CN112202762A - Game defense strategy optimization method and system for sensing edge cloud intelligent interference attack - Google Patents
Game defense strategy optimization method and system for sensing edge cloud intelligent interference attack Download PDFInfo
- Publication number
- CN112202762A CN112202762A CN202011039611.9A CN202011039611A CN112202762A CN 112202762 A CN112202762 A CN 112202762A CN 202011039611 A CN202011039611 A CN 202011039611A CN 112202762 A CN112202762 A CN 112202762A
- Authority
- CN
- China
- Prior art keywords
- cluster head
- attacker
- intelligent
- sensing equipment
- intelligent interference
- 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
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
- H04L63/1408—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic by monitoring network traffic
- H04L63/1416—Event detection, e.g. attack signature detection
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
- H04L63/1441—Countermeasures against malicious traffic
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/20—Network architectures or network communication protocols for network security for managing network security; network security policies in general
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Computer Hardware Design (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mobile Radio Communication Systems (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a game defense strategy optimization method and system for intelligent interference attack in sensing edge cloud. The method comprises the following steps: (1) acquiring a transmission power vector distributed to the computing task of an initial sensor equipment cluster head node set; (2) calculating a power distribution vector of the intelligent interference attacker when the game effect of the intelligent interference attacker is maximized according to a Stark-Boolean model; (3) calculating a transmission power distribution vector of the cluster head node of the sensing equipment when the Nash equilibrium point is reached according to a Stark Boolean model; (4) a decision configuration variable is determined. The system comprises an initialization module, an intelligent interference attacker prediction module, a defense strategy decision module and a configuration module. The invention can effectively defend the intelligent interference attacker with learning ability and provides a defense method for resisting the intelligent interference attack.
Description
Technical Field
The invention belongs to the technical field of Internet of things, and particularly relates to a game defense strategy optimization method and system for sensing edge cloud intelligent interference attack.
Background
The computing task of the sensing equipment can be unloaded to the edge service node through the service access point or the base station node, so that the resource consumption of the sensing equipment is greatly reduced, and the service quality of a user is improved. However, in an open environment, the sensing device computing task offloading process is vulnerable to intelligent interference attacks.
The sensing edge cloud system integrates sensing capability, control capability, communication capability and computing capability, and is widely applied to the field of industrial internet. An edge service node on the edge side in the sensing edge cloud system responds to a request of a sensing device node through an open wireless environment and receives a computing task from the sensing device node. In consideration of the complex wireless communication characteristics at the edge side, in the process of offloading the computing task between the sensing device and the edge service node, especially for the intelligent interference attack of the delay-sensitive computing task, the edge computing performance is reduced or the task offloading fails. Therefore, secure communication between the sensing devices and the edge service bed pole device cluster head nodes presents a significant challenge.
The cluster head node of the sensing equipment is used as an defender, because the channel gain of an intelligent interference attacker is difficult to capture, particularly the DDoS (distributed noise of service attack) attack initiated by the intelligent interference attacker on multiple channels, the more channels are attacked, the more the defender needs high calculation cost to obtain an optimized defense strategy. Bhattacharya et al formalizes a zero-sum chase-escape Game to calculate optimization strategies and performs a Game-interactive analysis of an analog jammying attack on a UAV communication network [ C ] against UAV (unmanned aircraft vehicle) interference attacks. Xiao et al consider that intelligent UAVs interfere with attackers, can specify attack types such as interference Attacks, eavesdropping Attacks, spoofing Attacks, etc., and defend Against these Attacks based on a power distribution strategy that is reinforcement learning (User-central View of Unmanned Aerial Vehicle Transmission Attacks). Xu et al considered incomplete channel state information and studied the competitive interaction process between UAV users and interference attackers using the Bayesian Stackelberg Game (a One-Leader Multi-Follower Bayesian-Stackelberg Game for Anti-learning Transmission in UAV Communication Networks J.). Xu et al, the defender uses the Stackelberg Anti-interference attack Game to evaluate the impact of the observation error of the intelligent interference attack on the defense performance, and obtains Nash equilibrium (A One-Leader Multi-Follower Bayesian-Stackelberg Game for Anti-Jamming Transmission in UAV Communication Networks).
These solutions present the following disadvantages:
(1) the proposed method has limited consideration on incomplete channel state information, so that selection of optimal strategies of game participants faces complexity, and when attack and defense strategies of both game participants change, the proposed method does not provide a quick reasoning function to realize defense strategy selection.
(2) Although the proposed solution designs a learning-based defense strategy, it does not consider how to defend against a learning-capable intelligent distracting attacker.
(3) The proposed solution needs to solve the complex problem of the multi-channel attack of the intelligent interference attacker to the computation task offloading, which greatly reduces the defense performance.
Disclosure of Invention
Aiming at the defects or improvement requirements of the prior art, the invention provides a game defense strategy optimization method and a game defense strategy optimization system for sensing intelligent interference attack of a marginal cloud, and aims to predict an attack strategy of an intelligent interference attacker according to a game model and intelligently optimize the defense strategy according to the predicted attack strategy, so that the technical problems that the prior art cannot defend the intelligent interference attacker with learning ability or cannot use a scheme due to too complex defense scheme because channel state information cannot be completely achieved.
In order to achieve the above object, according to an aspect of the present invention, there is provided a game defense strategy optimization method for sensing smart interference attack in a marginal cloud, including the following steps:
(1) acquiring a transmission power vector P distributed to the computing task of an initial sensor device cluster head node set: p ═ P1,P2,...,Pm) Wherein m is the number of available channel resources of the cluster head node of the sensing equipment;
(2) According to a Stark-Boolean model, a device cluster head node is taken as a leader, an intelligent interference attacker is taken as a follower, and according to channel gain vectors of n sensing device cluster head nodes attacked by the intelligent interference attacker, when the game effect of the intelligent interference attacker is maximized, the power distribution vector J of the intelligent interference attacker to the n attacked channels is calculatedNNAs a power allocation strategy for intelligent interference attackers;
(3) according to the Stark-Boolean model and on the premise that the intelligent interference attacker adopts the power distribution strategy in the step (2), calculating the game effectiveness of the maximized sensing equipment cluster head node according to the channel gain vectors of the m channels of the sensing equipment cluster head node so as to reach the Nash equilibrium point, and distributing a vector P to the transmission power of the m available channels by the sensing equipment cluster head nodeMMAs a power distribution strategy for the cluster head nodes of the sensing equipment;
(4) according to the power distribution strategy P of the cluster head nodes of the sensor equipment obtained in the step (3)MMAnd determining decision configuration variables and unloading tasks.
Preferably, in the method for optimizing a game defense strategy under intelligent interference attack in a sensing edge cloud, the method for calculating the game utility of the intelligent interference attacker in step (2) is as follows:
wherein n is the total number of channels attacked by the intelligent interference attacker, P is the transmission power vector allocated to the calculation task by the sensor device cluster head node set, and J is the transmission power vector J ═ of the intelligent interference attacker (J)1,J2,...,Jn);as,iThe using state of the ith channel of the cluster head node of the s-th sensing equipment is as as,iWhen the number is 1, the ith channel of the cluster head node of the s sensing equipment is used for unloading the calculation task, otherwise, as,i=0;hs,iAscending meter for ith channel of cluster head node of s-th sensing equipmentComputing task offload link channel gain, PiIs the transmission power of the cluster head node of the sensing equipment of the ith channel, n0,iIs the noise power of the ith channel, hJ,iChannel gain at ith channel for intelligent interference attackers, JiThe transmission power of the intelligent interference attacker in the ith channel, and gamma is the interference attack cost per unit interference power of the intelligent interference attacker.
Preferably, the game defense strategy optimization method under intelligent interference attack in the sensing edge cloud includes the steps of (2) calculating a power distribution strategy of the intelligent interference attacker when the game effect of the intelligent interference attacker is maximized, establishing an intelligent attack model by adopting a deep neural network, and establishing an intelligent attack model according to channel gain vectors H of channels of the cluster head nodes of the n sensing devices under attack by the intelligent interference attackers,i=(hs,1,hs,2,...,hs,n) And predicting a power allocation strategy of the intelligent interference attacker.
Preferably, in the method for optimizing game defense strategies under intelligent interference attacks in the sensing edge cloud, the game effectiveness of the maximized intelligent interference attacker in the step (2) is recorded as:
wherein, JmaxThe maximum transmission power for an intelligent interference attacker is constant.
