A traffic demand prediction method and system based on LSTM

Abstract

The present application discloses a traffic demand prediction method and system based on LSTM, which relates to the field of traffic prediction. The method comprises: obtaining first traffic usage information within a first time period of a target network and second traffic usage information within a second time period of the target network, wherein the first time period is less than the second time period; inputting the first traffic usage information into a first LSTM module in an LSTM traffic prediction model to obtain short-term traffic feature information; inputting the second traffic usage information into a second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information; inputting comprehensive traffic feature information into a feature fusion and prediction module in the LSTM traffic prediction model to obtain traffic demand prediction information, wherein the comprehensive traffic feature information comprises the short-term traffic feature information and the long-term traffic feature information.

Description

Technical Field

[0001] This specification relates to the field of traffic forecasting, and more specifically, the present application relates to a traffic demand forecasting method and system based on LSTM. Background Art

[0002] Predicting traffic demand within a data center can help optimize the allocation of server and storage resources and ensure efficient use of resources. By predicting the traffic demand of different servers or nodes, more effective load balancing strategies can be implemented to avoid overloading some nodes while others are idle. In addition, based on the predicted traffic data, cloud service providers can automatically adjust resource allocation, such as automatically increasing or decreasing virtual machine instances to meet service demand. With accurate predictions, cloud users can purchase and use cloud resources more efficiently, avoid wasting resources, and reduce operating costs.

[0003] The current traffic demand is predicted based on historical traffic usage over a period of time using time series analysis models, regression models and other methods. However, the current related technical traffic prediction methods are unable to cope with the current changing network demand and lack prediction accuracy.

[0004] Therefore, it is necessary to propose a more flexible and accurate flow demand prediction method and system. Summary of the invention

[0005] A series of simplified concepts are introduced in the Summary of the Invention section, which will be further described in detail in the Detailed Description of the Invention section. The Summary of the Invention section of this application does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the scope of protection of the claimed technical solution.

[0006] In the first aspect, the present application proposes a traffic demand prediction method based on LSTM, the method comprising:

[0007] Obtaining first traffic usage information of a target network within a first duration and second traffic usage information of a target network within a second duration, wherein the first duration is shorter than the second duration;

[0008] Input the first traffic usage information into the first LSTM module in the LSTM traffic prediction model to obtain short-term traffic feature information;

[0009] Input the second traffic usage information into the second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information;

[0010] The comprehensive traffic feature information is input into the feature fusion and prediction module in the above-mentioned LSTM traffic prediction model to obtain traffic demand prediction information. The above-mentioned comprehensive traffic feature information includes the above-mentioned short-term traffic feature information and the above-mentioned long-term traffic feature information.

[0011] In a feasible implementation, the first LSTM module includes a first input layer, a convolution layer, a pooling layer, a first LSTM layer and a first fully connected layer;

[0012] The second LSTM module includes a second input layer, a second LSTM layer, an attention layer, and a second fully connected layer;

[0013] The feature fusion and prediction module includes a feature fusion layer, a third fully connected layer and an output layer.

[0014] In a feasible implementation, the above LSTM traffic prediction model further includes a data enhancement module;

[0015] The obtaining of the first traffic usage information of the target network within the first duration and the second traffic usage information of the target network within the second duration includes:

[0016] Obtain the original traffic usage information of the target network;

[0017] Inputting the above-mentioned original traffic usage information into the above-mentioned data enhancement module to obtain traffic enhancement data, wherein the above-mentioned data increment module is used to perform random time slicing operations, time series translation operations and noise addition operations;

[0018] Based on the traffic enhancement data, first traffic usage information within a first duration and second traffic usage information within a second duration are obtained.

[0019] In a feasible implementation, the LSTM traffic prediction model further includes a data statistical feature extraction module, and the comprehensive traffic feature information further includes ARIMA residual feature information;

[0020] The above method further includes:

[0021] The above traffic enhancement data is input into the above data statistical feature extraction module to obtain the above ARIMA residual feature information.

[0022] In a feasible implementation, the LSTM traffic prediction model further includes other modal feature extraction modules, and the comprehensive traffic feature information further includes other modal feature information, and the other modal feature information includes feature information corresponding to weather, regional hot events, and large-scale activities;

[0023] The above method further includes:

[0024] Acquire other modal data, wherein the other modal data include weather data, regional hot event data and large-scale activity data;

[0025] The above-mentioned other modal data are sent to other modal feature extraction modules to obtain the above-mentioned other modal feature information, wherein the above-mentioned other modal feature information includes feature information corresponding to weather, regional hot events and large-scale activities.

