CN113163019A - Internet of things privacy protection service discovery system based on SDN and edge computing - Google Patents

Internet of things privacy protection service discovery system based on SDN and edge computing Download PDF

Info

Publication number
CN113163019A
CN113163019A CN202110588548.2A CN202110588548A CN113163019A CN 113163019 A CN113163019 A CN 113163019A CN 202110588548 A CN202110588548 A CN 202110588548A CN 113163019 A CN113163019 A CN 113163019A
Authority
CN
China
Prior art keywords
service
context
tree
internet
space
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
CN202110588548.2A
Other languages
Chinese (zh)
Other versions
CN113163019B (en
Inventor
周潘
周天翔
徐子川
付才
丁晓锋
吴静
胡钰林
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.)
Huazhong University of Science and Technology
Original Assignee
Huazhong 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 Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Priority to CN202110588548.2A priority Critical patent/CN113163019B/en
Publication of CN113163019A publication Critical patent/CN113163019A/en
Application granted granted Critical
Publication of CN113163019B publication Critical patent/CN113163019B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/51—Discovery or management thereof, e.g. service location protocol [SLP] or web services

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to an Internet of things privacy protection service discovery system based on SDN and edge computing, which comprises: the SDN controller is used for controlling each EN to serve as a cooperative learner to cache and process information in the MEC network; the context space model carries out context space partition on the preprocessed context records so as to mine and obtain similar context records; the service space model of the Internet of things covers the total service space through a service history tree and a service discovery tree, the service history tree and the service discovery tree are infinite binary trees, and each tree node corresponds to a service cluster in the total service space; the service history tree stores all historical service records of the Internet of things and marks the partition state of the total service space; the service discovery tree is constructed and refreshed according to the service history tree and the context space and is used for discovering an Internet of things service cluster suitable for the current user so as to select the optimal personalized service and recommend the optimal personalized service to the user; a balance is struck between accuracy of service discovery and the level of privacy protection.

Description

Internet of things privacy protection service discovery system based on SDN and edge computing
Technical Field
The invention relates to the technical field of internet of things, in particular to an internet of things privacy protection service discovery system based on SDN and edge computing.
Background
In the scene of the internet of things, a plurality of heterogeneous entities and intelligent devices are connected with each other under the support of radio frequency identification, wireless sensor networks and other technologies, and a large amount of data related to various physical environments are transmitted. With the help of ubiquitous deployment devices, internet of things services and applications are growing rapidly, covering almost every aspect of our lives (e.g., healthcare, smart home, industry, etc.). Over time, as smart devices continue to grow rapidly, more and more services will be provided to users.
The contextual information of the end user detected by the ambient sensors plays a crucial role in providing personal services. The service discovery system is an important component of an ecosystem of the Internet of things, and finds corresponding services from candidate services by using the context of a user instead of searching a plurality of services randomly. The service discovery can be applied to various scenes such as intelligent home automation, industrial automation and traffic, and the life of people is improved.
However, the highly heterogeneous nature of different users and the ever expanding number of services offered by many service providers in the internet of things greatly impact real-time services. To better meet the needs of individuals, the context of users (e.g., location, time, etc.) is widely used in the context of the internet of things, which raises privacy concerns. Also, many users have several different service requirements in each round, and therefore the service selection should be combined.
Disclosure of Invention
The invention provides an Internet of things privacy protection service discovery system based on SDN and edge calculation, aiming at the technical problems in the prior art, and solves the problems in the prior art.
The technical scheme for solving the technical problems is as follows: an internet of things privacy protection service discovery system based on SDN and edge computing, comprising: EN, SDN controller and service model;
the EN and the SDN controller are deployed in an MEC network; the SDN controller is used for controlling each EN to serve as a cooperative learner to cache and process information in the MEC network;
the service model comprises a context space model and an internet of things service space model;
the context space model carries out context space partition on the context records after preprocessing so as to mine and obtain similar context records;
the service space model of the Internet of things covers the total service space through a service history tree and a service discovery tree, the service history tree and the service discovery tree are infinite binary trees, and each tree node corresponds to a service cluster in the total service space; the service history tree stores all historical service records of the Internet of things and marks the partition state of the total service space; and the service discovery tree is constructed and refreshed according to the service history tree and the context space and is used for discovering an Internet of things service cluster suitable for the current user so as to select the optimal personalized service and recommend the optimal personalized service to the user.
The invention has the beneficial effects that: the invention provides a context-aware online algorithm supporting MEC, which is used for service discovery and supporting large-scale service in IoT, and numerical results show that compared with other context-aware online algorithms, the method has good performance and balances the accuracy of service discovery and the privacy protection level.
On the basis of the technical scheme, the invention can be further improved as follows.
Further, the process of the service model selecting the best personalized service to recommend to the user comprises the following steps:
extracting service vectors into the service space of the Internet of things until a user registers new Internet of things service to EN a in the MEC network
Figure RE-DEST_PATH_IMAGE001
To EN a, receiveUser's current context vector
Figure RE-DEST_PATH_IMAGE002
And carrying out pretreatment in EN a at the t round; wherein EN a represents the a-th EN, the superscript a represents the sequence number of the EN, and the subscript t represents the t-th round;
the context space model is in context space
Figure RE-DEST_PATH_IMAGE003
Finding a context subspace of the service record to obtain enough service records of the Internet of things for reference;
setting a reward estimated value B value representing tree nodes (h, i), and selecting an optimal Internet of things cluster corresponding to the highest B value
Figure RE-DEST_PATH_IMAGE004
；
For the user
Figure RE-238568DEST_PATH_IMAGE001
Selecting an optimal Internet of things cluster
Figure RE-DEST_PATH_IMAGE005
Set of internet of things services in
Figure RE-DEST_PATH_IMAGE006
In (1)
Figure RE-DEST_PATH_IMAGE007
N is the total number of the Internet of things services in the set of the Internet of things services;
if the number of services selected by the current user
Figure RE-DEST_PATH_IMAGE008
Not more than N, and
Figure RE-DEST_PATH_IMAGE009
is that
Figure RE-DEST_PATH_IMAGE010
Is selected from the group consisting of (a) a subset of,then is at
Figure RE-264030DEST_PATH_IMAGE010
Lirandon recommendation
Figure RE-DEST_PATH_IMAGE011
Providing the service to the user; otherwise, recommend N services to the user and ask for help from neighboring ENs to recommend others
Figure RE-106084DEST_PATH_IMAGE011
-N services;
and stopping the whole recommendation process after the user gives feedback.
Further, the process of obtaining similar context records by the context space model comprises:
mapping each of the user's contexts to one
Figure RE-DEST_PATH_IMAGE012
Dimensional context space
Figure RE-766873DEST_PATH_IMAGE003
(ii) a In the t-th round, the user has a context vector
Figure RE-174720DEST_PATH_IMAGE002
The context vector is encoded
Figure RE-DEST_PATH_IMAGE013
Sending to the context space model;
in each round, acquiring a valid record by mining the context information of the user, and taking the valid record as a reference for selecting the service type;
for the context space
Figure RE-793920DEST_PATH_IMAGE003
Partitioning into a plurality of context subspaces, and ensuring that the maximum number of context records in each space does not exceed a set threshold; finding the context vector
Figure RE-388850DEST_PATH_IMAGE013
The subspaces to which the context information belongs, and the appropriate context space containing the context-related information is obtained to obtain similar context records.
Further, the calculation formula of the reward estimation value B is as follows:
Figure RE-DEST_PATH_IMAGE014
；
wherein subscript h, i represents tree node (h, i);
Figure RE-DEST_PATH_IMAGE015
the average reward of experience is shown, and gamma is a parameter for balancing exploration and development;
Figure RE-DEST_PATH_IMAGE016
the expression of the Liphoz constant is shown,
Figure RE-DEST_PATH_IMAGE017
representing a context space
Figure RE-DEST_PATH_IMAGE018
The maximum distance of (a) is,
Figure RE-DEST_PATH_IMAGE019
represents the upper limit of the diameter of the subset region of tree nodes (h, i), infinity represents infinity;
Figure RE-DEST_PATH_IMAGE020
；
Figure RE-DEST_PATH_IMAGE021
；
Figure RE-DEST_PATH_IMAGE022
representing the number of times tree nodes (h, i) are selected at ena until round t,
Figure RE-DEST_PATH_IMAGE023
representing the set of ena and its single-hop neighbors,
Figure RE-DEST_PATH_IMAGE025
to indicate the function, n is the total number of rounds.
Further, the empirical average reward of the tree nodes (h, i) is:
Figure RE-DEST_PATH_IMAGE027
；
wherein the content of the first and second substances,
Figure RE-DEST_PATH_IMAGE028
is as follows.
Further, the service history tree is used in the service space model of the internet of things
Figure RE-DEST_PATH_IMAGE029
And context space
Figure RE-414181DEST_PATH_IMAGE013
Refreshing the service discovery tree
Figure RE-DEST_PATH_IMAGE030
And updating the B value, including:
for the
Figure RE-DEST_PATH_IMAGE031
To judge the existence
Figure RE-DEST_PATH_IMAGE032
Not exceeding a set threshold
Figure RE-DEST_PATH_IMAGE033
EN a seeks help from its neighboring hop EN, for EN a
Figure RE-DEST_PATH_IMAGE034
All ENs in time, the empirical average reward is calculated as:
Figure RE-DEST_PATH_IMAGE035
；
computing and updating the service discovery tree
Figure RE-DEST_PATH_IMAGE036
Average empirical consideration of
Figure RE-DEST_PATH_IMAGE037
Then, the B value is calculated and updated, and
Figure RE-DEST_PATH_IMAGE038
adding to the service discovery tree
Figure RE-597907DEST_PATH_IMAGE036
In (1).
Further, the service model also includes a local differential privacy mechanism
Figure RE-DEST_PATH_IMAGE039
To the user
Figure RE-653587DEST_PATH_IMAGE001
Initial context vector of
Figure RE-735813DEST_PATH_IMAGE013
And carrying out context random data perturbation processing, and calculating the influence of the perturbation processing on service discovery through a loss function.
Further, the local differential privacy mechanism
Figure RE-66300DEST_PATH_IMAGE039
The calculation formula of (2) is as follows:
Figure RE-DEST_PATH_IMAGE040
；
wherein the content of the first and second substances,
Figure RE-DEST_PATH_IMAGE041
as a context vector
Figure RE-855265DEST_PATH_IMAGE013
Via the local differential privacy mechanism
Figure RE-878584DEST_PATH_IMAGE039
Perturbing the set of post-operation mappings;
Figure RE-DEST_PATH_IMAGE042
，
Figure RE-DEST_PATH_IMAGE043
，
Figure RE-DEST_PATH_IMAGE044
representing a privacy factor;
the loss function is:
Figure RE-DEST_PATH_IMAGE045
。
further, after the service model selects the best personalized service to recommend to the user, the discovery system calculates a total reward value according to the server side reward and the user side reward:
Figure RE-DEST_PATH_IMAGE046
；
Figure RE-DEST_PATH_IMAGE047
and
Figure RE-DEST_PATH_IMAGE048
is a parameter, and
Figure RE-DEST_PATH_IMAGE049
；
Figure RE-DEST_PATH_IMAGE050
server-side rewards obtained from various ENs monitored by the SDN controller,
Figure RE-DEST_PATH_IMAGE051
and rewarding the user terminal determined according to the execution time, the response time and the reliability of the user.
Further, the discovery system performs offline update according to the completed service record:
in the service space model of the Internet of things, the service history tree is used for the service
Figure RE-974671DEST_PATH_IMAGE029
Each leaf tree node in (1)
Figure RE-DEST_PATH_IMAGE052
If, if
Figure RE-DEST_PATH_IMAGE053
(ii) a Updating the service history tree:
Figure RE-DEST_PATH_IMAGE054
；
updating the B value:
Figure RE-DEST_PATH_IMAGE055
；
in the context space model, if
Figure RE-DEST_PATH_IMAGE056
Then further dividing the context space
Figure RE-30221DEST_PATH_IMAGE003
；
Wherein the content of the first and second substances,
Figure RE-DEST_PATH_IMAGE057
representing context vectors
Figure RE-736009DEST_PATH_IMAGE013
The number of contexts in (a) and (b),
Figure RE-DEST_PATH_IMAGE058
and
Figure RE-DEST_PATH_IMAGE059
indicating that a threshold is set.
The beneficial effect of adopting the further scheme is that: only a single arm is considered to be selected in a round against the conventional CMAB algorithm, but many users have multiple requirements in a round and are therefore not applicable. The invention provides a context-aware online algorithm for local differential privacy and MEC support, which can process dynamic complex context problems, carry out balance between development and exploration after selecting proper arm, and select a group of arms by utilizing combined contextual multi-arm basis (CC-MAB) to meet different service requirements of each round of users. And the system can continuously learn the rewards of arms on the web, as the user context arrives, to maximize the total rewards.
Sensitive information of a user is protected on a user level instead of a service discovery system level by using a Local Differential Privacy (LDP) mechanism, random noise is introduced to interfere individual context by combining LDP and CC-MAB, the privacy of the user is ensured, and the balance between service discovery precision and local privacy/personal information utility is realized. The accuracy of the system is measured using the regret concept in multi-arm bandit, which is defined as the gap between the optimal service and the reward of the actually selected service, the sub-linear regret means that our method converges to the optimal service discovery strategy. The L2 loss was introduced as a practical measure to evaluate privacy loss and data utility during random perturbations. We have demonstrated theoretically that our algorithm can achieve a sub-linear regret, ensuring that promising local differential privacy is provided for individual ENs and users, but without significantly impacting the information utility.
Drawings
Fig. 1 is an interaction diagram of an internet of things privacy protection service discovery system based on SDN and edge computing according to an embodiment of the present invention;
fig. 2 is a schematic diagram of context space partitioning according to an embodiment of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
MEC (Mobile Edge Computing) is a technology based on 5G evolution architecture and integrates Mobile access networks with internet services deeply. The method has lower delay time so as to relieve network congestion and improve the experience quality of users. Data caching and processing may be performed on individual ENs (Edge Node services), where services may be deployed to end users with real-time responses. In addition, in order to realize a distributed ubiquitous computing scenario with high complexity and logic control on each EN, the Network complexity may be managed by using an SDN (Software Defined Network) that hides all internal details in the MEC Network, the control and the data plane are separated, and the programmable control is realized, so that the SDN controller can obtain a complete view and centralized control of the entire Network, and is more flexible and reliable. Furthermore, it can be deployed on eligible ENs in the MEC network, which is considered a promising solution to efficiently provide computing resources and coordinate the MEC network. Therefore, it is beneficial to cooperate the MEC supporting SDN with a service discovery system to provide real-time, low-latency services to users.
An internet of things privacy protection service discovery system based on SDN and edge computing, the discovery system comprising: EN, SDN controller, and service model.
The EN and the SDN controller are deployed in an MEC network; the SDN controller is used for controlling each EN to serve as a cooperative learner to cache and process information in the MEC network.
The service model comprises a context space model and an internet of things service space model.
And the context space model carries out context space partition on the preprocessed context information so as to mine and obtain similar context records.
The service space model of the Internet of things covers the total service space through a service history tree and a service discovery tree, wherein the service history tree and the service discovery tree are infinite binary trees; the service history tree stores all historical service records of the Internet of things and marks the partition state of the total service space; and the service discovery tree is constructed and refreshed according to the service history tree and the context space and is used for discovering an Internet of things service cluster suitable for the current user so as to select the optimal personalized service and recommend the optimal personalized service to the user.
The service history tree stores all historical service records of the Internet of things up to the t round, and the tree nodes in the service history tree can be gradually divided along with the increase of the history records. The service discovery tree may be refreshed based on a similar context space selected for obtaining a similar history and containing useful tree nodes in the service history tree. In the whole process, only the tree nodes containing the context related records of the service discovery tree in the service history tree are considered, and the whole service history tree is not considered. A particular internet of things service cluster is selected and the partitioning will be finer as the depth of the tree increases.
Example 1
Embodiment 1 provided in the present invention is an embodiment of a system for discovering internet of things privacy protection service based on SDN and edge computing, and as shown in fig. 1, an interaction diagram of the system for discovering internet of things privacy protection service based on SDN and edge computing provided in the embodiment of the present invention is shown, where the discovery system includes: EN, SDN controller, and service model.
The EN and the SDN controller are deployed in an MEC network; the SDN controller is used for controlling each EN to serve as a cooperative learner to cache and process information in the MEC network.
The service model comprises a context space model and an internet of things service space model.
And the context space model carries out context space partition on the preprocessed context information so as to mine and obtain similar context records.
The service space model of the Internet of things covers the total service space through a service history tree and a service discovery tree, wherein the service history tree and the service discovery tree are infinite binary trees; the service history tree stores all historical service records of the Internet of things and marks the partition state of the total service space; and the service discovery tree is constructed and refreshed according to the service history tree and the context space and is used for discovering an Internet of things service cluster suitable for the current user so as to select the optimal personalized service and recommend the optimal personalized service to the user.
Preferably, the process of the service model selecting the best personalized service recommendation to the user comprises:
step 1, when a new internet of things service is registered with EN a in the MEC network, extracting a service vector into the internet of things service space until a user
Figure RE-461388DEST_PATH_IMAGE001
To EN a, the current context vector of the user is received
Figure RE-659151DEST_PATH_IMAGE002
And carrying out pretreatment in EN a at the t round; EN a denotes the a-th ENE, the superscript a denotes the EN sequence number, and the subscript t denotes the t-th round.
Step 2, the context space model is in the context space
Figure RE-65862DEST_PATH_IMAGE003
Finds its context subspace to obtain enough service records of the internet of things for reference.
Specifically, the process of obtaining similar context records by the context space model includes:
mapping each of said user's contexts (which may be hobbies, location, time and duration, etc.) to one
Figure RE-32681DEST_PATH_IMAGE012
Dimensional context space
Figure RE-132224DEST_PATH_IMAGE003
(ii) a Mapping each of the user's contexts to one
Figure RE-676338DEST_PATH_IMAGE012
Dimensional context space
Figure RE-27685DEST_PATH_IMAGE003
(ii) a In the t-th round, the user has a context vector
Figure RE-973644DEST_PATH_IMAGE002
The context vector is encoded
Figure RE-385034DEST_PATH_IMAGE013
Sending to the context space model; context vector
Figure RE-150864DEST_PATH_IMAGE013
By
Figure RE-40323DEST_PATH_IMAGE012
And (5) modeling dimensional vectors.
FIG. 2 is a diagram illustrating a context space partition according to an embodiment of the present invention, where FIG. 2 is a diagram illustrating a context space partition
Figure RE-DEST_PATH_IMAGE060
And
Figure RE-DEST_PATH_IMAGE061
respectively represent
Figure RE-903106DEST_PATH_IMAGE060
And
Figure RE-610031DEST_PATH_IMAGE061
the time of the wheel is such that,
Figure RE-DEST_PATH_IMAGE062
、
Figure RE-DEST_PATH_IMAGE063
and
Figure RE-DEST_PATH_IMAGE064
respectively, Lc represents the Liphoz constant.
Normalizing the range of each dimension of a context record to [0,1 ]]Thus C is
Figure RE-DEST_PATH_IMAGE065
A context space. As can be seen in connection with the embodiment shown in FIG. 2, the three-dimensional context space contains users
Figure RE-DEST_PATH_IMAGE066
Service record of
Figure RE-DEST_PATH_IMAGE067
Up to
Figure RE-987791DEST_PATH_IMAGE060
Wheels, where dimensions may be age, monthly salary, and man-hours. For example,
Figure RE-DEST_PATH_IMAGE068
=0.20 means that the user is
Figure RE-133471DEST_PATH_IMAGE066
Is about 20 years old, and the maximum age is 100 years old, when
Figure RE-DEST_PATH_IMAGE069
When =0.89, he/she belongs to a high salary class because the threshold for the average payroll of 1200$ per month is 0.5. If it is not
Figure RE-DEST_PATH_IMAGE070
=0.26, maximum value 24, this indicates the user
Figure RE-460547DEST_PATH_IMAGE066
Work was done for about 6.2 hours per day.
In each round, a valid record is obtained by mining the context information of the user, and the valid record is used as a reference for selecting the service type.
For the context space
Figure RE-338373DEST_PATH_IMAGE003
Partitioning into a plurality of subspaces, and ensuring that the maximum number of context records in each space does not exceed a set threshold; finding the context vector
Figure RE-813217DEST_PATH_IMAGE013
The subspaces to which the context information belongs are obtained into an appropriate context space containing the context-related information so as to obtain similar context records for accurate selection.
For example, for the above-mentioned users
Figure RE-903532DEST_PATH_IMAGE066
Similar user records (high salary, young and easy to work) may be consulted.
Only the entire context space in the case of few history records is considered initially. As records increase, the context space can be divided more finely. T rounds since the recording is more complete
Figure RE-DEST_PATH_IMAGE071
The time becomes shorter. At the time of
Figure RE-881853DEST_PATH_IMAGE061
，
Figure RE-665001DEST_PATH_IMAGE071
Only a small subspace is covered, which contains more contextual users. Setting a threshold value
Figure RE-DEST_PATH_IMAGE072
Limiting the maximum number in each space when the number of records is greater than or equal to a threshold
Figure RE-373280DEST_PATH_IMAGE072
The space is further partitioned for accurate assessment. The context space only needs to be found
Figure RE-408232DEST_PATH_IMAGE071
The appropriate context space containing the context-related information can be obtained from the subspaces to which it belongs. If subspace
Figure RE-37797DEST_PATH_IMAGE071
Is greater than a threshold
Figure RE-991846DEST_PATH_IMAGE072
Then, then
Figure RE-441282DEST_PATH_IMAGE071
Will be partitioned and the context space will be updated. By using
Figure RE-DEST_PATH_IMAGE073
To represent
Figure RE-607821DEST_PATH_IMAGE071
The maximum distance of (c).
Step 3, setting the reward estimated value B of the tree node (h, i), and selecting the optimal Internet of things cluster corresponding to the highest B value
Figure RE-967258DEST_PATH_IMAGE004
. Each tree node corresponds to a service cluster in the total service space,
Figure RE-DEST_PATH_IMAGE074
。
preferably, the calculation formula of the reward estimate value B is:
Figure RE-685685DEST_PATH_IMAGE014
。
wherein, the superscript a represents the sequence number of EN, the subscript t represents the t-th round, and the subscript h, i represents the tree node (h, i);
Figure RE-763362DEST_PATH_IMAGE015
the average reward of experience is shown, and gamma is a parameter for balancing exploration and development;
Figure RE-530330DEST_PATH_IMAGE016
the expression of the Liphoz constant is shown,
Figure RE-744273DEST_PATH_IMAGE017
to represent
Figure RE-305705DEST_PATH_IMAGE018
The maximum distance of (a) is,
Figure RE-870678DEST_PATH_IMAGE019
represents the upper bound of the diameter of the subset region of tree nodes (h, i), infinity representing infinity.
Figure RE-DEST_PATH_IMAGE075
Is that
Figure RE-503654DEST_PATH_IMAGE018
The uncertainty of the context.
Exist of
Figure RE-DEST_PATH_IMAGE076
And 0 ≦ b ≦ 1, so for any tree node (h, i):
Figure RE-DEST_PATH_IMAGE077
，
Figure RE-DEST_PATH_IMAGE078
Figure RE-DEST_PATH_IMAGE079
indicating the diameter of the subset area.
The B value represents the reward estimate for the tree node (h, i), used as the basis for service discovery. The first item is based on the evaluation of service cluster performance previously recorded, i.e. the exploitation of past experience. The second term is the uncertainty of the cluster feedback, indicating the exploration level of the cluster. The last term represents the uncertainty of the tree node size.
The empirical average reward of the tree nodes (h, i) is:
Figure RE-555792DEST_PATH_IMAGE027
。
wherein the content of the first and second substances,
Figure RE-163491DEST_PATH_IMAGE028
is awarded for the t-th round.
Figure RE-340395DEST_PATH_IMAGE020
。
Figure RE-DEST_PATH_IMAGE080
。
Figure RE-386848DEST_PATH_IMAGE022
Representing the number of times tree nodes (h, i) are selected at ena until round t,
Figure RE-840963DEST_PATH_IMAGE023
representing the set of ena and its single-hop neighbors,
Figure RE-478618DEST_PATH_IMAGE025
to indicate the function, n is the total number of rounds.
Further, the service history tree is used in the service space model of the internet of things
Figure RE-142817DEST_PATH_IMAGE029
And context space
Figure RE-665066DEST_PATH_IMAGE003
Refreshing the service discovery tree
Figure RE-832742DEST_PATH_IMAGE030
And updating the B value, including:
for the
Figure RE-782243DEST_PATH_IMAGE031
To judge the existence
Figure RE-668160DEST_PATH_IMAGE032
Not exceeding a set threshold
Figure RE-994099DEST_PATH_IMAGE033
EN a, then considering that EN a is an inexperienced node, needs to learn from other nodes and let EN a seek help from its neighboring EN by one hop, for
Figure RE-281861DEST_PATH_IMAGE034
All ENs in time, the empirical average reward is calculated as:
Figure RE-261318DEST_PATH_IMAGE035
。
computing and updating the service discovery tree
Figure RE-775476DEST_PATH_IMAGE036
Average empirical consideration of
Figure RE-764160DEST_PATH_IMAGE037
Then, the B value is calculated and updated, and
Figure RE-781795DEST_PATH_IMAGE038
adding to the service discovery tree
Figure RE-197733DEST_PATH_IMAGE036
In (1).
The discovery system updates only the B-values of tree nodes containing relevant records and adds them to the service discovery tree, thereby reducing the number of candidates for service clusters and the likelihood of selecting to unrelated tree nodes.
Step 4, the user is provided with
Figure RE-933608DEST_PATH_IMAGE001
Selecting an optimal Internet of things cluster
Figure RE-DEST_PATH_IMAGE081
Set of internet of things services in
Figure RE-194825DEST_PATH_IMAGE006
In (1)
Figure RE-926020DEST_PATH_IMAGE007
And N is the total number of the Internet of things services in the set of the Internet of things services.
Step 5, if the number of the services selected by the current user
Figure RE-778439DEST_PATH_IMAGE008
Not more than N, and
Figure RE-1610DEST_PATH_IMAGE009
is that
Figure RE-66518DEST_PATH_IMAGE010
Is then in
Figure RE-58744DEST_PATH_IMAGE010
Lirandon recommendation
Figure RE-82064DEST_PATH_IMAGE011
Providing the service to the user; otherwise, recommend N services to the user and ask for help from neighboring ENs to recommend others
Figure RE-386006DEST_PATH_IMAGE011
-N services.
And 6, stopping the whole recommendation process after the user gives feedback.
The service model also includes a local differential privacy mechanism
Figure RE-395551DEST_PATH_IMAGE039
And step 2 is followed by: for the user
Figure RE-366918DEST_PATH_IMAGE001
Initial context vector of
Figure RE-29980DEST_PATH_IMAGE013
And carrying out context random data perturbation processing, and calculating the influence of the perturbation processing on better service discovery through a loss function.
The local differential privacy mechanism
Figure RE-227743DEST_PATH_IMAGE039
The calculation formula of (2) is as follows:
Figure RE-646173DEST_PATH_IMAGE040
。
wherein the content of the first and second substances,
Figure RE-612992DEST_PATH_IMAGE041
as a context vector
Figure RE-712535DEST_PATH_IMAGE013
Via the local differential privacy mechanism
Figure RE-256649DEST_PATH_IMAGE039
Perturbing the set of post-operation mappings;
Figure RE-607996DEST_PATH_IMAGE042
，
Figure RE-553955DEST_PATH_IMAGE043
，
Figure RE-965345DEST_PATH_IMAGE044
representing a privacy factor.
The loss function is:
Figure RE-731175DEST_PATH_IMAGE045
。
further, after service provisioning, the user explicitly or implicitly provides feedback to the system (user-side rewards if the service is selected by the user, the service-side rewards may be evaluated by data monitored by the MEC network, and if not, the latest history may be utilized.
After the service system selects the optimal personalized service to recommend to the user, the discovery system calculates a total reward value according to the reward of the service end and the reward of the user end:
Figure RE-620634DEST_PATH_IMAGE046
。
Figure RE-421100DEST_PATH_IMAGE047
and
Figure RE-862445DEST_PATH_IMAGE048
is a parameter, and
Figure RE-256517DEST_PATH_IMAGE049
。
Figure RE-808722DEST_PATH_IMAGE050
server-side rewards obtained from various ENs monitored by the SDN controller,
Figure RE-604639DEST_PATH_IMAGE051
the user terminal is awarded according to the execution time, the response time, the reliability and the like of the user.
Further, the discovery system also performs offline updating according to the completed service record:
in the service space model of the Internet of things, the service history tree is used for the service
Figure RE-482465DEST_PATH_IMAGE029
Each leaf tree node in (1)
Figure RE-363834DEST_PATH_IMAGE052
If, if
Figure RE-454149DEST_PATH_IMAGE053
(ii) a This indicates that the leaf node has been selected a sufficient number of times that the tree should be expanded to obtain more accurate regionsDomain and finer partitioning, the service history tree is updated:
Figure RE-229207DEST_PATH_IMAGE054
。
updating the B value:
Figure RE-153301DEST_PATH_IMAGE055
。
in the context space model, if
Figure RE-646599DEST_PATH_IMAGE056
Then further dividing the context space
Figure RE-415972DEST_PATH_IMAGE003
。
Wherein the content of the first and second substances,
Figure RE-45537DEST_PATH_IMAGE057
representing context vectors
Figure RE-140532DEST_PATH_IMAGE013
The number of contexts in (a) and (b),
Figure RE-855547DEST_PATH_IMAGE058
and
Figure RE-553244DEST_PATH_IMAGE059
indicating that a threshold is set.
The method specifically comprises the following steps:
Figure RE-DEST_PATH_IMAGE082
，
Figure RE-DEST_PATH_IMAGE083
，
Figure RE-DEST_PATH_IMAGE084
and
Figure RE-DEST_PATH_IMAGE085
to set the parameters.
The invention provides a context-aware online algorithm for local differential privacy and MEC support, which realizes sub-linear regret boundary and local differential privacy under the condition of not influencing the utility of context information and balances the accuracy of service discovery and the privacy protection level.
Only a single arm is considered to be selected in a round against the conventional CMAB algorithm, but many users have multiple requirements in a round and are therefore not applicable. The invention provides a context-aware online algorithm for local differential privacy and MEC support, which can process dynamic complex context problems, carry out balance between development and exploration after selecting proper arm, and select a group of arms by utilizing combined contextual multi-arm basis (CC-MAB) to meet different service requirements of each round of users. And the system can continuously learn the rewards of arms on the web, as the user context arrives, to maximize the total rewards.
Sensitive information of a user is protected on a user level instead of a service discovery system level by using a Local Differential Privacy (LDP) mechanism, random noise is introduced to interfere individual context by combining LDP and CC-MAB, the privacy of the user is ensured, and the balance between service discovery precision and local privacy/personal information utility is realized. The accuracy of the system is measured using the regret concept in multi-arm bandit, which is defined as the gap between the optimal service and the reward [35] of the actually selected service, the sub-linear regret means that our method converges to the optimal service discovery strategy. The L2 loss was introduced as a practical measure to evaluate privacy loss and data utility during random perturbations. We have demonstrated theoretically that our algorithm can achieve a sub-linear regret, ensuring that promising local differential privacy is provided for individual ENs and users, but without significantly impacting the information utility.
The invention provides a new context and combination bandit-based local differential privacy online learning method, which is used for service discovery in the Internet of things through an SDN. It guarantees that the sub-linear regret boundary indicates that the algorithm converges to the optimal strategy. Personalized service selection is possible because we take into account the end user's context and process the context information with adaptive context space partitioning. The performance of the IoT service is measured through the user side rewards and the service side rewards, so that the accuracy of the whole service discovery is improved. Under the cooperation of each EN and dynamic complex conditions in the Internet of things, the method supports a large data set which is continuously increased in an MEC scene. The privacy of the user is protected with local differential privacy while a balance is struck between the privacy protection level and the information utility.
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. An internet of things privacy protection service discovery system based on SDN and edge computing, the discovery system comprising: EN, SDN controller and service model;
the EN and the SDN controller are deployed in an MEC network; the SDN controller is used for controlling each EN to serve as a cooperative learner to cache and process information in the MEC network;
the service model comprises a context space model and an internet of things service space model;
the context space model carries out context space partition on the context records after preprocessing so as to mine and obtain similar context records;
the service space model of the Internet of things covers the total service space through a service history tree and a service discovery tree, the service history tree and the service discovery tree are infinite binary trees, and each tree node corresponds to a service cluster in the total service space; the service history tree stores all historical service records of the Internet of things and marks the partition state of the total service space; and the service discovery tree is constructed and refreshed according to the service history tree and the context space and is used for discovering an Internet of things service cluster suitable for the current user so as to select the optimal personalized service and recommend the optimal personalized service to the user.
2. The discovery system of claim 1 wherein said service model selects the best personalized service to recommend to said user comprises:
extracting service vectors into the service space of the Internet of things until a user registers new Internet of things service to EN a in the MEC network
Figure 599850DEST_PATH_IMAGE001
To EN a, the current context vector of the user is received
Figure 720253DEST_PATH_IMAGE002
And carrying out pretreatment in EN a at the t round; wherein EN a represents the a-th EN, the superscript a represents the sequence number of the EN, and the subscript t represents the t-th round;
the context space model is in context space
Figure 968831DEST_PATH_IMAGE003
Finding a context subspace of the service record to obtain enough service records of the Internet of things for reference;
setting a reward estimated value B value representing tree nodes (h, i), and selecting an optimal Internet of things cluster corresponding to the highest B value
Figure 82150DEST_PATH_IMAGE004
；
For the user
Figure 99784DEST_PATH_IMAGE001
Selecting an optimal Internet of things cluster
Figure 125509DEST_PATH_IMAGE005
Set of internet of things services in
Figure 110652DEST_PATH_IMAGE006
In (1)
Figure 512814DEST_PATH_IMAGE007
N is the total number of the Internet of things services in the set of the Internet of things services;
if the number of services selected by the current user
Figure 384955DEST_PATH_IMAGE008
Not more than N, and
Figure 96428DEST_PATH_IMAGE009
is that
Figure 54020DEST_PATH_IMAGE010
Is then in
Figure 259873DEST_PATH_IMAGE010
Lirandon recommendation
Figure 235788DEST_PATH_IMAGE011
Providing the service to the user; otherwise, recommend N services to the user and ask for help from neighboring ENs to recommend others
Figure 868895DEST_PATH_IMAGE011
-N services;
and stopping the whole recommendation process after the user gives feedback.
3. The discovery system of claim 1 wherein said process of obtaining similar context records by said context space model comprises:
mapping each of the user's contexts to one
Figure 579362DEST_PATH_IMAGE012
Dimensional context space
Figure 572595DEST_PATH_IMAGE003
(ii) a In the t-th round, the user has a context vector
Figure 419328DEST_PATH_IMAGE002
The context vector is encoded
Figure 223336DEST_PATH_IMAGE013
Sending to the context space model;
in each round, acquiring a valid record by mining the context information of the user, and taking the valid record as a reference for selecting the service type;
for the context space
Figure 404787DEST_PATH_IMAGE003
Partitioning into a plurality of context subspaces, and ensuring that the maximum number of context records in each space does not exceed a set threshold; finding the context vector
Figure 686864DEST_PATH_IMAGE013
The subspaces to which the context information belongs, and the appropriate context space containing the context-related information is obtained to obtain similar context records.
4. The discovery system of claim 2 wherein said reward estimate value, B, is calculated by the formula:
Figure 388104DEST_PATH_IMAGE014
；
wherein subscript h, i represents tree node (h, i);
Figure 612281DEST_PATH_IMAGE015
the average reward of experience is shown, and gamma is a parameter for balancing exploration and development;
Figure 766182DEST_PATH_IMAGE016
the expression of the Liphoz constant is shown,
Figure 117528DEST_PATH_IMAGE017
representing a context space
Figure 922542DEST_PATH_IMAGE018
The maximum distance of (a) is,
Figure 68353DEST_PATH_IMAGE019
represents the upper limit of the diameter of the subset region of tree nodes (h, i), infinity represents infinity;
Figure 709550DEST_PATH_IMAGE020
；
Figure 582697DEST_PATH_IMAGE021
；
Figure 524108DEST_PATH_IMAGE022
representing the number of times tree nodes (h, i) are selected at ena until round t,
Figure 840820DEST_PATH_IMAGE023
representing the set of ena and its single-hop neighbors,
Figure 218580DEST_PATH_IMAGE024
to indicate the function, n is the total number of rounds.
5. The discovery system according to claim 4, wherein the empirical average reward of a tree node (h, i) is:
Figure 646151DEST_PATH_IMAGE025
；
wherein the content of the first and second substances,
Figure 160177DEST_PATH_IMAGE026
is awarded for the t-th round.
6. The discovery system of claim 4, wherein said IOT service space model is based on said service history tree
Figure 913370DEST_PATH_IMAGE027
And context space
Figure 529159DEST_PATH_IMAGE013
Refreshing the service discovery tree
Figure 9688DEST_PATH_IMAGE028
And updating the B value, including:
for the
Figure 394533DEST_PATH_IMAGE029
To judge the existence
Figure 53047DEST_PATH_IMAGE030
Not exceeding a set threshold
Figure 421712DEST_PATH_IMAGE031
EN a seeks help from its neighboring hop EN, for EN a
Figure 705931DEST_PATH_IMAGE032
All ENs in time, the empirical average reward is calculated as:
Figure 210862DEST_PATH_IMAGE033
；
computing and updating the service discovery tree
Figure 40278DEST_PATH_IMAGE034
Average empirical consideration of
Figure 145506DEST_PATH_IMAGE035
Then, the B value is calculated and updated, and
Figure 718570DEST_PATH_IMAGE036
adding to the service discovery tree
Figure 812427DEST_PATH_IMAGE034
In (1).
7. The discovery system of claim 1, wherein said service model further comprises a local differential privacy mechanism
Figure 62012DEST_PATH_IMAGE037
To the user
Figure 139690DEST_PATH_IMAGE001
Initial context vector of
Figure 516444DEST_PATH_IMAGE013
And carrying out context random data perturbation processing, and calculating the influence of the perturbation processing on service discovery through a loss function.
8. The discovery system of claim 7 wherein said local differential privacy mechanism
Figure 979655DEST_PATH_IMAGE037
The calculation formula of (2) is as follows:
Figure 150874DEST_PATH_IMAGE038
；
wherein the content of the first and second substances,
Figure 715847DEST_PATH_IMAGE039
as a context vector
Figure 145561DEST_PATH_IMAGE013
Via the local differential privacy mechanism
Figure 479590DEST_PATH_IMAGE037
Perturbing the set of post-operation mappings;
Figure 821710DEST_PATH_IMAGE040
，
Figure 873979DEST_PATH_IMAGE041
，
Figure 841804DEST_PATH_IMAGE042
representing a privacy factor;
the loss function is:
Figure 30340DEST_PATH_IMAGE043
。
9. the discovery system of claim 1 wherein, after said service model selects the best personalized service to recommend to said user, said discovery system calculates a total reward value based on the server-side reward and the client-side reward:
Figure 543361DEST_PATH_IMAGE044
；
Figure 66615DEST_PATH_IMAGE045
and
Figure 323284DEST_PATH_IMAGE046
is a parameter, and
Figure 615594DEST_PATH_IMAGE047
；
Figure 299516DEST_PATH_IMAGE048
server-side rewards obtained from various ENs monitored by the SDN controller,
Figure 795219DEST_PATH_IMAGE049
and rewarding the user terminal determined according to the execution time, the response time and the reliability of the user.
10. The discovery system of claim 1 wherein said discovery system performs offline updates based on completed business records:
in the service space model of the Internet of things, the service history tree is used for the service
Figure 370426DEST_PATH_IMAGE027
Each leaf tree node in (1)
Figure 267975DEST_PATH_IMAGE050
If, if
Figure 388378DEST_PATH_IMAGE051
(ii) a Updating the service history tree:
Figure 902536DEST_PATH_IMAGE052
；
updating the B value:
Figure 15854DEST_PATH_IMAGE053
；
in the context space model, if
Figure 767909DEST_PATH_IMAGE054
Then further dividing the context space
Figure 324793DEST_PATH_IMAGE003
；
Wherein the content of the first and second substances,
Figure 44356DEST_PATH_IMAGE055
representing context vectors
Figure 446518DEST_PATH_IMAGE013
The number of contexts in (a) and (b),
Figure 53080DEST_PATH_IMAGE056
and
Figure 764553DEST_PATH_IMAGE057
indicating that a threshold is set.
CN202110588548.2A 2021-05-28 2021-05-28 Internet of things privacy protection service discovery system based on SDN and edge computing Active CN113163019B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110588548.2A CN113163019B (en) 2021-05-28 2021-05-28 Internet of things privacy protection service discovery system based on SDN and edge computing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110588548.2A CN113163019B (en) 2021-05-28 2021-05-28 Internet of things privacy protection service discovery system based on SDN and edge computing

Publications (2)

Publication Number Publication Date
CN113163019A true CN113163019A (en) 2021-07-23
CN113163019B CN113163019B (en) 2021-09-14

Family

ID=76877965

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110588548.2A Active CN113163019B (en) 2021-05-28 2021-05-28 Internet of things privacy protection service discovery system based on SDN and edge computing

Country Status (1)

Country Link
CN (1) CN113163019B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109543094A (en) * 2018-09-29 2019-03-29 东南大学 A kind of secret protection content recommendation method based on matrix decomposition
US20190138934A1 (en) * 2018-09-07 2019-05-09 Saurav Prakash Technologies for distributing gradient descent computation in a heterogeneous multi-access edge computing (mec) networks
CN110198302A (en) * 2019-04-26 2019-09-03 华中科技大学 A kind of method for secret protection and system for intelligent electric meter data publication
CN110650487A (en) * 2019-09-27 2020-01-03 常熟理工学院 Internet of things edge computing configuration method based on data privacy protection
US20200159961A1 (en) * 2017-12-28 2020-05-21 Ned M. Smith Privacy-preserving sanitization for visual computing queries
CN112035755A (en) * 2020-07-14 2020-12-04 中国科学院信息工程研究所 User-centered personalized recommendation privacy protection method and system

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200159961A1 (en) * 2017-12-28 2020-05-21 Ned M. Smith Privacy-preserving sanitization for visual computing queries
US20190138934A1 (en) * 2018-09-07 2019-05-09 Saurav Prakash Technologies for distributing gradient descent computation in a heterogeneous multi-access edge computing (mec) networks
CN109543094A (en) * 2018-09-29 2019-03-29 东南大学 A kind of secret protection content recommendation method based on matrix decomposition
CN110198302A (en) * 2019-04-26 2019-09-03 华中科技大学 A kind of method for secret protection and system for intelligent electric meter data publication
CN110650487A (en) * 2019-09-27 2020-01-03 常熟理工学院 Internet of things edge computing configuration method based on data privacy protection
CN112035755A (en) * 2020-07-14 2020-12-04 中国科学院信息工程研究所 User-centered personalized recommendation privacy protection method and system

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
DIFAN MU: "Privacy-Preserving MEC-Enabled Contextual Online Learning via SDN for Service Selection in IoT", 《 2019 IEEE 16TH INTERNATIONAL CONFERENCE ON MOBILE AD HOC AND SENSOR SYSTEMS (MASS)》 *
JUN WANG: "A Differentially Private Unscented Kalman Filter for Streaming Data in IoT", 《IEEE ACCESS》 *
李晖等: "移动互联服务与隐私保护的研究进展", 《通信学报》 *
苗秋成: "基于边缘辅助连接的位置差分隐私保护的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Also Published As

Publication number Publication date
CN113163019B (en) 2021-09-14

Similar Documents

Publication Publication Date Title
Poluru et al. An improved fruit fly optimization (IFFOA) based cluster head selection algorithm for internet of things
Lv et al. Big data driven hidden Markov model based individual mobility prediction at points of interest
Ahad et al. Neural networks in wireless networks: Techniques, applications and guidelines
US10217346B1 (en) Presence detection with neural networks
US9547827B2 (en) Terminal device, terminal control method, program and information processing system
WO2022000365A1 (en) Machine learning based downlink channel estimation and prediction
JP5536485B2 (en) Portable terminal, server, program, and method for estimating address / location as user moves
Liu et al. User recruitment for enhancing data inference accuracy in sparse mobile crowdsensing
CN110190918B (en) Cognitive wireless sensor network spectrum access method based on deep Q learning
US10275628B2 (en) Feature summarization filter with applications using data analytics
CN108829766B (en) Interest point recommendation method, system, equipment and computer readable storage medium
Gao et al. Distributed maintenance of cache freshness in opportunistic mobile networks
JP2018142957A (en) Management device, program for making computer execute, and computer-readable recording medium recording program
Papandrea et al. Location prediction and mobility modelling for enhanced localization solution
Wang et al. Attentional Markov model for human mobility prediction
JP6627496B2 (en) Management device, program to be executed by computer, and computer-readable recording medium recording the program
JP2015184764A (en) behavior prediction apparatus, behavior prediction method, and behavior prediction program
Yan et al. Two-dimensional task offloading for mobile networks: An imitation learning framework
Tuba et al. Energy efficient sink placement in wireless sensor networks by brain storm optimization algorithm
Long et al. Model-agnostic decentralized collaborative learning for on-device POI recommendation
Oh Using an adaptive search tree to predict user location
CN113163019B (en) Internet of things privacy protection service discovery system based on SDN and edge computing
CN107909498B (en) Recommendation method based on area below maximized receiver operation characteristic curve
Hasanin et al. [Retracted] Efficient Multiuser Computation for Mobile‐Edge Computing in IoT Application Using Optimization Algorithm
Chen et al. Context-aware online offloading strategy with mobility prediction for mobile edge computing

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