical data, i.e., each sample data in the historical data, the idea of ​​"shortest path" of Fermat's theorem is used to find its neighbors. The number of neighbors is dynamically determined and can be determined based on their distribution density in the dataset.

[0136] When generating data, the first step is to find the shortest path neighbor, aiming to find the neighbor closest to the sample data s, and thus determine the nearest neighbor data s′ to the sample data s; where the distance between two data points is calculated as follows:

[0137]

[0138] Where D(s, s′) is the distance between two data points, s i Let s' be the i-th feature of the sample data s. i Let be the i-th feature of the neighbor data s′;

[0139] Then, determine the number of neighbors, and let D be the average distance between all sample data and sample data s in the dataset consisting of historical data. avg Then the number of neighbors N d It can be:

[0140]

[0141] Here, ∈ is a very small constant to avoid division by zero;

[0142] Finally, after searching for N d After obtaining the neighbor data, the nearest neighbor can be selected to obtain the neighbor data s′ that is closest to the sample data s.

[0143] The third step is to interpolate and generate new samples.

[0144] s′ is the nearest neighbor of s, and the new sample data s new It is generated using the following interpolation formula:

[0145] s new =s + λ × (s′ - s)

[0146] Where λ is a random number in the range [0, 1];

[0147] Here, by combining the selected neighbors, i.e., the neighbor data s′, and using the interpolation method based on Fermat's theorem and the "shortest path" principle, new samples are generated, thus obtaining the new sample data s. new .

[0148] The fourth step is to inject diversity and determine the initial weights of the new sample data.

[0149] Assume W s and W s′ Let s and s' be the weights of the sample data s and the neighbor data s', respectively. Then, the new sample data s new initial weights W new It can be calculated as follows:

[0150]

[0151] Where η is a predetermined learning rate.

[0152] Step 5: Weight adjustment.

[0153] To ensure the new sample data s new In coordination with the distribution of the dataset composed of historically collected data, new sample data s are analyzed based on the principle of quantum fluctuation. newWeighting is adjusted; in physics, momentum is the product of an object's mass and velocity. In this embodiment, it can be interpreted as the "importance" of a data point and the "velocity" of generating new data points.

[0154] During weight adjustment, the principle of quantum fluctuation is used to correct the sample distribution. That is, each data sample is regarded as a "particle" and its possible locations are tried to be found, thereby improving data augmentation. Specifically, for each sample, its "fluctuation" is simulated so that it may appear in multiple places.

[0155] Here, the fluctuation intensity is defined as δ, based on the new sample data s. new Fluctuation sample data s after fluctuation q It can be calculated as follows:

[0156] s q =s new +δ×randn(1, 10)

[0157] Here, α and β are both hyperparameters used to control the intensity and range of quantum effects;

[0158] To ensure the accuracy of fluctuation sample data s after quantum fluctuations q The continuity and similarity between the original sample data s and the new sample data s, when updating the new sample data s. new When assigning weights, use the fluctuating sample data s q weight W q As the target weights for the new sample data, where:

[0159]

[0160] Here, γ and ζ are hyperparameters used to control the intensity and rate of weight updates, and are preset by humans.

[0161] Step 6, based on the new sample data s new The target weight Wq of the new sample data determines the generated data corresponding to the sample data s.

[0162] Based on the first to sixth steps above, new samples are generated for each sample data in the historical data collection, corresponding to S210-S250 above.

[0163] After that, steps one through six above can be repeated until the required number of new samples are generated, that is, the preset number of generated data corresponding to the preset number of sample data is obtained; then data integration is performed, that is, the historical collected data is merged with the preset number of generated data to obtain an expanded dataset, that is, the expanded dataset.

[0164] In this embodiment, the sample data example in Table 1 specifically given in S100 above can be expanded to obtain a corresponding generated data example, which is represented in tabular form as shown in Table 2:

[0165] Table 2

[0166]

[0167]

[0168] In this embodiment, to enhance model training, a SMOTE algorithm based on Fermat's theorem is proposed. Using this improved SMOTE algorithm for sample augmentation generates more realistic new samples, achieving high-quality sample augmentation and ensuring that the generated new samples have real paths within the dataset, thus guaranteeing the authenticity of the generated samples. Furthermore, by introducing the momentum principle, the distribution of the newly generated samples is ensured to be consistent with the original data. Unlike traditional random interpolation methods, this embodiment better ensures the authenticity of new samples in network traffic prediction tasks and makes the training data richer and more complete.