[0026] In a feasible implementation, the above-mentioned inputting of the comprehensive traffic feature information into the feature fusion and prediction module in the above-mentioned LSTM traffic prediction model to obtain traffic demand prediction information includes:

[0027] Obtaining the short-term feature weight corresponding to the above short-term traffic feature information;

[0028] Obtaining the long-term feature weight corresponding to the above long-term traffic feature information;

[0029] Obtain the residual feature weight corresponding to the above ARIMA residual feature information;

[0030] Obtain other feature weights corresponding to the above other modal features;

[0031] Each feature information in the above-mentioned traffic feature information and the weight corresponding to each feature information in the above-mentioned traffic feature information are input into the feature fusion and prediction module in the above-mentioned LSTM traffic prediction model to obtain traffic demand prediction information; each feature information in the above-mentioned traffic feature information includes: the above-mentioned short-time traffic feature information, the above-mentioned long-time traffic feature information, the above-mentioned ARIMA residual feature information and the above-mentioned other modal features; the weight corresponding to each feature information in the above-mentioned traffic feature information includes the above-mentioned short-time feature weight, the above-mentioned long-time feature weight, the above-mentioned residual feature weight and other feature weights.

[0032] In a feasible implementation manner, each of the short-time feature weights, the long-time feature weights, the residual feature weights and other feature weights includes a basic weight and a dynamic weight, the basic weight is determined based on the correlation analysis method, and the dynamic weight is obtained based on the gradient adjustment method and environmental change factors.

[0033] In a feasible implementation, the specific steps of determining the basic weight based on the correlation analysis method are:

[0034] Determine the Pearson correlation coefficient corresponding to each characteristic information in the above-mentioned traffic characteristic information according to the historical traffic demand data and the historical traffic forecast data; the Pearson correlation coefficient corresponding to each characteristic information in the above-mentioned traffic characteristic information includes: the Pearson correlation coefficient corresponding to the above-mentioned short-term traffic characteristic information, the Pearson correlation coefficient corresponding to the above-mentioned long-term traffic characteristic information, the Pearson correlation coefficient corresponding to the above-mentioned ARIMA residual characteristic information, and the Pearson correlation coefficient corresponding to the above-mentioned other modal characteristic information;

[0035] Based on the Pearson correlation coefficient corresponding to each characteristic information in the above-mentioned traffic characteristic information, the basic weight corresponding to each characteristic information in the above-mentioned traffic characteristic information is determined; the basic weight corresponding to each characteristic information in the above-mentioned traffic characteristic information includes: the basic weight corresponding to the above-mentioned short-time traffic characteristic information, the basic weight corresponding to the above-mentioned long-time traffic characteristic information, the basic weight corresponding to the above-mentioned ARIMA residual characteristic information and the basic weight corresponding to the above-mentioned other modal characteristic information.

[0036] In a feasible implementation, the specific steps of adjusting the acquired dynamic weight based on the gradient adjustment method and environmental change factors include:

[0037] Calculate in real time the gradient information of the weights corresponding to each characteristic information in the comprehensive flow characteristic information, and determine the first adjustment coefficient based on the gradient information;

[0038] Implement monitoring of external environment information, and determine a second adjustment coefficient based on the external environment information;

[0039] The dynamic weight corresponding to each feature information in the comprehensive flow feature information is determined based on the first adjustment coefficient and the second adjustment coefficient.

[0040] In the second aspect, the present application also proposes a traffic demand prediction system based on LSTM, including:

[0041] A first acquisition unit, configured to acquire first traffic usage information of a target network within a first duration and second traffic usage information of a target network within a second duration, wherein the first duration is shorter than the second duration;

[0042] A second acquisition unit is used to input the first traffic usage information into a first LSTM module in the LSTM traffic prediction model to obtain short-term traffic feature information;

[0043] A third acquisition unit is used to input the second traffic usage information into the second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information;

[0044] The fourth acquisition unit is used to input the comprehensive traffic feature information into the feature fusion and prediction module in the above-mentioned LSTM traffic prediction model to obtain traffic demand prediction information. The above-mentioned comprehensive traffic feature information includes the above-mentioned short-term traffic feature information and the above-mentioned long-term traffic feature information.

[0045] In summary, the LSTM-based traffic demand prediction method of the embodiment of the present application includes:

[0046] Obtain the first traffic usage information within the first duration of the target network and the second traffic usage information within the second duration of the target network, wherein the first duration is less than the second duration; input the first traffic usage information into the first LSTM module in the LSTM traffic prediction model to obtain short-term traffic feature information; input the second traffic usage information into the second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information; input the comprehensive traffic feature information into the feature fusion and prediction module in the LSTM traffic prediction model to obtain traffic demand prediction information, wherein the comprehensive traffic feature information includes the short-term traffic feature information and the long-term traffic feature information. The LSTM traffic prediction model in the present application scheme is more accurate in prediction due to its powerful ability in processing time series data, especially its ability to remember long-term dependencies. This is particularly important for coping with the growing and rapidly changing network traffic demand. The LSTM traffic prediction model can update its state in real time based on new data input, which means that the model can dynamically adapt to changes in network conditions instead of relying solely on historical data. This is particularly important for modern network environments, as network traffic may change rapidly due to various emergencies (such as social media activities, news events, or technical failures). The dual-module design (first LSTM module and second LSTM module) of the LSTM traffic prediction model in the present application enables the model to capture both short-term fluctuations and long-term trends. This method of combining short-term traffic feature information with long-term traffic feature information can provide a more comprehensive traffic forecast, thereby more effectively guiding resource allocation and network management. With more accurate traffic demand forecast information, data centers and cloud service providers can better plan resources, such as reasonably configuring server capacity, optimizing the deployment of virtual machines, and adjusting storage resources. This can not only improve resource utilization, but also reduce energy consumption and operating costs, thereby maximizing economic benefits. Accurate traffic demand forecast information enables network engineers to design more sophisticated load balancing strategies, such as allocating user requests to different servers based on expected traffic. This helps prevent some servers from being overloaded while other servers are underloaded, thereby improving the response speed and service quality of the entire network. The scalability and flexibility of the LSTM traffic prediction model enables it to adapt to future technology developments such as 5G, the Internet of Things (IoT), and the growth of machine learning applications, which will further increase the complexity and dynamics of network traffic.

