CN117994563A - Intelligent 5G slice planning method and system based on deep learning - Google Patents
Intelligent 5G slice planning method and system based on deep learning Download PDFInfo
- Publication number
- CN117994563A CN117994563A CN202311823232.2A CN202311823232A CN117994563A CN 117994563 A CN117994563 A CN 117994563A CN 202311823232 A CN202311823232 A CN 202311823232A CN 117994563 A CN117994563 A CN 117994563A
- Authority
- CN
- China
- Prior art keywords
- data
- slice
- packet
- neural network
- intelligent
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013439 planning Methods 0.000 title claims abstract description 27
- 238000013135 deep learning Methods 0.000 title claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 61
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 230000006870 function Effects 0.000 claims abstract description 19
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims description 20
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000004891 communication Methods 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 8
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 238000013075 data extraction Methods 0.000 claims description 3
- 238000013136 deep learning model Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000003062 neural network model Methods 0.000 claims description 2
- 238000007726 management method Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000004044 response Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 102100026278 Cysteine sulfinic acid decarboxylase Human genes 0.000 description 3
- 238000003491 array Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 108010064775 protein C activator peptide Proteins 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013468 resource allocation Methods 0.000 description 2
- 238000010008 shearing Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- NGBFQHCMQULJNZ-UHFFFAOYSA-N Torsemide Chemical compound CC(C)NC(=O)NS(=O)(=O)C1=CN=CC=C1NC1=CC=CC(C)=C1 NGBFQHCMQULJNZ-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013079 data visualisation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 229960005461 torasemide Drugs 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000007794 visualization technique Methods 0.000 description 1
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention belongs to the technical field of flow identification, and discloses an intelligent 5G slice planning method and system based on deep learning, wherein the method comprises the following steps of; step 1: converting the effective load information in the Pcap data packet into a visualized image, and preprocessing the image; step 2: training a CNN neural network, optimizing the CNN neural network, and calculating a loss function to measure the performance of the model; step 3: based on the electric service identification result, the slicing parameters are intelligently generated, the 5G slicing module is assisted to intelligently bind the network slices, and network resources are distributed according to needs. The invention can meet the real-time requirement, realize the high-precision identification of the electric service identification method and the system under the limited computing resource, and ensure the stable and efficient operation of the 5G intelligent power grid.
Description
Technical Field
The invention belongs to the technical field of flow identification, and particularly relates to an intelligent 5G slice planning method based on deep learning.
Background
In a 5G smart grid, electric service identification is a key link for ensuring efficient, stable and safe operation of the electric network. As the complexity of power systems increases, accurate identification and quick response to various electric services becomes particularly important. Although various electric service identification methods have been developed at present, most of them are used for the two-class detection of abnormal traffic;
The patent application with publication number CN103532776a identifies intrusion traffic based on data traffic; the patent application with publication number CN105184486a is based on SVM vector machine for flow identification.
However, in practical operation, there are still significant challenges and limitations due to the constraints of various factors such as model cost, computing resources, and real-time requirements. For example, high-precision business identification models require a large amount of computing resources, which is impractical in certain edge computing scenarios.
At present, 5G slice research facing the power grid is not mature. Current technical means are not able to meet the ever-increasing new service access requirements. In order to better serve modern power systems, there is a need to develop more efficient and flexible 5G slice planning and management approaches.
In analyzing challenges and limitations faced by the prior art (e.g., CN103532776a and CN105184486 a), one can do this from several points:
1. Model cost and computational resources: high-precision traffic recognition models, such as SVM-based traffic recognition models, typically require a significant amount of computational resources. This is impractical for environments with limited computing power, such as certain edge computing scenarios. This limits the wide deployment and application of models.
2. Real-time requirements: in key infrastructures such as power grids, the real-time requirement of 5G slicing is extremely high. The current technology cannot achieve real-time performance when processing large amounts of data, thereby affecting a quick response to an emergency.
3. Flexibility and adaptability: with the continuous access of new services, the 5G network needs to have high flexibility and adaptability to cope with the changing demands. The prior art has limitations in quickly adapting to new traffic types and traffic patterns.
4. Balance of accuracy and complexity: while high accuracy business identification models can provide better identification accuracy, this is often accompanied by higher model complexity. In resource constrained environments, how to balance accuracy and model complexity is an important challenge.
5. Energy consumption problem: in certain application scenarios (e.g., edge computing devices), energy efficiency is an important consideration. High complexity models result in excessive power consumption, which is detrimental to applications in these environments.
6. Security and privacy issues: in processing sensitive grid data, it is necessary to ensure the security of the data and the privacy of the user. The prior art needs to be further enhanced in terms of ensuring data security and privacy protection.
