Terminal side network slicing method and system based on load identification, electronic equipment and storage medium

Abstract

The invention discloses a terminal side network slicing method and system based on load identification, electronic equipment and a storage medium, and the method comprises the steps: obtaining the operation state, power consumption and power factor data of each piece of equipment, and carrying out the preprocessing and normalization of the obtained data; establishing a load characteristic data set, wherein the load characteristic data set comprises a load characteristic matrix, an adjacent matrix and a load type label; constructing a graph neural network load identification model, and training the graph neural network load identification model; taking the load characteristic matrix and the adjacent matrix as input of a graph neural network load identification model, and obtaining a load type corresponding to each node; setting a transmission priority of terminal data in combination with the load attribute; and according to the transmission priority of the terminal data, determining a URSP service mapping rule to complete slicing. The method enables the key load to obtain stable power supply, and improves the overall operation efficiency of the power grid.

Description

Technical Field

[0001] The present invention relates to the fields of load management terminal technology and terminal-side slicing technology, and in particular to a terminal-side network slicing method, system, electronic device and storage medium based on load identification. Background Art

[0002] With the rapid development and popularization of 5G, its characteristics, such as ultra-high speed, massive connections, and ultra-low latency, have been unprecedentedly demonstrated. The diversified business scenarios in the 5G era have brought complex and changing network requirements, including the coexistence of multiple services with different characteristics, such as enhanced mobile broadband (eMBB), massive machine type communication (mMTC), and ultra-reliable low latency communication (uRLLC), which has posed a challenge to the traditional single and static network architecture. As one of the key technologies of 5G, network slicing divides the physical network into multiple logically independent slices with different functions through software-defined networking and network function virtualization. Each slice can provide customized network services and resource guarantees for specific business scenarios. Therefore, static slicing strategies are difficult to cope with the differentiated needs of services. On the other hand, in the scenario of massive load control services, data from massive terminals, such as real-time load data, time series data, time-segment data, and data related to device attributes, have different requirements for network resources. Summary of the invention

[0003] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a terminal-side network slicing method, system, electronic device and storage medium based on load identification to solve the problem that existing slicing strategies are difficult to cope with the differentiated needs of services.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a terminal-side network slicing method based on load identification, comprising:

[0008] Obtain the operating status, power consumption and power factor data of each device, and pre-process and normalize the acquired data;

[0009] Based on the preprocessed and normalized data, a load characteristic data set is established, wherein the load characteristic data set includes a load characteristic matrix, an adjacency matrix, and a load type label;

[0010] Based on the load characteristic dataset, a graph neural network load identification model is constructed and trained;

[0011] The load feature matrix and adjacency matrix are used as the input of the graph neural network load identification model. The graph neural network load identification model is used to predict the current load type, and the predicted load probability matrix is ​​obtained. The load type corresponding to each node is obtained through one-hot decoding.

[0012] Based on the load type corresponding to each node and combined with the load attributes, the transmission priority of the terminal data is set;

[0013] According to the transmission priority of the terminal data, the URSP service mapping rules are determined to complete the slicing.

[0014] As a preferred solution of the terminal-side network slicing method based on load identification described in the present invention, wherein:

[0015] The preprocessing of the acquired data comprises the following steps:

[0016] Use nearest neighbor interpolation to clean up outliers and fill in missing values;

[0017] Set the value of the calculation interval Δt, and perform cubic spline interpolation processing on the sampling data that is greater than the calculation interval Δt;

[0018] In the calculation interval Δt, the characteristic information of various loads is extracted, including the effective value of current, average active power, average power factor, current crest factor and total current harmonic distortion rate.

[0019] As a preferred solution of the terminal-side network slicing method based on load identification described in the present invention, wherein:

[0020] The adjacency matrix includes abstracting the measurement devices as nodes of a graph, abstracting the communication relationships between the devices as edges of the graph, and representing the relationships between the nodes in the form of a matrix;

[0021] If there is a communication connection between any two nodes in the graph, the corresponding matrix element is set to 1;

[0022] If there is no direct communication between any two nodes in the graph, the corresponding matrix element is set to 0.

