CN115314925A - Intelligent power grid fusion network flow prediction method based on GNN and LSTM - Google Patents

Intelligent power grid fusion network flow prediction method based on GNN and LSTM Download PDF

Info

Publication number
CN115314925A
CN115314925A CN202210934264.9A CN202210934264A CN115314925A CN 115314925 A CN115314925 A CN 115314925A CN 202210934264 A CN202210934264 A CN 202210934264A CN 115314925 A CN115314925 A CN 115314925A
Authority
CN
China
Prior art keywords
lstm
gnn
base station
flow prediction
power grid
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
Application number
CN202210934264.9A
Other languages
Chinese (zh)
Inventor
温明才
黄茹玥
吴雨欣
曹泽
唐海强
时博
熊金齐
王超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Zhilianda Technology Co ltd
Original Assignee
Nanjing Zhilianda Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing Zhilianda Technology Co ltd filed Critical Nanjing Zhilianda Technology Co ltd
Priority to CN202210934264.9A priority Critical patent/CN115314925A/en
Publication of CN115314925A publication Critical patent/CN115314925A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/06—Testing, supervising or monitoring using simulated traffic
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0876—Network utilisation, e.g. volume of load or congestion level

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Artificial Intelligence (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Environmental & Geological Engineering (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention discloses a GNN and LSTM based intelligent power grid fusion network flow prediction method, which relates to the technical field of intelligent power grids and comprises the following steps: firstly, a smart grid processing center collects characteristic data related to terminal and 5G transmission through a power wireless private network, wherein the characteristic data comprises current required bandwidth, actual bandwidth, the number of connected terminals, service quality, coordinate position distribution of terminals, coordinate positions of base stations and grouping of the base stations, secondly, a GNN and LSTM-based network flow prediction model is designed by taking flow characteristics of the base stations as input, and the model is composed of a first part of LSTM recursive coding, a second part of GNN spatial context coding and a third part of LSTM decoding. According to the intelligent power grid fusion network flow prediction method based on the GNN and the LSTM, the GNN and the LSTM are used for completing flow prediction of a fusion network, a basis can be provided for resource allocation and network optimization of the intelligent power grid, and production efficiency of the power industry is improved.