In summary, although the prior art has made some progress in the study of 5G slices, there are significant challenges and limitations in terms of model cost, computational resources, real-time requirements, flexibility and adaptability, energy consumption, and safety, and further research and improvement are needed.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide an intelligent 5G slice planning method based on deep learning, which not only can meet the real-time requirement, but also can realize high-precision identification of electric service identification under limited computing resources, and ensures the stable and efficient operation of a 5G intelligent power grid.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
the intelligent 5G slice planning method based on deep learning comprises the following steps of;
Step 1: converting the effective load information (namely PACKET DATA) in the Pcap data packet into a visualized image, and preprocessing the image;
Step 2: training a CNN neural network, optimizing the CNN neural network, and calculating a loss function to measure the performance of the model; extracting key features of the preprocessed image by using a CNN neural network, identifying power grid business according to the extracted key features, binding a 5G slice template, and distributing network resources according to needs;
Step 3: based on the electric service identification result, the slicing parameters are intelligently generated, the 5G slicing module is assisted to intelligently bind the network slices, and network resources are distributed according to needs.
The step1 specifically comprises the following steps:
Step 1.1: payload information extraction
Analyzing the Pcap data packet and extracting effective load information by adopting python; extracting key information according to the structural characteristics of a Pcap data packet, wherein the Pcap data packet consists of a Pcap (Packet Capture) packet header, a packet header (PACKET HEADER) of the data packet and data packet content (PACKET DATA); determining the position and length of the data packet content by reading PACKET HEADER, thereby obtaining a number PACKET DATA; i.e. payload information;
Step 1.2: converting payload information into pictures
After extracting all PACKETDATA data from the Pcap data packet, converting the data content into a two-dimensional image with a specific size, namely a jpg format picture;
Step 1.3: data enhancement
And carrying out data enhancement on the two-dimensional image. Namely, rotating, reversing, zooming, translating and shearing the picture in the jpg format, and adding random noise into the picture in the jpg format; the method comprises the steps of importing needed libraries and modules, opening JPG image files, converting images into pixel arrays, traversing each pixel of the images, adding random noise to RGB channels of each pixel, limiting pixel values to 0-255, and saving the images with added noise as new JPG files so as to increase the robustness of data.
In the step 1.1, according to the characteristics of the Pcap data packet, extracting [36:40] bytes from the Pcap data packet directly to obtain PACKET DATA, and after data extraction, according to a predetermined image size (nxn), processing the extracted data to obtain a two-dimensional matrix meeting the requirements;
Defining the content of the data packet, namely the effective load information; firstly, judging whether the length of the data packet content is greater than n2, if the length of the data packet content is less than n2, indicating that the data is insufficient to fill the whole image, and therefore, performing zero filling operation; if the length of the data packet content is greater than n2, then the data packet content needs to be intercepted to ensure that the size of the two-dimensional matrix is nxn, and the most relevant information is reserved by intercepting the previous n2 data and remodelled into the two-dimensional matrix of nxn.
The step 1.2 specifically comprises the following steps:
Firstly, after a two-dimensional matrix containing Pcap Data is needed, recombining the two-dimensional matrix into a two-dimensional matrix according to a predefined size; if the extracted data length is smaller than the predefined size, the image is resized by means of zero padding to match the predefined size, which is then saved as a jpg formatted picture using the OpenCv library.
The step 2 specifically comprises the following steps:
step 2.1: light convolutional neural network
Adopting a convolutional neural network to identify power grid business, and realizing a light CNN model under the condition of ensuring the identification precision; the convolutional neural network comprises 3 convolutional layers, 2 pooling layers and 2 full connection layers;
Inputting the obtained image into a lightweight convolutional neural network, and classifying by using softmax;
The CNN model firstly processes input data through a convolution layer to generate a feature map with the size of 30 multiplied by 32, and then a pooling layer reduces the size of the feature map to 15 multiplied by 32, and then the data flows through a second convolution layer to obtain a feature map with the size of 13 multiplied by 64; after the pooling again, the feature map size is reduced to 6×6×64; the final convolution results in a 4 x 64 feature map; finally, the feature is mapped and flattened into a one-dimensional array, and then a classification result is finally obtained through two full-connection operations;
And inputting the jpg format pictures subjected to data processing into a CNN model, and finally classifying by using softmax, thereby realizing a simpler network model on the premise of ensuring the accuracy.
Step 2.2.: adopting an Adam optimizer in tensorflowkeras to optimize parameters of the CNN model so as to obtain a better training effect; and adopts Cross entropy loss function (Cross-EntropyLoss) and Accuracy (Accuracy) as performance indexes;
a model () function is used to compile the model, specify the Adam optimizer, select the Cross entropy loss function (Cross-EntropyLoss), and evaluate the index accuracy.
