CN112312496A - Vertical switching method based on neural network multi-attribute judgment - Google Patents

Vertical switching method based on neural network multi-attribute judgment Download PDF

Info

Publication number
CN112312496A
CN112312496A CN202011099898.4A CN202011099898A CN112312496A CN 112312496 A CN112312496 A CN 112312496A CN 202011099898 A CN202011099898 A CN 202011099898A CN 112312496 A CN112312496 A CN 112312496A
Authority
CN
China
Prior art keywords
neural network
network
layer
formula
values
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
CN202011099898.4A
Other languages
Chinese (zh)
Other versions
CN112312496B (en
Inventor
陈赓
孙红雨
谭晓楠
曾庆田
邵睿
徐先杰
张旭
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.)
Shandong University of Science and Technology
Original Assignee
Shandong University of Science and Technology
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 Shandong University of Science and Technology filed Critical Shandong University of Science and Technology
Priority to CN202011099898.4A priority Critical patent/CN112312496B/en
Publication of CN112312496A publication Critical patent/CN112312496A/en
Application granted granted Critical
Publication of CN112312496B publication Critical patent/CN112312496B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W36/00—Hand-off or reselection arrangements
    • H04W36/0005—Control or signalling for completing the hand-off
    • H04W36/0011—Control or signalling for completing the hand-off for data sessions of end-to-end connection
    • H04W36/0016—Hand-off preparation specially adapted for end-to-end data sessions
    • 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
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W36/00—Hand-off or reselection arrangements
    • H04W36/14—Reselecting a network or an air interface

Landscapes

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

Abstract

The invention discloses a vertical switching method based on neural network multi-attribute judgment, and particularly relates to the technical field of seamless switching of a communication network. The invention constructs respective three-layer BP neural network models by setting a network environment in which five major networks of UMTS, GSM, WLAN, 4G and 5G coexist, and carries out structural design of an input layer, a hidden layer and an output layer; the terminal collects values of the six attribute elements and inputs the values into a model for training, numerical approximation and prediction, finally performance evaluation is carried out on the whole network, relevant numerical values such as switching success rate and the like are obtained, and feedback information is provided for the operation flow of the whole algorithm according to evaluation analysis results. The method can effectively improve the success rate of vertical switching between wireless networks, reduce the ping-pong effect, further successfully realize seamless switching between the multi-element heterogeneous networks, and effectively maintain the high-quality operation of the wireless networks in the whole heterogeneous fusion network environment.

Description

Vertical switching method based on neural network multi-attribute judgment
Technical Field
The invention belongs to the technical field of seamless switching of communication networks, and particularly relates to a vertical switching method based on neural network multi-attribute judgment in a heterogeneous convergence network.
Background
In recent years, with rapid development of wireless communication network technology and rapid improvement of network access technology under a heterogeneous convergence architecture, wireless type networks of multiple different forms coexist. Different networks compete with each other and complement each other, resulting in coexistence of multiple network relationships. In order to meet the requirements of users at different levels and the development of various types of services, multiple comprehensive property networks in a heterogeneous convergence state are formed among wireless networks with different architectures, wherein the mutual convergence of a WLAN network and a Cellular series network is most representative. The WLAN network has relatively high bandwidth and low coverage, and the Cellular network (such as the current 3G/4G wireless network) has relatively low bandwidth and high coverage, so the fusion of the Cellular network and the WLAN network can complement each other, fully exert their own advantages and simultaneously inhibit their own limitations, and in addition, with the birth of the 5G technology, the fusion between multiple networks with different architectures will be revolutionarily developed.
Because multiple wireless networks with different architectures coexist under the heterogeneous converged network environment condition, at the juncture of the wireless networks with different architectures, the quality of the network will directly affect the user experience, so the vertical handover algorithm capable of providing an optional network with better performance is particularly important, and the existing ways for improving the network performance by some vertical handover algorithms in the market generally include: (1) a method combining hierarchical analysis and game theory; (2) fuzzy logic and hierarchical analysis combined method; (3) utilizing an IEEE related protocol and a multi-attribute decision-making auxiliary method; (4) fuzzy logic and multivariate attribute construction methods; (5) mixed Integer Linear Programming (MILP) model to perform multi-objective search; (6) genetic algorithms and markov chain selection. Although these algorithms are widely popular and have good practicability, the complexity is relatively high and the high level cannot be maintained for a long time in the aspects of improving the handover success rate, inhibiting the ping-pong effect and the like.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a vertical switching method based on neural network multi-attribute judgment, which solves the problems of low success rate and unstable performance of vertical switching algorithm switching under the heterogeneous fusion network environment condition.
In order to achieve the purpose, the invention adopts the following technical scheme:
a vertical switching method based on neural network multi-attribute judgment comprises a neural network model construction stage and a performance evaluation and feedback stage, wherein,
firstly, building a neural network model
Setting a network environment in which five types of networks including UMTS, GPRS, WLAN, 4G and 5G coexist, respectively building a three-layer BP neural network model of each network, wherein the building process is as follows:
step 1: the input layer structure design is that six attribute elements which can affect the performances of the five networks under the same communication condition are respectively set, the quantized values of the six attribute elements are input into the input layer, and the number of nodes of the input layer is set to be six;
step 2: designing a hidden layer structure, and determining the number of hidden layer neuron nodes by adopting a moderate principle according to an empirical formula of the number of hidden layer neurons;
and step 3: outputting the structural design of the layer and determining a related performance formula;
second, performance evaluation and feedback
And evaluating the performance of the whole network to obtain a relevant numerical value of the switching success rate, and providing feedback information for the operation flow of the whole algorithm according to the evaluation analysis result.
Preferably, in step 1, the values of the six attribute elements are corresponding to a maximum transmission rate, a minimum delay, an SINR, an error rate, a user moving speed, and a packet loss rate, respectively.
Preferably, in step 2, the empirical formula of the number of hidden layer neurons is:
Figure BDA0002724983870000021
wherein m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an arbitrary constant between 1 and 10.
Preferably, the step 3 specifically includes the following steps:
step 3.1: designing an output layer neuron of the neural network model according to respective network download rate predicted values of the five types of wireless networks, and selecting a wireless network with the optimal current quality and the most suitable environment according to the numerical values;
step 3.2: adopting a logsig type function as a transfer function of a neural network model and an L-M reverse error propagation algorithm as a training method to obtain a convergence rate, wherein the logsig type function formula is as follows:
Figure BDA0002724983870000022
step 3.3: the error is continuously reduced and approximated to the desired output value according to an error back propagation function, which has the formula:
Figure BDA0002724983870000023
wherein, tiAnd OiThe expected values of the network parameter results and the associated output values calculated for the neural network data, respectively.
Preferably, the general output formula of the neural network can be expressed as formula (4),
Figure BDA0002724983870000024
wherein a is a neuron threshold, w is a weight set corresponding to input data, x is an input data set of neurons, f is an activation function, and the expression is formula (5),
Figure BDA0002724983870000031
the output of the node in the hidden layer of the neural network is formula (6),
Figure BDA0002724983870000032
where net is the set of neuron variables for that layer;
the gradient of the error function of the hidden node of the neural network is formula (7),
Figure BDA0002724983870000033
wherein, OlThe set of predicted output values for the neural network, for which equation (8) is as follows,
Figure BDA0002724983870000034
the error of the neural network output layer node is formula (9),
Figure BDA0002724983870000035
wherein, tlA set of expected values for neural network parameters;
the weight adjustment of the BP neural network should be proportional to the negative gradient of the error E to achieve the minimization of the error E, expressed by equation (10),
Figure BDA0002724983870000036
wherein, TliA weight set which is corresponding to the current ith output layer neuron and has l hidden layer neurons is provided;
the gradient of the node error function of the neural network output layer is formula (11).
Figure BDA0002724983870000037
The invention has the following beneficial technical effects:
the invention provides a vertical switching method based on neural network multi-attribute judgment in a heterogeneous fusion network, which introduces six attribute elements which can most influence the quality of a wireless network from the viewpoint of considered decision factors, namely maximum transmission rate, minimum time delay, SINR, error rate, user moving speed and packet loss rate, combines the attribute elements with a BP neural network to ensure more comprehensive network attribute factors and lower complexity, and simultaneously introduces a 5G network to be fused with the original wireless networks such as UMTS, GPRS, WLAN, 4G and the like, thereby enriching the application types and the range of heterogeneous networks included by the algorithm of the invention, and further solving the problems of low success rate and unstable performance of vertical switching algorithm switching under the environmental condition of the heterogeneous fusion network.
Drawings
FIG. 1 is a flow chart of the algorithm operation of an embodiment of the present invention;
fig. 2 is a diagram of a BP neural network training model corresponding to a wireless network according to an embodiment of the present invention (taking a 5G network as an example);
FIG. 3 is a generalized three-layer neural network framework diagram according to an embodiment of the present invention.
Detailed Description
The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will occur to those skilled in the art upon reading the present disclosure and fall within the scope of the appended claims.
Fig. 1 shows a flowchart of the algorithm operation according to the embodiment of the present invention, which includes five processes:
(1) when the user terminal collects related network attribute parameter values in a heterogeneous converged network environment consisting of five wireless networks, the vertical handover algorithm starts to operate.
(2) And then, the acquired six types of network attribute parameters are sent to input layer neurons of a three-layer BP neural network model represented by the five networks respectively.
(3) And aiming at the network environment attribute acquisition value input into the neural network model, the BP neural network performs numerical calculation and function value approximation according to the previously input network attribute reference value and the expected value and finally determines the internal related weight of the neural network corresponding to each wireless network.
(4) And then, synthesizing the obtained network attribute acquisition values, and performing numerical prediction on the network downloading rate finally participating in decision making by using the trained and learned BP neural network model.
(5) And finally, according to the predicted network download rates of the five wireless networks, performing numerical comparison, and according to the compared numerical result, selecting the network with the optimal performance to execute a vertical switching decision.
The five steps are a complete process which needs to be operated when the vertical switching algorithm designed by the invention executes a switching decision. In the real environment that five heterogeneous networks are fused, when a user moves to different positions, six network attribute parameters set by the invention can change along with the change of the network environment around the user, therefore, the algorithm designed by the invention needs to continuously collect data to realize network switching in the real heterogeneous fusion network environment, namely, the five steps need to be continuously and circularly carried out to finish multiple judgments to adapt to the dynamic network environment in real life.
The method specifically comprises the following steps:
(1) when a user terminal collects related network attribute parameter values in a heterogeneous converged network environment consisting of five wireless networks, the vertical handover algorithm starts to operate: establishing a heterogeneous fusion network environment condition model, respectively setting six neuron nodes at input layers of respective BP neural network models of five wireless networks aiming at the six attribute elements, namely setting the number of the nodes of the input layers to be six, then respectively setting the six attribute elements of the five networks which can influence the respective performances under the same communication condition and inputting quantized values of the six attribute elements into the input layers;
(2) sending the collected six types of network attribute parameters into input layer neurons of a three-layer BP neural network model represented by the five networks respectively: the invention adopts the above mentioned six attribute elements, namely the values corresponding to the maximum transmission rate, the minimum time delay, the SINR, the bit error rate, the user moving speed and the packet loss rate, as the input values of the neuron parameters, and further participates in the data training and learning of the BP neural network, so that the algorithm comprehensively evaluates the network environment condition of the user;
(3) aiming at the network environment attribute acquisition value input into the neural network model, the BP neural network performs numerical calculation and function value approximation according to the previously input network attribute reference value and the expected value and finally determines the internal related weight of the neural network corresponding to each wireless network: adopting a moderate principle, avoiding two extreme situations of 'overfitting' and 'underfitting', under the premise of ensuring that the number of hidden layer neuron nodes is relatively well associated with the complexity of the actual environment of the heterogeneous fusion network, the number of the neuron nodes of a network input layer and a network output layer and an expected error value set by a BP (back propagation) neural network for vertical switching judgment, comprehensively considering that the number of the hidden layer neuron nodes is six, further adopting a logsig type function as a transfer function of a neural network model and an L-M (L-M) reverse error propagation algorithm as a training method so as to obtain a faster convergence rate, and continuously adjusting related weights in the neural network according to an error reverse propagation function so as to continuously reduce errors and approach the expected output value;
(4) and (3) synthesizing the obtained network attribute acquisition values, and performing numerical prediction on the network download rate finally participating in decision by using the trained and learned BP neural network model: setting the number of neurons of the output layer to be one for the neural network model corresponding to each wireless network at the output layer, representing the final network downloading rate, so as to perform neural network prediction calculation according to the obtained related network attribute acquisition values, and finally determining respective network downloading rate prediction values representing five types of wireless networks;
(5) according to the predicted network download rates of the five wireless networks, carrying out numerical comparison, and according to the compared numerical results, selecting the network with the optimal performance to execute a vertical switching decision: data of six attribute elements in five heterogeneous network environments collected by a terminal are finally respectively input into respective corresponding models of the five heterogeneous network environments, and are subjected to parameter training, numerical value approximation and prediction from an input layer, a hidden layer and an output layer, and finally, respective network download rate predicted values of the five heterogeneous network environments at the moment are generated and output by the output layer to participate in the next network judgment, and the wireless network with the most appropriate current performance environment and the best quality is selected to be switched by comparing the sizes of the respective network download rate predicted values so as to complete the algorithm process.
The following examples are provided to further illustrate and describe the design of the present invention.
First, the training principle of the BP neural network corresponding to the wireless network is explained. Fig. 2 shows a diagram of a training model of a BP neural network corresponding to a wireless network according to an embodiment of the present invention (taking a 5G network as an example), and it can be seen from the diagram that the neural network model structure of the algorithm of the present invention needs to be implemented by three steps, i.e., an input layer structure design, a hidden layer structure design, and an output layer structure design.
In the embodiment of the invention, the network weight is set to (W) according to the working principle of the BP neural network modelij，Tij) And appropriately setting a threshold for the relevant neuron, wherein X can be setjIs the jth input level node, YjIs the jth hidden layer node, OjIs the jth output layer sectionAnd E is an error function, the related calculation formula of the three-layer BP neural network model is as follows, wherein the input vector of the neuron is set as follows: x ═ X1,x2,x3,......,x6) Then the weight values corresponding to the input vectors in the input neurons are: w ═ W1,w2,w3,......,w6)。
(1) Input layer structure design of neural network model
Fig. 3 is a generalized general three-layer neural network framework diagram according to an embodiment of the present invention, in this embodiment, m, k, and n respectively indicate that an input layer has m neurons, a hidden layer has k neurons, and an output layer has n neurons, and for the vertical switching algorithm designed by the present invention, a BP neural network model combined with the vertical switching algorithm is constructed by using a general logsig-type transfer function, that is, formula (1).
Figure BDA0002724983870000061
Meanwhile, the error function of result reverse transmission, namely formula (2), can be used for continuously correcting the parameter weight of the neural network, so that the value of the error function is reduced, wherein t in the error functioniAnd OiThe expected values of the network parameter results and the related output values of the neural network data calculation are respectively, and the corresponding neural network model needs to be solved aiming at the operation of the algorithm.
Figure BDA0002724983870000062
In summary, the BP neural network is a neural network capable of performing back propagation, and by training the input sample parameters, the relevant network threshold or weight in the neuron is continuously corrected, and the value of the relevant error function is decreased in the direction of the negative gradient in the mathematical theory until approaching to the expected output value, so that it is very important to set the input layer structure of the BP neural network model.
Firstly, attribute elements which can affect the performance of each network under the same communication condition are respectively set for five types of networks, and the quantized values of the attribute elements are input into an input layer. In order to ensure that the designed algorithm comprehensively utilizes multiple network attributes and makes an effective switching strategy, the values of the six attribute elements, namely the values corresponding to the maximum transmission rate, the minimum time delay, the SINR, the error rate, the user moving speed and the packet loss rate, are adopted as the input values of the neuron parameters, so that the algorithm comprehensively evaluates the network environment condition of the user, and therefore, aiming at the six attribute elements, six neuron nodes are respectively arranged at the input layers of respective BP (back propagation) neural network models of five wireless networks, namely, the number of the nodes of the input layers is six.
The general output formula of the neural network can be expressed as formula (3).
Figure BDA0002724983870000063
Wherein, a is a neuron threshold, w is a weight set corresponding to input data, x is an input data set of neurons, and f is an activation function, and the expression thereof is formula (4).
Figure BDA0002724983870000064
(2) Neural network model hidden layer structure design
The number of hidden layer neurons is determined according to an empirical formula (5) for the number of hidden layer neurons, m is the number of input layer neurons, n is the number of output layer neurons, and a is an arbitrary constant between 1 and 10 (including 1 and 10).
Figure BDA0002724983870000071
As the hidden layer part of the BP neural network model, the neuron nodes of the hidden layer play a very critical role, therefore, in the model of the algorithm, the structural design of the hidden layer is particularly important, theoretically, a considerable number of hidden layer neuron nodes can carry out infinite numerical approximation on a nonlinear function with any small error precision, but the number of hidden layer neuron nodes cannot be set too much, otherwise, the complexity of network numerical calculation is deepened and the calculation amount is increased, meanwhile, the complex neural network calculation is very easy to generate the phenomenon of overfitting, on the contrary, if the number of hidden layer neurons is very small, the error is increased, the performance of the neural network is severely influenced, so that the difference between the performance of the neural network and the theoretical value of an expected decision result is too large, and the final switching decision of the algorithm is severely influenced by the two types of settings. Therefore, the number of the hidden layer neuron nodes needs to be set by adopting a proper principle, and the number of the hidden layer neuron nodes is determined to be six through comprehensive consideration in order to ensure that the number of the hidden layer neuron nodes is relatively well associated with the complexity of the actual environment of the heterogeneous fusion network, the number of the neuron nodes of the network input layer and the network output layer and the expected error value set by the BP neural network for the vertical switching judgment.
The output of the hidden layer node of the neural network is a formula (6), and net is a set of neuron variables of the layer.
Figure BDA0002724983870000072
Similarly, the gradient of the error function of the hidden node of the neural network is formula (7).
Figure BDA0002724983870000073
(3) Neural network model output layer structure design
Based on the two steps, the predicted network download rate values representing the five types of wireless networks are determined at the output layer as the final setting of the layer neurons. Therefore, for the neural network model corresponding to each wireless network, the number of neurons in the output layer is set to be one, finally, the output layer generates and outputs respective network download rate predicted values of the neurons at the moment for participating in the next network judgment, and the wireless network with the most suitable current performance environment and the best quality is selected to be switched by comparing the sizes of the respective network download rate predicted values so as to complete the algorithm process.
Therefore, the output of the neural network output layer node is formula (8).
Figure BDA0002724983870000074
Wherein, the error of the output layer node of the neural network is formula (9), tlIs a desired set of values for the neural network parameters.
Figure BDA0002724983870000081
Here, minimizing the error E is the final goal of training and learning of the BP neural network, so the weight adjustment of the BP neural network should be proportional to the negative gradient of the error E, expressed by equation (10), TliAnd the weight set which is corresponding to the current ith output layer neuron and has l hidden layer neurons is obtained.
Figure BDA0002724983870000082
Therefore, the gradient of the node error function of the neural network output layer is formula (11).
Figure BDA0002724983870000083
In summary, the BP neural network is a multi-layer feed-forward network trained according to an error back-propagation algorithm. In the algorithm, a small value is given to a network connection value, then a training sample consisting of sampling values of six attribute elements and corresponding network download rate theoretical values under a heterogeneous fusion network framework is selected to calculate an error gradient relative to the sample, then modification of network connection weight and continuous modification of application weight values and offset values are realized by using a mean square error and gradient descent method, and finally the actual output of the BP neural network is closer to the expected output.
In summary, the BP neural network model constructed by the present invention is combined with the proposed vertical switching algorithm, and the numbers of neurons in the input layer, hidden layer, and output layer are respectively 6, and 1. Data of six attribute elements in five heterogeneous network environments collected by a terminal are respectively input into respective models corresponding to the six attribute elements, then parameter training, numerical value approximation and prediction are carried out on the data through an input layer, a hidden layer and an output layer, then respective network download rate predicted values of the data at the moment are generated and output by the output layer to participate in the next network judgment, finally the whole vertical switching process of the algorithm is determined through all the steps, performance evaluation is carried out on the whole vertical switching method based on neural network multi-attribute judgment, relevant numerical values such as switching success rate and the like are obtained, and feedback information is provided for the operation process of the whole algorithm according to evaluation and analysis results.
In summary, the invention adopting the above technical scheme has the following beneficial effects: the invention provides a vertical switching method based on neural network multi-attribute judgment in a heterogeneous fusion network, which is characterized in that from the viewpoint of decision factors considered, six attribute elements which can most influence the quality of a wireless network, namely maximum transmission rate, minimum time delay, SINR, error rate, user moving speed and packet loss rate, are introduced and combined with a BP neural network, so that relatively comprehensive network attribute factors and relatively low complexity are ensured, and meanwhile, a 5G network is introduced and fused with the original UMTS, GPRS, WLAN, 4G and other wireless networks, so that the application types and the range of heterogeneous networks included by the algorithm are enriched, and the problems of low vertical switching algorithm switching success rate, unstable performance and the like under the environmental condition of the heterogeneous fusion network are solved.
It is to be understood that the above description is not intended to limit the present invention, and the present invention is not limited to the above examples, and those skilled in the art may make modifications, alterations, additions or substitutions within the spirit and scope of the present invention.

Claims (5)

1. A vertical switching method based on neural network multi-attribute judgment is characterized by comprising a neural network model building stage and a performance evaluation and feedback stage, wherein,
firstly, building a neural network model
Setting a network environment in which five types of networks including UMTS, GPRS, WLAN, 4G and 5G coexist, respectively building a three-layer BP neural network model of each network, wherein the building process is as follows:
step 1: the input layer structure design is that six attribute elements which can affect the performances of the five networks under the same communication condition are respectively set, the quantized values of the six attribute elements are input into the input layer, and the number of nodes of the input layer is set to be six;
step 2: designing a hidden layer structure, and determining the number of hidden layer neuron nodes by adopting a moderate principle according to an empirical formula of the number of hidden layer neurons;
and step 3: outputting the structural design of the layer and determining a related performance formula;
second, performance evaluation and feedback
And evaluating the performance of the whole network to obtain a relevant numerical value of the switching success rate, and providing feedback information for the operation flow of the whole algorithm according to the evaluation analysis result.
2. The vertical handover method according to claim 1, wherein in step 1, the values of the six attribute elements are corresponding to a maximum transmission rate, a minimum delay, an SINR, an error rate, a user moving speed, and a packet loss rate, respectively.
3. The vertical handover method based on neural network multi-attribute decision as claimed in claim 1, wherein in step 2, the hidden layer neuron number empirical formula is:
Figure FDA0002724983860000011
wherein m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is an arbitrary constant between 1 and 10.
4. The vertical handover method based on neural network multi-attribute decision as claimed in claim 1, wherein the step 3 specifically comprises the following steps:
step 3.1: designing an output layer neuron of the neural network model according to respective network download rate predicted values of the five types of wireless networks, and selecting a wireless network with the optimal current quality and the most suitable environment according to the numerical values;
step 3.2: adopting a logsig type function as a transfer function of a neural network model and an L-M reverse error propagation algorithm as a training method to obtain a convergence rate, wherein the logsig type function formula is as follows:
Figure FDA0002724983860000012
step 3.3: the error is continuously reduced and approximated to the desired output value according to an error back propagation function, which has the formula:
Figure FDA0002724983860000021
wherein, tiAnd OiThe expected values of the network parameter results and the associated output values calculated for the neural network data, respectively.
5. The vertical handover method based on neural network multi-attribute decision of claim 1, wherein the general output formula of the neural network is expressed as formula (4),
Figure FDA0002724983860000022
wherein a is a neuron threshold, w is a weight set corresponding to input data, x is an input data set of neurons, f is an activation function, and the expression is formula (5),
Figure FDA0002724983860000023
the output of the node in the hidden layer of the neural network is formula (6),
Figure FDA0002724983860000024
where net is the set of neuron variables for that layer;
the gradient of the error function of the hidden node of the neural network is formula (7),
Figure FDA0002724983860000025
wherein, OlThe set of predicted output values for the neural network, for which equation (8) is as follows,
Figure FDA0002724983860000026
the error of the neural network output layer node is formula (9),
Figure FDA0002724983860000027
wherein, tlA set of expected values for neural network parameters;
the weight adjustment of the BP neural network should be proportional to the negative gradient of the error E to achieve the minimization of the error E, expressed by equation (10),
Figure FDA0002724983860000028
wherein, TliA weight set which is corresponding to the current ith output layer neuron and has l hidden layer neurons is provided;
the gradient of the node error function of the neural network output layer is formula (11).
Figure FDA0002724983860000029
CN202011099898.4A 2020-10-15 2020-10-15 Vertical switching method based on neural network multi-attribute judgment Active CN112312496B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011099898.4A CN112312496B (en) 2020-10-15 2020-10-15 Vertical switching method based on neural network multi-attribute judgment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011099898.4A CN112312496B (en) 2020-10-15 2020-10-15 Vertical switching method based on neural network multi-attribute judgment

Publications (2)

Publication Number Publication Date
CN112312496A true CN112312496A (en) 2021-02-02
CN112312496B CN112312496B (en) 2022-05-24

Family

ID=74327162

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011099898.4A Active CN112312496B (en) 2020-10-15 2020-10-15 Vertical switching method based on neural network multi-attribute judgment

Country Status (1)

Country Link
CN (1) CN112312496B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113923736A (en) * 2021-09-29 2022-01-11 山东浪潮科学研究院有限公司 Base station switching method, device, equipment and product for terminal communication
CN116450708A (en) * 2023-06-13 2023-07-18 南京市城市数字治理中心 Enterprise data mining method and system

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104486806A (en) * 2014-11-14 2015-04-01 北京邮电大学 Carrying efficiency-based heterogeneous network combined carrying method and device
CN108235390A (en) * 2017-12-01 2018-06-29 吉林大学 Vertical handoff method based on Bayesian decision in a kind of heterogeneous wireless network

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104486806A (en) * 2014-11-14 2015-04-01 北京邮电大学 Carrying efficiency-based heterogeneous network combined carrying method and device
CN108235390A (en) * 2017-12-01 2018-06-29 吉林大学 Vertical handoff method based on Bayesian decision in a kind of heterogeneous wireless network

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
AYMEN BEN ZINEB ET AL: ""QoE-based vertical handover decision management for cognitive networks using ANN"", 《2017 SIXTH INTERNATIONAL CONFERENCE ON COMMUNICATIONS AND NETWORKING (COMNET)》 *
AYMEN BEN ZINEB ET AL: ""QoE-based vertical handover decision management for cognitive networks using ANN"", 《2017 SIXTH INTERNATIONAL CONFERENCE ON COMMUNICATIONS AND NETWORKING (COMNET)》, 1 April 2017 (2017-04-01) *
张艺峰等: "Application of the L-M optimized algorithm to predicting blast vibration parameters", 《ACTA SEISMOLOGICA SINICA(ENGLISH EDITION)》, no. 05, 15 October 2008 (2008-10-15) *
邸敬等: "无线异构网络中垂直切换机制与切换算法的研究", 《自动化与仪器仪表》 *
邸敬等: "无线异构网络中垂直切换机制与切换算法的研究", 《自动化与仪器仪表》, no. 07, 25 July 2018 (2018-07-25) *
马彬等: "异构无线网络中基于人工神经网络的自适应垂直切换算法", 《电子与信息学报》 *
马彬等: "异构无线网络中基于人工神经网络的自适应垂直切换算法", 《电子与信息学报》, no. 05, 28 February 2019 (2019-02-28), pages 1 - 7 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113923736A (en) * 2021-09-29 2022-01-11 山东浪潮科学研究院有限公司 Base station switching method, device, equipment and product for terminal communication
CN113923736B (en) * 2021-09-29 2023-05-16 山东浪潮科学研究院有限公司 Base station switching method, device, equipment and product for terminal communication
CN116450708A (en) * 2023-06-13 2023-07-18 南京市城市数字治理中心 Enterprise data mining method and system
CN116450708B (en) * 2023-06-13 2023-09-01 南京市城市数字治理中心 Enterprise data mining method and system

Also Published As

Publication number Publication date
CN112312496B (en) 2022-05-24

Similar Documents

Publication Publication Date Title
CN112312496B (en) Vertical switching method based on neural network multi-attribute judgment
CN112491712B (en) Data packet routing algorithm based on multi-agent deep reinforcement learning
CN108235390B (en) Vertical switching method based on Bayesian decision in heterogeneous wireless network
Tan et al. Vertical handover algorithm based on multi-attribute and neural network in heterogeneous integrated network
Dong et al. A learner based on neural network for cognitive radio
CN110225535A (en) Heterogeneous wireless network vertical handoff method based on depth deterministic policy gradient
Zineb et al. QoE-based vertical handover decision management for cognitive networks using ANN
CN114697229A (en) Construction method and application of distributed routing planning model
CN106789320A (en) A kind of multi-species cooperative method for optimizing wireless sensor network topology
CN113779302A (en) Semi-distributed cooperative storage method based on value decomposition network and multi-agent reinforcement learning
CN111917642A (en) SDN intelligent routing data transmission method for distributed deep reinforcement learning
CN114710439B (en) Network energy consumption and throughput joint optimization routing method based on deep reinforcement learning
Sun et al. Improving the scalability of deep reinforcement learning-based routing with control on partial nodes
CN115065728A (en) Multi-strategy reinforcement learning-based multi-target content storage method
Meng et al. Multi-colony ant algorithm using both generative adversarial nets and adaptive stagnation avoidance strategy
CN102300269A (en) Genetic algorithm based antenna recognition network end-to-end service quality guaranteeing method
CN114095940A (en) Slice resource allocation method and equipment for hybrid access cognitive wireless network
Zheng et al. Research on Multi-objective Shortest Path Based on Genetic Algorithm
CN105426960B (en) Aluminium electroloysis energy-saving and emission-reduction control method based on BP neural network Yu MBFO algorithms
Parija et al. Novel intelligent soft computing techniques for location prediction in mobility management
CN111967566A (en) Edge computing offloading decision making based on long-short term memory neural network in Internet of vehicles environment
Zhao et al. Optimizing radial basis probabilistic neural networks using recursive orthogonal least squares algorithms combined with micro-genetic algorithms
Hu Research on application of BP neural network based on genetic algorithm in multi-objective optimization
CN116306770B (en) Software defined network performance prediction method based on dense parent neural architecture search
CN115665765B (en) Multi-unmanned aerial vehicle sensor network multi-channel time delay optimization method based on fuzzy reasoning

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