Description

Intelligent power grid fusion network flow prediction method based on GNN and LSTM
Technical Field
The invention relates to the technical field of intelligent power grids, in particular to a GNN and LSTM-based intelligent power grid fusion network flow prediction method.
Background
Due to the characteristics of high speed, high capacity, ultralow time delay and the like, the 5G vertical power grid can not only meet the personalized service of individual users, but also meet the differentiated service requirements of vertical industries such as intelligent manufacturing, intelligent medical treatment, smart cities and smart power grids, and in the 5G vertical industry, the power industry is used as an indispensable basic industry of national civilian life and is highly matched with the 5G, and the information transmission characteristics such as ultrahigh reliability and low time delay of the 5G vertical power grid play an important role in power distribution automation, accurate load control, power differential protection and other power grid control services. However, the current 5G network has insufficient signal coverage capability, weak signal penetration capability and high traffic cost, which limits the application of the 5G network in the power scene, and the existing power wireless private network has the characteristics of high safety and low signal coverage cost. From the signal coverage condition, the electric power wireless private network effectively fills up the signal coverage blind area of the 5G network. Aiming at the current situation, a more effective implementation scheme is to fuse the 5G and the electric power wireless private network, so that the network construction and use cost can be saved, and the agile response and accurate control requirements of mass resources in the integration of power supply, power grid, load and energy storage can be met.
In the converged network, because the 5G public network belongs to a telecommunication operator, the smart grid needs to adopt advanced network management technology and means to ensure the communication quality of the power service, which has important significance in predicting the base station flow of the telecommunication 5G public network and can provide a basis for resource allocation and network optimization. If the smart grid processing center predicts that the flow of the 5G base station at the next time point exceeds the threshold, the service with low priority can reduce the code stream or enter a silent state according to the priority of the terminal service of the smart grid, and even the service is switched to a power private network for transmitting service data, so that the communication quality of the power service with high priority in the 5G bearer network can be ensured. The flow of the 5G public network is predicted, and the method belongs to the fusion of an application layer, and the intelligent power grid terminal has dual-mode communication capability, namely, the wireless private network and the 5G public network are simultaneously supported for access and communication. The intelligent power grid processing center collects state information of the terminals through a power private network, wherein the state information comprises the current required bandwidth of the base station, the actual bandwidth, the number of the connected terminals, the service quality, the coordinate position distribution of the terminals, the coordinate position of the base station and the grouping of the base station. And the intelligent power grid processing center performs statistical analysis on the terminal information acquired by the power private network by taking the base station as a unit, and predicts by adopting an algorithm model.
In order to predict the flow of the base station, the article "i.alawe, a.ksenti, y.hadjadj-Aoul and p.bertin," Improving Traffic Learning for 5G Core Network scalability. The paper "Y.Fu and X.Wang," Traffic Prediction-Enabled Energy Allocation in CRAN Based on Deep Learning, "in IEEE Open Journal of the Communications Society, vol.3, pp.159-175, 2022" proposes a method for predicting wireless Traffic Based on a two-dimensional CNN-LSTM model of time aggregation, which processes Traffic data as an image, thereby extracting spatial correlation between wireless network Traffic. The article "g.jang, n.kim, t.ha, c.lee and s.cho," Base Station Switching and Sleep Mode Optimization With LSTM-Based User Prediction, "in IEEE Access, vol.8, pp.222711-222723,2020 proposes to predict User traffic in a future short time slot Based on a long-short time memory (LSTM) model, thereby reducing overhead generated by information synchronization between a User and a Base Station and reducing power consumption of the Base Station to the maximum extent.
Compared with the traditional shallow algorithm, the traffic prediction method based on deep learning has obvious improvement on prediction accuracy, but the traffic prediction method is aimed at the traffic prediction of a single base station, the time correlation is obtained by taking the historical value of the time sequence of the single base station as training data, and for 5G, the coverage range of the base station is only about 300 meters generally, so that the traffic of adjacent base stations has obvious correlation. Currently, only a few traffic prediction methods take into account spatial correlation.
The paper "c.zhang, h.zhang, d.yuan and m.zhang," city wide Cellular Prediction Based on the noise Connected predictive network, "in IEEE Communications Letters, vol.22, no.8, pp.1656-1659, 2018" divides the area into fixed grids and then predicts the Traffic using a two-dimensional Convolutional Neural network, which is better able to capture the dynamic characteristics of the Traffic from a spatial perspective than a conventional time series model, whereas in a converged network, 5G base stations appear only in partial location concentration, apparently not well described with image-like data structures.