[0169] S300: The augmented dataset is used to extract features using a neural network algorithm optimized based on molecular stability, resulting in the first feature vector.

[0170] In optional implementations of this embodiment, such as Figure 4 The detailed process diagram shown below, S300 may include:

[0171] S310: Establish a data network based on the expanded dataset; the data network includes multiple data points and the connections between the data points.

[0172] S320: Calculate the affinity between each data point and other data points in the data network; where other data points are data points in the data network that have a connection relationship with the data point.

[0173] S330: Adjust the connection weights between data points and other data points based on affinity;

[0174] S340: Calculate the energy of each data point in the data network based on the connection weights to obtain the total energy of the data network;

[0175] S350: The data network is trained based on the total energy to obtain an optimized neural network algorithm, which is a neural network algorithm optimized based on molecular stability.

[0176] S360: The optimized neural network algorithm is used to extract features from each data point in the data network to obtain the first feature vector; the first feature vector includes the position of each data point in the data network and the relationship between the data point and other data points.

[0177] A neural network algorithm optimized based on molecular stability is used to extract features from the expanded dataset. Traditional neural network algorithms extract features by learning hidden patterns in the data. In this embodiment, inspired by molecular stability, data feature extraction is combined with molecular stability to achieve more stable feature learning in high-dimensional data environments.

[0178] Specifically, in physical chemistry, the stability of a molecule is related to its structure and the interactions between its atoms. Introducing this principle into neural networks, each data point is considered as an "atom," and the relationships between data points are simulated as "inter-atomic interactions." These interactions are represented by weights and connections in a neural network. To ensure the stability of the "molecule" (the combination of data points), an optimal mode of inter-atomic interaction is found, which corresponds to the weights and connections in the neural network.

[0179] Let the expanded dataset be D, and its attributes still include:

[0180] [t,P s P r B s B r , π, IP s IP d Port s Port d ]

[0181] A data network is built based on the extended dataset D. The data network includes multiple data points d, where each data point d is... i The 10 attributes are mapped to 10-dimensional vectors through normalization:

[0182] d i ={X i1 X i2 , ..., x ij , ..., X i10}

[0183] Where, x ij This represents the j-th attribute value of the i-th data point.

[0184] Based on the established data network, its multiple data points, and the connections between these data points, the feature extraction process is as follows:

[0185] The first step is initialization.

[0186] Randomly initialize the neural network weights and set an adaptive connection threshold and a maximum energy threshold.

[0187] The second step is forward propagation.

[0188] First, for each data point in the data network, calculate its "affinity" with surrounding data points; assuming a data sample d1 is:

[0189] d1={2023-09-0908:00:00,2000,2500,10000,15000,HTTP,192.168.1.100,10.0.0.50,80,8080}

[0190] Then calculate the affinity of the data, specifically for a data point d. i Other data points d j Affinity A ij The method can be expressed as:

[0191]

[0192] Here, ε is a very small constant to prevent the denominator from being 0;

[0193] Then, based on data point d i Other data points d j Affinity A ij Adjust the connection weights between data points based on the magnitude of affinity A; specifically, adjust the connection weights between data points based on affinity A. ij Adjust data point d i to data point d j Connection weight W ij , can be represented as:

[0194] W ij =tanh(A ij )

[0195] Next, the "energy" value of each data point and its connection point is calculated; specifically for each data point d i Calculate its energy E in the data network. i , can be represented as:

[0196]

[0197] Where N is the size of the expanded dataset D, which is the number of data points in the data network.

[0198] The third step is backpropagation.

[0199] After calculating the energy of each data point in the data network, the total energy of the data network can be calculated. Then, the total energy can be compared with the maximum energy threshold set in the first step. If the total energy exceeds the maximum energy threshold, the weights and connections can be adjusted. During the adjustment, the number of connections between data points should not exceed the adaptive connection threshold, and efforts should be made to minimize the total energy of the network.