The intelligent attack model established by adopting the deep neural network comprises an input layer, a normalization layer, a full connection layer, a data shaping layer, a convolution module, a pooling layer group and an output layer which are sequentially connected;
the input layer is used for inputting channel gain vectors H of channels of n sensing equipment cluster head nodes attacked by intelligent interference attackerss,i=(hs,1,hs,2,...,hs,n) To a normalization layer;
the normalization layer is used for converting channel gain vector H of cluster head node of the sensing equipments,i=(hs,1,hs,2,...,hs,n) Normalization processing is carried out to obtain normalized channel gain vectors of cluster head nodes of the sensing equipmentAnd input to the data shaping layer through the full connection layer;
the data shaping layer is used for passing the output of the full connection layerAfter shaping, the matrix is converted into a two-dimensional matrix and input to the convolutional layer.
The convolution module comprises two convolution layers, the two convolution kernels are respectively connected through Relu linear rectification functions, and each convolution kernel comprises n convolution kernels; the channel gain of the cluster head node of the shaped normalized sensing equipment is output through an activation function after the first convolution operation is carried out through a corresponding convolution kernel, is output to a pooling layer group through the activation function again after the second convolution operation is carried out through the corresponding convolution kernel;
the pooling layer group comprises n pooling layers connected in parallel and a full-connection layer;
the output layer is used for outputting intelligent interference attacker transmission power vectors (J) of Nash equilibrium points existing in the Stackelberg gameN,1,JN,2,...,JN,n) I.e. a power allocation strategy for the intelligent interference attacker.
Preferably, the game defense strategy optimization method under intelligent interference attack in the sensing edge cloud adopts an intelligent attack model established by a deep neural network, and is trained and obtained according to the following method:
randomly initializing a multi-channel training weight vectorTraining by using a gradient descent method, gradually transmitting in opposite phase to adjust the weight, and allowing an intelligent interference attacker to pass throughInteractively acquiring the transmission power of the cluster head node of the sensing equipment; the loss function of the intelligent attack model is represented as follows:
wherein alpha isJ,iWeight coefficient representing loss function, (1-alpha)J,i)tanh(|Ji-Jmax|) is a regularization term, the power constraint of the intelligent interference attacker participates in the training,
the weight value updating equation for training the intelligent attack model is as follows:
wherein, thetaJIndicating the learning rate.
Preferably, in the method for optimizing the game defense strategy under the intelligent interference attack in the sensing edge cloud, the method for calculating the game utility of the sensing device cluster head node in step (3) is as follows:
wherein m is the total number of channel resources available to the cluster head nodes of the sensor device, P is the transmission power vector allocated to the calculation task by the cluster head node set of the sensor device, and J is the transmission power vector J ═ of the intelligent interference attacker (J)1,J2,...,Jn);as,iThe using state of the ith channel of the cluster head node of the s-th sensing equipment is as as,iWhen the number is 1, the ith channel of the cluster head node of the s sensing equipment is used for unloading the calculation task, otherwise, as,i=0;hs,iUnloading link channel gain, P, for the calculation task of ith channel uplink of cluster head node of the s-th sensing equipmentiFor transmission of the ith channelTransmission power of cluster head node of inductive device, n0,iIs the noise power of the ith channel, hJ,iChannel gain at ith channel for intelligent interference attackers, JiAnd lambda is the transmission cost per unit transmission power of the cluster head node of the sensing equipment.
Preferably, in the game defense strategy optimization method under intelligent interference attack in the sensing edge cloud, when the game effectiveness of the sensing equipment cluster head nodes is calculated and maximized so as to reach the nash equilibrium point, the transmission power distribution strategy of the sensing equipment cluster head nodes adopts a deep neural network to establish an only-defense model, and a power distribution strategy of the sensing equipment cluster head nodes is obtained.
Preferably, in the method for optimizing game defense strategies under intelligent interference attack in the sensing edge cloud, the game utility of the maximized sensing device cluster head node in step (3) is recorded as:
wherein, PmaxThe maximum transmission power of the cluster head node of the sensing equipment is a constant.
The intelligent defense model established by adopting the deep neural network comprises an input layer, a normalization layer, a full connection layer, a data shaping layer, a convolution module, a pooling layer group and an output layer which are sequentially connected;
the input layer is used for inputting channel gain vectors H of m channels of the cluster head nodes of the sensing equipments,i=(hs,1,hs,2,...,hs,m) To a normalization layer;
the normalization layer is used for converting channel gain vector H of cluster head node of the sensing equipments,i=(hs,1,hs,2,...,hs,m) Performing normalization to obtain final productNormalized sensing equipment cluster head node channel gain vectorAnd input to the data shaping layer through the full connection layer;
the data shaping layer is used for passing the output of the full connection layerAfter shaping, the matrix is converted into a two-dimensional matrix and input to the convolutional layer.
The convolution module comprises two convolution layers, the two convolution kernels are respectively connected through Relu linear rectification functions, and each convolution kernel comprises m convolution kernels; the channel gain of the cluster head node of the shaped normalized sensing equipment is output through an activation function after the first convolution operation is carried out through a corresponding convolution kernel, is output to a pooling layer group through the activation function again after the second convolution operation is carried out through the corresponding convolution kernel;
the pooling layer group comprises m pooling layers connected in parallel and a full-connection layer;
the output layer is used for outputting an input power vector P of a cluster head node of the sensing equipment of a Nash equilibrium point existing in the Stackelberg gameMM=(PM,1,PM,2,...,PM,m) Namely, the power distribution strategy of the cluster head node of the sensor equipment is obtained.
Preferably, in the method for optimizing the game defense strategy under the intelligent interference attack in the sensing edge cloud, the intelligent defense model in the step (3) is obtained by training according to the following method:
randomly initializing a multi-channel training weight vectorTraining by using a gradient descent method, gradually transmitting and adjusting the weight in a reverse phase manner, and acquiring the transmission power of an intelligent interference attacker by a cluster head node of the sensing equipment through interaction; the loss function of the intelligent attack model is represented as follows:
wherein alpha issRepresenting weight coefficients of the loss function, balancing the influence of the constraint on the training process; (1-. alpha.) with a high degree of polymerizations)tanh(|P-Pmax|) is a regularization term, power constraints of cluster head nodes of the sensing equipment participate in training,by devitalizing the power in the loss function, we obtain:
the weight updating equation for training the intelligent defense model is as follows:
wherein, thetasIndicating the learning rate.
According to another aspect of the invention, a game defense strategy optimization system attacked by intelligent interference in a sensing edge cloud is provided, and comprises an initialization module, an intelligent interference attacker prediction module, a defense strategy decision module and a configuration module;
the initialization module is configured to acquire a transmission power vector P allocated to the computation task of an initial cluster head node set of the sensor device: p ═ P1,P2,...,Pm) And submitting the data to the intelligent interference attacker prediction module, wherein m is the number of available channel resources of the cluster head node of the sensing equipment;
the intelligent interference attacker prediction module is used for taking the equipment cluster head node as a leader and the intelligent interference attacker as a follower according to a Stark Boolean model, and calculating n attacking intelligent interference attackers according to the channel gain vectors of the channels of the n sensing equipment cluster head nodes attacked by the intelligent interference attacker when the game effect of the intelligent interference attacker is maximizedPower allocation vector J of channelNNThe power distribution strategy is submitted to the defense strategy decision module as a power distribution strategy of an intelligent interference attacker;
the defense strategy decision module is used for calculating and maximizing the game effectiveness of the cluster head nodes of the sensing equipment according to the channel gain vectors of m channels of the cluster head nodes of the sensing equipment on the premise that the intelligent interference attacker adopts a power distribution strategy according to a Stark-Boolean model so as to reach a Nash equilibrium point, and the cluster head nodes of the sensing equipment distribute vectors of transmission power of the m available channels as the power distribution strategy of the cluster head nodes of the sensing equipment and submit the power distribution strategies to the configuration module;
the configuration module is used for distributing the strategy P according to the power of the cluster head node of the sensor equipmentMMAnd determining decision configuration variables and unloading tasks. In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
in the invention, aiming at the problem of safety task unloading between the sensing equipment and the edge service node in the sensing edge cloud environment, the cluster head node of the sensing equipment with sufficient Stackelberg game resources optimizes a power distribution strategy and defends intelligent interference attack through learning, so that an intelligent interference attacker with learning capability can be effectively defended, and a defense method for resisting the intelligent interference attack is provided.
Preferably, the method is a Stackelberg game strategy learning process based on DNN (deep neural network), wherein a cluster head node of the sensing equipment is used as a leader role, and an intelligent interference attacker is used as a follower role. Firstly, an intelligent interference attacker acquires a transmission power allocation strategy of a cluster head node of the sensing equipment, learns an optimal power allocation strategy through own channel gain and maximizes the game effectiveness of the intelligent interference attacker. Secondly, the sensing equipment cluster head node acquires a power distribution strategy of the intelligent interference attacker, learns the optimal power distribution strategy through self channel gain, maximizes the game effectiveness, and can effectively perform defense decision under the condition that the power distribution strategy information of the intelligent interference attacker is incomplete
Drawings
FIG. 1 is a schematic diagram of a sensing edge cloud architecture to which the present invention is directed;
FIG. 2 is a schematic structural diagram of an intelligent attack model provided by an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a single-channel intelligent attack model provided by an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of an intelligent defense model provided by an embodiment of the invention;
fig. 5 is a schematic structural diagram of a single-channel intelligent defense model provided by an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The sensing edge cloud system comprises sensing equipment nodes, sensing equipment cluster head nodes, edge service nodes and intelligent interference attackers. The edge service node is composed of an AP access point and a micro cloud service. As shown in fig. 1. When the intelligent interference attack is received, the sensing equipment nodes unload the calculation tasks to the edge service nodes through the sensing equipment cluster head nodes, and defense strategies are implemented on intelligent interference attackers with learning capacity by utilizing the calculation capacity of the sensing equipment cluster head nodes.
The invention provides a game defense strategy optimization method under intelligent interference attack in a sensing edge cloud, which comprises the following steps:
(1) acquiring a transmission power vector P distributed to the computing task of an initial sensor device cluster head node set: p ═ P1,P2,...,Pm) Wherein m is the number of available channel resources of the cluster head node of the sensing equipment;
(2) according to the Stark-Boolean model, the device cluster head node is taken as a leader, and the intelligent interference attacker is taken as an attackerA follower calculates power distribution vector J of the intelligent interference attacker to the n channels attacked by the intelligent interference attacker when the game effect of the intelligent interference attacker is maximized according to the channel gain vectors of the channels of the n sensing equipment cluster head nodes attacked by the intelligent interference attackerNNAs a power allocation strategy for intelligent interference attackers.
The method for calculating the game utility of the intelligent interference attacker comprises the following steps:
wherein n is the total number of channels attacked by the intelligent interference attacker, P is the transmission power vector allocated to the calculation task by the sensor device cluster head node set, and J is the transmission power vector J ═ of the intelligent interference attacker (J)1,J2,...,Jn);as,iThe using state of the ith channel of the cluster head node of the s-th sensing equipment is as as,iWhen the number is 1, the ith channel of the cluster head node of the s sensing equipment is used for unloading the calculation task, otherwise, as,i=0;hs,iUnloading link channel gain, P, for the calculation task of ith channel uplink of cluster head node of the s-th sensing equipmentiIs the transmission power of the cluster head node of the sensing equipment of the ith channel, n0,iIs the noise power of the ith channel, hJ,iChannel gain at ith channel for intelligent interference attackers, JiThe transmission power of the intelligent interference attacker in the ith channel, and gamma is the interference attack cost per unit interference power of the intelligent interference attacker.
When the game effect of the intelligent interference attacker is maximized, the power distribution strategy of the intelligent interference attacker preferably adopts a deep neural network to establish an intelligent attack model, and channel gain vectors H of the channels of the n sensing equipment cluster head nodes attacked by the intelligent interference attacker are calculated according to the channel gain vectors Hs,i=(hs,1,hs,2,...,hs,n) Predicting a power distribution strategy of an intelligent interference attacker; the method comprises the following specific steps:
the game utility of the maximized intelligent interference attacker is recorded as:
wherein, JmaxThe maximum transmission power for an intelligent interference attacker is a constant.
The intelligent attack model established by adopting the deep neural network has a structure as shown in fig. 2 and comprises an input layer, a normalization layer, a full connection layer, a data shaping layer, a convolution module, a pooling layer group and an output layer which are sequentially connected;
the input layer is used for inputting channel gain vectors H of channels of n sensing equipment cluster head nodes attacked by intelligent interference attackerss,i=(hs,1,hs,2,...,hs,n) To a normalization layer;
the normalization layer is used for converting channel gain vector H of cluster head node of the sensing equipments,i=(hs,1,hs,2,...,hs,n) Normalization processing is carried out to obtain normalized channel gain vectors of cluster head nodes of the sensing equipmentAnd input to the data shaping layer through the full connection layer; the calculation method is as follows:
where E (-) represents expectation and D (-) represents variance.
The data shaping layer is used for passing the output of the full connection layerAfter shaping, the matrix is converted into a two-dimensional matrix and input to a convolutional layer。
The convolution module comprises two convolution layers, the two convolution kernels are respectively connected through Relu linear rectification functions, and each convolution kernel comprises n convolution kernels; the channel gain of the cluster head node of the shaped normalized sensing equipment is output through an activation function after the first convolution operation is carried out through a corresponding convolution kernel, is output to a pooling layer group through the activation function again after the second convolution operation is carried out through the corresponding convolution kernel;
the first convolution operation, using convolution kernel with 3 × 3 step length being 1, extracts n pieces of channel gain information of intelligent interference attacker, and the convolution operation output is as follows:
wherein the content of the first and second substances,as a weight, the output vector of the first convolution operation is
And (3) after the first convolution operation, adopting a Relu function as an activation function, and outputting a vector:wherein
The second convolution operation, using a convolution kernel with a 3 × 3 step size of 1, outputs the following:the output vector of the second convolution operation is
After the second convolution operation, a Relu function is adopted as an activation function, and output is carried outVector quantity:wherein
The pooling layer group comprises n pooling layers connected in parallel and a full-connection layer; preferably, in order to accelerate the training speed, the maximized pooling layer is used, and the output vector of the maximized pooling layer is obtained through a sliding window with 2 × 2 step length being 1
The output layer preferably adopts a Sigmoid function and outputs an intelligent interference attacker transmission power vector (J) of a Nash equilibrium point existing in the Stackelberg gameN,1,JN,2,...,JN,n) I.e. a power allocation strategy for the intelligent interference attacker; preferably a Sigmoid function is used as the output layer,
the intelligent attack model established by adopting the deep neural network is obtained by training according to the following method:
randomly initializing a multi-channel training weight vectorTraining by using a gradient descent method, gradually transmitting and adjusting the weight in a reverse phase manner, and obtaining the transmission power of the cluster head node of the sensing equipment by an intelligent interference attacker through interaction; the loss function of the intelligent attack model is represented as follows:
wherein alpha isJ,iWeight coefficient representing loss function, (1-alpha)J,i)tanh(|Ji-Jmax|) as regularization term, power constraint of intelligent interference attacker participates in trainingThe refining is carried out by the following steps,by devitalizing the power in the loss function, we obtain:
the weight value updating equation for training the intelligent attack model is as follows:
wherein, thetaJIndicating the learning rate.
After the intelligent attack model is trained, the trained intelligent attack model is used for predicting a transmission power strategy of an intelligent interference attacker, and when the channel gain vectors of the cluster head nodes of the sensing equipment of n channels attacked by the intelligent interference attacker are input, the power distribution strategy J of the intelligent interference attacker output by the intelligent attack modelNN=(JN,1,JN,2,...,JN,n)。
(3) According to a Stark-Boolean model, on the premise that the intelligent interference attacker adopts the power distribution strategy in the step (2), calculating the transmission power distribution vector of the sensing equipment cluster head node to the m available channels according to the channel gain vectors of the m channels of the sensing equipment cluster head node to maximize the game effectiveness of the sensing equipment cluster head node so as to reach a Nash equilibrium point, and taking the transmission power distribution vector of the sensing equipment cluster head node to the m available channels as the power distribution strategy of the sensing equipment cluster head node;
the method for calculating the game utility of the cluster head node of the sensing equipment comprises the following steps:
wherein m is the total number of channel resources available for the cluster head nodes of the sensor equipment, and P is the allocation of the cluster head node set of the sensor equipment to the cluster head nodesCalculating task transmission power vector, wherein J is transmission power vector J of intelligent interference attacker (J ═ J)1,J2,...,Jn);as,iThe using state of the ith channel of the cluster head node of the s-th sensing equipment is as as,iWhen the number is 1, the ith channel of the cluster head node of the s sensing equipment is used for unloading the calculation task, otherwise, as,i=0;hs,iUnloading link channel gain, P, for the calculation task of ith channel uplink of cluster head node of the s-th sensing equipmentiIs the transmission power of the cluster head node of the sensing equipment of the ith channel, n0,iIs the noise power of the ith channel, hJ,iChannel gain at ith channel for intelligent interference attackers, JiAnd lambda is the transmission cost per unit transmission power of the cluster head node of the sensing equipment.
When the game effectiveness of the sensing equipment cluster head nodes is calculated and maximized so as to reach a Nash equilibrium point, a transmission power distribution strategy of the sensing equipment cluster head nodes is preferably established by adopting a deep neural network to obtain a power distribution strategy of the sensing equipment cluster head nodes; the method comprises the following specific steps:
the game utility of the cluster head node of the maximized sensing equipment is recorded as follows:
wherein, PmaxThe maximum transmission power of the cluster head node of the sensing equipment is a constant.
The intelligent defense model established by adopting the deep neural network has a structure as shown in fig. 4, and comprises an input layer, a normalization layer, a full connection layer, a data shaping layer, a convolution module, a pooling layer group and an output layer which are sequentially connected;
the input layer is used for inputting cluster head nodes of the sensing equipmentOf m channels of the channel gain vector Hs,i=(hs,1,hs,2,...,hs,m) To a normalization layer;
the normalization layer is used for converting channel gain vector H of cluster head node of the sensing equipments,i=(hs,1,hs,2,...,hs,m) Normalization processing is carried out to obtain normalized channel gain vectors of cluster head nodes of the sensing equipmentAnd input to the data shaping layer through the full connection layer; the calculation method is as follows:
where E (-) represents expectation and D (-) represents variance.
The data shaping layer is used for passing the output of the full connection layerAfter shaping, the matrix is converted into a two-dimensional matrix and input to the convolutional layer.
The convolution module comprises two convolution layers, the two convolution kernels are respectively connected through Relu linear rectification functions, and each convolution kernel comprises m convolution kernels; the channel gain of the cluster head node of the shaped normalized sensing equipment is output through an activation function after the first convolution operation is carried out through a corresponding convolution kernel, is output to a pooling layer group through the activation function again after the second convolution operation is carried out through the corresponding convolution kernel;
extracting m pieces of channel gain information of the intelligent interference attacker by using a convolution kernel with a 3 multiplied by 3 step length of 1 in the first convolution operation, wherein the convolution operation outputs the following steps:
wherein the content of the first and second substances,as a weight, the output vector of the first convolution operation is
And (3) after the first convolution operation, adopting a Relu function as an activation function, and outputting a vector:wherein
The second convolution operation, using a convolution kernel with a 3 × 3 step size of 1, outputs the following:the output vector of the second convolution operation is
And (3) after the second convolution operation, adopting a Relu function as an activation function, and outputting a vector:wherein
The pooling layer group comprises m pooling layers connected in parallel and a full-connection layer; preferably, in order to accelerate the training speed, the maximized pooling layer is used, and the output vector of the maximized pooling layer is obtained through a sliding window with 2 × 2 step length being 1
The output layer is used for outputting an input power vector P of a cluster head node of the sensing equipment of a Nash equilibrium point existing in the Stackelberg gameMM=(PM,1,PM,2,...,PM,m) I.e. as said cluster of sensor devicesA power allocation policy of the head node; preferably a Sigmoid function is used as the output layer,
training and obtaining are carried out according to the following method:
randomly initializing a multi-channel training weight vectorTraining by using a gradient descent method, gradually transmitting and adjusting the weight in a reverse phase manner, and acquiring the transmission power of an intelligent interference attacker by a cluster head node of the sensing equipment through interaction; the loss function of the intelligent attack model is represented as follows:
wherein alpha issRepresenting weight coefficients of the loss function, balancing the influence of the constraint on the training process; (1-. alpha.) with a high degree of polymerizations)tanh(|P-Pmax|) is a regularization term, power constraints of cluster head nodes of the sensing equipment participate in training,by devitalizing the power in the loss function, we obtain:
the weight updating equation for training the intelligent defense model is as follows:
wherein, thetasIndicating the learning rate.
After the intelligent defense model is trained, the trained intelligent defense model is used for deciding the transmission power strategy of the cluster head nodes of the sensor equipment, and when m sensor equipment nodes are inputWhen the channel gain vector is obtained, the power distribution strategy P of the sensing equipment cluster head node output by the intelligent defense modelMM。
(4) According to the power distribution strategy P of the cluster head nodes of the sensor equipment obtained in the step (3)MMAnd determining decision configuration variables and unloading tasks.
The game defense strategy optimization system under the intelligent interference attack in the sensing edge cloud comprises an initialization module, an intelligent interference attacker prediction module, a defense strategy decision module and a configuration module;
the initialization module is configured to acquire a transmission power vector P allocated to the computation task of an initial cluster head node set of the sensor device: p ═ P1,P2,...,Pm) And submitting the data to the intelligent interference attacker prediction module, wherein m is the number of available channel resources of the cluster head node of the sensing equipment;
the intelligent interference attacker prediction module is used for calculating a power distribution vector J of the intelligent interference attacker to n channels attacked by the intelligent interference attacker according to the Steckelberg model, the device cluster head node as a leader, the intelligent interference attacker as a follower and channel gain vectors of the n sensing device cluster head nodes attacked by the intelligent interference attacker, wherein the channel gain vectors are used for maximizing the game effect of the intelligent interference attackerNNThe power distribution strategy is submitted to the defense strategy decision module as a power distribution strategy of an intelligent interference attacker;
the defense strategy decision module is used for calculating and maximizing the game effectiveness of the cluster head nodes of the sensing equipment according to the channel gain vectors of m channels of the cluster head nodes of the sensing equipment on the premise that the intelligent interference attacker adopts a power distribution strategy according to a Stark-Boolean model so as to reach a Nash equilibrium point, and the cluster head nodes of the sensing equipment distribute vectors of transmission power of the m available channels as the power distribution strategy of the cluster head nodes of the sensing equipment and submit the power distribution strategies to the configuration module;
the configuration module is used for distributing the strategy P according to the power of the cluster head node of the sensor equipmentMMDetermining a decision configurationAnd (5) carrying out variable task unloading.
The following are examples:
when the sensing equipment cluster head node unloads the delay sensitivity calculation task, the intelligent interference attacker increases the delay time and energy consumption of the calculation task unloading, and reduces the reliability of the channel and the unloading capacity of the calculation task. Let the decision configuration variable for the offloading of the computation task be a ═ as,iAnd | s belongs to M, i belongs to E }, wherein M represents a sensing equipment cluster head node set, and E represents a channel set of sensing equipment cluster head nodes. If the sensing equipment cluster head node s uses the channel i to unload the calculation task, as,i1, otherwise as,i0. Therefore, for a single channel i, the calculation task unloading capacity of the cluster head node s of the sensing equipment is as follows:
the sensing equipment cluster head node s is used as an defender against interference attack, and a certain transmission power is selected to unload a calculation task, so that the game effectiveness of the sensing equipment cluster head node s is as follows:
wherein, P represents the transmission power of the cluster head node s of the sensing equipment, and J represents the transmission power of the intelligent interference attacker. The power of each unit of the cluster head node of the sensing equipment and the intelligent interference attacker is lambda and gamma respectively. n is0Representing the noise power. h iss,iIndicating upstream computational tasks offload link channel gain, hJ,iRepresenting the channel gain of the intelligent interference attacker.
An intelligent interference attacker selects a certain power to interfere with the unloading process of the delay sensitive computing task of the cluster head node s of the sensing equipment. Thus, the game utility of the intelligent distracter attacker is:
when the sensing equipment cluster head node has m available channel resources, an intelligent interference attacker launches a multi-channel attack, so that the transmission of computation tasks unloaded on a plurality of channels fails. Let PiAnd JiIndicating the transmission power allocated to channel i by the cluster head nodes of the sensing devices and the intelligent interference attacker. Let P be (P)1,P2,...,Pm) A transmission power vector representing a sensing device cluster head node. J ═ J (J)1,J2,...,Jn) A transmission power vector representing an intelligent interference attacker. The transmission power of the cluster head node of the sensing equipment and the intelligent interference attacker is satisfiedAnd isUnder the m channel modes, the game effectiveness of the cluster head nodes of the sensing equipment is as follows:
under the multi-channel mode, an intelligent interference attacker carries out interference attack on n channels, and the game effectiveness of the intelligent interference attacker is as follows:
when the intelligent interference attacker with learning ability attacks, the method models the power allocation problem of the anti-interference attack into a Stackelberg game based on DNN. In the game model, the sensing equipment cluster head nodes and the intelligent interference attackers are game participants. The cluster head node of the sensing equipment is a leader, transmission power distribution is firstly carried out, an intelligent interference attacker is a follower, and interference attack is carried out on the calculation task unloading process of the cluster head node of the sensing equipment. The optimal power allocation strategy for each gaming participant is obtained through DNN inference. The invention designs a defense strategy aiming at an intelligent interference attacker with learning capability, the intelligent interference attacker can obtain the transmission power of the cluster head node of the sensing equipment through game interaction, and meanwhile, the power distribution strategy is deduced according to the self channel gain to maximize the game effectiveness. And similarly, the cluster head node of the sensing equipment is used as a defender, the power distribution strategy of the intelligent interference attacker can be obtained through game interaction, and meanwhile, the power distribution strategy of task unloading is calculated through inference according to the channel gain of the sensing equipment.
(1) Acquiring a transmission power vector P distributed to the computing task of an initial sensor device cluster head node set: p ═ P1,P2,...,Pm) Wherein m is the number of available channel resources of the sensing equipment cluster node;
(2) according to a Stark-Boolean model, a device cluster head node is taken as a leader, an intelligent interference attacker is taken as a follower, and according to channel gain vectors of n sensing device cluster head nodes attacked by the intelligent interference attacker, when the game effect of the intelligent interference attacker is maximized, the power distribution vector J of the intelligent interference attacker to the n attacked channels is calculatedNNAs a power allocation strategy for intelligent interference attackers;
when an intelligent interference attacker launches a multi-channel attack on an offload link, maximizing the game utility of the intelligent interference attacker can be formalized as follows:
the embodiment establishes a multi-channel intelligent attack model network MJnet to maximize the multi-channel attack game utility of the intelligent interference attacker. Meanwhile, the optimal attack strategy vector of the intelligent interference attacker in the multi-channel mode is deduced by training the MJnet.
MJnet architecture, as shown in fig. 2:
the MJnet processes the input and output steps of the intelligent interference attacker channel gain as follows:
firstly, inputting: the channel gain of the normalized smart interference attacker multi-channel attack is a standard normal distribution with a mean value of 0 and a variance of 1. The inputs to the MJnet multichannel gain normalization are:
② pass throughAfter shaping, the signal is converted into n two-dimensional matrixes to store channel gains.
Inputting the shaped data into a convolution layer, using n convolution kernels with the step length of 3 multiplied by 3 being 1 to carry out convolution operation for the first time, and extracting n pieces of channel gain information of the intelligent interference attacker.
The output of the first convolution layer is:whereinIs a weight value. Thus, the output vector of the first convolution is
The output of the first convolution layer uses relu as an activation function,for a single output of relu, the multiple output vector is
Sixthly, performing a second convolution operation, wherein the single output is as follows:multiple output vector is
Seventhly, using relu again as an activation function,for a single output of relu, the multiple output vector is
In order to accelerate the training speed, the maximization pooling layer is used, and the output vector of the maximization pooling layer is obtained through a sliding window with the step length of 2 multiplied by 2 being 1
Ninthly using Sigmoid function, a single output beingThe multiple output vector is (J)N,1,JN,2,...,JN,N)。
The MJnet training and reasoning process is as follows:
the intelligent interference attacker infers the multi-channel attack strategy through the MJnet. Therefore, the intelligent interference attacker first randomly initializes the multi-channel training weight vectorAfter initialization is completed, the MJnet is trained by using a random gradient descent method, and weight is gradually and reversely propagated and adjusted, so that the game utility value of an intelligent interference attacker is maximum. The intelligent interference attacker obtains the transmission power of the cluster head nodes of the sensing equipment through interaction. The loss function that maximizes the effectiveness of the game when an intelligent interference attacker makes a multi-channel attack is represented as follows:
wherein,αJ,iWeight coefficient representing loss function, (1-alpha)J,i)tanh(|Ji-Jmax|) is a regularization term, the power constraint of the intelligent interference attacker participates in the training,by aiming the power J in the loss functioniAnd (3) calculating a partial derivative to obtain:
the weight update equation for training the MJnet is as follows:
wherein, thetaJIndicating the learning rate. And after the MJnet training is finished, the trained MJnet is used for reasoning the transmission power strategy of the intelligent interference attacker multi-channel attack. Multi-channel gain vector when inputting intelligent interference attackerThe MJnet outputs an optimized multi-channel attack power policy vector.
When m-n-1, i.e. in the single channel attack mode, maximizing the game effectiveness of the smart disturbance attacker can be particularly simplified as follows:
the MJnet can be specifically simplified into a single-layer SJnet, and the response strategy of the intelligent interference attacker in the single-channel attack mode is learned and inferred by training the SJnet, so that the game effectiveness of the intelligent interference attacker is maximized.
The SJnet structure is shown in fig. 3: the network model consists of normalization layer, full connection layer, data shaping layer, convolution layer and pooling layer
The input and output steps of SJnet processing the intelligent interference attacker channel gain are as follows:
the method aims to accelerate the convergence speed of the game strategy learning of the intelligent interference attacker and ensure that the channel gains of the input intelligent interference attacker are distributed in the same way. Therefore, the input channel gain is normalized to a standard normal distribution with a mean of 0 and a variance of 1. The output of the intelligent interference attacker channel gain normalization is:
where E (-) represents expectation and D (-) represents variance. Normalizing the channel gain vector of the intelligent interference attackerInput SJnet full connectivity layer.
And secondly, data shaping. Output pass through of full connection layerAfter shaping, the matrix is transformed into a two-dimensional matrix.
Inputting the shaped data into a convolution layer, using a convolution kernel with the 3 multiplied by 3 step length of 1 to carry out the first convolution operation, and extracting the channel gain information of the key change of the intelligent interference attacker.
Sixthly, performing a second convolution operation, wherein the output of the convolution layer two is as follows:
To accelerate the training speed, the maximization pooling layer is used, the sliding window with the 2 x 2 step length of 1 is used, and the output of the maximization pooling layer is obtained
Ninthly, stretching the output of the maximized pooling layer into n multiplied by 1 vectors by using the full connection layer.
The SJnet training and reasoning process is as follows:
the intelligent interference attacker infers the attack strategy through SJnet. Therefore, the intelligent interference attacker first randomly initializes the training weightsAfter initialization is completed, a random gradient descent method is used for training SJnet, and weight is gradually and reversely propagated and adjusted, so that the game utility value of an intelligent interference attacker is the maximum. The intelligent interference attacker obtains the transmission power of the cluster head nodes of the sensing equipment through interaction. The loss function for an intelligent distracting attacker to maximize its gambling utility is represented as follows:
wherein alpha isJRepresenting the weight coefficients of the loss function. Item 2 is a regularization item, the interference power constraint of an intelligent interference attacker participates in training,
by devising the power J in the loss function, we obtain:
the weight update equation for training the SJnet is as follows:
wherein, thetaJIndicating the learning rate. After SJnet training is finished, the trained SJnet is used for reasoning the transmission power strategy of the intelligent interference attacker, and when the channel gain vector of the intelligent interference attacker is inputAnd the SJnet outputs an optimized single-channel attack power strategy.
(3) According to a Stark-Boolean model, on the premise that the intelligent interference attacker adopts the power distribution strategy in the step (2), calculating the transmission power distribution vector of the sensing equipment cluster head node to the m available channels according to the channel gain vectors of the m channels of the sensing equipment cluster head node to maximize the game effectiveness of the sensing equipment cluster head node so as to reach a Nash equilibrium point, and taking the transmission power distribution vector of the sensing equipment cluster head node to the m available channels as the power distribution strategy of the sensing equipment cluster head node;
and the sensor equipment cluster head node infers the defense strategy through a deep neural network. When an intelligent interference attacker launches a multi-channel attack on an unloading link, the game effectiveness of the cluster head node of the maximized sensing equipment can be formalized as follows:
and establishing a deep neural network defense model MSnet under a multi-channel attack mode to maximize the game effectiveness of the cluster head nodes of the sensing equipment. Meanwhile, the optimal defense strategy vector of the cluster head node of the sensing equipment in the multi-channel attack mode is deduced by training the MSnet. The MSnet structure is shown in fig. 4.
The MSnet processes the input and output steps of the cluster head node channel gain of the sensing equipment in the multi-channel attack mode as follows:
firstly, inputting: sensing equipment node multichannel gain vector with normalized mean value of 0 and variance of 1
And secondly, data shaping. ThroughAfter shaping, the matrix is transformed into m two-dimensional matrices, corresponding to m channels, respectively.
Inputting the shaped data into a convolution layer, performing a first convolution operation by using m convolution kernels with 3 multiplied by 3 step length of 1, and extracting multi-channel gain information of key change of the cluster head node of the sensing equipment.
The output of the convolution layer uses relu as an activation function,is reluSingle output, multiple output vector of
Sixthly, performing a second convolution operation, using m convolution kernels with 3 × 3 step length of 1, wherein the single output of the convolution layer is as follows:
Seventhly, using relu again as an activation function,for a single output of relu, the multiple output vector is
To accelerate the training speed, the maximization pooling layer is used, and the single output of the maximization pooling layer is obtained through a sliding window with the 2 x 2 step length of 1Multiple output vector is
The MSnet training and reasoning process is as follows:
and the sensing equipment cluster head node infers the defense strategy through the MSnet. Therefore, the cluster head node of the sensing equipment randomly initializes the training weight vector firstlyAfter initialization is completed, the MSnet is trained by using a random gradient descent method, and weight is gradually and reversely propagated and adjusted, so that the game effectiveness of the cluster head node of the sensing equipment is maximum. The sensing equipment cluster head node obtains the transmission power of the intelligent interference attacker through interaction. In the multi-channel attack mode, the loss function of the sensing equipment cluster head node for maximizing the game effectiveness is represented as follows:
wherein alpha iss,iRepresenting the weight coefficient of the loss function, the influence of balance constraint on the training process, (1-alpha)s,i)tanh(|Pi-Pmax|) is a regularization term, the power constraint of the cluster head nodes of the sensing equipment participates in training the weight of the MSnet.
By devitalizing the power in the loss function, we obtain:
the weight update equation for training the MSnet is as follows:
wherein, thetasIndicating the learning rate. And after MSnet training is finished, using the trained MSnet to infer the transmission power strategy vector used for defense by the cluster head node of the sensing equipment. In a multi-channel attack mode, when a channel gain vector of a cluster head node of a sensing device is inputThe MSnet outputs the optimized power strategy vector to enable the DNN to be in a convergence state through a random gradient descent method, and the Stackelberg game based on the DNN has Nash equilibrium point of (J)NN,PMM)。
At this time, under the multi-channel attack mode, the utility of the intelligent interference attacker and the cluster head node of the sensing equipment reaches the maximum.
When m is equal to n is equal to 1, namely in a single-channel attack mode, maximizing the game effectiveness of the cluster head node of the sensing device can be particularly simplified as follows:
the MSnet can be specifically simplified into a single-layer SSnet, and the SSnet is trained to learn and reason about the game strategy of the sensing device cluster head nodes in the single-channel attack mode, so that the game effectiveness of the sensing device cluster head nodes is maximized.
The structure of SSnet is shown in fig. 5: the network model is composed of normalization, full connection layer, data shaping, convolution layer, pooling layer, etc.
The SSnet processes the input and output steps of the cluster head node channel gain of the sensing equipment as follows:
firstly, in order to accelerate the convergence speed of the game strategy learning of the cluster head nodes of the sensing equipment and ensure that the channel gains of the cluster head nodes of the sensing equipment are distributed in the same way. Therefore, the input channel gain is normalized to a standard normal distribution with a mean of 0 and a variance of 1. The normalized output of the cluster head node channel gain of the sensing equipment is as follows:
where E (-) represents expectation and D (-) represents variance. Normalized sensing equipment cluster head node channel gain vectorThe SSnet full connectivity layer is entered.
And secondly, data shaping. Output pass through of full connection layerTransformed into a two-dimensional matrix after shaping。
Inputting the shaped data into a convolution layer, performing a first convolution operation by using a convolution kernel with a 3 multiplied by 3 step length of 1, and extracting channel gain information of key change of the cluster head node of the sensing equipment.
Sixthly, performing a second convolution operation, wherein the output of the convolution layer two is as follows:
To accelerate the training speed, the maximization pooling layer is used, the sliding window with the 2 x 2 step length of 1 is used, and the output of the maximization pooling layer is obtained
Ninthly, stretching the output of the maximized pooling layer into an mx 1 vector using the fully connected layer.
The SSnet training and reasoning process is as follows:
sensingAnd the equipment cluster head node infers the defense strategy through SSnet. Therefore, the cluster head node of the sensing equipment randomly initializes the training weight value firstlyAfter initialization is completed, the SSnet is trained by using a random gradient descent method, and the weight is gradually and reversely propagated and adjusted, so that the game effectiveness of the cluster head node of the sensing equipment is maximum. The sensing equipment cluster head node obtains the transmission power of the intelligent interference attacker through interaction. The loss function of the cluster head node of the sensing equipment for maximizing the game effectiveness is represented as follows:
wherein alpha issRepresenting the weight coefficient of the loss function, the influence of balance constraint on the training process, (1-alpha)s)tanh(|P-Pmax|) is a regularization term, the power constraint of the cluster head nodes of the sensing devices participates in training the weight of the SSnet.
By devitalizing the power P in the loss function, we obtain:
the weight update equation for training the SSnet is as follows:
wherein, thetasIndicating the learning rate. After SSnet training is finished, the trained SSnet is used for reasoning the transmission power strategy for defense of the cluster head node of the sensing equipment, and when the channel gain vector of the cluster head node of the sensing equipment is inputSSnet output optimized power policy PM. The DNN is in a convergent state by a random gradient descent method based onThe Stackelberg game of DNN has a Nash equilibrium point of (J)N,PM). At this time, under the single channel attack mode, the utility of the intelligent interference attacker and the sensing equipment cluster head node reaches the maximum.
4) According to the power distribution strategy P of the cluster head nodes of the sensor equipment obtained in the step (3)MMAnd determining decision configuration variables and unloading tasks.
In the embodiment, under the sensing edge cloud environment, when an attacker with learning ability attacks the sensing device computing task unloading link, the sensing device computing task unloading link is intelligently interfered, and low-complexity and accurate defense is realized. The method specifically comprises the following steps:
in the embodiment, a computation task unloading capacity model of a cluster head node of sensing equipment with sufficient resources is established for a computation task unloading scene of the sensing equipment. And designing a defense strategy aiming at the attack of an intelligent interference attacker with learning ability on the unloading process of the computing task.
The embodiment particularly lists the optimization problem of maximizing the game effectiveness of the cluster head nodes of the intelligent interference attacker and the sensing equipment under the single-channel attack mode. The method comprises the steps of establishing a deep neural network model SJnet for performing game strategy optimization by an intelligent interference attacker in a single-channel attack mode and a deep neural network model SSnet for performing game defense strategy optimization by a sensing equipment cluster head node, respectively training the SJnet and the SSnet by aiming at maximizing the effectiveness of game participants, and enabling the sensing equipment cluster head node to rapidly obtain an optimal power distribution strategy by SSnet reasoning in the single-channel attack mode to defend the intelligent interference attack when the intelligent interference attacker changes the single-channel attack strategy.
Generally, under a multi-channel attack mode, an attacker and a sensing equipment cluster head node are respectively formalized to interfere with the optimization problem of the game effectiveness of the attacker and the sensing equipment cluster head node to the maximum extent. The method comprises the steps of establishing a deep neural network model MJnet for an intelligent interference attacker to carry out multi-channel attack game strategy optimization in a multi-channel attack mode and a deep neural network model MSnet for a sensing equipment cluster head node to carry out multi-channel defense strategy optimization, respectively training the MJnet and the MSnet by taking the game effectiveness of the intelligent interference attacker and the sensing equipment cluster head node in the multi-channel mode as a target, and enabling the sensing equipment cluster head node to rapidly obtain an optimal power distribution strategy vector by MSnet reasoning to defend the multi-channel attack of the intelligent interference attacker in a multi-channel attack scene when the intelligent interference attacker changes the multi-channel attack strategy.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A game defense strategy optimization method for intelligent interference attack in sensing edge cloud is characterized by comprising the following steps:
(1) acquiring a transmission power vector P distributed to the computing task of an initial sensor device cluster head node set: p ═ P1,P2,...,Pm) Wherein m is the number of available channel resources of the cluster head node of the sensing equipment;
(2) according to a Stark-Boolean model, a device cluster head node is taken as a leader, an intelligent interference attacker is taken as a follower, and according to channel gain vectors of n sensing device cluster head nodes attacked by the intelligent interference attacker, when the game effect of the intelligent interference attacker is maximized, the power distribution vector J of the intelligent interference attacker to the n attacked channels is calculatedNNAs a power allocation strategy for intelligent interference attackers;
(3) according to a Stark-Boolean model, on the premise that the intelligent interference attacker adopts the power distribution strategy in the step (2), calculating the transmission power distribution vector of the sensing equipment cluster head node to the m available channels according to the channel gain vectors of the m channels of the sensing equipment cluster head node to maximize the game effectiveness of the sensing equipment cluster head node so as to reach a Nash equilibrium point, and taking the transmission power distribution vector of the sensing equipment cluster head node to the m available channels as the power distribution strategy of the sensing equipment cluster head node;
(4) the cluster head nodes of the sensor equipment obtained according to the step (3)Power allocation strategy PMMAnd determining decision configuration variables and unloading tasks.
2. The method for optimizing game defense strategy under intelligent interference attack in sensing edge cloud as claimed in claim 1, wherein the method for calculating game utility of intelligent interference attacker in step (2) is as follows:
wherein n is the total number of channels attacked by the intelligent interference attacker, P is the transmission power vector allocated to the calculation task by the sensor device cluster head node set, and J is the transmission power vector J ═ of the intelligent interference attacker (J)1,J2,...,Jn);as,iThe using state of the ith channel of the cluster head node of the s-th sensing equipment is as as,iWhen the number is 1, the ith channel of the cluster head node of the s sensing equipment is used for unloading the calculation task, otherwise, as,i=0;hs,iUnloading link channel gain, P, for the calculation task of ith channel uplink of cluster head node of the s-th sensing equipmentiIs the transmission power of the cluster head node of the sensing equipment of the ith channel, n0,iIs the noise power of the ith channel, hJ,iChannel gain at ith channel for intelligent interference attackers, JiThe transmission power of the intelligent interference attacker in the ith channel, and gamma is the interference attack cost per unit interference power of the intelligent interference attacker.
3. The method for optimizing game defense strategy attacked by intelligent interference in sensing edge cloud as claimed in claim 2, wherein the step (2) of calculating the power distribution strategy of the intelligent interference attacker when maximizing the game effect of the intelligent interference attacker, adopting a deep neural network to establish an intelligent attack model, and according to the channel gain vector H of the channel of the cluster head node of the n sensing devices attacked by the intelligent interference attackers,i=(hs,1,hs,2,...,hs,n) And predicting a power allocation strategy of the intelligent interference attacker.
4. The method for optimizing game defense strategy under intelligent interference attack in sensing edge cloud of claim 3, characterized in that the step (2) of maximizing the game effectiveness of intelligent interference attacker is marked as:
wherein, JmaxMaximum transmission power for an intelligent interference attacker;
the intelligent attack model established by adopting the deep neural network comprises an input layer, a normalization layer, a full connection layer, a data shaping layer, a convolution module, a pooling layer group and an output layer which are sequentially connected;
the input layer is used for inputting channel gain vectors H of channels of n sensing equipment cluster head nodes attacked by intelligent interference attackerss,i=(hs,1,hs,2,...,hs,n) To a normalization layer;
the normalization layer is used for converting channel gain vector H of cluster head node of the sensing equipments,i=(hs,1,hs,2,...,hs,n) Normalization processing is carried out to obtain normalized channel gain vectors of cluster head nodes of the sensing equipmentAnd input to the data shaping layer through the full connection layer;
the data shaping layer is used for passing the output of the full connection layerAfter shaping, the matrix is transformed into a two-dimensional matrix,input to the convolutional layer.
The convolution module comprises two convolution layers, the two convolution kernels are respectively connected through Relu linear rectification functions, and each convolution kernel comprises n convolution kernels; the channel gain of the cluster head node of the shaped normalized sensing equipment is output through an activation function after the first convolution operation is carried out through a corresponding convolution kernel, is output to a pooling layer group through the activation function again after the second convolution operation is carried out through the corresponding convolution kernel;
the pooling layer group comprises n pooling layers connected in parallel and a full-connection layer;
the output layer is used for outputting intelligent interference attacker transmission power vectors (J) of Nash equilibrium points existing in the Stackelberg gameN,1,JN,2,...,JN,n) I.e. a power allocation strategy for the intelligent interference attacker.
5. The method for optimizing game defense strategies under intelligent interference attacks in the sensor edge cloud as claimed in claim 4, wherein the intelligent attack model established by the deep neural network is obtained by training according to the following method:
randomly initializing a multi-channel training weight vectorTraining by using a gradient descent method, gradually transmitting and adjusting the weight in a reverse phase manner, and obtaining the transmission power of the cluster head node of the sensing equipment by an intelligent interference attacker through interaction; the loss function of the intelligent attack model is represented as follows:
wherein alpha isJ,iWeight coefficient representing loss function, (1-alpha)J,i)tanh(|Ji-Jmax|) is a regularization term, the power constraint of the intelligent interference attacker participates in the training,
the weight value updating equation for training the intelligent attack model is as follows:
wherein, thetaJIndicating the learning rate.
6. The method for optimizing game defense strategies under intelligent interference attacks in sensing edge clouds according to claim 1, wherein the method for calculating the game utility of the sensing device cluster head nodes in the step (3) is as follows:
wherein m is the total number of channel resources available to the cluster head nodes of the sensor device, P is the transmission power vector allocated to the calculation task by the cluster head node set of the sensor device, and J is the transmission power vector J ═ of the intelligent interference attacker (J)1,J2,...,Jn);as,iThe using state of the ith channel of the cluster head node of the s-th sensing equipment is as as,iWhen the number is 1, the ith channel of the cluster head node of the s sensing equipment is used for unloading the calculation task, otherwise, as,i=0;hs,iUnloading link channel gain, P, for the calculation task of ith channel uplink of cluster head node of the s-th sensing equipmentiIs the transmission power of the cluster head node of the sensing equipment of the ith channel, n0,iIs the noise power of the ith channel, hJ,iChannel gain at ith channel for intelligent interference attackers, JiAnd lambda is the transmission cost per unit transmission power of the cluster head node of the sensing equipment.
7. The method for optimizing game defense strategies under intelligent interference attacks in the sensor edge cloud as claimed in claim 1, wherein when the game effectiveness of the sensor device cluster head nodes is maximized through calculation so as to reach a nash equilibrium point, the transmission power distribution strategies of the sensor device cluster head nodes adopt a deep neural network to establish an energy-only defense model, and the power distribution strategies of the sensor device cluster head nodes are obtained.
8. The method for optimizing game defense strategies under intelligent interference attacks in sensing edge clouds according to claim 7, wherein the game effectiveness of the nodes of the cluster heads of the maximized sensing devices in the step (3) is recorded as:
wherein, PmaxThe maximum transmission power of the cluster head node of the sensing equipment is obtained;
the intelligent defense model established by adopting the deep neural network comprises an input layer, a normalization layer, a full connection layer, a data shaping layer, a convolution module, a pooling layer group and an output layer which are sequentially connected;
the input layer is used for inputting channel gain vectors H of m channels of the cluster head nodes of the sensing equipments,i=(hs,1,hs,2,...,hs,m) To a normalization layer;
the normalization layer is used for converting channel gain vector H of cluster head node of the sensing equipments,i=(hs,1,hs,2,...,hs,m) Normalization processing is carried out to obtain normalized channel gain vectors of cluster head nodes of the sensing equipmentAnd input to the data shaping layer through the full connection layer;
the data shaping layer is used for passing the output of the full connection layerAfter shaping, the matrix is converted into a two-dimensional matrix and input to the convolutional layer.
The convolution module comprises two convolution layers, the two convolution kernels are respectively connected through Relu linear rectification functions, and each convolution kernel comprises m convolution kernels; the channel gain of the cluster head node of the shaped normalized sensing equipment is output through an activation function after the first convolution operation is carried out through a corresponding convolution kernel, is output to a pooling layer group through the activation function again after the second convolution operation is carried out through the corresponding convolution kernel;
the pooling layer group comprises m pooling layers connected in parallel and a full-connection layer;
the output layer is used for outputting an input power vector P of a cluster head node of the sensing equipment of a Nash equilibrium point existing in the Stackelberg gameMM=(PM,1,PM,2,...,PM,m) Namely, the power distribution strategy of the cluster head node of the sensor equipment is obtained.
9. The method for optimizing game defense strategies under intelligent interference attacks in sensing edge clouds according to claim 8, wherein the intelligent defense model in the step (3) is obtained by training according to the following method:
randomly initializing a multi-channel training weight vectorTraining by using a gradient descent method, gradually transmitting and adjusting the weight in a reverse phase manner, and acquiring the transmission power of an intelligent interference attacker by a cluster head node of the sensing equipment through interaction; the loss function of the intelligent attack model is represented as follows:
wherein alpha issRepresenting weight coefficients of the loss function, balancing the influence of the constraint on the training process; (1-. alpha.) with a high degree of polymerizations)tanh(|P-Pmax|) is a regularization term, power constraints of cluster head nodes of the sensing equipment participate in training,by devitalizing the power in the loss function, we obtain:
the weight updating equation for training the intelligent defense model is as follows:
wherein, thetasIndicating the learning rate.
10. A game defense strategy optimization system attacked by intelligent interference in a sensing edge cloud is characterized by comprising an initialization module, an intelligent interference attacker prediction module, a defense strategy decision module and a configuration module;
the initialization module is configured to acquire a transmission power vector P allocated to the computation task of an initial cluster head node set of the sensor device: p ═ P1,P2,...,Pm) And submitting the data to the intelligent interference attacker prediction module, wherein m is the number of available channel resources of the cluster head node of the sensing equipment;
the intelligent interference attacker prediction module is used for calculating a power distribution vector J of the intelligent interference attacker to n channels attacked by the intelligent interference attacker according to the Steckelberg model, the device cluster head node as a leader, the intelligent interference attacker as a follower and channel gain vectors of the n sensing device cluster head nodes attacked by the intelligent interference attacker, wherein the channel gain vectors are used for maximizing the game effect of the intelligent interference attackerNNThe power distribution strategy is submitted to the defense strategy decision module as a power distribution strategy of an intelligent interference attacker;
the defense strategy decision module is used for calculating and maximizing the game effectiveness of the cluster head nodes of the sensing equipment according to the channel gain vectors of m channels of the cluster head nodes of the sensing equipment on the premise that the intelligent interference attacker adopts a power distribution strategy according to a Stark-Boolean model so as to reach a Nash equilibrium point, and the cluster head nodes of the sensing equipment distribute vectors of transmission power of the m available channels as the power distribution strategy of the cluster head nodes of the sensing equipment and submit the power distribution strategies to the configuration module;
the configuration module is used for distributing the strategy P according to the power of the cluster head node of the sensor equipmentMMAnd determining decision configuration variables and unloading tasks.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011039611.9A CN112202762B (en) | 2020-09-28 | 2020-09-28 | Game defense strategy optimization method and system for sensing edge cloud intelligent interference attack |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011039611.9A CN112202762B (en) | 2020-09-28 | 2020-09-28 | Game defense strategy optimization method and system for sensing edge cloud intelligent interference attack |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112202762A true CN112202762A (en) | 2021-01-08 |
CN112202762B CN112202762B (en) | 2022-07-08 |
Family
ID=74006838
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011039611.9A Active CN112202762B (en) | 2020-09-28 | 2020-09-28 | Game defense strategy optimization method and system for sensing edge cloud intelligent interference attack |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112202762B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112713942A (en) * | 2021-01-21 | 2021-04-27 | 电子科技大学 | MC-DBP algorithm-based method for jointly equalizing optical fiber signal damage |
CN112911600A (en) * | 2021-01-18 | 2021-06-04 | 长江大学 | Self-interference suppression method and device for in-band full-duplex cognitive wireless network |
CN113487870A (en) * | 2021-07-19 | 2021-10-08 | 浙江工业大学 | Method for generating anti-disturbance to intelligent single intersection based on CW (continuous wave) attack |
CN114501457A (en) * | 2022-01-25 | 2022-05-13 | 绍兴文理学院 | Invisible interference attack protection method and system for sensing edge cloud unloading link |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020062333A1 (en) * | 1998-06-12 | 2002-05-23 | Sanjay Anand | Method and computer program product for offloading processing tasks from software to hardware |
CN103532761A (en) * | 2013-10-18 | 2014-01-22 | 嘉兴学院 | Survivability evaluating method applicable to attacked wireless sensing network |
CN106446674A (en) * | 2016-07-27 | 2017-02-22 | 长春理工大学 | Attack prediction-based virtual machine monitoring resource allocation method in cloud computing environment |
CN109729504A (en) * | 2018-12-04 | 2019-05-07 | 深圳供电局有限公司 | A method of vehicle authentic authentication and caching based on block chain |
US20190251262A1 (en) * | 2015-03-10 | 2019-08-15 | AEMEA Inc. | Secure Non-Deterministic, Self-Modifiable Computing Machine |
CN110913357A (en) * | 2019-11-13 | 2020-03-24 | 绍兴文理学院 | Sensing cloud double-layer network defense system and method based on security situation awareness |
CN111641652A (en) * | 2020-05-29 | 2020-09-08 | 北京中超伟业信息安全技术股份有限公司 | Application security service platform based on cloud computing |
-
2020
- 2020-09-28 CN CN202011039611.9A patent/CN112202762B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020062333A1 (en) * | 1998-06-12 | 2002-05-23 | Sanjay Anand | Method and computer program product for offloading processing tasks from software to hardware |
CN103532761A (en) * | 2013-10-18 | 2014-01-22 | 嘉兴学院 | Survivability evaluating method applicable to attacked wireless sensing network |
US20190251262A1 (en) * | 2015-03-10 | 2019-08-15 | AEMEA Inc. | Secure Non-Deterministic, Self-Modifiable Computing Machine |
CN106446674A (en) * | 2016-07-27 | 2017-02-22 | 长春理工大学 | Attack prediction-based virtual machine monitoring resource allocation method in cloud computing environment |
CN109729504A (en) * | 2018-12-04 | 2019-05-07 | 深圳供电局有限公司 | A method of vehicle authentic authentication and caching based on block chain |
CN110913357A (en) * | 2019-11-13 | 2020-03-24 | 绍兴文理学院 | Sensing cloud double-layer network defense system and method based on security situation awareness |
CN111641652A (en) * | 2020-05-29 | 2020-09-08 | 北京中超伟业信息安全技术股份有限公司 | Application security service platform based on cloud computing |
Non-Patent Citations (4)
Title |
---|
DEJUN YANG ETAL: "《Coping with a Smart Jammer in Wireless Networks A Stackelberg Game Approach》", 《IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS》 * |
LIANG XIAO ETAL: "《Mobile Offloading Game Against Smart Attacks》", 《2016 IEEE CONFERENCE ON COMPUTER COMMUNICATIONS WORKSHOPS》 * |
安星硕等: "智慧边缘计算安全综述", 《电信科学》 * |
谢人超等: "移动边缘计算卸载技术综述", 《通信学报》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112911600A (en) * | 2021-01-18 | 2021-06-04 | 长江大学 | Self-interference suppression method and device for in-band full-duplex cognitive wireless network |
CN112911600B (en) * | 2021-01-18 | 2023-10-24 | 长江大学 | In-band full duplex cognitive wireless network self-interference suppression method and device |
CN112713942A (en) * | 2021-01-21 | 2021-04-27 | 电子科技大学 | MC-DBP algorithm-based method for jointly equalizing optical fiber signal damage |
CN113487870A (en) * | 2021-07-19 | 2021-10-08 | 浙江工业大学 | Method for generating anti-disturbance to intelligent single intersection based on CW (continuous wave) attack |
CN113487870B (en) * | 2021-07-19 | 2022-07-15 | 浙江工业大学 | Anti-disturbance generation method for intelligent single intersection based on CW (continuous wave) attack |
CN114501457A (en) * | 2022-01-25 | 2022-05-13 | 绍兴文理学院 | Invisible interference attack protection method and system for sensing edge cloud unloading link |
CN114501457B (en) * | 2022-01-25 | 2024-04-26 | 绍兴文理学院 | Invisible interference attack protection method and system for sensing edge cloud unloading link |
Also Published As
Publication number | Publication date |
---|---|
CN112202762B (en) | 2022-07-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112202762B (en) | Game defense strategy optimization method and system for sensing edge cloud intelligent interference attack | |
CN110347500B (en) | Task unloading method for deep learning application in edge computing environment | |
CN113449864B (en) | Feedback type impulse neural network model training method for image data classification | |
KR102034955B1 (en) | Method and apparatus for controlling transmit power in wireless communication system based on neural network | |
WO2021017227A1 (en) | Path optimization method and device for unmanned aerial vehicle, and storage medium | |
CN114818515A (en) | Multidimensional time sequence prediction method based on self-attention mechanism and graph convolution network | |
CN111401547B (en) | HTM design method based on circulation learning unit for passenger flow analysis | |
CN111224905B (en) | Multi-user detection method based on convolution residual error network in large-scale Internet of things | |
CN110472725A (en) | A kind of balance binaryzation neural network quantization method and system | |
CN111314928A (en) | Wireless ad hoc network performance prediction method based on improved BP neural network | |
CN117501245A (en) | Neural network model training method and device, and data processing method and device | |
CN115271099A (en) | Self-adaptive personalized federal learning method supporting heterogeneous model | |
CN115134778A (en) | Internet of vehicles calculation unloading method based on multi-user game and federal learning | |
Cui et al. | Multi-Agent Reinforcement Learning Based Cooperative Multitype Task Offloading Strategy for Internet of Vehicles in B5G/6G Network | |
CN117436485A (en) | Multi-exit point end-edge-cloud cooperative system and method based on trade-off time delay and precision | |
CN116663610A (en) | Scheduling network training method, task scheduling method and related equipment | |
CN114501457B (en) | Invisible interference attack protection method and system for sensing edge cloud unloading link | |
CN114879742B (en) | Unmanned aerial vehicle cluster dynamic coverage method based on multi-agent deep reinforcement learning | |
Li et al. | A UAV swarm sensing oriented distributed computing cooperation scheme | |
CN114219074A (en) | Wireless communication network resource allocation algorithm dynamically adjusted according to requirements | |
Jin et al. | Hector: A Reinforcement Learning-based Scheduler for Minimizing Casualties of a Military Drone Swarm | |
CN114077482A (en) | Intelligent calculation optimization method for industrial intelligent manufacturing edge | |
CN111787508A (en) | Radar communication integrated UAV network utility optimization method based on power control | |
CN102523055B (en) | Cooperation spectrum perception method under Nakagami-m fading channel | |
CN113615277A (en) | Power distribution method and device based on neural network |
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 |