[0047] The LSTM-based traffic demand prediction method proposed in this application and other advantages, objectives and features of this application will be partially reflected through the following description, and will also be partially understood by technical personnel in this field through research and practice of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present specification. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0049] Figure 1 A flow chart of a traffic demand prediction method based on LSTM provided in an embodiment of the present application;

[0050] Figure 2 A schematic diagram of the structure of a traffic demand prediction system based on LSTM is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0051] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0052] See also Figure 1 , which is a flow diagram of a traffic demand prediction method based on LSTM provided in an embodiment of the present application, which may specifically include:

[0053] S110. Obtain first traffic usage information within a first duration of the target network and second traffic usage information within a second duration of the target network, wherein the first duration is shorter than the second duration.

[0054] Exemplarily, first determine the first duration in the target network, which is a relatively short period of time. During this duration, the system collects network traffic usage data, i.e., first traffic usage information. This information reflects the traffic status of the network in a short period of time, which may include data transmission volume, number of connections, bandwidth usage rate, etc.

[0055] Traffic usage information is collected during a second period of time, where the second period of time is longer than the first period of time, providing a wider time window to observe long-term trends in traffic. The second traffic usage information helps reveal long-term network behavior and traffic patterns.

[0056] S120. Input the first traffic usage information into the first LSTM module in the LSTM traffic prediction model to obtain short-term traffic feature information.

[0057] Exemplarily, the first traffic usage information is input into the first LSTM module in the LSTM traffic prediction model. This module specializes in processing short-term data and outputs short-term traffic feature information by learning the changes and characteristics of short-term traffic. These short-term traffic feature information captures the dynamic changes of traffic in the short term and is crucial for understanding immediate or recent traffic behavior.

[0058] S130. Input the second traffic usage information into the second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information.

[0059] Exemplarily, the second traffic usage information is fed into the second LSTM module of the LSTM traffic prediction model. The second LSTM module is responsible for processing long-term data and extracting long-term traffic feature information from it. Long-term traffic feature information reflects the pattern and trend of network traffic over a long period of time and is the key to understanding and predicting long-term network behavior.

[0060] S140. Input the comprehensive traffic feature information into the feature fusion and prediction module in the above-mentioned LSTM traffic prediction model to obtain traffic demand prediction information. The above-mentioned comprehensive traffic feature information includes the above-mentioned short-term traffic feature information and the above-mentioned long-term traffic feature information.

[0061] Exemplarily, in this step, the short-term traffic feature information and long-term traffic feature information extracted above are input into the feature fusion and prediction module in the LSTM traffic prediction model. In this module, short-term and long-term traffic features are combined through a specific fusion strategy to form a comprehensive traffic feature view. The module uses these comprehensive features to generate the final traffic demand forecast information and predict the network traffic demand in a certain period of time in the future.

[0062] In summary, traditional time series analysis models and regression models usually rely on linear assumptions, which limits their ability to handle complex and nonlinear change patterns. The LSTM traffic prediction model in the present application scheme is more accurate due to its strong ability to process time series data, especially its ability to remember long-term dependencies. This is particularly important for coping with the growing and rapidly changing network traffic demand. The LSTM traffic prediction model can update its state in real time based on new data input, which means that the model can dynamically adapt to changes in network conditions instead of relying solely on historical data. This is particularly important for modern network environments because network traffic may change rapidly due to various emergencies (such as social media activities, news events, or technical failures). The dual-module design (first LSTM module and second LSTM module) of the LSTM traffic prediction model in the present application scheme enables the model to capture short-term fluctuations and long-term trends at the same time. This method of combining short-term traffic feature information and long-term traffic feature information can provide more comprehensive traffic predictions, thereby more effectively guiding resource allocation and network management. With more accurate traffic demand forecasting information, data centers and cloud service providers can better plan resources, such as reasonably configuring server capacity, optimizing the deployment of virtual machines, and adjusting storage resources. This not only improves resource utilization, but also reduces energy consumption and operating costs, maximizing economic benefits. Accurate traffic demand forecasting information enables network engineers to design more sophisticated load balancing strategies, such as allocating user requests to different servers based on expected traffic. This helps prevent some servers from being overloaded while others are underloaded, thereby improving the response speed and service quality of the entire network. The scalability and flexibility of the LSTM traffic forecasting model enables it to adapt to the development of future technologies, such as 5G, the Internet of Things (IoT), and the growth of machine learning applications, which will further increase the complexity and dynamics of network traffic.

[0063] In some examples, the first LSTM module includes a first input layer, a convolutional layer, a pooling layer, a first LSTM layer, and a first fully connected layer.