[0200] During feature extraction, for each data point, its position in the data network and its connections with other data points form its features. Using an energy minimization strategy, the most stable features can be extracted from the data network. The algorithm terminates when the total energy of the data network changes very little over multiple iterations or when the network weight update is less than a preset threshold. At this point, the data network is trained, resulting in an optimized neural network algorithm. After training, the optimized neural network algorithm is used to train data point d. i Feature extraction is performed to obtain the first feature vector F. i as follows:

[0201]

[0202] Based on this, for the aforementioned sample data d1, we first calculate its affinity with other data points, then adjust the connection weights, then calculate its energy, and finally extract a first feature vector F1. This first feature vector F1 includes the position of data point d1 in the data network and the relationship between data point d1 and other data points.

[0203] Inspired by molecular stability, this embodiment proposes a novel neural network feature extraction method. It treats the relationships between data points as interatomic interactions and simulates the stability of molecules and their structure, as well as the interactions between atoms, to extract features, thus achieving feature extraction based on molecular stability. This not only enriches the methods of feature extraction but also enables more stable feature learning in high-dimensional data environments.

[0204] S400: The first eigenvector is mapped using the droplet diffusion model to obtain the second eigenvector.

[0205] In an optional embodiment of this example, S400 may include:

[0206] S410: Define the feature mapping function using a droplet diffusion model and a data network;

[0207] The feature mapping function represents the mapping relationship of droplet diffusion based on the corresponding diffusion radius for each data point at multiple different time points;

[0208] S420: Based on the feature mapping function, perform feature mapping and concatenation on the first feature vector to obtain the second feature vector.

[0209] The expanded dataset is mapped to network traffic features based on the droplet diffusion model. The droplet diffusion model simulates the diffusion process of droplets in a medium. It considers each data point in the network traffic data as a droplet. When the network traffic increases, these "droplets" will spread out, reflecting different network traffic characteristics.

[0210] Specifically, firstly, for each data point d i Define its diffusion rate R i for:

[0211]

[0212] Where, x ij For data point d i The j-th attribute value;

[0213] Then, for each data point d i Define its diffusion radius S at time t. i (t) is:

[0214]

[0215] The diffusion phenomenon of droplets in a homogeneous medium was simulated, where the diffusion rate of the droplets is related to the average of their property values.

[0216] Secondly, the feature mapping function M(d) is defined using the droplet diffusion model. i )for:

[0217] M(d i )={S i (1), S i (2), ..., S i (10)}

[0218] The feature mapping function M(d) i This represents the data point d at 10 different time points. i The diffusion radius;

[0219] Finally, based on the feature mapping function M(d) i For the first eigenvector F i Perform feature mapping and concatenation to obtain the second feature vector G. i for:

[0220] G i =Con{F i M(d) i )}

[0221] Where Con{} represents the feature join operation.

[0222] In this embodiment, a droplet diffusion model is applied to the feature mapping of network traffic to simulate the diffusion process of droplets in a medium, providing a new perspective and description method for network traffic features and further enriching the means of feature extraction. Based on feature extraction based on molecular stability and feature mapping based on the droplet diffusion model, a stable feature learning method is proposed, which can capture the deep structure and characteristics of the data, making the model more likely to predict the true traffic behavior.

[0223] S500: Train the initial prediction model based on the second feature vector to obtain the target prediction model.

[0224] In an optional embodiment of this example, S500 may include:

[0225] S510: Generate a base model using an improved extreme learning machine algorithm based on symmetry structure; the base model consists of an input layer, a hidden layer and an output layer connected in sequence; the weights between the hidden layer and the output layer in the base model are structurally symmetric; the output weights of the base model are obtained by minimizing the error from the hidden layer to the output layer.

[0226] S520: The basic model is optimized using an optimization algorithm that combines symmetry constraints and antisymmetric regularization to obtain the initial prediction model; the weights of the hidden layers in the initial prediction model are obtained by optimizing the weights of the hidden layers in the basic model using a momentum factor optimization strategy.

[0227] Among them, the output weights of the initial prediction model are the optimal values ​​of the output weights of the base model.

[0228] S530: Input the second feature vector into the initial prediction model, train the model, and obtain the target prediction model.

[0229] After feature extraction, classification is performed using an improved Extreme Learning Machine (ELM) algorithm based on symmetric structure. In the standard ELM algorithm, output weights are obtained by minimizing the error between the hidden layer and the output layer. However, in some cases, the standard ELM algorithm may lead to overfitting. To overcome this problem, this embodiment proposes an improved ELM algorithm based on symmetric structure. The symmetric structure ensures that the weights between the hidden layer and the output layer are structurally symmetrical, thereby reducing the risk of overfitting and improving the model's generalization ability.

[0230] Specifically, the process for obtaining the improved extreme learning machine algorithm based on symmetric structure is as follows:

[0231] The first step is initialization.

[0232] Generate the base model and randomly generate the weight matrix W and bias b of the hidden layers; here, the second feature vector G is...i Let x be an input. After the input x is passed to the hidden layer, the output h(x) of the hidden layer is:

[0233] h(x) = σ(Wx + b)

[0234] Where σ is the Sigmoid activation function, defined as

[0235] The second step is to determine the symmetric weights.

[0236] The weights generated during the initialization phase can be represented as w. ij To maintain symmetry, a symmetric weight w is also generated for this weight. ji Based on the symmetric structure strategy, the weight matrix W is ensured to be symmetric, thus for each element w in the weight matrix W... ij , there is w ij =w ji .

[0237] The third step is to activate the hidden layer.

[0238] The input data x is passed through the hidden layer and processed using the activation function (Sigmoid function) to obtain the output of the hidden layer.

[0239] The fourth step is to calculate the output weights.

[0240] The output weight β is calculated by minimizing the error from the hidden layer to the output layer, which is obtained by minimizing the error E from the hidden layer to the output layer. This error E is the sum of prediction errors over all training data and can be expressed as:

[0241]

[0242] To increase symmetry, a symmetry penalty term E is added. sym :

[0243]

[0244] Where λ is the strength of the symmetry penalty;

[0245] Then the total error E can be obtained. total for:

[0246] E total =E+E sym

[0247] Among them, E sym This is a preset system error.

[0248] Correspondingly, the process of optimizing the base model can be viewed as the objective of training the improved extreme learning machine algorithm based on symmetry structure, specifically minimizing E. total And obtain the output weight β.

[0249] To accelerate the convergence speed of the model and improve its prediction accuracy, this embodiment introduces a momentum factor into the optimization strategy and uses the momentum factor optimization strategy to optimize the basic model.

[0250] Specifically, for each weight w ij Iterative momentum update v ij Given by the following formula:

[0251]

[0252] in, This refers to the momentum update in the t-th iteration, where γ is the momentum factor, typically set to 0.9, and η is the learning rate. It is the weight w ij For the total error E total The partial derivatives;

[0253] Furthermore, the weight update is given by the following formula:

[0254]

[0255] Thus, the optimization based on the momentum factor optimization strategy can effectively accelerate the weight update process and make the gradient descent direction more stable.

[0256] Fifth step: Antisymmetric regularization.

[0257] To further enhance the model's stability, an antisymmetric regularization term is introduced to encourage the model to ensure the antisymmetry of the weights when learning symmetric weights. In this embodiment, the antisymmetric regularization term E is introduced. anti-sym :

[0258]

[0259] Where μ is the strength of antisymmetric regularization;

[0260] Therefore, optimizing the basic model can be written as an optimization problem:

[0261]

[0262] By solving this optimization problem, the optimal β and W can be obtained.

[0263] The sixth step is to optimize.

[0264] By combining optimization algorithms with symmetry constraints and antisymmetric regularization, such as gradient descent or Adam, the optimal output weight β can be calculated. After determining the model parameters, an initial prediction model can be obtained.

[0265] Finally, the second feature vector G obtained based on S400 is... i The initial prediction model obtained above is used for model training to obtain the final trained target prediction model, which can be put into specific applications.

[0266] In this embodiment, an improved ELM algorithm based on symmetry structure is adopted, which can overcome the model overfitting problem that may be caused by the standard ELM algorithm. The algorithm introduces symmetry constraints, which can reduce overfitting and improve the generalization ability of the model, thereby improving the prediction effect of the model in the real environment. A symmetry penalty term is also added to enhance the symmetry of the model.

[0267] S600: Based on real-time collected data and target prediction models, network traffic is predicted to obtain traffic prediction results.

[0268] In an optional embodiment of this example, S600 may include:

[0269] S610: Use a neural network algorithm optimized based on molecular stability to extract features from real-time acquired data and obtain extracted feature vectors;

[0270] S620: Use the droplet diffusion model to perform feature mapping on the extracted feature vector to obtain the target feature vector;

[0271] S630: Input the target feature vector into the target prediction model to perform network traffic prediction and obtain the traffic prediction result.

[0272] Specifically, further implementation details of S610 and S620 can be found in the descriptions of the specific implementations of S300 and S400, which will not be repeated here for the sake of brevity. After obtaining the target feature vector, the network traffic prediction device can input the target feature vector into the target prediction model to perform network traffic prediction.

[0273] After S500 is completed, a model with optimal parameters, i.e., the target prediction model, is obtained. Then, the following formula can be used to predict network traffic based on new data, that is, to predict network traffic based on the real-time collected data obtained from S100, which can be expressed as:

[0274] y pred =βh(x new )

[0275] Where, x new It is the target feature vector corresponding to the real-time acquired data, y predIt is x new The corresponding traffic forecast results.

[0276] After network traffic prediction is completed, the persistence of the prediction results is crucial for their use by other systems and applications. In this embodiment, after the network traffic prediction is completed and the traffic prediction results are obtained, the traffic prediction results can be persistently stored in a relational database, such as MySQL or PostgreSQL, to facilitate access by other systems and applications. In this embodiment, the storage can be performed in the following manner:

[0277]

[0278] Wherein, ID is the unique identifier of the data, TimeStamp is the prediction timestamp, PredictedValue is the predicted traffic value, ActualValue is the actual traffic value, Difference is the difference between the actual value and the predicted value, and ModelVersion is the model version used to track and compare the performance of different models.

[0279] Network traffic prediction devices can also provide an easily accessible API interface, offering a structured way for other systems and applications to store and access traffic prediction results, making the entire traffic prediction process more complete and practical.

[0280] To allow other applications to easily access the traffic prediction results, this embodiment provides a RESTful API interface, as shown in the simple API interface below:

[0281] GET / predictions: Retrieves all prediction data;

[0282] GET / predictions / {id}: Retrieves prediction data for a specific ID;

[0283] POST / predictions: Submit new prediction data;

[0284] PUT / predictions / {id}: Updates the prediction data for a specific ID;

[0285] DELETE / predictions / {id}: Deletes the prediction data for a specific ID.

[0286] When a user needs to obtain all traffic prediction results, the client can send a request to the network traffic prediction device, and the network traffic prediction device can respond to the request and return all stored traffic prediction results to the client.

[0287] It should be noted that more implementation details of the above method steps can be found in the description of the specific implementation in Example 1. For the sake of brevity, these details will not be repeated here.

[0288] In this embodiment, network access traffic can be predicted more accurately, and the accuracy of traffic prediction can be improved. It effectively learns the inherent time-series characteristics of the network traffic and throughput curves of the application system, resulting in better performance when processing periodic data. Furthermore, the target prediction model used has been further improved in terms of robustness and generalization compared to existing models.

[0289] This embodiment presents a network traffic prediction method based on machine learning. By analyzing a large amount of network request data, it obtains a target detection model to predict network traffic. This method can be applied to scenarios involving traffic splitting, such as dual-active architecture within the same city and multi-active architecture across different locations, as well as request traffic splitting in a two-site, three-center architecture. It provides real-time and effective data for adjusting traffic allocation, enabling optimized traffic splitting strategies, thereby improving network resource utilization and enhancing user experience. Furthermore, this method is applicable to various industries, such as cloud service providers and financial institutions, contributing to improved overall network performance and business reliability.

[0290] Example 3

[0291] Based on the same inventive concept, referring to Figure 5 This application presents a first embodiment of a network traffic prediction device, which can be a virtual device and applied to a network traffic prediction equipment.

[0292] The following is combined Figure 5 The functional module diagram shown illustrates the network traffic prediction device provided in this embodiment in detail. The device may include:

[0293] The data acquisition module is used to acquire historical and real-time data.

[0294] The sample augmentation module is used to augment historical data to obtain an augmented dataset.

[0295] The feature extraction module is used to extract features from the expanded dataset using a neural network algorithm optimized based on molecular stability, to obtain the first feature vector.

[0296] The feature mapping module is used to perform feature mapping on the first feature vector using the droplet diffusion model to obtain the second feature vector;

[0297] The model training module is used to train the initial prediction model based on the second feature vector to obtain the target prediction model; wherein, the initial prediction model is constructed based on the improved extreme learning machine algorithm;

[0298] The traffic prediction module is used to predict network traffic based on real-time collected data and the target prediction model, and obtain the traffic prediction results.

[0299] Furthermore, the sample augmentation module may include:

[0300] The data generation unit is used to determine the nearest neighbor data of each sample data in the historical data collection using Fermat's Last Theorem; generate new sample data based on the sample data and the neighbor data; determine the initial weight of the new sample data based on the weight of the sample data and the weight of the neighbor data; adjust the weight of the new sample data using the quantum fluctuation principle to obtain the target weight of the new sample data; and obtain the generated data corresponding to the sample data based on the new sample data and the target weight.

[0301] The data synthesis unit is used to merge historically collected data with generated data corresponding to a preset number of sample data to obtain an expanded dataset.

[0302] Furthermore, the feature extraction module may include:

[0303] Establish network units to build a data network based on the expanded dataset; the data network includes multiple data points and the connections between them.

[0304] The algorithm optimization unit is used to calculate the affinity between each data point and other data points in the data network; where other data points are data points in the data network that have a connection relationship with the data point; adjust the connection weights between the data point and other data points according to the affinity; calculate the energy of each data point in the data network according to the connection weights to obtain the total energy of the data network; train the data network according to the total energy to obtain the optimized neural network algorithm;

[0305] The feature extraction unit is used to extract features from each data point in the data network using an optimized neural network algorithm to obtain a first feature vector. The first feature vector includes the position of each data point in the data network and the relationship between the data points and other data points.

[0306] Furthermore, the feature mapping module may include:

[0307] Function definition unit, used to define feature mapping functions using droplet diffusion models and data networks;

[0308] The feature mapping unit is used to perform feature mapping and concatenation on the first feature vector according to the feature mapping function to obtain the second feature vector.

[0309] Furthermore, the model training module may include:

[0310] The model generation unit is used to generate a base model using an improved extreme learning machine algorithm based on symmetry. The base model consists of an input layer, a hidden layer, and an output layer connected in sequence. The weights between the hidden layer and the output layer in the base model are structurally symmetric. The output weights of the base model are obtained by minimizing the error from the hidden layer to the output layer.

[0311] The parameter optimization unit is used to optimize the basic model using an optimization algorithm that combines symmetry constraints and antisymmetric regularization to obtain the initial prediction model. The weights of the hidden layers in the initial prediction model are obtained by optimizing the weights of the hidden layers in the basic model using a momentum factor optimization strategy.

[0312] The model training unit inputs the second feature vector into the initial prediction model to train the model and obtain the target prediction model.

[0313] Furthermore, the traffic prediction module may include:

[0314] The feature extraction submodule is used to extract features from real-time acquired data using a neural network algorithm optimized based on molecular stability, and obtain extracted feature vectors; then, the extracted feature vectors are mapped using a droplet diffusion model to obtain the target feature vector.

[0315] The traffic prediction submodule is used to input the target feature vector into the target prediction model to perform network traffic prediction and obtain the traffic prediction result.

[0316] It should be noted that the functions and corresponding technical effects of each module in the network traffic prediction device provided in this embodiment can be referred to the description of the specific implementation methods in the various embodiments of the network traffic prediction method of this application. For the sake of brevity, they will not be repeated here.

[0317] Example 4

[0318] Based on the same inventive concept, referring to Figure 2 The hardware structure diagram shows that this embodiment provides a network traffic prediction device. The network traffic prediction device may include a processor and a memory. The memory stores a network traffic prediction program. When the network traffic prediction program is executed by the processor, it implements all or part of the steps of the various embodiments of the network traffic prediction method of this application.

[0319] Specifically, network traffic prediction devices refer to terminal devices or network devices that can achieve network connectivity. These can be terminal devices such as mobile phones, computers, tablets, portable computers, and embedded industrial control computers, or network devices such as servers and cloud platforms.

[0320] It is understandable that network traffic prediction devices may also include a communication bus, a user interface, and a network interface. The communication bus is used to connect and communicate between these components; the user interface is used to connect to the client and communicate data with the client. The user interface may include output units such as a display screen and speakers, and input units such as a keyboard and microphone; the network interface is used to connect to the backend server and communicate data with the backend server. The network interface may include input / output interfaces, such as standard wired interfaces and wireless interfaces such as Wi-Fi interfaces; the memory is used to store various types of data. This data may include, for example, instructions for any application or method in the network traffic prediction device, as well as application-related data. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Random Access Memory (RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), and Electrically Erasable Programmable Read-Only Memory (EPROM). The memory may be an EEPROM, magnetic storage, flash memory, disk, or optical disk, etc.; optionally, the memory may also be a storage device independent of the processor; the processor is used to call the network traffic prediction program stored in the memory and execute the network traffic prediction method as described above. The processor may be an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic controller (PLC), a programmable logic device (PLD), a field-programmable gate array (FPGA), a controller, a microcontroller, a microprocessor, or other electronic components, used to execute all or part of the steps of the various embodiments of the network traffic prediction method described above.

[0321] It needs to be explained that, Figure 2The hardware structure shown does not constitute a limitation on the network traffic prediction device of this application. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0322] Example 5

[0323] Based on the same inventive concept, this embodiment provides a computer program product, including a computer program stored on a non-transitory computer-readable storage medium, such as a network traffic prediction program. The network traffic prediction program includes program instructions, which, when executed by a computer, can implement all or part of the steps of various embodiments of the network traffic prediction method of this application.

[0324] Example 6

[0325] Based on the same inventive concept, this embodiment provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, disk, optical disk, server, etc. The storage medium stores a computer program, which can be executed by one or more processors. When the computer program is executed by the processor, it can implement all or part of the steps of the various embodiments of the network traffic prediction method of this application.

[0326] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. The above embodiments are only optional embodiments of this application and do not limit the patent scope of this application. All equivalent structural or procedural transformations made based on the inventive concept of this application and the content of the specification and drawings of this application, or direct or indirect applications in other related technical fields, are included within the patent protection scope of this application.

Claims

1. A network traffic prediction method, characterized in that, The method includes: Acquire historical and real-time data; The historical data is augmented to obtain an augmented dataset; The augmented dataset is used to extract features using a neural network algorithm optimized based on molecular stability to obtain a first feature vector; The first feature vector is mapped using a droplet diffusion model to obtain the second feature vector; The initial prediction model is trained based on the second feature vector to obtain the target prediction model; Based on the real-time collected data and the target prediction model, network traffic prediction is performed to obtain the traffic prediction result; The step of training the initial prediction model based on the second feature vector to obtain the target prediction model includes: A base model is generated using an improved extreme learning machine algorithm based on symmetry structure. The base model includes an input layer, a hidden layer, and an output layer connected in sequence. The weights between the hidden layer and the output layer in the base model are structurally symmetric. The output weights of the base model are obtained by minimizing the error from the hidden layer to the output layer. The basic model is optimized using an optimization algorithm that combines symmetry constraints and antisymmetric regularization to obtain an initial prediction model; the weights of the hidden layers in the initial prediction model are obtained by optimizing the weights of the hidden layers in the basic model using a momentum factor optimization strategy. The second feature vector is input into the initial prediction model to train the model and obtain the target prediction model.

2. The network traffic prediction method as described in claim 1, characterized in that, The step of augmenting the historically collected data to obtain an augmented dataset includes: For each sample data in the historical data collection, Fermat's theorem is used to determine the nearest neighbor data of the sample data; New sample data is generated based on the sample data and the neighbor data; The initial weight of the new sample data is determined based on the weight of the sample data and the weight of the neighbor data. The new sample data is weighted using the principle of quantum fluctuation to obtain the target weight of the new sample data. Based on the new sample data and the target weight, generated data corresponding to the sample data is obtained; The historical data collected is merged with the generated data corresponding to a preset number of sample data to obtain an expanded dataset.

3. The network traffic prediction method as described in claim 1, characterized in that, The step of extracting features from the augmented dataset using a neural network algorithm optimized based on molecular stability to obtain the first feature vector includes: A data network is established based on the expanded dataset; the data network includes multiple data points and the connection relationships between the multiple data points. Calculate the affinity between each data point in the data network and other data points; wherein, the other data points are data points in the data network that have a connection relationship with the data point. Based on the affinity, the connection weight between the data point and the other data points is adjusted; The energy of each data point in the data network is calculated based on the connection weights to obtain the total energy of the data network. The data network is trained based on the total energy to obtain an optimized neural network algorithm; The optimized neural network algorithm is used to extract features from each data point in the data network to obtain a first feature vector; the first feature vector includes the position of each data point in the data network and the relationship between the data point and other data points.

4. The network traffic prediction method as described in claim 3, characterized in that, The step of using a droplet diffusion model to perform feature mapping on the first feature vector to obtain the second feature vector includes: A feature mapping function is defined using the droplet diffusion model and the data network; According to the feature mapping function, the first feature vector is connected by feature mapping to obtain the second feature vector.

5. The network traffic prediction method as described in claim 1, characterized in that, The step of performing network traffic prediction based on the real-time collected data and the target prediction model to obtain the traffic prediction result includes: The neural network algorithm optimized based on molecular stability is used to extract features from the real-time acquired data to obtain an extracted feature vector. The extracted feature vector is mapped using the droplet diffusion model to obtain the target feature vector; The target feature vector is input into the target prediction model to perform network traffic prediction and obtain the traffic prediction result.

6. A network traffic prediction device, characterized in that, The device includes: The data acquisition module is used to acquire historical and real-time data. The sample expansion module is used to expand the historical collected data to obtain an expanded dataset; The feature extraction module is used to extract features from the augmented dataset using a neural network algorithm optimized based on molecular stability, to obtain a first feature vector. The feature mapping module is used to perform feature mapping on the first feature vector using the droplet diffusion model to obtain the second feature vector; The model training module is used to train the initial prediction model based on the second feature vector to obtain the target prediction model; wherein the initial prediction model is constructed based on the improved extreme learning machine algorithm. The traffic prediction module is used to predict network traffic based on the real-time collected data and the target prediction model, and obtain the traffic prediction result. The model training module includes: A model generation unit is used to generate a base model using an improved extreme learning machine algorithm based on symmetry structure. The base model includes an input layer, a hidden layer and an output layer connected in sequence. The weights between the hidden layer and the output layer in the base model are structurally symmetric. The output weights of the base model are obtained by minimizing the error from the hidden layer to the output layer. The parameter optimization unit is used to optimize the basic model using an optimization algorithm that combines symmetry constraints and antisymmetric regularization to obtain an initial prediction model; the weights of the hidden layers in the initial prediction model are obtained by optimizing the weights of the hidden layers in the basic model using a momentum factor optimization strategy. The model training unit is used to input the second feature vector into the initial prediction model, perform model training, and obtain the target prediction model.

7. A network traffic prediction device, characterized in that, The network traffic prediction device includes a processor and a memory. The memory stores a network traffic prediction program. When the network traffic prediction program is executed by the processor, it implements the network traffic prediction method as described in any one of claims 1 to 5.

8. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, implement the network traffic prediction method as described in any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by one or more processors, implements the network traffic prediction method as described in any one of claims 1 to 5.

Images