CN112508408A - Mapping model construction method of wireless resource management index under edge calculation - Google Patents

Mapping model construction method of wireless resource management index under edge calculation Download PDF

Info

Publication number
CN112508408A
CN112508408A CN202011434366.1A CN202011434366A CN112508408A CN 112508408 A CN112508408 A CN 112508408A CN 202011434366 A CN202011434366 A CN 202011434366A CN 112508408 A CN112508408 A CN 112508408A
Authority
CN
China
Prior art keywords
social data
mapping
fuzzy
resource management
data
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.)
Granted
Application number
CN202011434366.1A
Other languages
Chinese (zh)
Other versions
CN112508408B (en
Inventor
林福宏
白亚莉
郭晋宁
周成成
李永军
赵玉萍
安凤平
许海涛
安建伟
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.)
Qinghai Zhongmeidi Mining Development Co ltd
Peking University
University of Science and Technology Beijing USTB
Huaiyin Normal University
Original Assignee
Qinghai Zhongmeidi Mining Development Co ltd
Peking University
University of Science and Technology Beijing USTB
Huaiyin Normal University
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 Qinghai Zhongmeidi Mining Development Co ltd, Peking University, University of Science and Technology Beijing USTB, Huaiyin Normal University filed Critical Qinghai Zhongmeidi Mining Development Co ltd
Priority to CN202011434366.1A priority Critical patent/CN112508408B/en
Publication of CN112508408A publication Critical patent/CN112508408A/en
Application granted granted Critical
Publication of CN112508408B publication Critical patent/CN112508408B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G06F18/24155—Bayesian classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y10/00—Economic sectors
    • G16Y10/75—Information technology; Communication
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/30—Control
    • G16Y40/35—Management of things, i.e. controlling in accordance with a policy or in order to achieve specified objectives

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Development Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Educational Administration (AREA)
  • Software Systems (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Operations Research (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Medical Informatics (AREA)
  • Mathematical Physics (AREA)
  • Accounting & Taxation (AREA)
  • Probability & Statistics with Applications (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method for constructing a mapping model of a wireless resource management index under edge calculation, which comprises the following steps: preprocessing original social data of the edge terminal based on an improved machine learning classification technology to obtain processed social data; mapping the social data based on a clustering analysis technology, and mapping the social data into information directly reflecting physical layer parameters; establishing a mathematical model corresponding to social data distribution and performing associated mapping to obtain a probability distribution model of physical layer parameters with the social data as variables; distributing different weights to QoS indexes of social data from different sources, and performing weight domain standard normalization processing; fuzzification processing is carried out on the standardized weight; de-fuzzification is realized on the fuzzified QoS index, and a fuzzy weight value is obtained; and selecting a mapping function of the QoE index according to the service requirement, and using the obtained fuzzy weight value in the mapping function to obtain the mapping weight value of the QoE index as a wireless resource management index.