Further, training a model using a model. Fit () method, importing training data, labels, and training rounds (epochs) and batch sizes (batchsize), and after training is completed, evaluating performance of the model using a model. Evaluation () method, importing test data and test labels;
The formula of the cross entropy function is as follows:
Wherein y i is the output value of sample i, p i is the probability that sample i is correct, J is the value of the loss function, and N is the number of results;
the accuracy formula is as follows:
accuracy = (tp+tn)/(tp+tn+fp+fn)
Here, TP (TruePositives) is the number of samples correctly predicted as positive class, TN (TrueNegatives) is the number of samples correctly predicted as negative class, FP (FalsePositives) is the number of samples incorrectly predicted as positive class (actually negative class), and FN (FalseNegatives) is the number of samples incorrectly predicted as negative class (actually positive class).
The step 3 specifically comprises the following steps of;
Step 3.1: intelligent slice parameter generation
The decision maker needs to determine which 5G traffic scenario the current classification result belongs to, the 5G traffic scenario being eMBB (enhanced mobile broadband), uRLLC (ultra-high reliability and low latency communication) and mMTC (large-scale machine type communication); the decision maker will automatically select or recommend the corresponding 5G network slice template to generate the corresponding 5G slice type id.
The purpose of the decision maker is to decide the network slice template that a particular power traffic class should be assigned. Considering that the 5G technology is to meet various service demands, three major service scenarios, that is eMBB (enhanced mobile broadband), uRLLC (ultra-high reliability and low latency communication) and mMTC (large-scale machine type communication) are specifically defined. These scenarios cover various demands from high-speed data transmission to low-latency, high-reliability communications.
The decision-making device sequentially judges whether the service is an eMBB, uRLLC, mMTc or other type of service; then, based on the characteristics and the requirements of the service scene, the decision maker can automatically bind the corresponding 5G network slice template. For example, if the power traffic demand is real-time reactive and highly reliable, the decision maker will select uRLLC as the best match and generate uRLLC network slice type id, which is passed to the slice binding module for slice binding.
The invention provides an intelligent 5G network slice planning system based on deep learning, which comprises the following modules:
1. A data preprocessing module configured to perform the following operations:
A payload information extracting unit, configured to extract payload information from the Pcap packet, where the extracting is based on the packet structure feature, and read the packet header of the packet from the Pcap packet header, the packet header of the packet, and the packet content to obtain the payload information;
an image conversion unit for converting the extracted payload information into a two-dimensional image of a set size;
and the data enhancement unit is used for performing data enhancement operation on the two-dimensional image obtained through conversion so as to improve the generalization capability and the robustness of the model.
2. A deep learning model training and optimization module configured to:
a Convolutional Neural Network (CNN) construction unit responsible for constructing a CNN neural network for processing the two-dimensional image;
the training unit is used for training the CNN neural network and learning by utilizing the preprocessed image data;
The optimizing unit is used for optimizing parameters of the CNN neural network and measuring and improving the performance of the model by calculating a loss function;
And the feature extraction unit is used for extracting key features from the trained CNN neural network so as to realize accurate identification of power grid business.
A 3.5G slice parameter generation and binding module configured to:
The recognition result processing unit is used for receiving the power grid service recognition result output by the CNN neural network;
the slice parameter generating unit intelligently generates corresponding 5G slice parameters according to the identification result;
and the slice binding unit is used for applying the generated slice parameters to the 5G slice module to realize intelligent binding of the power grid service and the network slice so as to achieve the on-demand distribution of network resources.
The system realizes full-flow automation from data packet analysis to intelligent planning of the 5G network slice through integrating data preprocessing, deep learning model training and optimizing and 5G slice parameter generation and binding modules, and remarkably improves the efficiency and accuracy of 5G network slice planning.
In combination with the technical scheme and the technical problems to be solved, the technical scheme to be protected has the following advantages and positive effects:
First, the invention effectively aims at the high complexity of the 5G smart grid business through the deep pretreatment means such as effective load information extraction, information conversion, data enhancement and the like. Meanwhile, the depth feature extraction is carried out by adopting a light model, the comprehensive features of the power grid service are further captured, and fine division is carried out by combining a fully-connected neural network, so that the accurate identification of the power grid service is ensured.
The invention further analyzes network structure characteristics of the power communication network, different service types and different communication requirements of various services carried in the network, and provides a method suitable for identifying and classifying the power service in the power communication network, thereby realizing high-accuracy, high-speed and fine classification of the power communication network service, improving the visualization degree of the service in the power data communication network, providing an important reference basis for fine and intelligent 5G power grid slice management, and improving the flexibility and resource scheduling management and control capability of the power data communication network.
Second, the significant technical progress brought by the intelligent 5G slice planning method based on deep learning is mainly realized in the following aspects:
1) The innovative data processing mode is as follows: the method for converting the payload information of the Pcap data packet into an image is an innovative data processing mode. This transformation allows complex network data to be more intuitively analyzed and processed in the form of images, exploiting the techniques of image processing to mine deep features of the data.
2) Efficient feature extraction: the Convolutional Neural Network (CNN) is applied to image processing and feature extraction, so that key features in the power grid service can be more effectively identified and extracted. The use of CNNs improves the accuracy and efficiency of data processing, particularly when processing large-scale data.
3) Intelligent 5G slice planning: and the intelligent slice parameter generation based on the power grid service identification result provides an efficient solution for the resource allocation of the 5G network. The method can dynamically adjust network resources according to the real-time data, and realize more flexible and efficient network management.
4) On-demand allocation of network resources: through intelligent slice module binding, network resources are allocated according to needs, the utilization rate of the network resources is improved, and network congestion and delay are reduced.
5) Application of data enhancement: and the data enhancement technology is applied in the data preprocessing stage, so that a training data set can be enlarged, and the generalization capability of the model can be improved. This helps to improve the stability and accuracy of the model in practical applications.
In general, the intelligent 5G slice planning method based on deep learning not only improves the efficiency and accuracy of data processing, but also greatly improves the performance and reliability of the 5G network through intelligent network resource management.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is an overall flow chart provided by an embodiment of the present invention;
FIG. 2 is a flow chart of extracting payloads provided by an embodiment of the present invention;
FIG. 3 is a flow chart for converting a payload into a grayscale image according to an embodiment of the present invention;
FIG. 4 is a decision flow chart of a decision maker provided by an embodiment of the present invention;
Fig. 5 is a CNN model framework diagram provided by an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Example 1: video streaming service optimization in 5G networks
1) Data preparation: and collecting Pcap data packets of the video streaming service from the 5G network, and extracting effective load information.
2) Image conversion: the payload information is converted into a two-dimensional image and data enhancement is performed to expand the data set.
3) CNN training: a Convolutional Neural Network (CNN) is trained, optimized specifically for video streaming service data, to extract key features.
4) Service identification: and analyzing the image by using the trained CNN model, and identifying the characteristics of the video streaming service.
5) Slice parameter generation: and according to the identification result, intelligently generating 5G slice parameters suitable for video streaming service, such as bandwidth allocation, delay optimization and the like.
6) Network slice binding: and applying the generated slice parameters to a 5G network, and optimizing the transmission quality and speed of the video streaming service.
Example 2: emergency service quick response in 5G networks
1) Data preparation: pcap data packets in the 5G network relating to emergency services (e.g., emergency, alarm) are collected and payload information is extracted.
2) Image conversion: and converting the payload information into a two-dimensional image and performing data enhancement.
3) CNN training: the convolutional neural network is trained to be able to identify data features associated with emergency services.
4) Service identification: and analyzing the image by using the CNN model, and accurately identifying the characteristics of emergency service.
5) Slice parameter generation: based on the identification result, slice parameters for emergency services, such as priority adjustment, fast path configuration, etc., are generated.
6) Network slice binding: applying these parameters to a 5G network ensures that emergency services can get a fast and reliable network response.
The two embodiments show the practical application of the intelligent 5G slice planning method based on deep learning in different application scenes, and the efficiency and quality of 5G network service are improved through intelligent data processing and network resource management.
The invention is described in further detail below with reference to the accompanying drawings.
The 5G technology has great potential, but lacks an intelligent slicing scheme aiming at the massive business requirements of the power grid. Based on the above, the invention provides an intelligent 5G slice planning method based on deep learning, as shown in fig. 1, mainly comprising the following steps of;
Firstly, through converting a Pcap data packet of a power grid into an image format, a Convolutional Neural Network (CNN) is utilized to automatically extract key features, which are key steps of feature engineering, and the complex data are suitable for machine learning. These features are then used to identify different grid traffic types, and the 5G slice templates are automatically bound according to the electrical traffic identification results. Finally, the model performance is accurately measured by optimizing an algorithm and calculating a loss function, and finally, the classification accuracy rate of up to 98% is achieved.
According to the standard of DL/T5391-2007, the electric power service can be divided into four types of data service, voice service, video service and multimedia service according to the different service flow types.
The eight-level VPN services mainly comprise information, communication, scheduling, video, IMS, voice, network management and IPv6 test point services.
The present invention employs ISCXVPN2016 dataset, which was developed by Information Security Centre ofExcellence (ISCX) laboratory at the university of new torsemide, canada, and designed specifically for network traffic identification and analysis. ISCXVPN2016 the data set is directed to providing a variety of VPN traffic intended to simulate VPN usage scenarios common in contemporary networks. The data set contains a large number of normal and network traffic samples encrypted by VPN, covering 8 communication modes for numerous applications such as instant messaging, mail, file transfer, video streaming and P2P. After extracting the payload information and converting it into a picture, it is marked and as per 9: the scale of 1 is divided into training and test sets, while cross-validation is used during the experiment in order to avoid interference.
The overall flow is shown in fig. 1. After the electric traffic (such as the unmanned aerial vehicle inspection traffic) is collected, the collected Pcap data packet is subjected to visual processing and is input into a classifier. The classifier automatically recognizes the traffic as one of the three types eMBB, ulllc or mMTc and passes its classification result to the decision maker. According to the decision of the decision maker, the corresponding 5G slice template is bound to the service, so that the optimal management and resource allocation of the service are realized.
The invention selects the lightweight CNN as a basic model. Before inputting the model, the invention firstly carries out preprocessing steps such as effective load information extraction, information conversion and the like on the Pcap data packet, processes the data into pictures with the size of 32 multiplied by 32, then carries out data enhancement on the pictures, and inputs the pictures into the CNN model.
Step 1: and the input data is subjected to image visualization processing, most of service data types are Pcap data packets because of complex and diversified power grid services, and key features are automatically extracted by utilizing CNN by visualizing effective load information in the Pcap data packets as images, so that the processing efficiency and accuracy of the power grid service data are improved.
The overall flow of data visualization is shown in fig. 2. First, data part data in each packet is extracted according to the structure characteristics of the Pcap data packet. And then, converting the extracted data into a regular two-dimensional matrix according to the size of the output picture, and performing operations such as rotation, overturning, deformation and the like on the two-dimensional matrix after obtaining the two-dimensional matrix. And finally, storing the processed two-dimensional matrix into jepg picture format.
The method comprises the following specific steps:
Step 1.1: payload information extraction
Python is used to parse the Pcap packet and extract the key information. The invention extracts key information according to the structural characteristics of the Pcap data packet. The Pcap packet consists of a Pcap header, a header (PACKET HEADER) of the packet, and a packet content (PACKET DATA). To extract all the data portion of the PCAP packet, the location and length of the packet contents need to be determined first. By reading PACKETHEADER. To acquire PACKET DATA, the data length of PACKET HEADER needs to be acquired first.
Specifically, since the first 36 bytes of the packet are made up of the PCAP packet header and part of the bytes of PACKET HEADER, the [36:40] bytes can be extracted directly from the packet to obtain PACKET DATA. After data extraction, the extracted data needs to be processed according to a predetermined image size (nxn) to obtain a two-dimensional matrix meeting the requirements. This process can be seen with reference to fig. 3. First, it is determined whether PACKET DATA is longer than n≡2. If PACKET DATA is less than n 2, the specification data is insufficient to fill the entire image, and therefore a zero-fill operation is required. If PACKET DATA is longer than n 2, then the data needs to be truncated to ensure that the size of the two-dimensional matrix is nxn. The most relevant information is retained by intercepting the first n 2 data and remodelling it into a two-dimensional matrix of nxn.
Step 1.2: converting payload information into pictures
After all the data portions are extracted from the PCAP data packet, the next step is to convert the data into an image. The goal here is to convert the data content into a two-dimensional image of a specific size, such as 100 (10 x 10), 256 (16 x 16), 625 (25 x 25), or 1024 (32 x 32) pixels. To achieve this, it is first necessary to acquire pixel values of the image data and reorganize them to a predefined size. If the extracted data length is smaller than the predefined size, the image is resized by means of zero padding to match the predefined size. This is then saved as a jpg formatted picture using the OpenCv library.
Step 1.3: data enhancement
The method comprises the steps of rotating, inverting, scaling, translating and shearing a picture in a JPG format, adding some random noise into the picture in the JPG format, importing needed libraries and modules, opening a JPG image file, converting an image into a pixel array, traversing each pixel of the image, adding the random noise to an RGB channel of each pixel, limiting the pixel value to be between 0 and 255, and saving the image with added noise as a new JPG file so as to increase the robustness of data;
step 2: the identification of the power grid service is realized:
step 2.1: light convolutional neural network
The invention adopts the simplified CNN convolutional neural network to identify the power grid business, and realizes a light CNN model under the condition of ensuring the identification precision.
The CNN model framework is shown in fig. 5, which consists of 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. The model first processes the input data through a convolutional layer to produce a feature map of 30 x 32 in size, with 320 parameters participating in the calculation of the layer. Next, a pooling layer halves the feature map size to 15 x 32. The data then flows through a second convolution layer, resulting in a 13 x 64 feature map, which has 18496 parameters. After the next pooling, the feature map size is reduced to 6×6×64. The last convolution yields a4 x 64 feature map containing 36928 parameters. The next operation of the model is to flatten the feature map into a 1024-dimensional one-dimensional array, then through two full-connection operations, firstly reduce to 64 dimensions and then reduce to 3 dimensions, and finally obtain the classification result.
According to the invention, the pictures subjected to data processing are input into the CNN model, and finally classified by using softmax, so that a simpler network model is realized on the premise of ensuring the accuracy.
Step 2.2: adopting an Adam optimizer in tensorflow keras to optimize parameters of the neural network model so as to obtain a better training effect; and a Cross entropy loss function (Cross-EntropyLoss) and an Accuracy rate (Accuracy) are adopted as performance indexes.
A model () function is used to compile the model, specify the Adam optimizer, select the Cross entropy loss function (Cross-EntropyLoss), and evaluate the index accuracy.
The model is trained using model. Fit () method, incoming training data, tags, and training rounds (epochs) and lot sizes (batchsize). After training is complete, model.evaluation () method is used to evaluate the performance of the model, incoming test data and test tags.
The formula of the cross entropy function is as follows:
Where y i is the output value of sample i, p i is the probability that sample i is correct, J is the value of the loss function, and N is the number of results.
The accuracy formula is as follows:
accuracy = (tp+tn)/(tp+tn+fp+fn)
Wherein TP (TruePositives) is the number of samples correctly predicted as positive class, TN (TrueNegatives) is the number of samples correctly predicted as negative class, FP (FalsePositives) is the number of samples incorrectly predicted as positive class (actually negative class), FN (FalseNegatives) is the number of samples incorrectly predicted as negative class (actually positive class).
Step 3: based on the electric service identification result, the slicing parameters are intelligently generated, the 5G slicing module is assisted to intelligently bind the network slices, and network resources are distributed according to needs.
Step 3.1: intelligent slice parameter generation
The invention designs a simple decision maker, and the main purpose of the decision maker is to determine the network slicing templates to be allocated for specific power service classes. Considering that the 5G technology is to meet various service demands, three major service scenarios, that is eMBB (enhanced mobile broadband), uRLLC (ultra-high reliability and low latency communication) and mMTC (large-scale machine type communication) are specifically defined. These scenarios cover various demands from high-speed data transmission to low-latency, high-reliability communications.
The whole flow is shown in fig. 4, and the decision device decides whether the service is an eMBB, a ul lc, mMTc or other type service according to the above; then, based on the characteristics and the requirements of the service scene, the decision maker can automatically bind the corresponding 5G network slice template. For example, if the power traffic demand is real-time reactive and highly reliable, the decision maker will select uRLLC as the best match and bind uRLLC the network slice template accordingly.
Classical convolution classifiers Alexnet, reanext, DNN (DenseNet 121) contrast experiments were chosen using the same open source dataset and the same data processing and visualization methods. The experimental results are shown in table 1, and it can be seen that the CNN network provided by the present application can obtain the same accuracy as the CNN network and the CNN network with smaller memory and network complexity, and the integrated accuracy can reach about 98% although the loss is not very low, but the error is within an acceptable range.
TABLE 1
| Algorithm | CNN | Alexnet | Resnet50 | DNN |
| Accuracy | 98.7％ | 98.5％％ | 97.6％ | 84.9％ |
| Loss | 0.0104 | 0.0077 | 0.0126 | 0.5726 |
| Size | 847KB | 247M | 96M | 33M |
In the application scenario of smart grids, 5G network slicing techniques are used to optimize network support for different grid services. First, data is collected from sensors and devices in the grid, and different business requirements are identified from the data content classifier, such as real-time monitoring, fault detection, or data analysis. The determiner then decides the most suitable 5G slice type based on these traffic demands and the current network state, such as selecting a low delay slice for real-time monitoring or a high speed data transmission slice for failure detection. The network then automatically configures the corresponding network resources, including bandwidth, delay, and priority parameters, based on these decisions. The method not only ensures the high-efficiency and reliable operation of the power grid service, but also can dynamically adjust the network resources according to the change of the real-time performance and the service demand, thereby improving the operation efficiency of the power grid and simultaneously providing the capability of quick response for emergency.
It should be noted that the embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or special purpose design hardware. Those of ordinary skill in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device of the present invention and its modules may be implemented by hardware circuitry, such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., as well as software executed by various types of processors, or by a combination of the above hardware circuitry and software, such as firmware.
The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the invention is not limited thereto, but any modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention will be apparent to those skilled in the art within the scope of the present invention.
Claims (10)
1. An intelligent 5G slice planning method based on deep learning is characterized by comprising the following steps of;
Step 1: converting the effective load information (namely PACKET DATA) in the Pcap data packet into a visualized image, and preprocessing the image;
Step 2: training a CNN neural network, optimizing the CNN neural network, and calculating a loss function to measure the performance of the model; extracting key features of the preprocessed image by using a CNN neural network, identifying power grid business according to the extracted key features, binding a 5G slice template, and distributing network resources according to needs;
Step 3: based on the electric service identification result, the slicing parameters are intelligently generated, the 5G slicing module is assisted to intelligently bind the network slices, and network resources are distributed according to needs.
2. The intelligent 5G slice planning method based on deep learning of claim 1, wherein the step 1 specifically includes the steps of:
Step 1.1: payload information extraction
Analyzing the Pcap data packet and extracting effective load information by adopting python; extracting key information according to the structural characteristics of a Pcap data packet, wherein the Pcap data packet consists of a Pcap packet header, a packet header of the data packet and data packet contents; acquiring the content of the data packet, namely payload information, by reading the packet header of the data packet;
step 1.2: converting the payload information into a picture
Converting the content of the data packet into a two-dimensional image with a set size;
Step 1.3: data enhancement
And carrying out data enhancement on the two-dimensional image.
3. The intelligent 5G slice planning method based on deep learning of claim 2, wherein in step 1.1, [36:40] bytes are directly extracted from the Pcap data packet according to the Pcap data packet characteristics to obtain the data packet content, and after the data extraction, the extracted data is processed according to the set image size to obtain the two-dimensional matrix meeting the requirements.
4. A deep learning based intelligent 5G slice planning method according to claim 3, wherein the data packet content, i.e. the payload information, is defined;
Firstly, judging whether the length of the data packet content is greater than n 2, and if the length of the data packet content is less than n 2, performing zero padding operation; if PACKET DATA is longer than n 2, then the data packet content needs to be intercepted to ensure that the size of the two-dimensional matrix is nxn, and the most relevant information is reserved by intercepting the previous n 2 data and remodelled into a one-dimensional matrix of nxn.
5. The intelligent 5G slice planning method based on deep learning of claim 4, wherein the step 1.2 specifically comprises:
Firstly, acquiring a two-dimensional matrix containing Pcap Data, and recombining the two-dimensional matrix into a two-dimensional matrix according to a predefined size; if the extracted data length is smaller than the predefined size, the image is resized by means of zero padding to match the predefined size, which is then saved as a jpg formatted picture using the OpenCv library.
6. The intelligent 5G slice planning method based on deep learning of claim 1, wherein the step 2 specifically comprises the following steps:
step 2.1: light convolutional neural network
The power grid business is identified by adopting a convolutional neural network, wherein the convolutional neural network comprises 3 convolutional layers, 2 pooling layers and 2 full connection layers;
Inputting the obtained image into a lightweight convolutional neural network, and classifying by using softmax;
Step 2.2: the Adam optimizer in tensorflow keras is used to optimize the parameters of the neural network model, and Cross entropy loss function (Cross-EntropyLoss) and Accuracy (Accuracy) are used as performance indexes.
7. The intelligent 5G slice planning method based on deep learning of claim 6, wherein the convolutional neural network first processes the input data through a convolutional layer to generate a feature map with a size of 30 x 32, and then a pooling layer halves the feature map to a size of 15 x 32, and then the data flows through a second convolutional layer to obtain a feature map of 13 x 64; after the pooling again, the feature map size is reduced to 6×6×64; the final convolution results in a 4 x 64 feature map; finally, the feature is mapped and flattened into a one-dimensional array, and then the classification result is finally obtained through two full-connection operations.
8. The deep learning based intelligent 5G slice planning method of claim 6, wherein model () method is used to train the model, incoming training data, tags, and training rounds (epochs) and batch sizes (batchsize), and model.
The formula of the cross entropy function is as follows:
Wherein y i is the output value of sample i, p i is the probability that sample i is correct, J is the value of the loss function, and N is the number of results;
the accuracy formula is as follows:
accuracy = (tp+tn)/(tp+tn+fp+fn)
Here, TP (TruePositives) is the number of samples correctly predicted as positive class, TN (TrueNegatives) is the number of samples correctly predicted as negative class, FP (FalsePositives) is the number of samples incorrectly predicted as positive class (actually negative class), and FN (FalseNegatives) is the number of samples incorrectly predicted as negative class (actually positive class).
9. The intelligent 5G slice planning method based on deep learning of claim 1, wherein the step 3 specifically comprises the following steps;
The decision maker needs to determine which 5G traffic scenario the current classification result belongs to, the 5G traffic scenario being eMBB (enhanced mobile broadband), uRLLC (ultra-high reliability and low latency communication) and mMTC (large-scale machine type communication); the decision maker will automatically select or recommend the corresponding 5G network slice template to generate the corresponding 5G slice type id.
10. A deep learning-based intelligent 5G network slice planning system implementing the deep learning-based intelligent 5G slice planning method of any one of claims 1-9, the deep learning-based intelligent 5G network slice planning system comprising:
1) A data preprocessing module configured to perform the following operations:
A payload information extracting unit, configured to extract payload information from the Pcap packet, where the extracting is based on the packet structure feature, and read the packet header of the packet from the Pcap packet header, the packet header of the packet, and the packet content to obtain the payload information;
an image conversion unit for converting the extracted payload information into a two-dimensional image of a set size;
the data enhancement unit is used for performing data enhancement operation on the two-dimensional image obtained through conversion so as to improve generalization capability and robustness of the model;
2) A deep learning model training and optimization module configured to:
a Convolutional Neural Network (CNN) construction unit responsible for constructing a CNN neural network for processing the two-dimensional image;
the training unit is used for training the CNN neural network and learning by utilizing the preprocessed image data;
The optimizing unit is used for optimizing parameters of the CNN neural network and measuring and improving the performance of the model by calculating a loss function;
the feature extraction unit is used for extracting key features from the trained CNN neural network so as to realize accurate identification of power grid business;
3) A 5G slice parameter generation and binding module configured to:
The recognition result processing unit is used for receiving the power grid service recognition result output by the CNN neural network;
the slice parameter generating unit intelligently generates corresponding 5G slice parameters according to the identification result;
and the slice binding unit is used for applying the generated slice parameters to the 5G slice module to realize intelligent binding of the power grid service and the network slice so as to achieve the on-demand distribution of network resources.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311823232.2A CN117994563A (en) | 2023-12-27 | 2023-12-27 | Intelligent 5G slice planning method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311823232.2A CN117994563A (en) | 2023-12-27 | 2023-12-27 | Intelligent 5G slice planning method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117994563A true CN117994563A (en) | 2024-05-07 |
Family
ID=90889773
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311823232.2A Pending CN117994563A (en) | 2023-12-27 | 2023-12-27 | Intelligent 5G slice planning method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117994563A (en) |
-
2023
- 2023-12-27 CN CN202311823232.2A patent/CN117994563A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Learning IoT in edge: Deep learning for the Internet of Things with edge computing | |
| Chen et al. | An edge traffic flow detection scheme based on deep learning in an intelligent transportation system | |
| US11676408B2 (en) | Identification of neural-network-generated fake images | |
| CN108875540B (en) | Image processing method, device and system and storage medium | |
| WO2019221965A1 (en) | Unsupervised cross-domain distance metric adaptation with feature transfer network | |
| US10388009B2 (en) | Machine-learning measurements of quantitative feature attributes | |
| WO2020253127A1 (en) | Facial feature extraction model training method and apparatus, facial feature extraction method and apparatus, device, and storage medium | |
| CN111368636B (en) | Object classification method, device, computer equipment and storage medium | |
| CN113989583A (en) | Method and system for detecting malicious traffic of internet | |
| CN114170482B (en) | Document pre-training model training method, device, equipment and medium | |
| Nigade et al. | Clownfish: Edge and cloud symbiosis for video stream analytics | |
| CN112861575A (en) | Pedestrian structuring method, device, equipment and storage medium | |
| CN110532959B (en) | Real-time violent behavior detection system based on two-channel three-dimensional convolutional neural network | |
| CN113807399A (en) | Neural network training method, neural network detection method and neural network detection device | |
| WO2021099844A1 (en) | Systems and methods for object detection and recognition | |
| CN110796039B (en) | Face flaw detection method and device, electronic equipment and storage medium | |
| CN114429675A (en) | Motion recognition method, model training method and device and electronic equipment | |
| Ren et al. | A novel deep learning method for application identification in wireless network | |
| Tan et al. | Deep learning on mobile devices through neural processing units and edge computing | |
| Qureshi et al. | Neurocomputing for internet of things: object recognition and detection strategy | |
| CN117994563A (en) | Intelligent 5G slice planning method and system based on deep learning | |
| Gong et al. | Research on mobile traffic data augmentation methods based on SA-ACGAN-GN | |
| Xiao et al. | Optimization methods of video images processing for mobile object recognition | |
| Sun et al. | Lightweight Traffic Classification Model Based on Deep Learning | |
| Zan et al. | Defect Identification of Power Line Insulators Based on a MobileViT‐Yolo Deep Learning Algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |