Metropolitan area network flow prediction application method based on neural network

Abstract

The invention relates to a metropolitan area network traffic prediction application method based on a neural network, and belongs to the technical field of network traffic. Dividing a training set, a verification set and a test set; 70%, 20% and 10% of data are used for model training, optimization and evaluation; inputting the traffic data into the double-layer stacked LSTM network, obtaining hidden state information and extracting time sequence characteristics and dependency of network traffic; capturing a long-term and short-term dependency relationship in the flow by using an encoder; global dependency features are extracted through a self-attention mechanism; gradually predicting flow data of a plurality of time points in the future by adopting a decoder; an early stop strategy is adopted in model training to prevent overfitting; carrying out anti-standardization processing on a model prediction result, and recovering an original scale; and evaluating a model prediction result. According to the invention, training and optimization of the deep learning model are carried out on the historical data and the feature data of the metropolitan area network, so that the accuracy of IP network traffic prediction is remarkably improved.

Description

Technical Field

[0001] The invention belongs to the technical field of network flow, and in particular relates to a metropolitan area network flow prediction application method based on neural network. Background Art

[0002] As the network scale grows, the network complexity becomes higher and higher. In the future, the network must meet the business requirements of high bandwidth, low latency and on-demand services. It is difficult to provide the optimal solution in real time using traditional methods in traffic scheduling and path planning.

[0003] With the significant acceleration of the construction of new metropolitan area networks, operators need to gradually cut over old metropolitan area network equipment to the new metropolitan area network environment. For example, scenarios such as home broadband services and route changes will be gradually migrated from the old metropolitan area network to the new metropolitan area network, which will inevitably face a large number of network cutovers. However, the key operations of the current cutover process, such as cutover risk identification, cutover impact range confirmation, cutover command execution, and cutover verification, are extremely dependent on manual work. Although there is gradually process-based and programmable support, there is still a lack of ability to timely assess the impact of cutover services and cutover risk identification. It is necessary to introduce intelligent means to predict network traffic trends, support dynamic baseline judgment, timely understand cutover risks, and ensure the stability of network quality and service quality after cutover.

[0004] Currently, network traffic prediction mainly relies on traditional statistical models and machine learning methods. Common methods include:

[0005] (1) Time series analysis methods: Linear prediction models based on time series data, such as ARIMA and SARIMA models, have limited modeling capabilities for complex nonlinear relationships and need to rely on expert experience and knowledge.

[0006] (2) Machine learning methods: Common methods include decision trees, linear regression, support vector machines, random forests, etc., which are suitable for classification and regression problems. However, they have low computational efficiency when processing large amounts of complex data.

[0007] (3) Deep learning methods: including models such as recurrent neural networks and long short-term memory networks, which have powerful representation learning capabilities and can handle complex nonlinear relationships, but are computationally intensive and require a large amount of data for training, debugging, and optimization.

[0008] The prior art has the following defects:

[0009] (1) Insufficient prediction accuracy: Traditional models have difficulty in processing complex, nonlinear, and long-term series data when dealing with complex network traffic. Their prediction accuracy is limited and they are unable to cope with the drastic fluctuations and diversity of traffic.

[0010] (2) Low computational efficiency: Some models have low computational efficiency when processing large-scale data, making them difficult to apply in real time in actual network environments.

[0011] (3) Difficulty in capturing complex features: Traditional models lack an effective model update mechanism and are difficult to adapt to dynamic changes in the network environment. Summary of the invention

[0012] In view of the above-mentioned deficiencies in the prior art, the purpose of the invention is to provide a metropolitan area network traffic prediction application method based on neural network, which significantly improves the accuracy of IP network traffic prediction by training and optimizing the deep learning model on the historical data and feature data of the metropolitan area network.

[0013] The present invention proposes a metropolitan area network traffic prediction application method based on a neural network, comprising:

[0014] S1, data collection and preprocessing, including: collecting device IP, port, timestamp, inbound traffic, outbound traffic, and bandwidth data from the historical traffic data of the metropolitan area network; performing data cleaning, processing missing values ​​and outliers; standardizing network traffic data; extracting time features and classification features to provide input for subsequent model training;

[0015] S2, training the LSTM neural network model based on historical traffic data, including: firstly dividing the data set in chronological order, including training set, validation set and test set; wherein, the training set is used for model parameter learning, the validation set is used for model hyperparameter tuning and evaluation, and the test set is used for final performance evaluation of the model;

[0016] S3, inputs the traffic data into a two-layer LSTM network to obtain hidden state information and feature representation, uses the LSTM encoder to capture the long-term and short-term dependencies in the network traffic data, and extracts the temporal features and related dependencies;

[0017] S4, introduces multiple Transformer blocks and extracts global dependency features through self-attention mechanism;

[0018] S5, uses the trained two-layer LSTM model to predict network traffic, inputs the recent historical traffic data into the LSTM decoder, and gradually generates the predicted traffic sequence at multiple time points in the future;

[0019] S6, during the LSTM neural network model training process, an early stopping strategy is used to prevent overfitting. When the performance of the validation set no longer improves within a preset number of epochs, the training is automatically stopped. After the training is completed, the optimal model is saved in .h5 format and applied to the test set for prediction.

[0020] S7, performing de-standardization processing on the model prediction results to restore the original scale, including: based on the parameters used in data standardization, restoring the prediction results to the scale of the original data; wherein the parameters used in the standardization include: minimum value, maximum value, mean value, and standard deviation;

[0021] S8, evaluate the model prediction results, including: compare the prediction results with the actual network traffic data, evaluate the prediction performance of the model, and combine time series visualization analysis to check the prediction error and performance of the model in different time periods

[0022] Furthermore, in S2, the first 70% of the data is used as the training set, the next 20% of the data is used as the validation set, and the last 10% of the data is used as the test set.

[0023] Furthermore, in S5, the LSTM decoder uses the zero vector input and combines the hidden state information of the LSTM encoder to gradually generate a predicted traffic sequence for multiple future time points, including: predicting the traffic for the next 24 hours based on historical data at 5-minute intervals, that is, generating prediction results for 288 time points.

[0024] Furthermore, the LSTM decoder outputs the prediction results step by step through a multi-layer LSTM network and generates the final prediction value through a time-distributed dense layer.

[0025] Furthermore, after obtaining the model prediction result, the real-time monitoring traffic is compared with the model prediction result to dynamically adjust the application strategy.

[0026] Furthermore, after the application is implemented, the actual traffic and predicted traffic are compared to evaluate the application implementation effect, locate problems and take corrective measures.

[0027] Furthermore, it also provides model deployment: providing RESTFUL API for other systems to call, including:

[0028] Use HTTP protocol and define standardized RESTful API interface to facilitate other systems to call model capabilities;

[0029] Through the POST method, current or historical traffic data is passed in and future traffic forecast values ​​are returned;

[0030] Provide API interface to support monitoring the running status of the model;

[0031] Through capability plug-ins, API functions are embedded into various network management systems of operators.

[0032] Furthermore, according to the changes in the network environment, the LSTM neural network model is regularly and dynamically adjusted, including: adjusting the model parameters and optimizing the model structure.

[0033] Furthermore, the process of adjusting the model parameters includes:

[0034] Fine-tune existing pre-trained models for different network environments, including freezing some layers of the model and retraining the last layer or layers of the model to adapt to the new data distribution, thereby improving the prediction performance of the model in the new environment;

[0035] The adaptive optimization algorithm is used to dynamically adjust the learning rate to adapt the LSTM neural network model to different data characteristics.

[0036] Furthermore, the optimization and adjustment of the model structure includes:

[0037] Use the automatic architecture search NAS method to automatically select the model architecture that best suits the current network environment and optimize the model structure based on hardware resources and performance requirements;

[0038] Decompose the LSTM neural network model into multiple functional modules and dynamically combine the above functional modules according to different network conditions;

[0039] The gating mechanism is introduced to enable the LSTM neural network model to automatically select different processing paths according to the characteristics of the input traffic.

[0040] The beneficial effects of the present invention are as follows:

[0041] (1) Automated prediction: This paper uses LSTM and Transformer models combined with metropolitan area network traffic characteristics to realize the automation of network traffic prediction. Compared with traditional methods, no manual intervention is required, making the prediction process more efficient and intelligent. By comprehensively applying LSTM and Transformer models, adaptive feature extraction and efficient prediction are achieved, solving the problems of insufficient prediction accuracy and generalization ability in traditional methods.

[0042] (2) Improve prediction accuracy: Through the training and optimization of deep learning models, the present invention significantly improves the accuracy of IP network traffic prediction. In key scenarios, such as network cutover, accurate traffic prediction can improve the success rate of cutover, reduce the scope of business impact, and ensure the stability and quality of network cutover. The present invention dynamically adjusts model parameters and structures to adapt to changes in different network environments, providing operators with efficient and accurate traffic prediction, thereby optimizing network resource allocation and greatly improving network operation efficiency.

[0043] (3) Multi-model integration: Innovatively combining LSTM to capture the temporal characteristics of traffic and extracting global contextual relationships through Transformer, significantly improving the accuracy and stability of traffic prediction.

[0044] (4) Adaptive Optimization: By comprehensively applying the LSTM and Transformer models, adaptive feature extraction and efficient prediction are achieved, which solves the problems of insufficient prediction accuracy and generalization ability in traditional methods, and significantly improves the ability to capture drastic fluctuations such as traffic peaks in special time periods.

[0045] (5) Practicality and efficiency: The present invention can dynamically adjust model parameters and structures to adapt to changes in different network environments, provide operators with efficient and accurate traffic prediction, thereby optimizing network resource allocation and greatly improving network operation efficiency.

[0046] (6) Combining LSTM and Transformer models: By combining the advantages of LSTM and Transformer, the powerful ability of LSTM in capturing long-term and short-term dependencies in time series and the superiority of Transformer in parallel computing and global dependency feature extraction are fully utilized to solve the limitations of a single model in processing complex time series data.

[0047] (7) Efficient training and reasoning: The self-attention mechanism of the Transformer in the present invention can speed up training and reduce training time. Due to the recursive structure of the traditional LSTM model, data must be processed sequentially, resulting in a slow training speed and difficulty in fully utilizing the parallel computing capabilities of modern GPUs. However, the Transformer can process data in parallel. At the same time, combined with the sequence processing capabilities of LSTM, the model ensures that it has efficient reasoning performance in practical applications, which is very suitable for scenarios such as real-time network traffic prediction in metropolitan area networks.

[0048] (8) Ability to process diverse data: Traditional LSTM neural network models may have certain limitations when processing highly nonlinear and complex data patterns. However, the Transformer used in the present invention can better process diverse data patterns due to its flexible structure and strong adaptability. Therefore, the combination of the two can greatly enhance the model's ability to process diverse data.

[0049] (9) Stability and robustness of the model: Although the traditional LSTM neural network model has good robustness, it often does not work well when dealing with complex changes. However, due to the characteristics of the Transformer, the present invention is more stable when dealing with data fluctuations and anomalies, and can better adapt to changes in network traffic data such as metropolitan area networks.

[0050] (10) Dynamic iterative update model: The present invention proposes a unified traffic prediction API specification, provides services to various systems through capability plug-ins, and continuously adds new features and historical data to iteratively update the model. This dynamic update mechanism ensures continuous optimization of the model and high-precision prediction capabilities.

[0051] (11) Applicable to a wide range of application scenarios: The present invention is not only applicable to metropolitan area network environments, but also has broad application prospects and can be extended to other types of network operation and maintenance scenarios, thereby enhancing the technical support capabilities for network management and optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0053] Figure 1 A flow chart of a method for predicting metropolitan area network traffic based on a neural network according to an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of a model building process according to an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of the model convergence process of an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of the model prediction result evaluation according to an embodiment of the present invention;

[0057] Figure 5 A schematic diagram of the application of the prediction application method according to an embodiment of the present invention in a cutover process;

[0058] Figure 6 A schematic diagram of a real-time alarm reminder according to an embodiment of the present invention;

[0059] Figure 7 This is a diagram of the LSTM working principle of an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0061] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0062] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0063] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0064] The technical terms involved in the present invention are explained below:

[0065] Neural Network: A computing model that mimics the structure and function of the human brain, consisting of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function, called an activation function. The connection between each two nodes represents a weighted value for the signal passing through the connection, called a weight, which is equivalent to the memory of the artificial neural network. The output of the network varies depending on the connection method of the network, the weight value and the activation function. The network itself is usually an approximation of a certain algorithm or function in nature, or it may be an expression of a logical strategy, and is often used in tasks such as pattern recognition and data classification.

[0066] Recurrent Neural Network (RNN): It is a type of recursive neural network that takes sequence data as input, recurses in the direction of sequence evolution, and all nodes (recurrent units) are connected in a chain. It is one of the deep learning algorithms. Recurrent neural networks have memory, parameter sharing, and Turing completeness, so they have certain advantages in learning the nonlinear features of sequences. Recurrent neural networks are used in natural language processing (NLP), such as speech recognition, language modeling, machine translation, and other fields, and are also used for various time series forecasts.

[0067] Long Short-Term Memory (LSTM): A special recursive neural network with excellent performance on time series. The LSTM neural network model makes up for the problems of gradient vanishing, gradient explosion, and insufficient long-term memory capacity of the recurrent neural network (RNN). It is very suitable for solving data prediction with time series changes, such as network traffic, stock market, traffic and other traffic predictions.

[0068] Autoregressive Integrated Moving Average (ARIMA) model: A common time series forecasting model. When performing time series analysis, the ARIMA model finds the data change model (moving average, periodic component) based on the change pattern of historical data, thereby achieving future predictions.

[0069] Metropolitan Area Network (MAN) is a computer communication network established within a city, used to connect different local area networks and wide area networks. Because it uses local area network technology with active switching elements, the transmission delay in the network is small. Its transmission medium mainly uses optical cables, and the transmission rate is above 100 megabits per second.

[0070] Traffic Prediction: Predict future network traffic based on historical data and models, aiming to improve network resource utilization and service quality.

[0071] Mean Squared Error (MSE): It is a measure of the difference between the estimator and the estimated quantity, and can be used to measure the prediction error.

[0072] The present invention proposes a metropolitan area network traffic prediction application method based on neural network, which involves the field of network business support, deep learning and network traffic prediction technology. The method provided by the present invention mainly includes the following process:

[0073] (1) Data preprocessing: Collect and process network traffic data, perform standardization and feature extraction.

[0074] (2) Model training: Train the neural network model based on historical traffic data.

[0075] (3) Model prediction: Use the trained model to predict future network traffic.

[0076] (3) Model deployment: Provide a RESTful API for other systems to call.

[0077] (4) Model update: Update the model regularly to adapt to changes in the network environment.

[0078] The present invention combines the long short-term memory network (LSTM) and the Transformer model to achieve high-precision prediction of network traffic data.

[0079] like Figure 1 and Figure 2 As shown, the neural network-based metropolitan area network traffic prediction application method of the embodiment of the present invention includes the following steps:

[0080] S1, data collection and preprocessing, including: collecting device IP, port, timestamp, inbound traffic, outbound traffic, and bandwidth data from the historical traffic data of the metropolitan area network; performing data cleaning, processing missing values ​​and outliers; standardizing network traffic data; extracting time features and classification features to provide input for subsequent model training.

[0081] In this step, the Z-score method is used to standardize the network traffic data.

[0082] In this step, time features and classification features are extracted. Time features include: hourly, daily, weekly periodic features, etc. Classification features include: device IP and port and the level of the device in the network, etc.

[0083] S2, training the LSTM neural network model based on historical traffic data, including: firstly dividing the data set in chronological order, including training set, validation set and test set; among which, the training set is used for model parameter learning, the validation set is used for model hyperparameter tuning and evaluation, and the test set is used for the final performance evaluation of the model.

[0084] Specifically, the preprocessed network traffic data is divided into a training set, a validation set, and a test set at a ratio of 7:2:1. That is, the first 70% of the data is used as the training set, the next 20% of the data is used as the validation set, and the last 10% of the data is used as the test set.

[0085] S3, inputs the traffic data into a two-layer LSTM network to obtain hidden state information and feature representation, uses the LSTM encoder to capture the long-term and short-term dependencies in the network traffic data, and extracts timing features and related dependencies.

[0086] Specifically, taking the prediction of traffic every 5 minutes in the next day as an example, the traffic data of the most recent day is input into the two-layer LSTM network to obtain hidden state information and feature representation. The LSTM encoder can capture the long-term and short-term dependencies in network traffic data and extract time series features and related dependencies. This information can provide rich context for subsequent Transformer processing.

[0087] The following is an introduction to the LSTM neural network model:

[0088] Long Short-Term Memory (LSTM) has excellent performance on time series. The LSTM neural network model makes up for the problems of gradient vanishing, gradient explosion, and insufficient long-term memory capacity of the Recurrent Neural Network (RNN). It is very suitable for solving data prediction with time series changes, such as network traffic, stock market, traffic and other traffic predictions.

[0089] The LSTM neural network model introduces a gating mechanism, which consists of memory neurons, input gates, output gates, and forget gates. The calculation principles of each control gate are as follows:

[0090] 1. Input gate: remember some current information. Calculate the input gate The value of the candidate state value a of the input cell at time t t , Each value in is between 0 and 1, and a t The value is between -1 and 1.

[0091]

[0092] in, Represents the corresponding weight.

[0093] a t =tanh(W c ×(h t-1 ,X t )+b c ), (2)

[0094] in, Represents the corresponding offset.

[0095] 2. Forget gate: controls which information is discarded. Calculate the activation value f of the forget gate at time t t , f t Each value in is between 0 and 1.

[0096] f t =σ(W f ×(h t-1 ,X t )+b f ), (3)

[0097] Among them, W f ,b f They represent the weight and bias of the forget gate respectively, and σ represents the Sigmoid function.

[0098] 3. Neuron state update: According to the calculation results of the input gate and the forget gate, the neuron state is updated to obtain the cell state update value C at time t t :

[0099]

[0100] 4. Output gate: controls which information needs to be output. Updates the value C based on the calculated cell state t , we can get the calculation formula of the output gate:

[0101] h t =σ(W0×(h t-1 ,X t )+b0)×tanh(C t ) (5)

[0102] Among them, W0,b0 represents the weight and bias of the output gate, h t is the output value of the current unit.

[0103] like Figure 7 As shown, the working principle of the LSTM neural network model is as follows:

[0104] The current neuron receives the output h of the previous neuron. t-1 With the current input X t , calculate the forget gate f t , Input Gate Output gate o t and the hidden state C to be updated t :

[0105] f t =σ(W f [h t-1 ,x t-1 ]+b f )

[0106]

[0107] o t =σ(W o [h t-1 ,x t-1 ]+b o )

[0108]

[0109] b. Hidden state C at the last moment t Multiply it with the forget gate, multiply the hidden state to be updated with the update gate, and add the two results to get the current neuron hidden state C t :

[0110]

[0111] c. The hidden state of the neuron at the current moment C t Through the tanh activation function and multiplying with the output gate, we get the output h of the neuron at the current moment. t :

[0112] h t =o t *tanh(C t ) (8)

[0113] Compared with the traditional recurrent neural network, the LSTM neural network model has more hidden states of neurons C t, used to select forgetting and updating information in the neuron cyclic transmission, is the main channel for information transmission in LSTM neurons, and is calculated by the forget gate and the input gate, while h t is the hidden state C of the current neuron t After generation, through C t Calculated by the output gate as the output of the current neuron.

[0114] S4 introduces multiple Transformer blocks and extracts global dependency features through the Transformer self-attention mechanism.

[0115] Specifically, in this step, the Transformer block is applied: on the basis of the LSTM encoder, multiple Transformer blocks are introduced to further extract global dependency features through the self-attention mechanism. The Transformer block includes a multi-head attention layer and a feedforward neural network layer. The multi-head attention layer can process different parts of data in parallel and enhance the expressiveness of the model, while the residual connection and layer normalization technology improve the stability and training efficiency of the model.

[0116] Transformer is an advanced neural network model that is particularly suitable for processing time series and natural language processing tasks. Unlike traditional recurrent neural networks (RNNs), Transformer relies on a self-attention mechanism and does not need to process input data in sequence, thereby better parallelizing calculations. This feature significantly improves the efficiency of Transformer when processing large-scale data, while solving the gradient vanishing and gradient exploding problems in RNNs and enhancing the ability to capture long-term dependent information.

[0117] In network traffic forecasting, the Transformer model can effectively analyze complex time series data and achieve high-precision forecasting by capturing the medium-term and short-term dependency characteristics of network traffic. This gives it significant advantages in optimizing network resource allocation, improving network stability, and reducing operation and maintenance costs, thereby helping operators manage network resources more efficiently and improving overall operational efficiency and service quality.

[0118] The Transformer model processes sequence data by introducing a self-attention mechanism and a fully feedforward neural network. Its main components include a multi-head self-attention mechanism, a feedforward neural network, and position encoding.

[0119] 1. Multi-head self-attention mechanism: multiple self-attention heads are introduced to capture the dependencies between different positions in the sequence

[0120] Query, Key, and Value calculations:

[0121] Q=XW Q , K=XW K , V=XW V (9)

[0122] Where X is the input sequence, W Q , W K , W V are the weight matrices for query, key, and value, respectively.

[0123] 2. Calculation of Self-Attention:

[0124]

[0125] Among them, d k is the dimension of the key.

[0126] 3. Calculation of Multi-Head Self-Attention**:

[0127] MultiHead(Q,K,V)=Concat(head1,head2,…,head h )W O

[0128] head1= Attention(QW Q ,KW k ,VW v ) , (11)

[0129] Among them, W O is the output weight matrix.

[0130] 4. Feedforward Neural Network: Apply fully connected layers independently at each position to enhance the expressiveness of the model

[0131] FFN(x)=max(0,xW1+b1)W2+b2 (12)

[0132] Among them, W1 and W2 are weight matrices, b1 and b2 are bias vectors, and max() is the ReLU activation function.

[0133] 5. Position encoding: In order to retain the position information in the sequence, position encoding is used to inject the position information into the input

[0134]

[0135] Where pos is the position, i is the dimension index, and d modelis the model dimension.

[0136] S5, the trained two-layer LSTM model performs network traffic prediction, inputs the recent historical traffic data into the LSTM decoder, and gradually generates the predicted traffic sequence at multiple time points in the future.

[0137] Specifically, for example, based on a 5-minute interval, the LSTM decoder is used to predict the traffic volume for the next day. The LSTM decoder is fed with a zero vector and combined with the hidden state information of the LSTM encoder to gradually generate traffic forecast data for the next 288 time points. That is, based on the historical data at a 5-minute interval, the traffic volume for the next 24 hours is predicted, and the prediction results for 288 time points are generated.

[0138] The LSTM decoder outputs the prediction results step by step through a multi-layer LSTM network and generates the final prediction value through a time-distributed dense layer.

[0139] S6, during the training of the LSTM neural network model, an early stopping strategy is used to prevent overfitting. When the performance of the validation set no longer improves within a preset number of epochs, the training is automatically stopped. After the training is completed, the optimal model is saved in .h5 format and applied to the test set for prediction.

[0140] The preset number is 5. That is, during the training of the LSTM neural network model, an early stopping strategy is used to prevent overfitting, and training is automatically stopped when the performance of the validation set does not improve within 5 consecutive epochs.

[0141] In this step, the training cycle is initially set to 30 epochs, the batch size is set to 64, and the validation set split ratio is 0.20.

[0142] Specifically, during the model training process, an early stopping strategy is used to prevent overfitting, and the training cycle, batch size, and validation set split ratio are set to ensure that the model performs well on both the training set and the validation set. After the training is completed, the model is saved and applied to the test data set to make predictions and evaluate the model performance.

[0143] Figure 3 Schematic diagram of the model convergence process of an embodiment of the present invention. Figure 3As shown in the figure, during the model convergence process, we can see that the training loss (Loss) and validation loss (val_loss) have dropped significantly from the initial high values ​​to a level close to 0, which indicates that the model has successfully reduced the prediction error during the training process and also shows good prediction accuracy on unseen validation data. The continuous decline in loss is a key indicator of successful model training. Although the average accuracy (map@0.5) and the average accuracy of the validation set (val_map) have low starting values, they gradually improve as training progresses. Although the numerical increase seems small, considering that the average accuracy is in the range of [0,100], its improvement actually reflects the enhancement of the model's recognition and prediction capabilities.

[0144] At the same time, as the training progresses, the changes in each Epoch tend to be stable, and there is no sign of large fluctuations or overfitting (i.e., a sudden increase in validation loss). This indicates that the model maintains good stability during the learning process. And the model converges relatively quickly, which can be seen from the rapid decrease in loss value and the gradual increase in accuracy. This is very beneficial for model training in practical applications, especially when large-scale data needs to be processed. In general, the training process shows good performance improvement and stability of the model. The rapid reduction in loss and the continuous improvement in accuracy indicate that the model has good learning and generalization capabilities. These results indicate that the model may perform well in practical applications.

[0145] S7, performing denormalization processing on the prediction results of the LSTM neural network model to restore the original scale, including: restoring the prediction results to the scale of the original data according to the parameters used in data normalization. The parameters used in the normalization include: minimum value, maximum value, mean value, and standard deviation.

[0146] Specifically, denormalization is performed based on the minimum value, maximum value, mean or standard deviation used when normalizing the data, and the Z-score normalization method is used to convert the data into a distribution with a mean of 0 and a standard deviation of 1. Denormalization is to restore the data to its original mean and standard deviation. The mathematical formula is: The standardized data The original data is restored to its original scale according to its mean μ and standard deviation σ.

[0147] The prediction results are denormalized to restore the original scale for comparison and evaluation with real data. Through this hybrid method combining LSTM and Transformer models, this solution effectively improves the accuracy and stability of network traffic prediction and is suitable for complex time series prediction tasks.

[0148] S8,evaluates the prediction results of the LSTM neural network model, including: comparing the prediction results with the actual network traffic data, and evaluating the prediction performance of the model, and combining time series visualization analysis to check the prediction error and performance of the model in different time periods.

[0149] Specifically, the prediction results are compared with the actual network traffic data, and the prediction performance of the model is evaluated using indicators such as mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE). Time series visualization analysis is also used to check the prediction error and performance of the model in different time periods.

[0150] like Figure 4 As shown in the figure, the evaluation of the model prediction results shows that the overall trends of the inflow and outflow are relatively consistent. The predicted flow curve is basically consistent with the actual flow curve, indicating that the model prediction effect is good. The change trends and peak values ​​of the two flows in certain specific time periods are very close, indicating that the model can accurately capture the trend of flow changes in most cases. In some time periods, the predicted flow deviates slightly from the actual flow, which may be caused by emergencies or abnormal conditions. Overall, the prediction results of the model are highly consistent with the actual data, and the prediction performance of the model is good.

[0151] After obtaining the model prediction results, the real-time monitoring traffic is compared with the model prediction results to dynamically adjust the application strategy. After the application is implemented, the actual traffic is compared with the predicted traffic to evaluate the application implementation effect, locate the problem and take corrective measures.

[0152] In addition, the present invention also includes providing model deployment: providing a RESTFUL API for other systems to call.

[0153] Specifically, API design: Use HTTP protocol to define a standardized RESTful API interface to facilitate other systems to call model capabilities.

[0154] Prediction request: Use the POST method to pass in current or historical traffic data and return future traffic prediction values.

[0155] Model status monitoring: Provides an API interface to support monitoring the running status of the model, such as whether it needs to be updated or tuned.

[0156] Capability expansion: Through capability plug-ins, API functions can be embedded into various network management systems of operators, such as resource scheduling systems, network cutover systems, etc. The flexibility of plug-ins enables the model to provide customized services for different business needs, and identity authentication mechanisms can be added as needed to ensure the security of the interface.

[0157] The present invention supports model updating: the model is updated regularly to adapt to changes in the network environment.

[0158] Specifically, according to changes in the network environment, the LSTM neural network model is adjusted regularly, including: model fine-tuning, retraining model parameters, and optimization and adjustment of the model structure.

[0159] 1. Model fine-tuning and retraining

[0160] In different network environments and scenarios, by fine-tuning model parameters and appropriate retraining, the model can better adapt to new data distribution and network requirements. For example, in the IPRAN network, for the optimization of bandwidth rules, the model can be fine-tuned and retrained to analyze historical traffic data and predict traffic for a certain period of time in the future. Based on the prediction results, the bandwidth allocation of other services can be optimized to ensure reasonable allocation of resources and avoid excessive or insufficient problems.

[0161] For capacity planning issues in base station diagnosis, fine-tuning and retraining can be used to collect historical traffic data (such as daily peak traffic, trough traffic, and average traffic), analyze the time distribution and seasonal changes of traffic, and predict peak and trough time periods. Based on the prediction results, adjust resource allocation (such as spectrum allocation, power control, user access strategy, etc.), optimize resource utilization, and avoid network congestion during peak hours. At the same time, based on demand, decide whether to add new base stations or upgrade the hardware facilities of existing base stations (such as adding transmission lines, upgrading antennas, or enhancing server processing capabilities).

[0162] Therefore, by regularly fine-tuning and retraining, the model can adapt to changes in different network environments and scenarios. The specific steps are as follows:

[0163] Fine-tune model parameters, including:

[0164] Transfer learning and fine-tuning: Use existing pre-trained models and fine-tune them for different network environments. Freeze most layers and only adjust the parameters of the last few layers to adapt to the new data distribution.

[0165] For example: the parameters of the existing pre-trained model are θ pretrained , the training data of the new environment is D new , then the fine-tuned parameter θ fine-tune It can be obtained by optimizing the following objective function:

[0166]

[0167] in, is the loss function and λ is the regularization parameter that controls the fine-tuning procedure.

[0168] Adaptive learning rate: Use adaptive optimization algorithms (such as Adam and AdaGrad) to dynamically adjust the learning rate so that the model can quickly adapt to different data characteristics.

[0169] For example, when using the Adam optimization algorithm, the learning rate α is dynamically adjusted as follows:

[0170]

[0171] in, is the sliding average of the squared gradient and ∈ is a smoothing parameter to ensure numerical stability.

[0172] Model retraining: Regularly retrain the model using the latest traffic data to ensure that the model can adapt to environmental changes at any time. Use online learning to update model parameters every time new data is obtained.

[0173] 2. Optimization and adjustment of model structure

[0174] Automatic Architecture Search (NAS): Use the NAS method to automatically select the model architecture that best suits the current network environment. This method can optimize the model structure based on hardware resources and performance requirements.

[0175] For example: NAS can be defined as an optimization problem, whose goal is to find the optimal model structure A * :

[0176]

[0177] in, is the space of all possible architectures, and Cost(A) is the computational cost of the architecture, such as latency or memory usage.

[0178] Modular design: The model is decomposed into multiple modules, such as multi-layer LSTM layers and Transformer layers, and selectively combined according to different network conditions and task requirements. For example, LSTM layers, Transformer layers, and fully connected layers. Each module can be selected through NAS or manual strategy. In different network environments, the strategy function π is used to select the optimal module combination M. * :

[0179] M * =π(NetworkCondition,ResourceConstraint) (17)

[0180] In this way, if the network is in a high latency, reduce the number of LSTM layers n LSTM , increase the number of Transformer layers n Transformer .

[0181] Gating mechanism: The introduction of a gating mechanism allows the model to automatically select different processing paths based on the characteristics of the input traffic. When traffic surges, more Transformer layers can be activated to capture complex time series features.

[0182] 3. Feedback-based adaptive mechanism

[0183] Real-time monitoring and feedback: After the model is deployed, the prediction performance is monitored in real time, and the model weights are adjusted or retrained based on the feedback. For example, when the prediction error of the model is large in certain periods, some parameters of the model can be adjusted based on the feedback, or the weight of historical data can be increased. The real-time adjustment mechanism can be used to adjust the prediction performance of the model based on the feedback error ∈ t To guide the update of the model, the specific parameter update formula is:

[0184]

[0185] Where γ is the feedback adjustment step size, The model prediction error.

[0186] Meta-learning: Using meta-learning methods, an auxiliary model is trained to learn how to adjust the parameters of the main model so that it can quickly adapt to the new network environment. For example, a meta-learner g(·) is trained, whose input is the current network state and model performance, and the output is the adjustment strategy:

[0187] Δθ=g(NetworkCondition,ModelPerformance) (19)

[0188] Reference below Figure 5 The application of the method of the present invention in a practical scenario is described by taking the application example of traffic prediction in network cutover.

[0189] Process description: Before, during and after the cutover, the network quality, service quality, and related equipment alarms of the ports involved in the cutover are monitored in real time, and the traffic prediction model is used to automatically determine whether there are abnormal trends in indicators such as port traffic.

[0190] Preparation before cutover: Before network cutover, use the traffic prediction model to predict the traffic in the upcoming cutover period, select the traffic trough period for cutover, and plan resource allocation in advance.

[0191] Monitoring during the cutover process: During the cutover process, by comparing the real-time monitoring of traffic changes with the predicted results, the cutover strategy is dynamically adjusted to ensure a smooth cutover process.

[0192] Post-cutover evaluation: After the cutover is completed, the actual traffic flow is compared with the predicted traffic flow to evaluate the cutover effect, quickly locate the problem and take corrective measures.

[0193] Scenario 1: Route change: The service of one port is transferred to another port through route change, and the traffic of the current port is added to the target port. Before the cutover, the traffic prediction model is used to predict the traffic carrying situation after the cutover, and potential risks such as traffic overload are discovered in time.

[0194] Scenario 2: Dynamic monitoring: The dynamic baseline is calculated based on the trend of the data traffic forecast for the day and the traffic data of the previous day, week, and month compared with the same period last year and month last year. The traffic after the cutover is predicted and compared with the actual traffic. If the deviation is too large, a warning is issued in a timely manner.

[0195] like Figure 6 As shown, by comparing the predicted value with the actual real-time traffic value collected, a large difference appears around 12:00, and a real-time alarm reminder is issued.

[0196] The application of the method provided by the present invention in traffic prediction in network cutover can achieve the following functions:

[0197] Real-time dynamic monitoring: During the network cutover process, the impact of the cutover can be dynamically understood through real-time traffic forecasting, and actual traffic data can be collected and compared with the forecast data. Error thresholds can be set, and warnings can be triggered when the error exceeds the threshold. The cause of the error can be analyzed by combining historical data and the network alarm system, and corresponding measures can be taken.

[0198] Plan resources in advance: Before upgrading or cutting network equipment, determine the required equipment and bandwidth resources through traffic forecasting and deploy them in advance to ensure that traffic peaks are effectively managed during the cutover process and avoid network congestion caused by sudden traffic. At the same time, formulate emergency response plans based on traffic forecasts and real-time monitoring results to ensure that backup resources can be quickly called upon when sudden traffic increases to ensure the continuity of network services.

[0199] Reduce cutover risks: By predicting future traffic, identify periods of low traffic in the future, evaluate the potential risks of cutover in different time periods, develop risk assessment reports, and select the most appropriate time for cutover. At the same time, develop detailed emergency plans for high-risk periods to ensure rapid response when emergencies occur. During the cutover process, monitor actual traffic in real time to ensure that it is consistent with the predicted results. Based on real-time data, dynamically adjust the cutover plan and resource allocation to reduce cutover risks.

[0200] Improve operation and maintenance efficiency: Automated traffic prediction and real-time monitoring systems can identify potential problems in advance, reduce the frequency and severity of emergencies, and thus improve the efficiency of network operation and maintenance. At the same time, the traffic prediction system can automatically generate reports and warnings, reducing the workload and pressure of operation and maintenance personnel, allowing them to focus on higher-value tasks and decisions.

[0201] The present invention solves the shortcomings of traditional methods in processing complex and nonlinear data, and improves the accuracy and adaptability of traffic prediction. By introducing LSTM and Transformer models, the time dependency and nonlinear relationship in the data can be better captured, and accurate prediction of metropolitan area network traffic can be achieved, thereby improving the utilization rate of network resources and service quality.

[0202] The present invention combines advanced neural network technologies such as LSTM (Long Short-Term Memory Recurrent Neural Network) and Transformer model with the traffic characteristics of metropolitan area network (IP) network to provide a professional model for metropolitan area network traffic prediction. The model significantly improves the accuracy of IP network traffic prediction by training and optimizing the deep learning model based on the historical data and feature data of the metropolitan area network.

[0203] In the operator's network cutover and other operation and maintenance scenarios, this model can effectively reduce the reliance on manual intervention. Through accurate traffic prediction, the model improves the success rate of network cutover and reduces the scope of the impact of network cutover on the business, thereby ensuring the stability and quality of network cutover. For example, when performing network cutover operations, accurate traffic prediction can help operators plan resource allocation in advance, understand the cutover results and impacts in real time, reduce the occurrence of emergencies, and ensure the continuity and stability of network services.

[0204] The present invention also enables operators to manage network resources more efficiently. By accurately predicting network traffic, operators can optimize the allocation and use of network resources, reduce resource waste, and improve overall network operation efficiency and quality. This technology is not only applicable to metropolitan area network environments, but also has broad application prospects and can be extended to other types of network operation and maintenance scenarios, providing strong technical support for network management and optimization.

[0205] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting metropolitan area network traffic based on a neural network, characterized in that: include: S1, data collection and preprocessing, including: collecting device IP, port, timestamp, inbound traffic, outbound traffic, and bandwidth data from the historical traffic data of the metropolitan area network; performing data cleaning, processing missing values ​​and outliers; standardizing network traffic data; extracting time features and classification features to provide input for subsequent model training; S2, training the LSTM neural network model based on historical traffic data, including: firstly dividing the data set in chronological order, including training set, validation set and test set; wherein, the training set is used for model parameter learning, the validation set is used for model hyperparameter tuning and evaluation, and the test set is used for final performance evaluation of the model; S3, inputs the traffic data into a two-layer LSTM network to obtain hidden state information and feature representation, uses the LSTM encoder to capture the long-term and short-term dependencies in the network traffic data, and extracts the temporal features and related dependencies; S4, introduces multiple Transformer blocks and extracts global dependency features through the Transformer self-attention mechanism; S5, uses the trained two-layer LSTM model to predict network traffic, inputs the recent historical traffic data into the LSTM decoder, and gradually generates the predicted traffic sequence at multiple time points in the future; S6, during the LSTM neural network model training process, an early stopping strategy is used to prevent overfitting. When the performance of the validation set no longer improves within a preset number of epochs, the training is automatically stopped. After the training is completed, the optimal model is saved in .h5 format and applied to the test set for prediction. S7, performing de-standardization processing on the model prediction results to restore the original scale, including: restoring the prediction results to the scale of the original data according to the parameters used in data standardization; wherein the parameters used in the standardization include: minimum value, maximum value, mean value, and standard deviation; S8, evaluates the model prediction results, including: comparing the prediction results with the actual network traffic data, evaluating the prediction performance of the model, and combining time series visualization analysis to check the prediction error and performance of the model in different time periods.

2. The method for predicting metropolitan area network traffic based on neural network according to claim 1, characterized in that: In S2, the first 70% of the data is used as the training set, the next 20% of the data is used as the validation set, and the last 10% of the data is used as the test set.

3. The method for predicting metropolitan area network traffic based on neural network according to claim 1, characterized in that: In S5, the LSTM decoder gradually generates traffic prediction data for multiple future time points through zero vector input and combined with the hidden state information of the LSTM encoder, including: predicting the traffic in the next 24 hours based on historical data at 5-minute intervals, that is, generating prediction results for 288 time points.

4. The method for predicting metropolitan area network traffic based on neural network according to claim 3, characterized in that: The LSTM decoder outputs the prediction results step by step through a multi-layer LSTM network and generates the final prediction value through a time-distributed dense layer.

5. The method for predicting metropolitan area network traffic based on neural network according to claim 1, characterized in that: After obtaining the model prediction results, the real-time monitoring traffic is compared with the model prediction results to dynamically adjust the application strategy.

6. The method for predicting metropolitan area network traffic based on neural network according to claim 1, characterized in that: After the application is implemented, the actual traffic and predicted traffic are compared to evaluate the application implementation effect, locate problems and take corrective measures.

7. The method for predicting metropolitan area network traffic based on neural network according to claim 1, characterized in that: It also provides model deployment: providing a RESTFUL API for other systems to call, including: Using HTTP protocol, define standardized RESTful API interface to facilitate other systems to call model capabilities; Through the POST method, current or historical traffic data is passed in and future traffic forecast values ​​are returned; Provide API interface to support monitoring the running status of the model; Through capability plug-ins, API functions are embedded into various network management systems of operators.

8. The method for predicting metropolitan area network traffic based on neural network according to claim 1, characterized in that: According to changes in the network environment, the LSTM neural network model is regularly and dynamically adjusted, including: adjusting model parameters and optimizing model structure.

9. The method for predicting metropolitan area network traffic based on neural network according to claim 8, characterized in that: The process of adjusting model parameters includes: Fine-tune existing pre-trained models for different network environments, including freezing some layers of the model and retraining the last layer or layers of the model to adapt to the new data distribution, thereby improving the prediction performance of the model in the new environment; The adaptive optimization algorithm is used to dynamically adjust the learning rate to adapt the LSTM neural network model to different data characteristics.

10. The method for predicting metropolitan area network traffic based on neural network according to claim 8, characterized in that: The optimization and adjustment of the model structure includes: Use the automatic architecture search NAS method to automatically select the model architecture that best suits the current network environment and optimize the model structure based on hardware resources and performance requirements; Decompose the LSTM neural network model into multiple functional modules and dynamically combine the above functional modules according to different network conditions; The gating mechanism is introduced to enable the LSTM neural network model to automatically select different processing paths according to the characteristics of the input traffic.

Images