Description

Mapping model construction method of wireless resource management index under edge calculation
Technical Field
The invention relates to the technical field of edge calculation and artificial intelligence, in particular to a method for constructing a mapping model of a wireless resource management index under edge calculation.
Background
In recent years, with the rapid development of artificial intelligence and internet of things, a series of new application scenes are promoted, and the demand industries such as smart home, medical health, vehicle-mounted communication, urban informatization and the like are increased in a large scale. Machine Type Communication (MTC) has undergone blowout-Type development, and the number of devices accessing a network and the amount of data generated by the devices have rapidly increased. However, the centralized data processing mode based on cloud computing is difficult to meet the service requirements of terminal equipment such as real-time performance and low power consumption, and therefore edge computing is in force. Since the resources of the edge computing environment are limited and the terminal devices are always carried and used by people, the social data of the terminal users naturally become an important basis for the management of the wireless resource allocation. How to efficiently and reasonably allocate wireless resources in an edge computing environment and how to ensure the experience priority of an edge terminal user in the resource using process to the maximum extent is very valuable for exploration and research.
In an edge computing environment, mass connectivity machine type communication has the following features: the transmission data volume is small and is infrequent; there is a large difference in Quality of service (QoS) between different device classes. The massive device connection and diversified services of the internet of things bring new technical challenges to wireless resource allocation in the existing edge computing environment. The existing edge computing wireless resource management takes one or more independent QoS indexes as an optimization target, objective QoS indexes cannot completely reflect the subjective feeling of a user on the provided service, and part of the existing QoE indexes cannot completely and comprehensively consider the user experience. In the edge computing environment, especially due to the heterogeneity of the network and the diversity of the services, the traditional radio resource management index lacks pertinence, so that the optimization effect of the radio resource management model is limited, and even unnecessary resource waste is caused. Therefore, in the current edge computing environment, it is important to provide a QoE mapping model for edge computing because it should be the final basis for radio resource management to improve the QoE (Quality of experience).
Disclosure of Invention
The invention aims to overcome the defects in the existing wireless resource management technology, and provides a mapping model construction method of wireless resource management indexes under edge computing, which can perform data mining on user social data under an edge computing environment, obtain physical layer parameter model distribution and a QoS-QoE mapping model with the social data as variables, and take the physical layer parameter model distribution and the QoS-QoE mapping model as indexes of edge computing wireless resource management, highlight the prior experience of edge users, and better meet the requirements of heterogeneous type and service diversity of a network.
To solve the above technical problem, an embodiment of the present invention provides the following solutions:
a mapping model construction method of wireless resource management indexes under edge calculation comprises the following steps:
acquiring original social data of edge terminal equipment;
preprocessing the original social data based on an improved machine learning classification technology to obtain processed social data, wherein the social data comprises: social data of individual users and social data among users;
mapping the social data based on an association analysis and cluster analysis technology, and mapping the social data into information directly reflecting physical layer parameters;
establishing a mathematical model corresponding to social data distribution and performing associated mapping to obtain a probability distribution model of physical layer parameters with the social data as variables, and taking the probability distribution model as an optimization target of a wireless resource management problem;
distributing different weights to QoS indexes of social data from different sources, and performing standard normalization processing of weight discourse domain;
fuzzifying the standardized weight;
selecting a proper membership function for the QoS index after fuzzification processing, and realizing defuzzification according to a membership maximization principle to obtain a fuzzy weight value;
and selecting a mapping function of the QoE index according to the service requirement, and using the obtained fuzzy weight value in the mapping function to obtain the mapping weight value of the QoE index as a wireless resource management index.
Preferably, the preprocessing the raw social data based on the improved machine learning classification technique specifically includes:
improving a naive Bayes algorithm, and classifying the input social data based on the improved naive Bayes algorithm;
the data are equalized by utilizing an oversampling technology and based on a machine learning algorithm KNN, and the steps are as follows:
calculating Euclidean distances between a few classes of samples:
Figure BDA0002827649830000021
in the formula xi, xjRespectively representing minority samples, wherein n is the number of the minority samples, and the value of n is 1 and 2;
setting the K value of the KNN nearest neighbor to be 5, selecting 5 minimum values from the Euclidean distance set table obtained by the formula, and selecting 10 data values in the samples from the selected 5 nearest neighbor samples in a random mode;
the data in the original sample and the 10 data values are processed as follows to generate a new sample value:
xnew＝xi+rand(0,1)×dist
and obtaining the social data of the individual users and the social data among the users after preprocessing.
Preferably, the mapping social data based on the association analysis and cluster analysis technology, the mapping social data to information directly reflecting physical layer parameters specifically includes:
determining whether the corresponding social data can be directly converted with the QoS index or not according to whether the social data from the user individuals and among the users are associated with the physical layer parameters or not;
for the social data capable of being converted, mapping the social data into information directly reflecting physical layer parameters based on a cluster analysis technology;
the social networking level feature vector formed by the social data before mapping is alpha, and the network after mapping isThe physical layer parameter feature vector is beta, and the set of the social layer feature vectors before mapping is D (alpha)1,α2,...,αK) The feature vector set after the clustering analysis is as follows:
Figure BDA0002827649830000031
preferably, the mapping social data to information directly reflecting physical level parameters specifically includes:
the social data is mapped to information related to time, distance, speed, location, signal strength that directly reflects physical level parameters.
Preferably, the allocating different weights to the QoS indicators of the social data from different sources, and the performing standard normalization processing of the weight discourse domain specifically includes:
according to the source of the obtained social data: from the social network, from the big data, from the physical layer monitoring results of the edge environment, different weights are allocated to the QoS indexes of the social data;
the weight value of the QoS index distribution of the three sources is u1,u2,u3Wherein u is1,u2,u3Satisfies the following conditions:
u1+u2+u3＝1
in order to highlight the experience of the terminal user, the QoS index which is set from the big data and reflects the differentiated communication characteristics of the user occupies higher weight.
Preferably, the QoS indicator of each source is further classified according to the network model of its physical layer, and is divided into: an application layer QoS index, a network layer QoS index, and an underlying QoS index.
Preferably, the fuzzifying the normalized weight specifically includes:
the method comprises the steps that a fuzzy operator of weighted average is used, the influence of each index on a model is comprehensively considered, all information of single index decision is kept, and a fuzzy weight set of each index is constructed based on an improved analytic hierarchy process;
and constructing a target-factor judgment matrix to check the reliability of the fuzzy weight set.
Preferably, the selecting a proper membership function for the QoS index after fuzzification processing, and implementing defuzzification according to a membership maximization principle, to obtain a fuzzy weight value specifically includes:
according to a fuzzy rule, selecting a proper membership function for the QoS index after fuzzification processing, and calculating the membership function to obtain a fuzzy matrix R with membership as an element;
and performing fuzzy operation on the domain weight U and the fuzzy matrix R to obtain output after applying a fuzzy theory:
Figure BDA0002827649830000041
and (4) according to the membership maximization principle, defuzzification is realized to obtain a fuzzy weight value.
Preferably, the membership functions are obtained in the following way:
obtaining a membership function according to a fuzzy statistical experiment method;
or building a neural network training membership function, and realizing mapping from the influencing factors to membership degrees by building a three-layer convolutional neural network.
Preferably, the mapping function of the QoE metric includes: non-real-time service description based on a logarithmic function; real-time service description based on an exponential function; voice service description based on a linear function; and a comprehensive service description based on the piecewise function and the multiple functions.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
the mapping model construction method of the wireless resource management index in the edge computing environment can perform data mining on the user social data in the edge computing environment, obtain the physical layer parameter model distribution taking the social data as a variable and the QoS-QoE mapping model, and serve as the index of the edge computing wireless resource management, so that the priority experience of an edge user can be highlighted, and the requirements of the heterogeneous type and the business diversity of a network can be met.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of a method for constructing a mapping model of a radio resource management indicator under edge computing according to an embodiment of the present invention;
FIG. 2 is a flow chart of the pre-processing of raw social data based on the improved NB algorithm provided by the embodiments of the present invention;
FIG. 3 is a flow chart of determining QoS index input domain weight in edge computing environment according to the present invention;
fig. 4 is a flowchart of mapping model weight determination based on fuzzy mathematics provided by the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
An embodiment of the present invention provides a method for constructing a mapping model of a radio resource management indicator under edge computing, as shown in fig. 1, the method includes the following steps:
acquiring original social data of edge terminal equipment;
preprocessing the original social data based on an improved machine learning classification technology to obtain processed social data, wherein the social data comprises: social data of individual users and social data among users; the social data of the user individuals refers to habits, preferences and the like of the users obtained based on mining technology; social data among users is obtained by performing association analysis on the social network.
Mapping the social data based on an association analysis and cluster analysis technology, and mapping the social data into information directly reflecting physical layer parameters;
establishing a mathematical model corresponding to social data distribution and performing associated mapping to obtain a probability distribution model of physical layer parameters with the social data as variables, and taking the probability distribution model as an optimization target of a wireless resource management problem;
the QoE index is the synthesis of multiple QoS indexes, different weights are distributed to the QoS indexes of the social data from different sources, and standard normalization processing of weight discourse domains is carried out;
fuzzifying the standardized weight, and fuzzifying the input numerical value according to different service types and different time;
selecting a proper membership function for the QoS index after fuzzification processing, and realizing defuzzification according to a membership maximization principle to obtain a fuzzy weight value;
and selecting a mapping function of the QoE index according to the service requirement, and using the obtained fuzzy weight value in the mapping function to obtain the mapping weight value of the QoE index as a wireless resource management index.
The method can carry out data mining on the user social data under the edge computing environment, obtain the physical layer parameter model distribution and the QoS-QoE mapping model which take the social data as variables, and take the physical layer parameter model distribution and the QoS-QoE mapping model as the index of edge computing wireless resource management, so that the method not only can highlight the prior experience of edge users, but also can meet the requirements of the heterogeneous type and the service diversity of a network.
As a specific embodiment of the present invention, a specific construction method of the radio resource management indicator mapping model in the edge computing environment is as follows:
step 1, obtaining original social data on edge terminal equipment.
And 2, preprocessing the original social data.
As shown in fig. 2, the traditional Naive Bayes (NB) algorithm is improved by the implementation of the present invention, the input original social data is classified based on the improved NB algorithm, and the improved NB algorithm can effectively solve the problem of classification result deviation caused by the inability of the original algorithm to learn features of each category equally. The data are equalized by utilizing an oversampling technology based on a machine learning algorithm-KNN (K-Nearest Neighbor), and the steps are as follows:
first, the euclidean distances between the minority class samples are calculated:
Figure BDA0002827649830000061
in the formula xi，xjRespectively represent a minority class sample, n is the number of the minority class samples, and the value of n is 1 and 2.
Then, setting the K value nearest to KNN to be 5, and selecting 5 minimum values according to the Euclidean distance set table. Selecting 10 data values in the samples from 5 nearest neighbor samples in a random mode, and generating new sample values by processing the data in the original samples and the 10 data values as follows:
xnew＝xi+rand(0,1)×dist
and obtaining the social data of the individual users and the social data among the users after preprocessing.
And step 3, as shown in fig. 3, performing cluster analysis and weight normalization processing on the preprocessed social data. And (3) dividing the social data obtained in the step (2) into information which is related to time, distance, speed, position, signal strength and the like and directly reflects physical layer parameters.
First, social data is divided into three sources: big data, social network, and network layer monitoring results. And respectively assigning weights u to the three categories1,u2,u3The sum of the weights is 1,
Figure BDA0002827649830000062
in the second classification, data sources under this property are classified as time-related, position-related, distance-related, velocity-related, and the like. The data source under the social network property is set to be L, the data source under the big data property is set to be M, the data source under the network layer monitoring property is set to be K, and the following parameter constraints are respectively met:
u1＝u1,1+u1,2+...+u1,L-1+u1,L
u2＝u2,1+u2,2+...+u2,M-1+u2,M
u3＝u3,1+u3,2+...+u3,k-1+u3,k
and 4, regarding the intimacy among the users as directly reflecting the comprehensive channel quality information, and defining a channel gain coefficient h of a comprehensive physical layer link and a social layer link.
Wherein, with h1The channel quality information of the actual physical layer link is represented, and the quality of the physical layer link is embodied; h is2And reflecting the channel quality information of the link of the social layer, wherein the channel quality information is determined by the social relationship between the source node and the target node. The comprehensive channel quality shows that the final communication performance is determined by the physical level and the social level together, and the deterioration of the gain of any one of the two can cause the performance of the system to be reduced, thereby directly reflecting the user experience quality:
h＝h1×h2
Gh(dB)＝Gh1(dB)+Gh2(dB)
channel quality gain G integrating physical and social levelsh(dB) can also be used for describing physical layer parameter indexes such as data transmission rate, cell coverage, transmission power and the like, and the indexes also contain social information.
Step 5, as shown in fig. 4, the embodiment of the present invention constructs a mapping model based on fuzzy mathematics, and forms an influencing factor and channel quality set U, V.
As can be seen from step 3, U ═ U1,u2,u3The first layer is the influence factor set of big data, social network and network layer monitoring results, wherein uiIs composed of the following set:
u2＝{u21,u22,...,u2M-1,u2M}
the second layer is a social network property with a data source of L, a big data property with a data source of M, and a network layer monitoring property with a data source of K. Set of channel qualities V ═ V1,v2,v3,v4In which v isjJ-1, 2,3,4 respectively represent that the channel quality is very good, better, and generally poor.
And 6, in order to more remarkably reflect the experience of the terminal user, the mapping model needs to set a QoS index which is from big data and reflects the differentiated communication characteristics of the user to occupy higher weight. Wherein A is(1)For the first layer influencing factor fuzzy weight set, A(2)For the second layer of contributor fuzzy weight sets, each layer of contributor fuzzy weight set is represented as follows:
A(1)＝(a1,a2,a3)
Figure BDA0002827649830000081
A1 (2)＝(a11,a12,...,a1L)
A2 (2)＝(a21,a22,...,a2M)
A3 (2)＝(a31,a32,...,a3K)
step 7, the embodiment of the invention is based on the principle of maximum membership, and for a fuzzy subset A on a domain of discourse U, i fuzzy sub-objects V are provided1,V2,......ViPreferentially determining V under the fuzzy constraint condition of AiThe following conditions are satisfied:
Figure BDA0002827649830000082
by uiOne-factor formed evaluation vector RiThe evaluation matrix formed based on the membership functions is:
Figure BDA0002827649830000083
wherein R isi＝[ri1 ri2 ri3 ri4]Channel quality mapping values corresponding to the respective influencing factors, wherein
Figure BDA0002827649830000084
And 8, constructing a fuzzy decision matrix according to the influence factor weight set of each layer in the step 6. The fuzzy operator used in the embodiment of the invention is a weighted average type, the influence of each factor on the model is comprehensively considered, and simultaneously, all information of single factor decision is reserved.
Making a decision according to the first-layer influence factor set, wherein the decision object is the membership degree of elements in the decision set, and then obtaining a first-level fuzzy decision matrix B(1)Comprises the following steps:
Figure BDA0002827649830000085
according to the result of the primary decision matrix, carrying out comprehensive decision on the inner-layer influence factor set, carrying out comprehensive decision among all factors, and carrying out R of the primary fuzzy weight decision(1)A two-stage fuzzy weight decision matrix R with single-factor set membership(2)As a result of a first-level fuzzy decision, i.e. R(2)＝B(1)Deriving a two-level fuzzy weight decision matrix:
Figure BDA0002827649830000091
and 9, constructing a fuzzy weight set of each factor based on an improved analytic hierarchy process. Targeting a, the set of influence factors is U ═ U1,u2,u3,u4}，uijIs uiFor u is pairedjUsing scale discrete values 1-9 to represent the factor u respectivelyiAnd ujThe degree of importance of the comparison is positively correlated. For each uiPerforming the analysis to obtain a judgment matrix W which is formed into an A-U judgment matrix:
Figure BDA0002827649830000092
from the W matrix, the overall standard deviation is calculated:
Figure BDA0002827649830000093
if deltaijWhen all the weights are less than 1, the comprehensive decision of each factor can be considered to have uniformity, and at the moment, the invention uses the arithmetic mean value of the secondary fuzzy weight decision matrix as the final weight decision value, namely B(2)＝[bij]max(L,K,M)Wherein
Figure BDA0002827649830000094
To check the consistency of the results, the IAHP (improved analysis technology) method is used to construct the matrix W*＝[wij *]max(L,M,K)Wherein
Figure BDA0002827649830000095
And obtaining a vector corresponding to the maximum eigenvalue of the construction matrix, namely a fuzzy weight judgment result of each influence factor, and translating by standard deviation to obtain a normalized relative weight result.
If delta existsijIf the decision weight is more than or equal to 1, the comprehensive decision of each factor is not consistent, and the arithmetic mean of the decision weight obtained in the step 8 as the final decision result is unreliable. The invention calculates an optimal transfer matrix, wherein max is max (L, M, K):
Figure BDA0002827649830000096
obtaining the optimal transfer matrix B ═ B with the smallest J based on the extreme methodij]max×max. Reference deltaijAnd if the weight is less than 1, the weight of each influence factor can be obtained, and the relative weight value is obtained after normalization.
Step 10, according to step 7, determining a membership function based on the fuzzy rule, wherein the membership function determines r in the evaluation matrixij(i-1, 2, 3; j-1, 2,3, 4). The embodiment of the invention adopts the following 2 comprehensive ways to obtain the membership functions for different state quantities:
1) obtaining a membership function according to a fuzzy statistical experiment method, wherein the experiment method mainly aims at channel influence factors with high use frequency, and the calculation of membership has certain subjective properties;
2) and (3) building a neural network training membership function, and building three layers of convolutional neural networks to complete mapping from influencing factors to membership. When a bp (back deployment) network is built, the number of hidden layer nodes can be according to a formula: i + o + a. Wherein i and o are the unit numbers of the input layer and the output layer respectively, and c is a value between 1 and 10.
And 11, outputting the membership function and the judgment matrix weight obtained in the steps 9 and 10, and obtaining a corresponding fuzzy weight output value according to the decision matrix calculation formula in the step 8.
Step 12, applying the fuzzy weight value to a mapping function of QoE, including the following: non-real-time service description based on a logarithmic function; real-time service description based on an exponential function; voice service description based on a linear function; and a comprehensive service description based on the piecewise function and the multi-function; the method is used for describing a plurality of QoS indexes, and finally mapping from the QoS indexes to the QoE indexes reflecting user differences is obtained.
In conclusion, the invention can perform data mining on the user social data in the edge computing environment, obtain the physical layer parameter model distribution and the QoS-QoE mapping model which take the social data as variables, and take the physical layer parameter model distribution and the QoS-QoE mapping model as the indexes of the edge computing wireless resource management, highlight the prior experience of the edge user, and better meet the requirements of the heterogeneous type and the service diversity of the network.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (10)

1.一种边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，包括以下步骤：1. a mapping model construction method of radio resource management index under edge computing, is characterized in that, comprises the following steps: 获取边缘终端设备的原始社交数据；Obtain raw social data of edge terminal devices; 基于改进的机器学习分类技术对所述原始社交数据进行预处理，得到处理后的社交数据，所述社交数据包括：用户个体的社交数据和用户间的社交数据；The original social data is preprocessed based on the improved machine learning classification technology to obtain the processed social data, and the social data includes: the social data of individual users and the social data between users; 基于关联分析及聚类分析技术对社交数据进行映射，将社交数据映射为直接反映物理层面参数的信息；Mapping social data based on association analysis and cluster analysis technology, and mapping social data into information that directly reflects parameters at the physical level; 建立社交数据分布相应的数学模型并进行关联映射，获得以社交数据为变量的物理层面参数的概率分布模型，将其作为无线资源管理问题的优化目标；Establish a corresponding mathematical model of social data distribution and perform correlation mapping to obtain a probability distribution model of physical-level parameters with social data as a variable, and use it as the optimization goal of the wireless resource management problem; 对不同来源的社交数据的QoS指标分配不同的权值，进行权值论域的标准归一化处理；Assign different weights to the QoS indicators of social data from different sources, and perform standard normalization of the weight universe; 对标准化的权值进行模糊化处理；Fuzzy the standardized weights; 对模糊化处理后的QoS指标选择合适的隶属函数，并根据隶属度最大化原则实现去模糊化，得到模糊权重值；Select the appropriate membership function for the QoS index after fuzzification, and realize the defuzzification according to the principle of maximizing membership degree, and obtain the fuzzy weight value; 根据业务需求选取QoE指标的映射函数，将得到的模糊权重值用于所述映射函数，得到QoE指标的映射权重值，作为无线资源管理指标。The mapping function of the QoE indicator is selected according to the service requirements, and the obtained fuzzy weight value is used in the mapping function to obtain the mapping weight value of the QoE indicator, which is used as the radio resource management indicator. 2.根据权利要求1所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，所述基于改进的机器学习分类技术对所述原始社交数据进行预处理具体包括：2. The method for constructing a mapping model of wireless resource management indicators under edge computing according to claim 1, wherein the preprocessing of the original social data based on the improved machine learning classification technology specifically comprises: 对朴素贝叶斯算法进行改进，基于改进的朴素贝叶斯算法对输入的社交数据进行分类；Improve the Naive Bayes algorithm, and classify the input social data based on the improved Naive Bayes algorithm; 利用过采样技术，基于机器学习算法KNN对数据进行均衡处理，步骤如下：Using oversampling technology, the data is balanced based on the machine learning algorithm KNN. The steps are as follows: 计算少数类样本之间的欧氏距离：Compute the Euclidean distance between minority class samples:
Figure FDA0002827649820000011
Figure FDA0002827649820000011
 式中xi，xj分别代表少数类样本，n为少数类样本个数，n取值为1，2；In the formula, x i , x j represent the minority class samples respectively, n is the number of minority class samples, and n is 1, 2; 设置KNN最邻近的K值为5，从上述式子所得的欧式距离集合表中选取5个最小值，从选取的5个最近邻样本中以随机方式选择样本中的10个数据值；Set the K value of the KNN nearest neighbor to 5, select 5 minimum values from the Euclidean distance set table obtained by the above formula, and randomly select 10 data values in the sample from the selected 5 nearest neighbor samples; 将原样本中的数据与这10个数据值作如下处理产生新样本值：The data in the original sample and these 10 data values are processed as follows to generate new sample values: xnew＝xi+rand(0,1)×distx new = x i +rand(0,1)×dist 预处理后得到用户个体及用户间的社交数据。After preprocessing, the social data of individual users and users is obtained. 3.根据权利要求1所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，所述基于关联分析及聚类分析技术对社交数据进行映射，将社交数据映射为直接反映物理层面参数的信息具体包括：3. The method for constructing a mapping model of wireless resource management indicators under edge computing according to claim 1, wherein the social data is mapped based on association analysis and cluster analysis technology, and the social data is mapped to directly reflect physical data. The information of the layer parameters specifically includes: 根据来自于用户个体及用户间的社交数据是否与物理层面参数相关联，决定相应的社交数据是否能够直接与QoS指标相转化；Determine whether the corresponding social data can be directly converted to QoS indicators according to whether the social data from individual users and between users is associated with physical-level parameters; 对于能够转化的社交数据，基于聚类分析技术，将社交数据映射为直接反映物理层面参数的信息；For social data that can be transformed, based on cluster analysis technology, the social data is mapped to information that directly reflects the parameters at the physical level; 映射前社交数据构成的网络社交层面特征向量为α，映射后网络物理层面参数特征向量为β，映射前网络社交层面特征向量集合为D(α1,α2,...,αK)，经过聚类分析后的特征向量集合为：The network social level feature vector composed of social data before mapping is α, the network physical level parameter feature vector after mapping is β, and the network social level feature vector set before mapping is D(α 1 ,α 2 ,...,α K ), The set of eigenvectors after cluster analysis is: D(α1,1,α1,2,...,α1,L)D(α 1,1 ,α 1,2 ,...,α 1,L ) D(α2,1,α2,2,...,α2,M)其中L+M+N＝KD(α 2,1 ,α 2,2 ,...,α 2,M ) where L+M+N=K D(α3,1,α3,2,...,α3,N)D(α 3,1 ,α 3,2 ,...,α 3,N ) ......。 … 4.根据权利要求3所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，所述将社交数据映射为直接反映物理层面参数的信息具体包括：4. The method for constructing a mapping model of wireless resource management indicators under edge computing according to claim 3, wherein the mapping of social data to information that directly reflects physical level parameters specifically includes: 将社交数据映射为与时间、距离、速度、位置、信号强度相关的直接反映物理层面参数的信息。Map social data into information related to time, distance, speed, location, signal strength that directly reflects physical-level parameters. 5.根据权利要求1所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，所述对不同来源的社交数据的QoS指标分配不同的权值，进行权值论域的标准归一化处理具体包括：5. The method for constructing a mapping model of wireless resource management indicators under edge computing according to claim 1, wherein the QoS indicators of social data from different sources are assigned different weights, and the standard of the weight universe is carried out. The normalization process specifically includes: 根据获得的社交数据的来源：来自于社交网络、来自于大数据、来自于对边缘环境的物理层监控结果，对社交数据的QoS指标分配不同的权值；According to the source of the obtained social data: from social networks, from big data, from the physical layer monitoring results of the edge environment, assign different weights to the QoS indicators of social data; 上述三种来源的QoS指标分配的权值分别为u1,u2,u3，其中u1,u2,u3满足：The weights assigned to the QoS indicators from the above three sources are u 1 , u 2 , and u 3 respectively, where u 1 , u 2 , and u 3 satisfy: u1+u2+u3＝1u 1 +u 2 +u 3 =1 其中，为突出体现终端用户的体验，设置来自于大数据的反映用户差异化通信特征的QoS指标占据更高的权重。Among them, in order to highlight the experience of the end user, the QoS indicators that reflect the differentiated communication characteristics of users from big data are set to occupy a higher weight. 6.根据权利要求5所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，对每个来源的QoS指标进一步按照其物理层面的网络模型进行分类，分为：应用层QoS指标，网络层QoS指标，以及底层QoS指标。6. the mapping model construction method of wireless resource management index under edge computing according to claim 5, is characterized in that, the QoS index of each source is further classified according to the network model of its physical layer, is divided into: application layer QoS Metrics, network layer QoS metrics, and underlying QoS metrics. 7.根据权利要求5所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，所述对标准化的权值进行模糊化处理具体包括：7. The method for constructing a mapping model of wireless resource management indicators under edge computing according to claim 5, wherein the fuzzification processing on the standardized weights specifically comprises: 使用加权平均的模糊算子，综合考虑各个指标对模型的影响，同时保留单个指标决策的全部信息，基于改进的层次分析法构建各指标模糊权重集；Using the weighted average fuzzy operator, comprehensively considering the impact of each index on the model, while retaining all the information of a single index decision, and constructing the fuzzy weight set of each index based on the improved AHP; 构造目标-因素判断矩阵对所述模糊权重集的可靠性进行检验。A target-factor judgment matrix is constructed to test the reliability of the fuzzy weight set. 8.根据权利要求1所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，所述对模糊化处理后的QoS指标选择合适的隶属函数，并根据隶属度最大化原则实现去模糊化，得到模糊权重值具体包括：8. The method for constructing a mapping model of a wireless resource management index under edge computing according to claim 1, wherein the described QoS index after the fuzzification is selected a suitable membership function, and is realized according to the principle of maximizing membership degree Defuzzification, the fuzzy weight value obtained specifically includes: 根据模糊规则，对模糊化处理后的QoS指标选择合适的隶属函数，通过隶属函数的计算得到以隶属度为元素的模糊矩阵R；According to the fuzzy rules, select the appropriate membership function for the QoS index after fuzzy processing, and obtain the fuzzy matrix R with membership degree as the element through the calculation of the membership function; 将论域权重U与模糊矩阵R进行模糊运算，得到应用模糊理论后的输出:Perform the fuzzy operation on the universe weight U and the fuzzy matrix R, and get the output after applying the fuzzy theory:
Figure FDA0002827649820000031
Figure FDA0002827649820000031
 根据隶属度最大化原则实现去模糊化，得到模糊权重值。Defuzzification is achieved according to the principle of maximizing membership degree, and the fuzzy weight value is obtained. 9.根据权利要求8所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，采取以下方式获得隶属函数：9. The method for constructing a mapping model of a wireless resource management index under edge computing according to claim 8, wherein the membership function is obtained in the following manner: 根据模糊统计实验法得到隶属函数；The membership function is obtained according to the fuzzy statistical experiment method; 或者，搭建神经网络训练隶属函数，通过构建三层卷积神经网络实现从影响因素到隶属度之间的映射。Or, build a neural network training membership function, and realize the mapping from influencing factors to membership degrees by building a three-layer convolutional neural network. 10.根据权利要求1所述的边缘计算下无线资源管理指标的映射模型构建方法，其特征在于，所述QoE指标的映射函数包括：基于对数函数的非实时业务描述；基于指数函数的实时业务描述；基于线性函数的语音业务描述；以及基于分段函数和多函数的综合业务描述。10. The method for constructing a mapping model of a wireless resource management index under edge computing according to claim 1, wherein the mapping function of the QoE index comprises: a logarithmic function-based non-real-time service description; an exponential function-based real-time service description; Service description; voice service description based on linear function; and comprehensive service description based on piecewise function and multi-function.
CN202011434366.1A 2020-12-10 2020-12-10 Mapping model construction method of radio resource management index under edge calculation Active CN112508408B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011434366.1A CN112508408B (en) 2020-12-10 2020-12-10 Mapping model construction method of radio resource management index under edge calculation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011434366.1A CN112508408B (en) 2020-12-10 2020-12-10 Mapping model construction method of radio resource management index under edge calculation

Publications (2)

Publication Number Publication Date
CN112508408A true CN112508408A (en) 2021-03-16
CN112508408B CN112508408B (en) 2024-01-05

Family

ID=74970467

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011434366.1A Active CN112508408B (en) 2020-12-10 2020-12-10 Mapping model construction method of radio resource management index under edge calculation

Country Status (1)

Country Link
CN (1) CN112508408B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114154780A (en) * 2021-10-19 2022-03-08 港华能源投资有限公司 Evaluation method and device, electronic equipment and related product
DE202022104275U1 (en) 2022-07-28 2022-08-25 Ahmed Alemran System for intelligent resource management for distributed machine learning tasks
CN116483585A (en) * 2023-06-16 2023-07-25 江西师范大学 Multi-service pre-deployment method based on edge computing

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103874132A (en) * 2014-03-25 2014-06-18 北京工业大学 Heterogeneous wireless network access selection method based on users
CN104866558A (en) * 2015-05-18 2015-08-26 中国科学院计算技术研究所 Training method of social networking account mapping model, mapping method and system
CN107516015A (en) * 2017-08-29 2017-12-26 武汉大学 Comprehensive evaluation method of composite insulator aging state based on multiple characteristic quantities
CN109639452A (en) * 2018-10-31 2019-04-16 深圳大学 Social modeling training method, device, server and storage medium
CN110532436A (en) * 2019-07-17 2019-12-03 中国人民解放军战略支援部队信息工程大学 Across social network user personal identification method based on community structure

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103874132A (en) * 2014-03-25 2014-06-18 北京工业大学 Heterogeneous wireless network access selection method based on users
CN104866558A (en) * 2015-05-18 2015-08-26 中国科学院计算技术研究所 Training method of social networking account mapping model, mapping method and system
CN107516015A (en) * 2017-08-29 2017-12-26 武汉大学 Comprehensive evaluation method of composite insulator aging state based on multiple characteristic quantities
CN109639452A (en) * 2018-10-31 2019-04-16 深圳大学 Social modeling training method, device, server and storage medium
CN110532436A (en) * 2019-07-17 2019-12-03 中国人民解放军战略支援部队信息工程大学 Across social network user personal identification method based on community structure

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114154780A (en) * 2021-10-19 2022-03-08 港华能源投资有限公司 Evaluation method and device, electronic equipment and related product
DE202022104275U1 (en) 2022-07-28 2022-08-25 Ahmed Alemran System for intelligent resource management for distributed machine learning tasks
CN116483585A (en) * 2023-06-16 2023-07-25 江西师范大学 Multi-service pre-deployment method based on edge computing
CN116483585B (en) * 2023-06-16 2023-09-15 江西师范大学 Multi-service pre-deployment method based on edge computing

Also Published As

Publication number Publication date
CN112508408B (en) 2024-01-05

Similar Documents

Publication Publication Date Title
CN111866954B (en) User selection and resource allocation method based on federal learning
CN112508408A (en) Mapping model construction method of wireless resource management index under edge calculation
CN103108382A (en) Heterogeneous network multi-attribute decision-making method based on network analytic hierarchy process
CN112511342A (en) Network slicing method and device, electronic equipment and storage medium
US11888703B1 (en) Machine learning algorithms for quality of service assurance in network traffic
CN110505644A (en) Joint optimization method of user task offloading and resource allocation in 5G ultra-dense heterogeneous network
CN112804702B (en) Multi-link air-ground data exchange link performance evaluation method based on utility function
CN113014340A (en) Satellite spectrum resource dynamic allocation method based on neural network
CN112819054A (en) Slice template configuration method and device
CN114860462B (en) Intelligent computing resource distribution system and method for double-path rack-mounted server
CN113988160B (en) A semi-asynchronous hierarchical federated learning update method based on timeliness
CN114040272B (en) Path determination method, device and storage medium
CN113886454B (en) LSTM-RBF-based cloud resource prediction method
CN113038612B (en) Cognitive radio power control method based on deep learning
CN115967990A (en) A Classification and Prediction Based Edge-End Collaborative Service Offloading Method
Gao et al. A deep learning framework with spatial-temporal attention mechanism for cellular traffic prediction
CN116916386A (en) A large model-assisted edge task offloading method considering user competition and load
CN113055229A (en) Wireless network self-selection protocol method based on DDQN
Nikoloska et al. Data selection scheme for energy efficient supervised learning at IoT nodes
CN114995964A (en) Combination service reconstruction method, device, equipment and computer readable medium
CN114205238B (en) Network resource optimization and model training method and device, storage medium and equipment
CN117811935A (en) Smart city Internet of things system based on wireless communication and operation method
CN115208470B (en) Resource allocation method of hybrid multiple access VLC network based on fuzzy logic
Hoiles et al. Risk-averse caching policies for YouTube content in femtocell networks using density forecasting
CN112867101B (en) Public security heterogeneous wireless private network selection method based on comprehensive gray evaluation value

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
GR01 Patent grant
GR01 Patent grant