Disclosure of Invention
The invention aims to provide a GNN and LSTM-based intelligent power grid fusion network flow prediction method, and aims to solve the problem that the existing method for flow prediction in the background art only uses historical flow characteristic information of a single base station, cannot utilize flow correlation between adjacent base stations, and is low in prediction accuracy.
In order to achieve the purpose, the invention provides the following technical scheme: a GNN and LSTM-based intelligent power grid fusion network flow prediction method comprises the following steps:
step 1: describing the characteristics;
step 1.1: setting the length and the width of an area covered by the smart power grid and the coverage range of a base station;
step 1.2: and acquiring the current required bandwidth, actual bandwidth, the number of connected terminals, service quality, coordinate position distribution of the terminals, the coordinate position of the base station and grouping of the base station to form a characteristic vector.
Step 2: a first part of a network flow prediction model based on GNN and LSTM is designed, LSTM recursive encoding is carried out, and the specific steps are as follows:
step 2.1: the input of the whole neural network is
Figure BDA0003782851200000041
The maximum base station number of each group is N, and T is the length of the characteristic sequence. The feature vector of each time step is subjected to implicit projection to a high-dimensional embedding space through linear transformation, the input dimension is 17 dimensions, the output dimension is 512 dimensions, and x is used t To represent;
step 2.2: the neuron at the current moment receives the output h of the neuron at the previous moment t-1 Embedding x with base station of current time in high dimension t Calculating the forgetting gate G forget Output gate G for selecting information to be forgotten output Selecting information to be memorized and a temporary cell state
Figure BDA0003782851200000042
G forget ＝σ(W f [h t-1 ,x t-1 ]+b f )
G update ＝σ(W u [h t-1 ,x t-1 ]+b u )
G output ＝σ(W o [h t-1 ,x t-1 ]+b o )
Figure BDA0003782851200000043
Wherein, σ is a Sigmoid activation function, and tanh is also an activation function;
step 2.3: calculating the cell state at the current time and the output h of the neuron at the current time t ：
Figure BDA0003782851200000044
h t ＝G output *tanh(C t )。
And step 3: designing a second part of a network flow prediction model based on GNN and LSTM, and GNN spatial context coding, wherein the specific steps are as follows:
step 3.1: the graph is composed of nodes and edges, and symbols are represented by G = { V, E }, where
Figure BDA0003782851200000045
For the set of base stations in the group, E = { E = { (E) i,j Is the set of edges when v i And v j In the direct vicinity, a side e is added between the two base stations i,j ；
Step 3.2: the characteristics of the edges and nodes are alternately updated, thereby enabling information to be passed through the graph. Let K be the total number of iterations, and in the kth step, v is connected i And v j Edge e of i,j Is updated to
Figure BDA0003782851200000046
Wherein
Figure BDA0003782851200000051
f edge The sensor is a multilayer sensor, and the information of the edges is updated in sequence according to the node information; when k =1, the feature vector v of each node (1) Is output h of LSTM t ；
Step 3.3: in the (k + 1) th step, updating the characteristics of the node according to the original characteristics of the node and the average characteristics of the edges adjacent to the node, specifically as follows:
Figure BDA0003782851200000052
f node the multi-layer sensor outputs the characteristic information of each base station after K iterations
Figure BDA0003782851200000053
And 4, step 4: designing a third part of a network flow prediction model based on GNN and LSTM, and decoding the LSTM, wherein the specific steps are as follows:
implicit eigenvectors per base station for the output of GNNs
Figure BDA0003782851200000054
Inputting the data into LSTM for decoding, adopting double-layer LSTM structure, and outputting predicted flow of each base station at each time step
Figure BDA0003782851200000055
And 5: training and testing the model, and specifically comprises the following steps:
step 5.1: setting the learning rate of training to be 0.01;
step 5.2: using the mean square error MSE of the predicted and actual flows of the base station as a learned error function, i.e.
Figure BDA0003782851200000056
Wherein b is i For the actual traffic of the base station,
Figure BDA0003782851200000057
is the predicted base station traffic;
step 5.3: and optimizing the network parameters by adopting an Adam optimizer, and gradually reducing the loss value.
Step 6: the system deployment and application specifically comprises the following steps:
step 6.1: collecting characteristic data of base stations in the area;
step 6.2: inputting the characteristic data into a trained GNN and LSTM-based network flow prediction model;
step 6.3: and outputting the final flow prediction result of each base station.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the intelligent power grid fusion network flow prediction method based on the GNN and the LSTM, the electric wireless private network acquisition terminal 5G is used for transmitting relevant characteristic data including the position distribution of the terminal, the type of the carried service, the bandwidth requirement, the actual bandwidth and the communication quality.