[0023] As a preferred solution of the terminal-side network slicing method based on load identification described in the present invention, wherein:

[0024] The graph neural network load identification model includes two graph convolutional layers and one output layer, and a skip connection is introduced between the graph convolutional layers;

[0025] The output layer uses Softmax as the activation function and takes the probability matrix P of each node belonging to each type of load as the output.

[0026] As a preferred solution of the terminal-side network slicing method based on load identification described in the present invention, wherein:

[0027] The training of the graph neural network load identification model includes the following steps:

[0028] Onehot encode the load type label L to form the label matrix L (1) ;

[0029] The cross entropy loss function is used to calculate the load probability matrix P and label matrix L predicted by the graph neural network load identification model. (1) The loss value between

[0030] Introducing cost-sensitive weights based on misclassification ratio into the loss function of the graph neural network load identification model, dynamically adjusting the loss according to the imbalance ratio of labels in the sample set;

[0031] Using the training set to train the graph neural network load identification model and update the model parameters;

[0032] Evaluate model performance based on the validation set to detect whether the model is overfitting and adjust the optimal hyperparameters for model training;

[0033] When the loss function converges, the training stops.

[0034] As a preferred solution of the terminal-side network slicing method based on load identification described in the present invention, wherein:

[0035] The cost-sensitive weight based on the misclassification ratio is introduced into the loss function of the graph neural network load identification model, and the loss function of the model is expressed as:

[0036]

[0037] Among them, N is the number of samples; C is the cost-sensitive matrix; K represents the number of categories; Y ic represents the predicted probability that the i-th observed sample belongs to category c; C ic Represents sample Y i The cost of misclassification; T i Represents sample Y i corresponding target;

[0038] The element C in the cost-sensitive matrix C(i,j) It is expressed as:

[0039]

[0040] Among them, c ij represents the number of samples in the i-th class that are misclassified into the j-th class, w i is the weight considering the proportion of misclassification, p i is the importance of different categories, and p i ∈[0,1];

[0041] The weight w considering the misclassification ratio i It is expressed as:

[0042]

[0043] Among them, (n Y≠T ) i It represents the total number of misclassified samples in the i-th category.

[0044] As a preferred solution of the terminal-side network slicing method based on load identification described in the present invention, wherein:

[0045] Determining the transmission priority of the terminal data includes obtaining the load type corresponding to each ID number and dividing the load type into transmission levels;

[0046] Set the data from public charging piles, distributed photovoltaics, and energy storage devices to high priority and mark them as A;

[0047] The data from the dedicated charging pile, transformer and area master meter are given a higher priority and marked as B;

[0048] Set the data from high-power resistive, high-power inductive and high-power mixed devices to a low priority, marked as C;

[0049] Data from low power loads etc. is given the lowest priority and is marked as D.

[0050] In a second aspect, the present invention provides a terminal-side network slicing system based on load identification, including:

[0051] The preprocessing module is used to obtain the operating status, power consumption and power factor data of each device, and preprocess and normalize the acquired data;

[0052] Establishing a data set module, used to establish a load characteristic data set based on the preprocessed and normalized data, wherein the load characteristic data set includes a load characteristic matrix, an adjacency matrix, and a load type label;

[0053] Establish a model module to build a graph neural network load identification model based on the load characteristic data set and train the graph neural network load identification model;

[0054] The input module is used to use the load feature matrix and the adjacency matrix as the input of the graph neural network load identification model, use the graph neural network load identification model to predict the current load type, obtain the predicted load probability matrix, and obtain the load type corresponding to each node through one-hot decoding;

[0055] The priority setting module is used to set the transmission priority of terminal data based on the load type corresponding to each node and the load attributes;

[0056] The slicing module is used to determine the URSP service mapping rules to complete slicing according to the transmission priority of the terminal data.

[0057] In a third aspect, the present invention provides a computing device, comprising:

[0058] Memory, used to store programs;

[0059] A processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the terminal-side network slicing method based on load identification.

[0060] In a fourth aspect, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the terminal-side network slicing method based on load identification are implemented.

[0061] Beneficial effects of the invention: The present invention proposes a network slicing method based on load identification, constructs a topology diagram based on the communication relationship between measurement devices in the area, dynamically adjusts network resources according to terminal load types and actual business needs, and uses a load identification model based on a graph neural network to predict the real-time load type of the equipment, thereby performing slicing processing based on the equipment type and communication priority, enhancing the information processing capabilities of the power distribution and power consumption links, thereby ensuring that critical loads are supplied with stable power and improving the overall operation efficiency of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0063] Figure 1 A basic flow chart of a terminal-side network slicing method based on load identification provided by an embodiment of the present invention;

[0064] Figure 2 A basic process model diagram of a terminal-side network slicing method based on load identification provided by an embodiment of the present invention;

[0065] Figure 3 A graph neural network model training process diagram of a terminal-side network slicing method based on load identification provided by an embodiment of the present invention;

[0066] Figure 4 A test set confusion matrix diagram of a terminal-side network slicing method based on load identification provided by an embodiment of the present invention;

[0067] Figure 5 A priority-based network slice mapping diagram of a terminal-side network slicing method based on load identification provided by an embodiment of the present invention;

[0068] Figure 6 A terminal device communication topology diagram of a terminal-side network slicing method based on load identification is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0069] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0070] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0072] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0073] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0074] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0075] Example 1

[0076] Reference Figure 1-5 , is an embodiment of the present invention, and provides a terminal side network slicing method based on load identification, including Figure 1 , as shown in 2, comprising the following steps:

[0077] S1: Obtain the operating status, power consumption and power factor data of each device, and preprocess and normalize the acquired data;

[0078] In the embodiment of the present application, since the measuring equipment is not uniform, the parameters that can be measured by different equipment are different. The data that can be collected mainly include equipment power consumption, three-phase voltage and current, active and reactive power, power factor, harmonics, etc.; and the sampling frequency of different equipment is also different, and the sampling time ranges from 15s to 15min. By collecting sufficient historical data of different time periods and different types of loads, sufficient data is provided for subsequent model training.

[0079] In the embodiment of the present application, distributed photovoltaic, charging piles, energy storage batteries, user loads, transformers, substation meters and other equipment are equipped with power data collection and monitoring equipment for monitoring and collecting data such as the operating status, power consumption, power factor and other data of each device, and sending it to the management side through the communication network.

[0080] In the embodiment of the present application, due to the influence of external disturbances and noise, the measured data will have abnormal values ​​and missing data, so the measured data should be cleaned to deal with abnormal values ​​and missing values. Since the acquisition accuracy of different devices is different and the collected data is also different, the cleaned data cannot be used directly, and further features need to be extracted and the data format is unified. In order to improve the quality and integrity of the measurement data, the extracted features are normalized using the standard score normalization method.

[0081] In the embodiment of the present application, the data preprocessing and normalization steps are as follows:

[0082] S1.1: Replace the outliers in the data with the nearest neighbor interpolation method and complete the missing values ​​to complete the data cleaning. Considering the different acquisition frequencies of different measuring equipment, 5 minutes is selected as the calculation interval Δt, and the sampling data greater than the calculation interval Δt is processed by cubic spline interpolation to make its sampling interval within the calculation interval Δt.

[0083] S1.2: Load feature extraction. Within the calculation interval Δt, the load features including current effective value, average active power, average power factor, current crest factor, and total current harmonic distortion rate are calculated through sampling values. The specific formula is as follows:

[0084] Current effective value I rms :

[0085] Wherein, m is the number of sampling points of the real-time current within Δt.

[0086] Average active power P avg :

[0087] Wherein, m is the number of sampling points of real-time power within Δt.

[0088] Average power factor PF avg :

[0089] Wherein, m is the number of sampling points of the real-time power factor within Δt.

[0090] Current crest factor I c :

[0091] Wherein, m is the number of sampling points of the real-time current within Δt.

[0092] Total harmonic distortion of current THD I :

[0093]

[0094] in, is the effective value of the fundamental current, is the effective value of each harmonic current.

[0095] The final structure is as follows: x = [I rms P avg PF avg I c THD I T], where T represents the data collection time point, T∈[1,24].

[0096] It should be noted that, considering that not all measuring devices can report all data, the missing data that cannot be calculated are filled with 0. Through load feature extraction, the sampling data can be effectively reduced in dimension, the problem of inconsistent sampling frequency can be solved, and rich load feature information can be extracted, which can also make the identification model not too complicated.

[0097] S1.2: Data normalization:

[0098] In order to improve the quality and integrity of the measurement data, the standard score standardization method is used to normalize it, and the formula is:

[0099]

[0100] Among them, X is the data to be normalized, X mean is the mean of X, X std is the standard deviation of X, Y is the normalized data, Y mean is the target mean, Y std is the target standard deviation.

[0101] S2: Based on the preprocessed and normalized data, a load characteristic data set is established, wherein the load characteristic data set includes a load characteristic matrix, an adjacency matrix, and a load type label;

[0102] In the embodiment of the present application, the load identification data set construction includes:

[0103] Take an area with n measurement devices as an example.

[0104] The load characteristic matrix X is the set of load characteristics monitored by n measuring devices at a certain moment, X = [x1, x2, ..., x i ]′, where x i represents the load characteristics of the i-th load, which is obtained by the above S1 method.

[0105] The adjacency matrix A represents the communication connection relationship between the measurement devices. From the perspective of graph theory, the measurement devices can be abstracted as nodes of the graph, and the communication relationship between the devices can be abstracted as the edges of the graph. The relationship between the nodes is represented in the form of a matrix, which is the adjacency matrix A. If node i has a communication relationship with node j, then the element a in the adjacency matrix A is ij =a ji =1; if there is no communication relationship, it is 0.

[0106] The load type label L records the load type of each node at the current moment, including charging piles, large power equipment, high-power equipment, etc. L = [l1, l2, ... l n ], the corresponding relationship between load name and load label is shown in Table 1:

[0107] Table 1 Comparison table of load names and load labels

[0108]

[0109]

[0110] The load feature matrix X, the adjacency matrix A, and the type label L together constitute a piece of data in the data set. As much measurement data as possible at different times is obtained to build a complete load identification data set, and the data set is divided in an 8:1:1 manner to construct a training set, a validation set, and a test set.

[0111] S3: Based on the load feature dataset, a graph neural network load identification model is constructed and trained;

[0112] In the embodiment of the present application, the construction of the identification model includes the following steps:

[0113] The load identification model mainly consists of two graph convolutional layers with skip connections and one output layer.

[0114] Specifically, the graph convolution layer Gcl aggregates node information and extracts high-dimensional features to update node expressions by performing convolution operations on the graph structure. The convolution operation of the convolution layer is:

[0115]

[0116] Among them, X (i) is the feature matrix of the i-th layer input, W (i) is the learnable weight of the i-th layer.

[0117] The skip connection operation between graph convolutional layers is:

[0118]

[0119] Among them, Z (i)is the output after the i-th layer skip connection, →(·) represents the output of the dimension X (i) To Dimension Mapping operation.

[0120] Specifically, the graph convolution layers learn the hierarchical structure and complex relationships between nodes through hierarchical propagation, further capturing the local and global information of graph features; the output layer uses Softmax as the activation function to output the probability matrix of each node belonging to each type of load. The overall hierarchical propagation formula of the model is:

[0121]

[0122] in, is the degree matrix of the adjacency matrix, and I is the identity matrix of the same dimension as the adjacency matrix A.

[0123] The model takes the load characteristic matrix X and the adjacency matrix A as input, and takes the probability matrix P of each node belonging to each type of load as output, where p ij It represents the probability that the i-th node is predicted to be the j-th type of load.

[0124] In the embodiments of the present application, Figure 3 As shown in FIG. 1 , the training of the load identification model includes the following steps:

[0125] The load identification task is a multi-classification task, which usually uses the cross entropy loss function to calculate the loss value between the load probability matrix predicted by the model and the one-hot encoded true load label matrix to speed up the convergence of the model.

[0126] Since the labels of the load identification data set composed of different types of loads are not balanced due to the influence of conditions such as number, usage time, and usage frequency, a cost-sensitive weight based on the misclassification ratio is introduced into the loss function of the model. The loss is dynamically adjusted according to the imbalance ratio of the labels in the sample set to further enhance the recognition ability of the model.

[0127] The loss function of the model is expressed as:

[0128]

[0129] Among them, N is the number of samples; C is the cost-sensitive matrix; K represents the number of categories; Y ic represents the predicted probability that the i-th observed sample belongs to category C; C ic Represents sample Y i The cost of misclassification; i represents sample Y i corresponding target;

[0130] The element C in the cost-sensitive matrix C (i,j) It is expressed as:

[0131]

[0132] Among them, c ij represents the number of samples in the i-th class that are misclassified into the j-th class, w i It is the weight considering the proportion of misclassification, and the misclassification of imbalanced samples can be easily obtained through the confusion matrix, p i is the importance of different categories, and p i ∈[0,1];

[0133] The weight w considering the misclassification ratio i It is expressed as:

[0134]

[0135] Among them, (n Y≠T ) i It represents the total number of misclassified samples in the i-th category.

[0136] The load identification model is trained using the training set to update the model parameters; the model performance is evaluated based on the validation set to detect whether the model is overfitting and adjust the optimal hyperparameters of the model training; when the loss function is trained to convergence or the training rounds are completed, the generalization ability of the model is evaluated using the test set.

[0137] S4: The load feature matrix and adjacency matrix are used as inputs of the graph neural network load identification model. The graph neural network load identification model is used to predict the current load type, and the predicted load probability matrix is ​​obtained. The load type corresponding to each node is obtained through one-hot decoding.

[0138] In the embodiment of the present application, load identification and labeling includes the following steps:

[0139] S4.1: Preprocessing of data to be identified

[0140] Sequentially number the measuring devices in the area, assign a unique ID to each device, and construct the adjacency matrix A based on the communication relationship between the devices. Obtain the load characteristic matrix X within the calculation interval Δt through the above steps 1 and 2, and perform data preprocessing.

[0141] S4.2: Load identification and labeling

[0142] With the adjacency matrix A and the load characteristic matrix X as input, the identification model is used to predict the current load type to obtain the predicted load probability matrix, and then the load type corresponding to each node is obtained through one-hot decoding.

[0143] S5: Based on the load type corresponding to each node and combined with the load attributes, the transmission priority of the terminal data is set;

[0144] In the embodiment of the present application, the steps of determining the transmission priority according to the load label and the load attribute are as follows:

[0145] After obtaining the load type corresponding to each ID number, the transmission level can be divided according to Table 2. A high priority is set for data from public charging piles, distributed photovoltaics, and energy storage devices, marked as A; a higher priority is set for data from dedicated charging piles, transformers, and substation meters, marked as B; a low priority is set for data from high-power resistive, high-power inductive, and high-power hybrid devices, marked as C; and the lowest priority is set for data from low-power loads, marked as D.

[0146] Table 2 Data transmission priority level table

[0147]

[0148]

[0149] S6: Determine the URSP service mapping rule to complete the slicing based on the transmission priority of the terminal data.

[0150] In the embodiments of the present application, Figure 4 As shown in Figure 5, in the network slice, the service descriptor TD contains DNN, application type APP ID, IP descriptor (including destination address, destination port and protocol type), FQDN and connection characteristics. According to the priorities distinguished in Table 2, each priority is mapped to a corresponding APP ID to determine the priority-based service descriptor TD identifier. On the network side, the network allocates slices according to the APP ID, such as Figure 5 shown.

[0151] This embodiment also provides a terminal-side network slicing system based on load identification, including:

[0152] The preprocessing module is used to obtain the operating status, power consumption and power factor data of each device, and preprocess and normalize the acquired data;

[0153] Establishing a data set module, used to establish a load characteristic data set based on the preprocessed and normalized data, wherein the load characteristic data set includes a load characteristic matrix, an adjacency matrix, and a load type label;

[0154] Establish a model module to build a graph neural network load identification model based on the load characteristic data set and train the graph neural network load identification model;

[0155] The input module is used to use the load feature matrix and the adjacency matrix as the input of the graph neural network load identification model, use the graph neural network load identification model to predict the current load type, obtain the predicted load probability matrix, and obtain the load type corresponding to each node through one-hot decoding;

[0156] The priority setting module is used to set the transmission priority of terminal data based on the load type corresponding to each node and the load attributes;

[0157] The slicing module is used to determine the URSP service mapping rules to complete slicing according to the transmission priority of the terminal data.

[0158] Furthermore, it also includes:

[0159] Memory, used to store programs;

[0160] A processor is used to load the program to execute the terminal-side network slicing method based on load identification.

[0161] This embodiment also provides a computer-readable storage medium, which stores a program. When the program is executed by a processor, it implements the terminal-side network slicing method based on load identification.

[0162] The storage medium proposed in this embodiment and the terminal-side network slicing method based on load identification proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0163] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0164] Example 2

[0165] Reference Figure 6, as an embodiment of the present invention, provides a terminal-side network slicing method based on load identification. In order to verify its beneficial effects, scientific demonstration is carried out through specific implementation methods and implementation effects.

[0166] The present embodiment is as follows:

[0167] Step 1: Load terminal data measurement and collection

[0168] like Figure 6 As shown, in a certain target area, smart meters and other measuring devices are installed at the substation total meter, 1 public charging pile, 2 dedicated charging piles, 1 distributed photovoltaic, 1 energy storage device, 2 substation transformers and 5 user-side devices. The measuring equipment is used to obtain sufficient historical measurement data in the target area, and the data is collected and summarized on the management side.

[0169] Step 2: Data preprocessing and normalization;

[0170] Step 2.1: First, perform data cleaning, use interpolation to complete the obviously abnormal and missing data, and interpolate the data that does not meet the calculation interval Δt.

[0171] Step 2.2: Secondly, calculate the load characteristics corresponding to the sampling value according to the formula, including the effective value of current, average active power, average power factor, current crest factor, total harmonic distortion rate of current, etc.

[0172] The dimension of the load characteristic vector is 6, that is, x = [I rms P avg PF avg I c THD I T], where T represents the data collection time point, T∈[1,24].

[0173] Step 2.3: Finally, the data is normalized using the following formula to form load characteristic data of uniform size.

[0174] Step 3: Build and train the graph neural network

[0175] Step 3.1: Construct load characteristic data set

[0176] The adjacency matrix A represents the communication connection relationship between measurement devices. From the perspective of graph theory, the measurement devices can be abstracted as nodes of the graph, and the communication relationship between devices can be abstracted as the edge of the graph. The relationship between nodes is represented in the form of a matrix, which is the adjacency matrix A. If node i has a communication relationship with node j, then the element aij=aji=1 in the adjacency matrix A, and if there is no communication relationship, it is 0. Figure 6As shown, there are 13 measuring devices in the target area, which are numbered from top to bottom and from left to right as ID1 to 13. The adjacency matrix A is constructed in this way, and A is a symmetric matrix with a main diagonal of 0. The constructed side adjacency matrix A is as follows:

[0177]

[0178] The load characteristic matrix X is the set of load characteristics monitored by 13 measuring devices at a certain moment, X = [x1, x2, ..., x1, ... x 13 ]′, where x i The load characteristics of the i-th load are obtained by the above steps 1 and 2. Then the size of the load characteristic matrix X is [13×6].

[0179] The load type label L records the load type of each node at the current moment, L = [l1, l2, ... l 13 ], the correspondence between load names and load labels is shown in Table 1. There are 16 types of load labels in total.

[0180] The load characteristic matrix X, the adjacency matrix A, and the type label L together constitute a piece of data in the data set. Take the historical measurement data of different time periods to construct 8,000 load characteristic data to form a load characteristic data set, and divide the data set into 8:1:1 to construct the training set, validation set, and test set.

[0181] Step 3.2: constructing an identification model according to the method of the present invention;

[0182] The model takes the load characteristic matrix X and the adjacency matrix A as input, and takes the probability matrix P of each node belonging to each type of load as output, where p ij It represents the probability that the i-th node is predicted to be the j-th type of load. The dimension of the probability matrix P is [13×16].

[0183] Step 3.3 Training of load identification model

[0184] Specifically, the batch size of the model is set to 64, the initial value of the piecewise learning rate is 0.01, and the learning rate is reduced to half of the original value after every 50 rounds of training.

[0185] The loss function of the model can be specifically expressed as:

[0186]

[0187] Among them, the number of samples N is 64, the cost-sensitive matrix is ​​C; the number of categories K is 16; Y ic Represents the i-th observation sample Y i The predicted probability of belonging to category c; C ic Represents sample Yi The cost of misclassification; T i Represents sample Y i The corresponding target.

[0188] For the element C in the cost-sensitive matrix C (i,j) have:

[0189]

[0190] Among them, c ij represents the number of samples in the i-th class that are misclassified into the j-th class, w i is the weight considering the proportion of misclassification, p i is the importance of different categories, and p i ∈[0,1];

[0191] The weight w considering the misclassification ratio i It is expressed as:

[0192]

[0193] Among them, (n Y≠T ) i It represents the total number of misclassified samples in the i-th category.

[0194] The load identification model is trained using the training set to update the model parameters; the model performance is evaluated based on the validation set to detect whether the model is overfitting and adjust the optimal hyperparameters of the model training; when the loss function is trained to convergence or the training rounds are completed, the generalization ability of the model is evaluated using the test set.

[0195] After the training is completed, the accuracy of the test set is 97.6321%, and the F1-score is 0.97668. It can be seen that the load identification model of the present invention has a high identification ability for various types of loads.

[0196] Step 4: Load identification and labeling

[0197] Step 4.1: Preprocessing of data to be identified

[0198] The measuring equipment in the target area is numbered with the same sequential ID as in step 3.1, and the adjacency matrix A is constructed. The measurement data within Δt before the current moment is taken, and the characteristic data within the calculation interval Δt is obtained by steps 1 and 2, and the data is preprocessed and normalized to construct the load characteristic matrix X.

[0199] Step 4.2: Load identification and labeling

[0200] With the adjacency matrix A and the load characteristic matrix X as input, the identification model is used to predict the current load type to obtain the predicted load probability matrix, and then the load type corresponding to each node is obtained through one-hot decoding. Taking a certain moment in the winter evening as an example, the labels corresponding to the ID numbers predicted by the model are [APL, CS-1, TR, BESS, CS-2, TR, NPL, CS-2, RPL-3, RPL-2, IPL-1, RPL-2, SPL]. It can be seen that at this time, the distributed photovoltaic is in a no-load state due to the lack of light input, the energy storage equipment is in a load state for replenishing energy in the target area, there are cars replenishing energy at both the public charging piles and the dedicated charging piles, users 9 to 12 have high-power electric heating equipment running, and user 13 may not be using high-power equipment.

[0201] Step 5: Transmission priority setting on the terminal side

[0202] After obtaining the load type corresponding to each ID number, the transmission level can be divided according to Table 2. 5G network slices are identified according to the transmission priority, and the information transmission level corresponding to the ID number is [B, A, B, A, B, B, D, B, C, C, C, C, D].

[0203] Step 6: Network slicing on the terminal side

[0204] The URSP service mapping rule is determined based on the transmission priority on the terminal side to complete the slicing. Each rule contains a service descriptor, which is associated with different service flows. According to the transmission priority vector [B, A, B, A, B, B, D, B, C, C, C, C, D] in step 5, each priority is mapped to a corresponding APP ID, and the priority-based service descriptor TD identifier is determined. On the network side, the network allocates slices based on the APP ID.

[0205] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A terminal-side network slicing method based on load identification, characterized in that: include: Obtain the operating status, power consumption and power factor data of each device, and preprocess and normalize the acquired data; Based on the preprocessed and normalized data, a load characteristic data set is established, wherein the load characteristic data set includes a load characteristic matrix, an adjacency matrix, and a load type label; Based on the load characteristic dataset, a graph neural network load identification model is constructed and trained; The load feature matrix and adjacency matrix are used as the input of the graph neural network load identification model. The graph neural network load identification model is used to predict the current load type, and the predicted load probability matrix is ​​obtained. The load type corresponding to each node is obtained through one-hot decoding. Based on the load type corresponding to each node and combined with the load attributes, the transmission priority of the terminal data is set; According to the transmission priority of the terminal data, the URSP service mapping rules are determined to complete the slicing.

2. The method for terminal-side network slicing based on load identification according to claim 1, characterized in that: The preprocessing of the acquired data comprises the following steps: Use nearest neighbor interpolation to clean up outliers and fill in missing values; Set the value of the calculation interval Δt, and perform cubic spline interpolation processing on the sampling data that is greater than the calculation interval Δt; In the calculation interval Δt, the characteristic information of various loads is extracted, including the effective value of current, average active power, average power factor, current crest factor and total current harmonic distortion rate.

3. The terminal-side network slicing method based on load identification according to claim 1 or 2, characterized in that: The adjacency matrix includes abstracting the measurement devices as nodes of a graph, abstracting the communication relationships between the devices as edges of the graph, and representing the relationships between the nodes in the form of a matrix; If there is a communication connection between any two nodes in the graph, the corresponding matrix element is set to 1; If there is no direct communication between any two nodes in the graph, the corresponding matrix element is set to 0.

4. The method for terminal-side network slicing based on load identification according to claim 3, characterized in that: The graph neural network load identification model includes two graph convolutional layers and one output layer, and a skip connection is introduced between the graph convolutional layers; The output layer uses Softmax as the activation function and takes the probability matrix P of each node belonging to each type of load as the output.

5. The method for terminal-side network slicing based on load identification according to claim 4, characterized in that: The training of the graph neural network load identification model includes the following steps: Onehot encode the load type label L to form the label matrix L (1) ; The cross entropy loss function is used to calculate the load probability matrix P and label matrix L predicted by the graph neural network load identification model. (1) The loss value between A cost-sensitive weight based on the misclassification ratio is introduced into the loss function of the graph neural network load identification model, and the loss is dynamically adjusted according to the imbalance ratio of the labels in the sample set. Using the training set to train the graph neural network load identification model and update the model parameters; Evaluate model performance based on the validation set to detect whether the model is overfitting and adjust the optimal hyperparameters for model training; When the loss function converges, the training stops.

6. The method for terminal-side network slicing based on load identification according to claim 4, characterized in that: The cost-sensitive weight based on the misclassification ratio is introduced into the loss function of the graph neural network load identification model, and the loss function of the model is expressed as: Among them, N is the number of samples; C is the cost-sensitive matrix; K represents the number of categories; Y ic represents the predicted probability that the i-th observed sample belongs to category c; C ic Represents sample Y i The cost of misclassification; T i Represents sample Y i corresponding target; The element C in the cost-sensitive matrix C (i,j) It is expressed as: Among them, c ij represents the number of samples in the i-th class that are misclassified into the j-th class, w i is the weight considering the proportion of misclassification, p i is the importance of different categories, and p i ∈[0,1]; The weight w considering the misclassification ratio i It is expressed as: Among them, (n Y≠T ) i It represents the total number of misclassified samples in the i-th category.

7. The method for terminal-side network slicing based on load identification according to claim 6, characterized in that: Determining the transmission priority of the terminal data includes obtaining the load type corresponding to each ID number and dividing the load type into transmission levels; Set the data from public charging piles, distributed photovoltaics, and energy storage devices to high priority and mark them as A; The data from the dedicated charging pile, transformer and area master meter are given a higher priority and marked as B; Set the data from high-power resistive, high-power inductive and high-power mixed devices to a low priority, marked as C; Data from low power loads etc. is given the lowest priority and is marked as D.

8. A system based on the load identification-based terminal-side network slicing method according to claim 1, characterized in that: The preprocessing module is used to obtain the operating status, power consumption and power factor data of each device, and preprocess and normalize the acquired data; Establishing a data set module, used to establish a load characteristic data set based on the preprocessed and normalized data, wherein the load characteristic data set includes a load characteristic matrix, an adjacency matrix, and a load type label; Establish a model module to build a graph neural network load identification model based on the load characteristic data set and train the graph neural network load identification model; The input module is used to use the load feature matrix and the adjacency matrix as the input of the graph neural network load identification model, use the graph neural network load identification model to predict the current load type, obtain the predicted load probability matrix, and obtain the load type corresponding to each node through one-hot decoding; The priority setting module is used to set the transmission priority of terminal data based on the load type corresponding to each node and the load attributes; The slicing module is used to determine the URSP service mapping rules to complete slicing according to the transmission priority of the terminal data.

9. An electronic device, characterized in that: include: Memory, used to store programs; A processor, used to load the program to execute the steps of the terminal-side network slicing method based on load identification as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a program, characterized in that: When the program is executed by the processor, the steps of the terminal-side network slicing method based on load identification as described in any one of claims 1 to 7 are implemented.

Images