[0064] The second LSTM module includes a second input layer, a second LSTM layer, an attention layer and a second fully connected layer.

[0065] The feature fusion and prediction module includes a feature fusion layer, a third fully connected layer and an output layer.

[0066] The first LSTM module includes:

[0067] First input layer: This layer is responsible for receiving and processing the first traffic usage information of the target network within the first time period. This information usually includes the network traffic data in a short period of time, such as data transmission volume, number of connections, bandwidth usage, etc.

[0068] Convolutional layer: The convolutional layer is used to extract spatial features in the first traffic usage information, helping the model capture local dependencies in time series data.

[0069] Pooling layer: The pooling layer is used to reduce the dimension of the convolutional layer output, enhance the generalization ability of the model, and reduce the amount of calculation.

[0070] First LSTM layer: The first LSTM layer processes the features extracted by the convolution and pooling layers to learn long-term dependencies in the data and generate short-term traffic feature information.

[0071] First fully connected layer: This layer is used to integrate the features obtained from the first LSTM layer to form a more compact and expressive feature representation, preparing for the feature fusion and prediction module.

[0072] The second LSTM module includes:

[0073] Second input layer: receives and processes the second traffic usage information of the target network within the second time period, which usually covers the traffic data within a longer time range.

[0074] The second LSTM layer: specializes in processing long time series data, learns long-term traffic patterns and trends from the data received from the second input layer, and generates long-term traffic feature information.

[0075] Attention layer: The attention layer is used to enhance the importance of the features most relevant to the prediction target in the output of the second LSTM layer and improve the representation ability of the features.

[0076] The second fully connected layer: further integrates and abstracts the features obtained from the attention layer and prepares them for output to the feature fusion and prediction module.

[0077] The feature fusion and prediction module includes:

[0078] Feature fusion layer: The main task of this layer is to effectively fuse the short-term traffic feature information and long-term traffic feature information output by the first LSTM module and the second LSTM module. Through specific fusion strategies, such as weighted summation and splicing, a comprehensive traffic feature view is formed.

[0079] The third fully connected layer: This layer further processes the fused traffic feature information and extracts the key features for the final prediction through a series of neural network layers.

[0080] Output layer: Generates the final traffic demand forecast information based on the output of the third fully connected layer. This information predicts the network traffic demand in a certain period of time in the future, which is crucial for network resource management and planning.

[0081] This traffic demand forecasting method can accurately predict future network traffic and support network managers to make more data-driven decisions on resource allocation and network optimization.

[0082] In some examples, the above LSTM traffic prediction model also includes a data enhancement module;

[0083] The obtaining of the first traffic usage information of the target network within the first duration and the second traffic usage information of the target network within the second duration includes:

[0084] Get the raw traffic usage information of the target network.

[0085] The above-mentioned original traffic usage information is input into the above-mentioned data enhancement module to obtain traffic enhancement data, wherein the data increment module is used to perform random time slicing operations, time series translation operations, and noise addition operations.

[0086] Based on the above traffic enhancement data, first traffic usage information within a first time period is obtained, and based on the above traffic enhancement data, second traffic usage information within a second time period is obtained.

[0087] Exemplarily, raw traffic usage information is obtained from the target network, which includes unprocessed network traffic data and reflects the usage of the network under various conditions. The raw traffic usage information is input into the data enhancement module to obtain traffic enhancement data. The data enhancement module is used to perform random time slicing operations, time series translation operations, and noise addition operations to generate more diverse and rich traffic enhancement data. These operations can simulate different network conditions and enhance the robustness and generalization ability of the model.

[0088] It should be noted that the random time slicing operation is to randomly extract time periods from the original traffic usage information to generate multiple short time slices. The time series shift operation is to shift the time series data to generate traffic data in different time periods. The noise addition operation is to add noise to the original traffic data to simulate random fluctuations and abnormalities in the real network.

[0089] The traffic enhancement data generated by the data enhancement module is used to obtain the first traffic usage information within the first time period. These enhanced data provide more diverse short-term traffic features, which help the LSTM traffic prediction model to better learn the traffic change pattern in the short term. The traffic enhancement data generated by the data enhancement module is used to obtain the second traffic usage information within the second time period. These enhanced data provide more diverse long-term traffic features, which help the LSTM traffic prediction model to better learn the traffic change trend in the long term.

[0090] The method proposed in the embodiment of the present application utilizes the traffic enhancement data generated by the data enhancement module, combined with the first LSTM module and the second LSTM module in the LSTM traffic prediction model, to effectively improve the accuracy and robustness of the traffic demand prediction.

[0091] In some examples, the LSTM traffic prediction model also includes a data statistical feature extraction module, and the traffic feature information also includes ARIMA residual feature information.

[0092] The above method further includes:

[0093] The above traffic enhancement data is input into the above data statistical feature extraction module to obtain the above ARIMA residual feature information.

[0094] Exemplarily, the traffic enhancement data is input to the data statistical feature extraction module, which includes an input layer, an ARIMAEN layer, a residual extraction layer, and a feature conversion layer. The input layer is responsible for receiving the traffic data after data enhancement. The ARIMA layer is responsible for applying the ARIMA model to the input traffic enhancement data and selecting appropriate model parameters (such as p, d, q), which represent the number of autoregressive terms, difference orders, and moving average terms, respectively. The residual extraction layer is responsible for extracting the residuals between the ARIMA prediction and the actual data. These residuals represent the part of the data that is not explained by the ARIMA model, which is very important for capturing emergencies and abnormal patterns. The feature conversion layer converts the residuals into features that can be used by the LSTM model. The feature conversion layer includes encoding, normalizing, or further statistical analysis of the residual data to generate ARIMA residual feature information. The output layer outputs the processed ARIMA residual feature information to the main body of the LSTM model. These features will be used together with other features in the LSTM model (such as short-term and long-term traffic features) for the final traffic prediction.

[0095] The method proposed in the embodiment of the present application utilizes the traffic enhancement data generated by the data enhancement module, combined with the ARIMA residual feature information generated by the data statistical feature extraction module, and the first LSTM module and the second LSTM module in the LSTM traffic prediction model, to effectively improve the accuracy and robustness of the traffic demand prediction.

[0096] In some examples, the LSTM traffic prediction model also includes other modal feature extraction modules, and the comprehensive traffic feature information also includes other modal feature information, and the other modal feature information includes feature information corresponding to weather, regional hot events and large-scale activities.

[0097] The above method further includes:

[0098] Acquire other modal data, wherein the other modal data include weather data, regional hot event data and large-scale event data.

[0099] The above-mentioned other modal data are sent to other modal feature extraction modules to obtain the above-mentioned other modal feature information, wherein the above-mentioned other modal feature information includes feature information corresponding to weather, regional hot events and large-scale activities.

[0100] Exemplarily, other modal data are first obtained, which are not limited to the network traffic itself, but include external factors that may affect the network traffic. Specifically, the above other modal data include weather data, regional hot event data and large-scale event data. These data types are selected because they often have a significant direct or indirect impact on network usage patterns. For example:

[0101] Weather data: Can influence people’s online behavior, such as increasing online video streaming consumption during rainy days or extreme weather conditions.

[0102] Regional hot event data: Local holidays or political events may affect network traffic in a specific area.

[0103] Large event data: These events, such as sporting events or music festivals, can cause spikes in traffic, especially on sites that broadcast these events live.

[0104] Once the above other modal data are acquired, they are then fed into the other modal feature extraction module, which is tasked with extracting useful features from these non-traditional data sources, which are then used to enhance the predictive power of the traffic prediction model.

[0105] The main functions of other modal feature extraction modules include:

[0106] Data preprocessing: standardize and clean the data, such as handling missing values, standardizing the format, etc.

[0107] Feature extraction: Extract meaningful features from raw data, such as temperature, humidity, and wind speed from weather data; event type, duration, impact range, etc. from regional hot event data; event scale, number of participants, activity category, etc. from large-scale event data.

[0108] Feature transformation: Apply statistical methods or machine learning algorithms to transform or synthesize new features so that these features can better express the potential impact of data on network traffic.

[0109] Once features are extracted from these multimodal data, these so-called other modal feature information is integrated into the traffic prediction model. In this way, the model not only relies on historical traffic data, but also utilizes these external data to provide more comprehensive and accurate traffic predictions.

[0110] In this implementation, the proposed LSTM traffic forecasting model can not only process traditional time series data, but also utilize data from multiple sources to provide a multi-dimensional forecasting framework. This makes the model more accurate in predicting network traffic demand, especially in the face of traffic fluctuations caused by large-scale or unusual events.

[0111] In some examples, the above-mentioned comprehensive traffic feature information is input into the feature fusion and prediction module in the above-mentioned LSTM traffic prediction model to obtain traffic demand prediction information, including:

[0112] Obtain the short-term feature weight corresponding to the above short-term traffic feature information.

[0113] Obtain the long-term feature weight corresponding to the above long-term traffic feature information.

[0114] Get the residual feature weights corresponding to the above ARIMA residual feature information.

[0115] Get other feature weights corresponding to the above other modal features.

[0116] Each feature information in the above-mentioned traffic feature information and the weight corresponding to each feature information in the above-mentioned traffic feature information are input into the feature fusion and prediction module in the above-mentioned LSTM traffic prediction model to obtain traffic demand prediction information; each feature information in the above-mentioned traffic feature information includes: the above-mentioned short-time traffic feature information, the above-mentioned long-time traffic feature information, the above-mentioned ARIMA residual feature information and the above-mentioned other modal features; the weight corresponding to each feature information in the above-mentioned traffic feature information includes the above-mentioned short-time feature weight, the above-mentioned long-time feature weight, the above-mentioned residual feature weight and other feature weights.

[0117] Exemplarily, short-term traffic feature information is provided by the first LSTM module, reflecting the traffic dynamics of the network in a shorter time scale. Long-term traffic feature information is provided by the second LSTM module, expressing traffic trends and patterns in a longer time scale. ARIMA residual feature information is provided by the data statistical feature extraction module, capturing fluctuations that cannot be explained by the classical time series prediction model. Other modal feature information is provided by the other modal feature extraction modules, including relevant features of weather data, regional hot event data, and large-scale event data.

[0118] The feature fusion layer combines different traffic feature information and their corresponding weights based on the weighted average fusion method to generate the final traffic feature information, which can be calculated by the following formula:

[0119] F final =w s ×F s +w l ×F l +w a ×F a +w o ×F o

[0120] in:

[0121] F final It is the final traffic characteristic information.

[0122] w s ,w l, w a ,w o They are short-term feature weights, long-term feature weights, residual feature weights, and other feature weights.

[0123] F s ,F l ,F a ,F o They are short-term traffic characteristic information, long-term traffic characteristic information, ARIMA residual characteristic information and other modal characteristic information.

[0124] The third fully connected layer further extracts and transforms the fused final traffic feature information, so as to integrate and optimize the comprehensive traffic feature information in preparation for the final prediction.

[0125] The output layer generates the final traffic demand forecast result, outputs the final traffic demand forecast, and provides future traffic demand information.

[0126] In some examples, each of the short-time feature weights, the long-time feature weights, the residual feature weights and other feature weights includes a basic weight and a dynamic weight, the basic weight is determined based on the correlation analysis method, and the dynamic weight is obtained based on the gradient adjustment method and environmental change factors.

[0127] Exemplarily, each feature weight consists of two parts: a basic weight and a dynamic weight.

[0128] The base weights are determined by correlation analysis based on the statistical correlation between each feature and the predicted target. The base weights reflect the inherent importance of each feature without the influence of external changes.

[0129] Dynamic weights are obtained through gradient adjustment and environmental change factor adjustment. Dynamic weights enable the model to adapt to the latest data and external changes, such as weather changes, social events, etc.

[0130] Statistical methods (such as Pearson correlation coefficient) are used to evaluate the strength of the relationship between each feature and the target variable (such as traffic). Based on these analysis results, an initial basic weight is set for each feature, which reflects the impact of the feature on the predicted target under static conditions.

[0131] During model training, weights are adjusted based on the magnitude of the gradients calculated during backpropagation. The magnitude of the gradient indicates the usefulness of the feature weight in reducing prediction error; features with larger gradients should have their weights increased more. Weights are adjusted based on changes in external environmental factors. For example, if the weather forecast indicates that a large storm is approaching, which may affect network usage patterns, the weights of the corresponding weather-related features will be increased.

[0132] Combine the base weights with the dynamic weights to form the final weight for each feature. This can be achieved using a weighted average approach, for example:

[0133] Final Weight=α×Base Weight+(1-α)×Dynamic Weight

[0134] Among them, α is an adjustment parameter, Final Weight is the final weight, Base Weight is the basic weight, and Dynamic Weight is the dynamic weight.

[0135] In some examples, the specific steps of determining the basic weight based on the correlation analysis method are:

[0136] According to the historical traffic demand data and the historical traffic forecast data, the Pearson correlation coefficient corresponding to each characteristic information in the above-mentioned traffic characteristic information is determined; the above-mentioned Pearson correlation coefficient corresponding to each characteristic information in the above-mentioned traffic characteristic information includes: the Pearson correlation coefficient corresponding to the above-mentioned short-time traffic characteristic information, the Pearson correlation coefficient corresponding to the above-mentioned long-time traffic characteristic information, the Pearson correlation coefficient corresponding to the above-mentioned ARIMA residual characteristic information and the Pearson correlation coefficient corresponding to the above-mentioned other modal characteristic information.

[0137] Based on the Pearson correlation coefficient corresponding to each characteristic information in the above-mentioned traffic characteristic information, the basic weight corresponding to each characteristic information in the above-mentioned traffic characteristic information is determined; the basic weight corresponding to each characteristic information in the above-mentioned traffic characteristic information includes: the basic weight corresponding to the above-mentioned short-time traffic characteristic information, the basic weight corresponding to the above-mentioned long-time traffic characteristic information, the basic weight corresponding to the above-mentioned ARIMA residual characteristic information and the basic weight corresponding to the above-mentioned other modal characteristic information.

[0138] For example, historical traffic demand data includes the network traffic data actually observed in the past, recording the actual usage of the network in the past period of time. Historical traffic forecast data includes the traffic data predicted by the traffic forecast model in the past, reflecting the accuracy of historical forecasts and the performance of the forecast model.

[0139] The traffic characteristic information includes short-term traffic characteristic information, long-term traffic characteristic information, ARIMA residual characteristic information and other modal characteristic information. The Pearson correlation coefficient is used to measure the correlation between the above short-term traffic characteristic information, long-term traffic characteristic information, ARIMA residual characteristic information and other modal characteristic information and the historical traffic demand data.

[0140] The corresponding basic weights are set according to the values ​​of each Pearson correlation coefficient. The higher the correlation coefficient, the stronger the correlation between the feature and the historical traffic demand data, so the basic weight assigned to the feature should also be higher. These basic weights reflect the inherent importance of each feature to predicting traffic demand in a stable environment without external interference.

[0141] In this way, the traffic forecasting model can not only effectively use the information learned from historical data to optimize the forecasting performance, but also ensure that the weights of each feature reflect their actual relevance to traffic demand, thereby improving the accuracy and reliability of the forecast. This data-based weight determination method provides the model with a dynamic adjustment capability, making it more flexible and adaptable to different environmental conditions.

[0142] In some examples, specific steps of adjusting the acquired dynamic weight based on the gradient adjustment method and environmental change factors include:

[0143] The gradient information of the weight corresponding to each characteristic information in the above-mentioned comprehensive flow characteristic information is calculated in real time, and the first adjustment coefficient is determined based on the above-mentioned gradient information.

[0144] The external environment information is monitored and a second adjustment coefficient is determined based on the external environment information.

[0145] The dynamic weight corresponding to each feature information in the comprehensive flow feature information is determined based on the first adjustment coefficient and the second adjustment coefficient.

[0146] Exemplarily, during the model training process, the gradient information of each feature weight is calculated in real time. The gradient information indicates the degree of influence of each feature weight on the prediction error under the current model parameters, that is, how the prediction error will change if this weight is changed. Gradient information is used to evaluate the importance of features and their influence on the prediction target. A high gradient value means that a small change in the corresponding feature may lead to a significant change in the prediction result, which indicates that the feature is of high importance in the current model.

[0147] The first adjustment factor is determined based on the gradient information. This adjustment factor is used to quantify how the feature weight should be adjusted according to its gradient to optimize model performance. The first adjustment factor is usually a function of the gradient, which may include amplifying or reducing the gradient, as well as adjusting the gradient direction to more effectively reduce the prediction error.

[0148] Continuously monitor external environmental information that affects network traffic, such as weather conditions, major social events, holidays, etc. This information has a significant impact on traffic patterns, especially in extreme events or unusual situations. Changes in the external environment can significantly affect the predictive importance of certain features. For example, a large event may cause a surge in traffic in a short period of time, and features related to the event become more important.

[0149] The second adjustment coefficient is determined based on external environment information. This adjustment coefficient reflects how the weights of each feature should be adjusted to adapt to changes in the external environment. The second adjustment coefficient enables the model to dynamically respond to changes in the external environment and adjust feature weights to better capture the impact of these changes on traffic.

[0150] The dynamic weight is calculated by combining the first adjustment coefficient and the second adjustment coefficient. The dynamic weight reflects the comprehensive results of real-time data analysis and external environment monitoring, ensuring that the model weight can not only reflect the current statistical importance of the feature, but also adapt to environmental changes.

[0151] In the second aspect, the present application proposes a traffic demand prediction system based on LSTM, including:

[0152] A first acquisition unit 21 is used to acquire first traffic usage information within a first duration of a target network and second traffic usage information within a second duration of the target network, wherein the first duration is less than the second duration;

[0153] A second acquisition unit 22, configured to input the first traffic usage information into a first LSTM module in an LSTM traffic prediction model to obtain short-term traffic feature information;

[0154] A third acquisition unit 23 is used to input the second traffic usage information into the second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information;

[0155] The fourth acquisition unit 24 is used to input the comprehensive traffic feature information into the feature fusion and prediction module in the LSTM traffic prediction model to obtain traffic demand prediction information, wherein the comprehensive traffic feature information includes the short-term traffic feature information and the long-term traffic feature information.

[0156] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0157] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0158] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0159] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0160] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0161] An embodiment of the present application also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device executes the process of predicting the decomposition rate in alumina production in the corresponding embodiment.

[0162] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or a data center that includes one or more available media integration. The available medium can be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state drive (SSD)), etc.

[0163] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0164] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0165] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0166] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0167] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0168] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A traffic demand prediction method based on LSTM, comprising: Acquire first traffic usage information within a first duration of the target network and second traffic usage information within a second duration of the target network, wherein the first duration is less than the second duration; Inputting the first traffic usage information into a first LSTM module in an LSTM traffic prediction model to obtain short-term traffic feature information; Inputting the second traffic usage information into the second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information; Inputting the comprehensive traffic feature information into the feature fusion and prediction module in the LSTM traffic prediction model to obtain traffic demand prediction information, wherein the comprehensive traffic feature information includes the short-term traffic feature information and the long-term traffic feature information; The first LSTM module includes a first input layer, a convolution layer, a pooling layer, a first LSTM layer and a first fully connected layer; The second LSTM module includes a second input layer, a second LSTM layer, an attention layer, and a second fully connected layer; The feature fusion and prediction module includes a feature fusion layer, a third fully connected layer and an output layer. The feature fusion layer is used to fuse the short-term traffic feature information and the long-term traffic feature information output by the first LSTM module and the second LSTM module to form a comprehensive traffic feature view.

2. The LSTM-based traffic demand prediction method according to claim 1, characterized in that: The LSTM traffic prediction model also includes a data enhancement module; The obtaining of first traffic usage information of a target network within a first duration and second traffic usage information of a target network within a second duration includes: Obtain the original traffic usage information of the target network; Inputting the original traffic usage information into the data enhancement module to obtain traffic enhancement data, wherein the data enhancement module is used to perform a random time slicing operation, a time series shifting operation, and a noise adding operation; The first traffic usage information within the first duration and the second traffic usage information within the second duration are obtained based on the traffic enhancement data.

3. The LSTM-based traffic demand prediction method according to claim 2 is characterized in that: The LSTM traffic prediction model also includes a data statistical feature extraction module, and the comprehensive traffic feature information also includes ARIMA residual feature information; The method further comprises: The traffic enhancement data is input into the data statistical feature extraction module to obtain the ARIMA residual feature information.

4. The LSTM-based traffic demand prediction method according to claim 3 is characterized in that: The LSTM traffic prediction model also includes other modal feature extraction modules, and the comprehensive traffic feature information also includes other modal feature information, and the other modal feature information includes feature information corresponding to weather, regional hot events and large-scale activities; The method further comprises: Acquire other modal data, wherein the other modal data includes weather data, regional hot event data, and large-scale activity data; The other modal data is sent to other modal feature extraction module to obtain the other modal feature information.

5. The LSTM-based traffic demand prediction method according to claim 4 is characterized in that: The step of inputting the comprehensive traffic feature information into the feature fusion and prediction module in the LSTM traffic prediction model to obtain traffic demand prediction information includes: Obtaining a short-term feature weight corresponding to the short-term traffic feature information; Obtaining a long-term feature weight corresponding to the long-term traffic feature information; Obtaining the residual feature weight corresponding to the ARIMA residual feature information; Obtaining other feature weights corresponding to the other modal features; Each feature information in the traffic feature information and the weight corresponding to each feature information in the traffic feature information are input into the feature fusion and prediction module in the LSTM traffic prediction model to obtain traffic demand prediction information; each feature information in the traffic feature information includes: the short-time traffic feature information, the long-time traffic feature information, the ARIMA residual feature information and the other modal features; the weight corresponding to each feature information in the traffic feature information includes the short-time feature weight, the long-time feature weight, the residual feature weight and other feature weights.

6. The LSTM-based traffic demand prediction method according to claim 5 is characterized in that: Each of the short-time feature weights, the long-time feature weights, the residual feature weights and other feature weights includes a basic weight and a dynamic weight, wherein the basic weight is determined based on a correlation analysis method, and the dynamic weight is obtained based on a gradient adjustment method and environmental change factors.

7. The LSTM-based traffic demand prediction method according to claim 6, characterized in that: The specific steps for determining the basic weights based on the correlation analysis method include: Determine the Pearson correlation coefficient corresponding to each characteristic information in the traffic characteristic information according to the historical traffic demand data and the historical traffic forecast data; the Pearson correlation coefficient corresponding to each characteristic information in the traffic characteristic information includes: the Pearson correlation coefficient corresponding to the short-term traffic characteristic information, the Pearson correlation coefficient corresponding to the long-term traffic characteristic information, the Pearson correlation coefficient corresponding to the ARIMA residual characteristic information, and the Pearson correlation coefficient corresponding to the other modal characteristic information; Based on the Pearson correlation coefficient corresponding to each characteristic information in the traffic characteristic information, the basic weight corresponding to each characteristic information in the traffic characteristic information is determined; the basic weight corresponding to each characteristic information in the traffic characteristic information includes: the basic weight corresponding to the short-time traffic characteristic information, the basic weight corresponding to the long-time traffic characteristic information, the basic weight corresponding to the ARIMA residual characteristic information and the basic weight corresponding to the other modal characteristic information.

8. The LSTM-based traffic demand prediction method according to claim 6, characterized in that: The specific steps of adjusting the dynamic weights obtained based on the gradient adjustment method and environmental change factors include: Calculate in real time the gradient information of the weights corresponding to each characteristic information in the comprehensive flow characteristic information, and determine the first adjustment coefficient based on the gradient information; Implement monitoring of external environment information, and determine a second adjustment coefficient based on the external environment information; The dynamic weight corresponding to each feature information in the comprehensive flow feature information is determined based on the first adjustment coefficient and the second adjustment coefficient.

9. A traffic demand prediction system based on LSTM, characterized in that: include: A first acquisition unit, configured to acquire first traffic usage information of a target network within a first duration and second traffic usage information of a target network within a second duration, wherein the first duration is shorter than the second duration; A second acquisition unit, used for inputting the first traffic usage information into a first LSTM module in the LSTM traffic prediction model to obtain short-term traffic feature information; A third acquisition unit, used for inputting the second traffic usage information into a second LSTM module in the LSTM traffic prediction model to obtain long-term traffic feature information; A fourth acquisition unit is used to input the comprehensive traffic feature information into the feature fusion and prediction module in the LSTM traffic prediction model to obtain traffic demand prediction information, wherein the comprehensive traffic feature information includes the short-term traffic feature information and the long-term traffic feature information; The first LSTM module includes a first input layer, a convolution layer, a pooling layer, a first LSTM layer and a first fully connected layer; The second LSTM module includes a second input layer, a second LSTM layer, an attention layer, and a second fully connected layer; The feature fusion and prediction module includes a feature fusion layer, a third fully connected layer and an output layer. The feature fusion layer is used to fuse the short-term traffic feature information and the long-term traffic feature information output by the first LSTM module and the second LSTM module to form a comprehensive traffic feature view.

Images