2. According to the intelligent power grid fusion network flow prediction method based on the GNN and the LSTM, time sequence features are modeled by adopting a recurrent neural network LSTM at first, and then information fusion is carried out on hidden layers of LSTM codes by adopting a graph neural network, so that the time and space correlation of base station flow can be effectively captured.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a schematic diagram of the fusion of a private power network and a 5G public network according to the present invention;
FIG. 3 is a schematic diagram of a GNN and LSTM based feature encoding network structure according to the present invention;
FIG. 4 is a schematic diagram of the LSTM recursive encoding network structure of the present invention;
fig. 5 is a schematic diagram of a base station traffic prediction decoding end LSTM network structure according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
As shown in FIG. 1, the invention provides a GNN and LSTM-based smart grid fusion network flow prediction method, which comprises the following steps:
step 1: description of features and problems;
step 1.1: a power private network and a 5G public network are fused as shown in fig. 2, and the length and the width of an area covered by the smart power grid and the coverage range of a base station are set;
step 1.2: acquiring the current required bandwidth, actual bandwidth, the number of connected terminals, service quality, coordinate position distribution of terminals, the coordinate position of a base station and grouping of the base station, determining the service quality according to the proportion of the bandwidth of the base station to the actual required bandwidth, counting the number of the terminals within 1000 meters at intervals of 100 for the coordinate position distribution of the terminals, counting the number of the terminals at 10 dimensions in total, combining the collected characteristic data to obtain a characteristic vector f with the length of 17, namely f belongs to R 17 ；
Step 1.3: defining the flow prediction of a single base station, and given a characteristic sequence F of the single base station with the length T, predicting the required flow of the next time step of the base station, wherein the problem is defined as:
Figure BDA0003782851200000071
wherein, the first and the second end of the pipe are connected with each other, a single base station signature sequence of length T is defined as F = { F 1 ,f 2 ,…,f T }；
Step 1.4: defining a group base station traffic prediction. In a converged network, 5G base stations only cover a partial area, and in a coverage area, the 5G base stations are usually deployed in groups, the base stations in the groups are adjacent to each other to complete coverage of the whole local area, and given characteristics of N base stations sampled simultaneously at the same sampling interval, the objective is to predict the traffic demand of the next S time steps for the N base stations, which is defined as:
Figure BDA0003782851200000072
and 2, step: as shown in fig. 3, a first part of a GNN and LSTM based network traffic prediction model is designed, and LSTM is recursively encoded, and the specific steps are as follows:
step 2.1: the input of the whole neural network is
Figure BDA0003782851200000073
The maximum base station number of each group is N, the length of the characteristic vector corresponding to each base station at a sampling point is 17 dimensions, firstly the characteristic vector of each time step is hidden to a high-dimensional embedding space to describe the relation between the vectors in the characteristic vector, the process is realized by linear transformation, the input is 17 dimensions, the output is 512 dimensions, and x is used t To represent;
step 2.2: as shown in fig. 4, the characteristic high dimension of each base station at each time step is embedded into x t Inputting the input data into an LSTM module, obtaining the fusion information of the sampling point and the historical sampling point at the moment through the combined action of a forgetting gate, an updating gate, an output gate and a memory unit, and receiving the output h of the neuron at the previous moment by the neuron at the current moment t-1 Embedding x with base station of current time in high dimension t Calculating out the forgetting gate G forget Output gate G for selecting information to be forgotten output Selecting information to be memorized and temporary cell states
Figure BDA0003782851200000081
G forget ＝σ(W f [h t-1 ,x t-1 ]+b f )
G update ＝σ(W u [h t-1 ,x t-1 ]+b u )
G output ＝σ(W o [h t-1 ,x t-1 ]+b o )
Figure BDA0003782851200000082
Wherein σ is a Sigmoid activation function, and tanh is the same activation function;
step 2.3: calculating the cell state and neuron output at the current timeGo out of h t ：
Figure BDA0003782851200000083
h t ＝G output *tanh(C t )
A 2-layer LSTM structure is adopted, the hidden layer of the LSTM is 512 dimensions, and the final output h of the characteristics of the time context of the base station is fused t Is 512 dimensions.
And 3, step 3: a second part of a network traffic prediction model based on GNN and LSTM is designed, GNN spatial context coding is carried out, and the specific steps are as follows:
step 3.1: obtaining a time context code h for each base station t Then, because the traffic of the adjacent base stations has strong correlation, each base station is taken as a node in the graph, the adjacent base stations are connected through edges, and then the GNN model is used for completing message flow between the nodes, so that the characteristics of the time context of the base stations can be further fused with the space context information. Let the graph be composed of nodes and edges, and the notation is G = { V, E }, where
Figure BDA0003782851200000091
For the set of base stations in the group, E = { E = { (E) i,j Is the set of edges when v i And v j In the direct vicinity, a side e is added between the two base stations i,j ；
Step 3.2: alternately updating the characteristics of the edges and the nodes so as to enable information to be transmitted in the graph, setting the total number of iterations as K, and connecting v at the kth step i And v j Edge e of i,j Is updated to
Figure BDA0003782851200000092
Wherein
Figure BDA0003782851200000093
f edge Is a multilayer perceptron, and is based on nodes in sequenceThe information of the information update side. When k =1, the feature vector v of each node (1) Is output h of LSTM t ；
Step 3.3: in the (k + 1) th step, updating the characteristics of the node according to the original characteristics of the node and the average characteristics of the edges adjacent to the node, specifically as follows:
Figure BDA0003782851200000094
f node the multi-layer sensor outputs the characteristic information of each base station after K iterations
Figure BDA0003782851200000095
In implementation f edge And f node All adopt 3 layers of linear transformation, and then follow the structure of batch normal, and the empirical value of K is 3.
And 4, step 4: the third part of the network traffic prediction model based on the GNN and the LSTM is designed, the LSTM decodes, as shown in FIG. 5, the steps are as follows:
implicit eigenvectors per base station for the output of GNNs
Figure BDA0003782851200000096
The data is input into an LSTM and decoded to output the predicted flow of each time step, a double-layer LSTM structure is adopted, and the number of units is 512.
And 5: training and testing the model;
step 5.1: setting the learning rate of training to be 0.01;
step 5.2: using the mean square error MSE of the predicted and actual flows of the base station as a learned error function, i.e.
Figure BDA0003782851200000101
Wherein b is i For the actual traffic of the base station,
Figure BDA0003782851200000102
is the predicted base station traffic;
step 5.3: and optimizing the network parameters by adopting an Adam optimizer, and gradually reducing the loss value.
Step 6: the system deployment and application specifically comprises the following steps:
step 6.1: collecting characteristic data of base stations in the area;
step 6.2: inputting the characteristic data into a trained GNN and LSTM-based network flow prediction model;
step 6.3: and outputting the final flow prediction result of each base station.
Examples
The length and width of an area covered by the smart grid are 10000 meters, the coverage range of the base stations is 300 meters, the bandwidth of the base stations is 1G, the number of the base stations in each group is uniformly distributed, the range is (9, 49), the base stations are continuously deployed to completely cover the area, and the whole area comprises 15 groups of base stations. The types of the terminals in the area range are 3 types, the first type is a mobile terminal type, the mobile terminal conforms to an RWP model, the flight speed of an unmanned aerial vehicle is referred to, the moving speed V of the terminal ranges from 10 meters per second to 25 meters per second, the longest operation time Pmax of the terminal in a place is 1-10 minutes, V and Pmax are uniformly sampled, the mobile terminal mainly bears routing inspection services, the bandwidth required by each terminal is 10M, and the coverage area is 2000 mobile terminals in total. The second type is a video monitoring service of a fixed terminal, the required bandwidth of each terminal is 10M, and the system has 20 terminals of the type. The third type is an acquisition type terminal, which acquires 1 time every 10 minutes, and the environment has 200 terminals in total, and the bandwidth required by each acquisition is 1M.
Each terminal sends 5G communication data to the smart grid processing center every 1 minute, the communication data comprise currently connected base stations, the smart grid processing center performs statistics by taking the base stations as a unit to obtain the characteristic vector of each base station, the data are specifically the current required bandwidth, the actual bandwidth, the number of connected terminals, the service quality, the coordinate position distribution of the terminals, the coordinate position of the base stations and the grouping of the base stations, and for the service quality, the data are determined according to the ratio of the bandwidth of the base stations to the actual required bandwidth. For the position distribution of the terminals, counting the number of the terminals within 1000 meters at intervals of 100, and finally obtaining the dimension of the feature vector of each base station as 17.
After sampling for many times, the training sample and the test sample are divided according to a certain proportion, and compared with the current mainstream time sequence learning algorithm in order to verify the effectiveness of the method, wherein the training sample and the test sample are respectively a Linear recurrent neural network (Linear RNN), an LSTM and a Seq2Seq LSTM. Wherein, the Linear RNN directly adopts a Linear function to model the historical state and the current input, and for the Linear RNN and the LSTM, the state h which is corresponding to the last time step t and is packaged with the historical information is directly used t It is decoded into the predicted traffic of the following 4 consecutive time steps, and Seq2Seq LSTM is also based on LSTM. Based on the completely same training set and test data, the prediction of the flow of each method for the next 4 time steps is obtained, table 1 shows the comparison of the prediction error results of the 3 methods and the method provided by the invention, and as can be seen from table 1, the prediction result of the method provided by the invention is obviously improved compared with other methods.
TABLE 1 Bandwidth error for 4 time steps predicted by different methods
Algorithm Step 1 Step 2 Step 3 Step 4
Linear RNN 19.13 22.53 65.22 80.09
LSTM 13.01 15.41 41.33 48.53
Seq2Seq-LSTM 11.78 13.42 35.77 39.29
GNN-LSTM 8.39 10.02 29.11 31.05
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims (7)

1. A GNN and LSTM-based intelligent power grid fusion network flow prediction method is characterized by comprising the following steps:
step 1: the intelligent power grid processing center collects characteristic data related to the transmission of the terminals and the 5G through the electric power wireless private network, wherein the characteristic data comprises the current required bandwidth of the base station, the actual bandwidth, the number of connected terminals, the service quality, the coordinate position distribution of the terminals, the coordinate position of the base station and the grouping of the base station;
and 2, step: the traffic characteristics of a base station are taken as input, a network traffic prediction model based on GNN and LSTM is designed, and the model consists of a first part of LSTM recursive encoding, a second part of GNN spatial context encoding and a third part of LSTM decoding;
and step 3: training a network flow prediction model based on GNN and LSTM, testing on a test set, and deploying the trained model to a smart grid processing center;
and 4, step 4: and finally, collecting characteristic data of the base stations in the area, inputting the characteristic data into the trained network traffic prediction model, and obtaining a final traffic prediction result.
2. The intelligent power grid fusion network flow prediction method based on GNN and LSTM as claimed in claim 1, wherein: the method based on the characteristic description in the step 1 comprises the following steps:
step 1: setting the length and width of an area covered by the smart power grid and the coverage range of a base station;
step 2: and acquiring the current required bandwidth, actual bandwidth, the number of connected terminals, service quality, coordinate position distribution of the terminals, the coordinate position of the base station and grouping of the base station to form a characteristic vector f.
3. The intelligent power grid fusion network flow prediction method based on GNN and LSTM according to claim 1, wherein: based on the LSTM recursive encoding in step 2, the method comprises the following steps:
step 1: the input of the whole neural network is
Figure FDA0003782851190000011
The maximum base station number of each group is N, T is the length of the characteristic sequence, the characteristic vector of each time step is subjected to hidden projection to a high-dimensional embedding space through linear transformation, the input dimension is 17 dimensions, the output dimension is 512 dimensions, and x is used t To represent;
step 2: the LSTM neuron at the current moment receives the output h of the neuron at the previous moment t-1 High-dimensional embedding x with current time t Calculating the forgetting gate G forget Output gate G for selecting information to be forgotten output Selecting information to be memorized and a temporary cell state
Figure FDA0003782851190000021
G forget ＝σ(W f [h t-1 ,x t-1 ]+b f )
G update ＝σ(W u [h t-1 ,x t-1 ]+b u )
G output ＝σ(W o [h t-1 ,x t-1 ]+b o )
Figure FDA0003782851190000022
Wherein σ is a Sigmoid activation function, and tanh is the same activation function;
and step 3: calculating the cell state at the current time and the output h of the neuron at the current time t ：
Figure FDA0003782851190000023
h t ＝G output *tanh(C t )。
4. The intelligent power grid fusion network flow prediction method based on GNN and LSTM according to claim 1, wherein: the GNN spatial context coding in step 2 comprises the following steps:
step 1: the graph is composed of nodes and edges, and symbols are represented by G = { V, E }, where
Figure FDA0003782851190000024
For the set of base stations in the group, E = { E = { (E) i,j Is the set of edges when v i And v j When directly adjacent, add edge e between the two base stations i,j ；
Step 2: the characteristics of the edges and nodes are alternately updated, thereby enabling information to be passed through the graph. Let K be the total number of iterations, and in the kth step, v is connected i And v j Edge e of i,j Is updated to
Figure FDA0003782851190000025
And step 3: in the (k + 1) th step, updating the characteristics of the node according to the original characteristics of the node and the average characteristics of the edges adjacent to the node, specifically as follows:
Figure FDA0003782851190000026
f node the multi-layer sensor outputs the characteristic information of each base station after K iterations
Figure FDA0003782851190000027
5. The intelligent power grid fusion network flow prediction method based on GNN and LSTM according to claim 1, wherein the LSTM decoding in step 2 specifically comprises: implicit eigenvectors v per base station for the output of GNNs i K Inputting the data into LSTM for decoding, adopting double-layer LSTM structure, outputting the predicted flow of each base station at each time step
Figure FDA0003782851190000031
6. The intelligent power grid fusion network flow prediction method based on GNN and LSTM as claimed in claim 1, wherein: the training and testing model based on the step 3 comprises the following steps:
step 1: setting the learning rate of training to be 0.01;
step 2: using the mean square error MSE of the predicted and actual flows of the base station as a learned error function, i.e.
Figure FDA0003782851190000032
Wherein b is i For the actual traffic of the base station,
Figure FDA0003782851190000033
is the predicted base station traffic;
and step 3: and optimizing the network parameters by adopting an Adam optimizer, and gradually reducing the loss value.
7. The intelligent power grid fusion network flow prediction method based on GNN and LSTM according to claim 1, wherein the flow prediction in step 4 specifically comprises the following steps:
step 1: collecting characteristic data of base stations in the area;
step 2: inputting the characteristic data into a trained network flow prediction model based on GNN and LSTM;
and step 3: and outputting the final flow prediction result of each base station.
CN202210934264.9A 2022-08-04 2022-08-04 Intelligent power grid fusion network flow prediction method based on GNN and LSTM Pending CN115314925A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210934264.9A CN115314925A (en) 2022-08-04 2022-08-04 Intelligent power grid fusion network flow prediction method based on GNN and LSTM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210934264.9A CN115314925A (en) 2022-08-04 2022-08-04 Intelligent power grid fusion network flow prediction method based on GNN and LSTM

Publications (1)

Publication Number Publication Date
CN115314925A true CN115314925A (en) 2022-11-08

Family

ID=83858800

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210934264.9A Pending CN115314925A (en) 2022-08-04 2022-08-04 Intelligent power grid fusion network flow prediction method based on GNN and LSTM

Country Status (1)

Country Link
CN (1) CN115314925A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115987816A (en) * 2022-12-15 2023-04-18 中国联合网络通信集团有限公司 Network flow prediction method and device, electronic equipment and readable storage medium
CN116153069A (en) * 2023-02-09 2023-05-23 东南大学 Traffic flow model and data fusion driven traffic state estimation method and device

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115987816A (en) * 2022-12-15 2023-04-18 中国联合网络通信集团有限公司 Network flow prediction method and device, electronic equipment and readable storage medium
CN116153069A (en) * 2023-02-09 2023-05-23 东南大学 Traffic flow model and data fusion driven traffic state estimation method and device
CN116153069B (en) * 2023-02-09 2024-01-30 东南大学 Traffic flow model and data fusion driven traffic state estimation method and device

Similar Documents

Publication Publication Date Title
CN108053080B (en) Method, device, equipment and medium for predicting regional user number statistics
CN115314925A (en) Intelligent power grid fusion network flow prediction method based on GNN and LSTM
CN112902969B (en) Path planning method of unmanned aerial vehicle in data collection process
CN108123828B (en) Ultra-dense network resource allocation method based on access user mobility prediction
CN108134979B (en) Small base station switch control method based on deep neural network
CN111246552B (en) Base station dormancy method based on mobile network flow prediction
CN111343650B (en) Urban scale wireless service flow prediction method based on cross-domain data and loss resistance
CN112819236B (en) Low-orbit satellite Internet of things active terminal prediction method based on neural network
CN113099474B (en) Method and device for predicting short-term traffic demand of satellite internet user terminal
CN115278698B (en) Unmanned aerial vehicle base station dynamic deployment method and device based on dynamic user distribution prediction
Shinkuma et al. Data assessment and prioritization in mobile networks for real-time prediction of spatial information with machine learning
CN113610329B (en) Short-time rainfall approaching forecasting method of double-current convolution long-short term memory network
Gu et al. A Spatial-Temporal Transformer Network for City-Level Cellular Traffic Analysis and Prediction
CN116723527B (en) Base station power consumption optimization method based on digital twin
CN116961805A (en) Short wave prediction method based on space-time diagram neural network
Martini et al. Challenges and perspectives in neuromorphic-based visual IoT systems and networks
Zhong et al. Traffic prediction of converged network for smart gird based on GNN and LSTM
Lou et al. Terrain-Based UAV Deployment: Providing Coverage for Outdoor Users
Abdullah et al. Weather-aware fiber-wireless traffic prediction using graph convolutional networks
Plazas et al. Proposal of a computational intelligence prediction model based on Internet of Things technologies
Eller et al. Propagation-aware Gaussian process regression for signal-strength interpolation along street canyons
Hussein Data Reduction Algorithms for Wireless Sensor Networks Applications
Pattanayak et al. Identification of spectrum holes using ANN model for cognitive radio applications
Subramanian et al. A study of mobile user movements prediction methods
Wang et al. Deep Learning Based Traffic Prediction in Mobile Network-A Survey

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination