A security protection method and terminal based on the MEC system of the active distribution network

Abstract

The present invention discloses a method and a terminal for security protection based on an active distribution network MEC system. By establishing a system model, obtaining the maximum calculation speed expression, the location privacy reward function, and the usage pattern privacy reward function to protect the privacy of calculation tasks, and then using a neural network to learn the comprehensive model constructed by the maximum calculation speed expression and the privacy protection level expression and the preset wireless channel energy gain, a corresponding training model is obtained, so that the system can obtain the optimal additional redundant information amount and the offloading decision in the training model through the currently obtained wireless channel energy gain between the server and the user during operation. At the same time, the time allocation and the offloading decision are optimized, the privacy protection ability of the system for users is improved, and the calculation delay is reduced, realizing that the system can maintain a high calculation speed while protecting the privacy of users.

Description

Technical Field

[0001] The present invention relates to the field of power grid security protection, and in particular to a security protection method and terminal based on an active distribution network MEC system. Background Art

[0002] As an important support for the smart grid, the active distribution network has become increasingly important in terms of its information data. When the active distribution network realizes the highly intelligent integration of "energy flow and information flow", it also has the characteristics of different types of complementary sources, power distribution and transmission of the power distribution network, and dynamic access of energy storage devices and users. The complex access environment, flexible access methods, highly volatile distributed energy, and a large number of access terminals make the structure of the active distribution network more complex than that of the traditional distribution network, the interaction between the energy flow and the information flow is more frequent, and the mutual dependence is stronger. The interaction makes the integration between the information network and the power network closer, but it also makes the security of the information network directly affect the security of the power network.

[0003] To meet the growing demand for electricity, large interconnected power grids are becoming larger in scale and more complex in structure. It is necessary to strengthen the dynamic monitoring of the system. The Wide Area Measurement System (WAMS) has emerged as a result. It realizes the collection of data from each node in the whole network based on the phasor measurement unit (PMU). After the PMU marks the collected data with accurate time stamps by means of the high-precision time synchronization signal provided by GPS, the data is transmitted to the monitoring and control center through a high-speed communication network, thus realizing the monitoring of the dynamic behavior of the whole network and further taking stability control measures. The upload and download of various service data in the power system are inseparable from the power communication system. The power communication system has become an important part of the modern power system and is the basic support for realizing the stability control function of the power system. All signals of the PMU system are directly collected through hard wiring, and the data is not obtained from other network systems by means of network communication. The data acquisition screen converges the collected data to the centralized processing screen through optical fiber communication and then accesses the real-time service switch of the dispatching data network through optical fiber communication. This system is not connected to any external network. The key point of security protection is to strengthen the security management of the data centralized processing workstation. It is necessary to strengthen the management of the access of mobile storage media and laptop computers to the workstation to avoid introducing malicious codes such as viruses into the power dispatching data network. The security protection work of the power dispatching data network should adhere to the principles of security zoning, network specialization, horizontal isolation, and vertical authentication. The key point of security protection is to ensure the security of the power dispatching data network. The overall goals include: 1) preventing the interruption of power dispatching data network services and affecting power production; 2) preventing the equipment of the power dispatching data network from being attacked by viruses and hackers; and 3) preventing malicious damage and attacks on the entire power dispatching data network from the service end.

[0004] As the core architecture of the fifth-generation communication technology, Mobile Edge Computing (MEC) provides convenience for the transmission and processing of emerging services and the exponentially growing massive data thereon. The mobile edge computing system can be divided into two major parts in terms of architecture: the edge network and the edge terminal device. By sinking the computing tasks to the network edge or terminal device, it can make full use of the "local" computing resources to process nearby, reducing the large performance overhead caused by the long-distance transmission of massive data. To ensure the correct, stable, and efficient operation of the system, the mobile edge computing system also needs to effectively defend against the security threats of each part of the system. This includes: 1) the accurate identification and elimination of known threats; 2) the improvement of the resilience or passive defense ability of the security of software and hardware systems with potential defects and vulnerabilities; and 3) the fine-grained protection of different types of data on the mobile intelligent terminal platform where the tasks are sunk to avoid the leakage of important sensitive information.

[0005] With the rapid development of smart distribution networks, modern communication technologies and computer network technologies are widely applied in smart distribution networks. Information interaction in real time can be carried out between the power grid and users through a two-way communication digital network. At this time, terminals directly exposed to users (such as smart meters) are most likely to become the targets of attacks. As long as one terminal is compromised, attacks can be launched on other devices in the network, directly affecting the stable operation of the distribution network. With the high-speed development of smart distribution networks, higher requirements are put forward for data security and communication technologies.

[0006] Due to the large number of devices in the distribution network, the number of sensors monitoring their states is large. These sensors collect users' information at any time, generating various data, which contain various privacy of users. If these privacy information is leaked, it is very likely to cause losses to users' spirits or property. However, only targeting the privacy protection of users will lead to problems such as a reduction in the system calculation speed and system data transmission delay. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a security protection method and terminal based on an active distribution network MEC system, which improve the privacy protection of users while reducing system calculation delay.

[0008] To solve the above technical problem, the technical solution adopted by the present invention is as follows:

[0009] A security protection method based on an active distribution network MEC system, including the steps of:

[0010] Establishing a system model, where the system model includes a variety of preset offloading decisions;

[0011] Establishing a calculation model according to the system model to obtain a maximum calculation speed expression;

[0012] Setting a location privacy reward function and a usage pattern privacy reward function according to the calculation model to obtain a privacy protection level expression;

[0013] Obtaining a comprehensive model according to the maximum calculation speed expression and the privacy protection level expression;

[0014] Training a neural network according to the comprehensive model and a preset wireless channel energy gain to obtain a training model;

[0015] Obtaining the wireless channel energy gain between the current server and the user, and outputting the optimal additional redundant information amount and the offloading decision according to the training model.

[0016] To solve the above technical problems, another technical solution adopted by the present invention is as follows:

[0017] A security protection terminal based on an active distribution network MEC system, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, each step in a security protection method based on an active distribution network MEC system as described above is implemented.

[0018] The beneficial effects of the present invention are as follows: By establishing a system model including various offloading decisions, and establishing a corresponding calculation model based on the system model to obtain the maximum calculation speed expression. At the same time, a location privacy reward function and a usage pattern privacy reward function are set based on the system model to protect the privacy of calculation tasks. Then, through a neural network, a comprehensive model constructed by the maximum calculation speed expression and the privacy protection level expression, and a preset wireless channel energy gain are learned to obtain a corresponding training model. During the operation of the system, the system can obtain the optimal additional redundant information amount and the offloading decision in the training model through the currently obtained wireless channel energy gain between the server and the user. And when the wireless channel gain between the device and the MEC server is large, the device is more inclined to offload tasks to the MEC server. At the same time, the time allocation and offloading decision are optimized, the privacy protection ability of the system for users is improved, and the calculation delay is reduced, realizing that the system can maintain a high calculation speed while protecting the privacy of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the steps of a security protection method based on an active distribution network MEC system according to an embodiment of the present invention;

[0020] Figure 2 It is a schematic structural diagram of a security protection terminal based on an active distribution network MEC system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To describe in detail the technical content, the achieved objectives, and the effects of the present invention, the following is described in conjunction with the embodiments and with reference to the drawings.

[0022] Please refer to Figure 1 , a security protection method based on an active distribution network MEC system, comprising the steps of:

[0023] Establish a system model, where the system model includes a preset variety of offloading decisions;

[0024] Establish a calculation model according to the system model to obtain the maximum calculation speed expression;

[0025] Set a location privacy reward function and a usage pattern privacy reward function according to the calculation model to obtain a privacy protection level expression;

[0026] Obtain a comprehensive model according to the maximum calculation speed expression and the privacy protection level expression;

[0027] Train a neural network according to the comprehensive model and a preset wireless channel energy gain to obtain a trained model;

[0028] Obtain the wireless channel energy gain between the current server and the user, and output the optimal additional redundant information amount and the offloading decision according to the trained model.

[0029] As can be seen from the above description, the beneficial effects of the present invention are as follows: By establishing a system model including various offloading decisions, and establishing a corresponding calculation model according to the system model to obtain a maximum calculation speed expression. At the same time, on the basis of the system model, a location privacy reward function and a usage pattern privacy reward function are set to protect the privacy of the calculation task. Then, the neural network is used to learn the comprehensive model constructed by the maximum calculation speed expression and the privacy protection level expression and the preset wireless channel energy gain to obtain a corresponding trained model. During the operation of the system, the system can obtain the optimal additional redundant information amount and the offloading decision in the trained model through the currently obtained wireless channel energy gain between the server and the user. And when the wireless channel gain between the device and the MEC server is large, the device is more inclined to offload the task to the MEC server. At the same time, the time allocation and offloading decision are optimized, the privacy protection ability of the system for users is improved, and the calculation delay is reduced, realizing that the system can maintain a high calculation speed while protecting the privacy of users.

[0030] Further, the obtaining the maximum calculation speed expression by establishing a calculation model according to the system model includes:

[0031] Obtain terminal device information and establish a local calculation model;

[0032] Calculate the local maximum calculation speed expression according to the local calculation model;

[0033] Obtain the interaction information between the device and the server to obtain a mobile edge computing model;

[0034] Calculate the local maximum task offloading speed expression according to the mobile edge computing model;

[0035] Obtain the total maximum calculation speed expression of the terminal device according to the local maximum calculation speed expression and the maximum task offloading speed expression.

[0036] As described above, corresponding local computing models and mobile edge computing models are established based on the information of the terminal device obtained and the interaction information between the device and the server, and the local maximum computing speed expression and the maximum task offloading speed expression are calculated respectively, so that the corresponding maximum computing speed expression can be obtained under different offloading decisions.

[0037] Further, the obtaining of the terminal device information and the establishment of the local computing model include:

[0038] Obtain the charging power, energy consumption, and data processing speed of the terminal device in each preset time slot, and establish a local computing model.

[0039] As described above, by obtaining the charging power, energy consumption, and data processing speed of the device in each preset time slot, the local computing speed can be accurately maximized, and the accuracy of the local computing model can be improved.

[0040] Further, the obtaining of the interaction information between the terminal device and the server and the obtaining of the mobile edge computing model include:

[0041] Obtain the communication bandwidth, transmission power of the terminal device, noise power, and data transmission ratio between the terminal device and the server in each preset time slot, and obtain a mobile edge computing model.

[0042] As described above, by obtaining the communication bandwidth, transmission power of the terminal device, noise power, and data transmission ratio between the device and the server in each preset time slot, the server can provide the maximum computing speed expression for the device, and the accuracy of the mobile edge computing model can be improved.

[0043] Further, the setting of the location privacy reward function and the usage pattern privacy reward function according to the computing model to obtain the privacy protection level expression includes:

[0044] According to the local maximum computing speed expression and the maximum task offloading speed expression, obtain the average computing speed expression;

[0045] According to the average computing speed expression and the maximum task offloading speed expression, obtain the location privacy reward function;

[0046] Obtain the data transmission ratio without privacy protection and the data transmission ratio under privacy protection, and obtain the data transmission ratio difference;

[0047] According to the data transmission ratio difference and the maximum task offloading speed expression, obtain the usage pattern privacy reward function;

[0048] According to the location privacy reward function and the usage pattern privacy reward function, a privacy protection level expression is obtained.

[0049] As can be seen from the above description, by calculating the average computing speed expression under location privacy protection and combining it with the maximum task offloading speed expression, the location privacy reward function is obtained. And by calculating the difference in data transmission ratio under usage pattern privacy protection and combining it with the maximum task offloading speed expression, the usage pattern privacy reward function is obtained. This enables the combination of the two privacy protection models with the maximum computing speed expression, so that during the data transmission process, the system needs to comprehensively consider privacy protection and the maximum computing speed to obtain the best offloading decision, that is, while protecting user privacy, the system can maintain a relatively high computing speed.

[0050] Further, the obtaining of the privacy protection level expression according to the location privacy reward function and the usage pattern privacy reward function includes:

[0051] Set the location privacy weight and the usage pattern privacy weight;

[0052] Adjust the privacy protection level expression according to the location privacy weight and the usage pattern privacy weight.

[0053] As can be seen from the above description, by setting the location privacy weight and the usage pattern privacy weight to adjust the privacy protection level, when it is necessary to improve the device computing speed, the privacy protection level can be correspondingly reduced, and when it is necessary to improve the privacy protection level, the computing speed is correspondingly reduced, so as to form a balance between the computing speed of task offloading and privacy protection in the system.

[0054] Further, the comprehensive model obtained according to the maximum computing speed expression and the privacy protection level expression is:

[0055] Based on the maximum computing speed expression and the privacy protection level expression, a comprehensive model is obtained:

[0056]

[0057]

[0058] Where h = {h m ∣m ∈ M}, x = {x m ∣m ∈ M}, v = {v m ∣m ∈ M} and b = {b m ∣m ∈ M}, x m represents an indicator variable, β 1 represents the location privacy weight, β 2Let \( \omega \) denote the privacy weight of the usage pattern, \( M \) denote the set of terminal devices, i.e., \( M=\{0,1,\cdots,m\} \), \( aT(0\lt a\leqslant1) \) denote the charging duration within a time slot, \( T \) denote the length of the time slot, \( h \) m denote the channel energy gain between the current terminal device \( m \) and the MEC server, \( k \) m is the energy efficiency factor, where \( B \) denotes the communication bandwidth between device \( m \) and the MEC server, \( N \) 0 denotes the noise power, \( v \) m denote the data transmission ratio, \( E \) is the expression of the average computing speed, denote the number of CPU cycles required to process 1 bit of data, \( b \) m \( T \) denotes the time required for offloading tasks; \( \mu \) denotes the charging efficiency factor, \( P \) denotes the energy transmission power of wireless charging.

[0059] As can be seen from the above description, a comprehensive model is obtained from the maximum computing speed expression and the privacy protection level expression. After subsequent learning of the model, the system can obtain the optimal additional redundant information and the offloading decision in the training model through the currently obtained wireless channel energy gain between the server and the user.

[0060] Furthermore, the training of the neural network according to the comprehensive model and the preset wireless channel energy gain to obtain the training model includes:

[0061] Obtain the preset additional redundant information, the offloading decision, and the wireless channel energy gain, and train the neural network according to the comprehensive model to obtain the first training set;

[0062] Perform secondary learning on the neural network according to the first training set and the wireless channel energy gain to obtain the training model.

[0063] As can be seen from the above description, first training the neural network with the preset additional redundant information, the offloading decision, and the wireless channel energy gain can transform the non-convex optimization problem of mixed integer programming in the original comprehensive model into a convex problem, that is, solving two unknowns, the maximum computing speed and the privacy protection level, through the above three known quantities, and obtaining the corresponding first training set; then performing secondary learning by combining the obtained first training set with the wireless channel energy gain, which not only simplifies the problem of solving the maximum value of the comprehensive model, but also improves the accuracy of the solution.

[0064] Furthermore, the obtaining of the preset additional redundant information, the offloading decision, and the wireless channel energy gain, and training the neural network according to the comprehensive model to obtain the first training set includes:

[0065] Perform a Lagrangian transformation on the comprehensive model to obtain a Lagrangian dual model;

[0066] Solve the Lagrangian dual model according to the preset additional redundant information amount, the offloading decision, and the wireless channel energy gain to obtain the first training set.

[0067] As can be seen from the above description, by introducing Lagrange multipliers to transform the comprehensive model, the comprehensive model is easier to solve, and a more accurate first training set is obtained.

[0068] A security protection terminal based on an active distribution network MEC system includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements each step in the above-mentioned security protection method based on an active distribution network MEC system.

[0069] The above-mentioned security protection method and terminal based on the active distribution network MEC system of the present invention can be applied to an active distribution network, protect the privacy of data transmission between servers and devices in the distribution network, and improve the computing speed of the system at the same time. The following is described through specific embodiments:

[0070] Please refer to Figure 1 , a security protection method based on an active distribution network MEC system, includes the steps:

[0071] S1. Establish a system model, where the system model includes a variety of preset offloading decisions;

[0072] In an optional implementation manner, the MEC system model includes an MEC server and m terminal devices; the MEC server is powered by a stable power supply, and the terminal devices are wirelessly charged by the MEC server through radio frequency technology (RF); and the MEC server can charge multiple terminal devices at the same time, and the terminal devices can store this power for computing tasks and transmission tasks; at the same time, it is set that the terminal devices only charge once at the beginning of each time slot, so that the terminal devices run more stably; if the length of each time slot is preset as T, the set of time slots is defined as T = {0, 1,...}, and the set of terminal devices is defined as M = {0, 1,...m}; in any time slot, each terminal device will generate a computing task, and these tasks can be locally computed or offloaded to the MEC server; it is set that the tasks generated by the terminal devices in the system model are not divisible, that is, the tasks generated by the terminal devices in one time slot can only be processed locally or offloaded to the MEC server for processing to improve the security of data in one time slot; define M 0Let \(M\) be the set of terminal devices selected to execute tasks locally. 1 Let \(M_0\) be the set of devices selected to offload tasks to the MEC server. It can be seen that \(M_0\) and \(M_1\) are two independent sets, that is, \(M_0\cup M_1 = M\). It can be known that this system model only provides a specific implementation scenario for the solution of this application, and does not further limit this application. 0 ∪M 1 =M; It can be seen that this system model only provides a specific implementation scenario for the solution of this application, and does not further limit this application.

[0073] Five preset offloading decision flags are used to represent different processing decisions of tasks by the terminal device in each time slot, which are: and

[0074] Offloading decision means that the terminal device executes the tasks generated in this time slot locally.

[0075] Offloading decision means that the terminal device offloads the tasks generated in this time slot to the MEC server, but does not protect the user's privacy.

[0076] Offloading decision means that the terminal device offloads the tasks generated in this time slot to the MEC server, and only protects the user's location privacy.

[0077] Offloading decision means that the terminal device offloads the tasks generated in this time slot to the MEC server, and only protects the user's usage pattern privacy.

[0078] Offloading decision means that the terminal device offloads the tasks generated in this time slot to the MEC server, and simultaneously protects the user's location privacy and usage pattern privacy. That is, the offloading decision set of the terminal device in a time slot can be expressed as

[0079] And the following relationship can be obtained: and In addition, these decision metrics should also meet the following operational constraints:

[0080] ∑ m∈M α m =M; (1)

[0081] At the beginning of each time slot, the terminal device prepares for charging for local computing or transmission tasks; the energy obtained by the device in a time slot can be expressed as:

[0082] E m =μPh m aT; (2)

[0083] where, μ represents the charging efficiency factor, p represents the energy transmission power of wireless charging, and h m represents the channel energy gain between the current terminal device m and the MEC server, and aT (0 < a ≤ 1) represents the charging duration within a time slot; it is assumed that h m remains unchanged during the process of device task offloading and result downloading, that is, h m remains unchanged in magnitude within a time slot, that is, h m is a fixed parameter within a time slot;

[0084] Based on the established system model above, when the wireless channel gain between the terminal device and the MEC server is large, the terminal device is more inclined to offload tasks to the MEC server; therefore, the location privacy information of the terminal device is highly correlated with the channel gain; when an untrusted MEC server obtains the offloaded data information (i.e., tasks) of a certain terminal device, it can thereby obtain the distance between the terminal device and it; moreover, by comparing the offloaded data of the terminal device in different time slots, the untrusted MEC server can crack the moving trajectory of the terminal device. When multiple MEC servers collude with each other, the accurate location of the terminal device will be exposed; therefore, to protect the location privacy of the terminal device, the terminal device can deliberately offload tasks to the server when the channel quality is poor to disguise its decision;

[0085] Meanwhile, in any time slot, the magnitude of the channel energy gain between the terminal device and the server is relatively stable; and the task data under the same working mode has the same data format, which leads to different usage patterns of the terminal device generating data with different sizes and formats; therefore, by analyzing the size of the offloaded data volume of the terminal device in this time slot, the untrusted MEC server can crack the usage pattern information of its operation; therefore, to protect the usage pattern privacy of the terminal device, artificially changing the size of the data volume by having the terminal device transmit redundant privacy protection information is an effective method;

[0086] S2. Establish a calculation model according to the system model to obtain the maximum calculation speed expression;

[0087] Specifically, S21. Obtain terminal device information and establish a local calculation model;

[0088] S22. Calculate the local maximum calculation speed expression according to the local calculation model;

[0089] S23. Obtain the interaction information between the device and the server to obtain the mobile edge calculation model;

[0090] S24. Calculate the expression of the maximum task offloading speed of the local area according to the mobile edge computing model;

[0091] S25. Obtain the expression of the total maximum computing speed of the terminal device according to the expression of the maximum computing speed of the local area and the expression of the maximum task offloading speed;

[0092] S3. Set the location privacy reward function and the usage pattern privacy reward function according to the computing model to obtain the expression of the privacy protection level;

[0093] S4. Obtain the comprehensive model according to the expression of the maximum computing speed and the expression of the privacy protection level;

[0094] S5. Train the neural network according to the comprehensive model and the preset wireless channel energy gain to obtain the trained model;

[0095] S6. Obtain the wireless channel energy gain between the current server and the user, and output the optimal additional redundant information amount and the offloading decision according to the trained model.

[0096] Embodiment 2

[0097] The difference between this embodiment and the embodiment lies in defining how to establish the local computing model and the mobile edge computing model;

[0098] Step S21 includes:

[0099] S211. Obtain the charging power, consumed energy, and data processing speed of the terminal device in each preset time slot, and establish a local computing model; specifically, in any time slot, when the task is locally executed, the energy consumed by the device is where k m is the energy efficiency factor, f m is the operating frequency of the local CPU, τ m represents the duration of local computing within a time slot; the charging power is E m ;

[0100] Specifically in step S22, according to the law of conservation of energy, the energy consumed by the device must be less than or equal to its charging power; that is, the data size of local operation can be obtained as f m τ m / φ, where represents the CPU cycles required to process 1 bit of data; so the computing speed achieved by local computing in one time slot is:

[0101]

[0102] To obtain the maximum local computing data volume and the maximum local computing speed, it is assumed that the wireless device depletes the energy it has obtained during this time slot, that is and it has been performing operations throughout the time slot, that is τ m = T, and the calculation formula for the operation frequency of the local CPU can be obtained:

[0103]

[0104] and the expression for the maximum local computing speed:

[0105]

[0106] where

[0107] Step S23 includes:

[0108] S231. Obtain the communication bandwidth between the terminal device and the server, the transmission power of the terminal device, the noise power, and the data transmission ratio in each preset time slot to obtain a mobile edge computing model. Specifically:

[0109] Assume that all terminal devices use the same frequency band to communicate with the MEC server. Then, due to the limitation of time division multiplexing, the server can only receive data from one terminal device at a time. Therefore, after charging is completed, these terminal devices will unload tasks to the MEC server in sequence. Specifically, in any time slot, the time required for terminal device m to unload tasks is defined as b m T(b m ∈[0,1)), that is, the size of the data volume unloaded by terminal device m is:

[0110]

[0111] where B represents the communication bandwidth between device m and the MEC server, represents the transmission power of terminal device m, N 0 represents the noise power, V m represents the data transmission ratio (the ratio of the data transmitted by the terminal device to the original data; since there will be some protocol header information or privacy protection information in the final transmitted data compared to the initial data, the data volume of the privacy protection information will also increase accordingly when the data is encrypted), and V m ≥1; assume that the computing resources and data transmission capabilities of the MEC server are much greater than those of the terminal device. Therefore, the time required for data to be operated on the server, the time and energy consumption for downloading the operation results can be ignored;

[0112] Specifically in step S24, based on the above model, a and b can be obtainedm The relationship between:

[0113]

[0114] Therefore, it can be obtained that the maximum computing speed that the MEC server can provide to a device is equal to the task offloading speed of the device; so in order to obtain the maximum task offloading speed, it is assumed that the device uses all the power obtained in this time slot for task offloading, that is;

[0115]

[0116] In this way, the expression of the maximum task offloading speed that device m can obtain can be obtained:

[0117]

[0118] Where

[0119] Specifically in step S25, combining the above calculation expressions, the expression of the total maximum computing speed of M devices can be obtained:

[0120]

[0121] At the same time, in order to improve the readability of the formula, an indicator variable x is introduced m :

[0122]

[0123] That is, formula (8) is converted into the following form:

[0124]

[0125] Embodiment III

[0126] The difference between this embodiment and Embodiment 1 or 2 is that it specifically defines how to obtain the privacy protection level expression;

[0127] Step S3 includes:

[0128] S31. According to the local maximum computing speed expression and the maximum task offloading speed expression, obtain the average computing speed expression; the average computing speed expression represents: when the terminal device changes the initial local computing decision x' m = 0 to the offloading computing decision x m = 1, the improvement in the computing speed obtained during the process divided by the mean value of its offloading computing speed, denoted as E;

[0129] S32. Obtain the location privacy reward function according to the average calculation speed expression and the maximum task offloading speed expression;

[0130] The obtained location privacy reward function is:

[0131]

[0132] S33. Obtain the data transmission ratio without privacy protection and the data transmission ratio with privacy protection, and obtain the data transmission ratio difference, that is:

[0133]

[0134] where V′ m represents the ratio between the data transmitted without using the mode privacy protection strategy and the original data, and I(x m =1) and I(x‘ m =1) represent the decision states;

[0135] S34. Obtain the usage mode privacy reward function according to the data transmission ratio difference and the maximum task offloading speed expression:

[0136]

[0137] S35. Obtain the privacy protection level expression according to the location privacy reward function and the usage mode privacy reward function:

[0138] P m = P l,m + P u,m ;

[0139] where the location privacy weight β 1 and the usage mode privacy weight β 2 are also set;

[0140] Adjust the privacy protection level expression according to the location privacy weight and the usage mode privacy weight:

[0141] P m = β 1 P l，m + β 2 P u，m ; (13)

[0142] Meanwhile, in an optional implementation manner, the multiple offloading decisions correspond to the privacy protection level; that is and respectively correspond to the values of β 1 and β 2 , such as 00, 01, 10, and 11;

[0143] Based on the maximum computing speed expression and the privacy protection level expression, a comprehensive model is obtained:

[0144]

[0145] s.t. (6), v m ≥ 1, b m ≥ 0, a ≥ 0, x m ∈ {0, 1}

[0146] where h = {h m |m ∈ M}, x = {x m |m ∈ M}, v = {v m |m ∈ M} and b = {b m |m ∈ M}, and the remaining parameters are fixed values (such as: P, B, etc.).

[0147] Example 4

[0148] The difference between this implementation example and the first, second, or third example is that it specifically defines how to obtain the training model;

[0149] It can be seen from the above comprehensive model formula that the comprehensive model is a problem of mutual trade - off between the privacy protection level and the computing speed. Therefore, in the case of known channel gain h m , the problem to be solved is how to maximize the weighted sum of the local computing speed, the offloading computing speed, and the privacy protection level, as shown in the following formula:

[0150]

[0151] Since P1 is a non - convex optimization problem of mixed - integer programming, it is difficult to solve using conventional methods; nevertheless, as long as the values of x and v (i.e., x m and v m ) can be obtained in advance, P1 can be transformed into a convex optimization problem:

[0152]

[0153] Such a change enables problem P2 to be solved by establishing a task offloading algorithm for protecting user privacy based on deep reinforcement learning;

[0154] Step S5 includes:

[0155] S51. Obtain the preset additional redundant information amount, the offloading decision, and the wireless channel energy gain, and train the neural network according to the comprehensive model to obtain the first training set;

[0156] That is, solve P2 in the context of known h m and multiple sets of {x m , v m} pairs, and save the corresponding optimal (the one that maximizes Q(h, b, a)) h m and {x m , v m} pairs to obtain the first training set and perform further learning;

[0157] Specifically, it includes the steps:

[0158] S511. Perform a Lagrangian transformation on the comprehensive model to obtain a Lagrangian dual model;

[0159] Introduce a Lagrange multiplier ξ to restrict formula (6) to obtain the Lagrangian form of P2:

[0160]

[0161] And its Lagrangian function dual form can be obtained as:

[0162] min ξ {max (a,b) L(a, b, ξ) | ξ ≥ 0}; (18)

[0163] S512. Solve the Lagrangian dual model according to the preset additional redundant information amount, the offloading decision, and the wireless channel energy gain to obtain the first training set;

[0164] When {xm, vm} is known, P2 is a convex optimization problem, so its corresponding dual problem (18) is solvable. And the final result should satisfy the following restrictions:

[0165]

[0166] The relationship between the parameter b m , a, and ξ is:

[0167]

[0168] where W(x) represents the Lambert-W function; making a simple transformation of formula (20) can be used to solve for the value of b m :

[0169]

[0170] where

[0171] Moreover, due to the limitation of (19), by combining formula (19) and (21), a closed-form α value can be obtained:

[0172]

[0173] Since both Φ m (ξ) and Γ(ξ) are functions of ξ, the key to solving P2 is to obtain the value of ξ, that is, to solve the following equation:

[0174]

[0175] where the solution method of v m is as follows: When the user needs to protect the usage pattern privacy of the device, the terminal device m can generate two different privacy protection messages, which correspond to two different usage pattern privacy protection levels; therefore, the Logistic function can be used, that is as the activation function of the output layer, so that the final output can be divided into two categories, that is, as a classifier in machine learning, the samples are divided into two categories by comparing with the threshold, representing the lower-level usage pattern privacy protection ability and the higher-level usage pattern privacy protection ability respectively; and the usage pattern privacy protection decision based on the output of the algorithm at a certain time slot t is v m t , and K1 (K1 ≤ M + 1) optimal solution candidates v k1 can be obtained; first, according to the usage pattern privacy protection decision v m t directly output by the neural network, v 1 can be obtained:

[0176]

[0177] After that, based on v 1 , the remaining (K1 - 1) optimal solution candidates can be obtained, and the specific method is as follows:

[0178]

[0179] Each element of each group of v is composed of 1.1 or 1.3, where v1, v2, v3, and v4 belong to the candidate solutions, and the arrangement of 1.1 or 1.3 is determined by formula (25); for example, when K1 = 4 and vt = [0.1, 0.3, 0.6, 0.9], four candidate solutions v1 = [1.1, 1.1, 1.3, 1.3], v2 = [1.3, 1.3, 1.3, 1.3], v3 = [1.1, 1.3, 1.3, 1.3], and v4 = [1.1, 1.1, 1.1, 1.3] are obtained;

[0180] Regarding x m Solution method: Learning the distribution of x m and the network for learning the distribution of v m have the same output layer activation function, so the optimal x m The solution method for the distribution is similar to that of v m ; that is, the obtained x k2 :

[0181]

[0182] S52. Perform secondary learning on the neural network according to the first training set and the wireless channel energy gain to obtain a training model;

[0183] Use the first training set for secondary algorithm training to enable it to learn faster the optimal {x m m , v m} that maximizes Q(h, b, a);

[0184] Neural network learning model: In this system, there are two networks respectively used to learn the distributions of x m and v m under the given h m Once the amount of data stored in the training set is greater than the amount of data required for batch learning, these two networks will immediately start training; performing batch training on the neural network can greatly improve the learning efficiency. When the training set is full, the newly generated training data will replace the old training data; therefore, the neural network only learns from the latest and better training data; because only one training data is added or updated in one time slot, there is almost no difference between the training sets in adjacent time slots; therefore, it is more efficient to train again after several time slots; The pseudocode is shown in Table 1;

[0185]

[0186]

[0187] Example Five

[0188] Please refer to Figure 2 , a security protection terminal based on the active distribution network MEC system, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements each step in a security protection method based on the active distribution network MEC system as described in any one of Examples 1 to 4.

[0189] ​In summary, a security protection method and terminal based on the MEC system of the active distribution network provided by the present invention establish a system model including various offloading decisions, establish a corresponding calculation model based on the system model to obtain the maximum calculation speed expression, and at the same time set a location privacy reward function and a usage pattern privacy reward function on the basis of the system model to protect the privacy of the calculation task. Then, the neural network is used to learn the comprehensive model constructed by the maximum calculation speed expression and the privacy protection level expression and the preset wireless channel energy gain to obtain the corresponding training model. During the operation of the system, the system can obtain the optimal additional redundant information amount and the offloading decision in the training model through the currently obtained wireless channel energy gain between the server and the user. Therefore, when the user is far from the edge server and the wireless channel quality deteriorates, the offloading decision is disguised, and the real data with mixed privacy protection information is sent to the server to achieve privacy protection. At the same time, the time allocation and offloading decision are optimized, the privacy protection ability of the system for users is improved, and the calculation delay is reduced, realizing that the system can maintain a high calculation speed while protecting the privacy of users.

[0190] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in the related technical field, shall be included in the patent protection scope of the present invention by the same token.

Claims

1. A security protection method based on the MEC system of an active distribution network, characterized in that, it includes the steps of: establishing a system model, where the system model includes a variety of preset offloading decisions; establishing a calculation model according to the system model to obtain a maximum calculation speed expression; setting a location privacy reward function and a usage pattern privacy reward function according to the calculation model to obtain a privacy protection level expression; obtaining a comprehensive model according to the maximum calculation speed expression and the privacy protection level expression; training a neural network according to the comprehensive model and a preset wireless channel energy gain to obtain a trained model; acquiring the wireless channel energy gain between the current server and the user, and outputting the optimal additional redundant information amount and the offloading decision according to the trained model; The setting the location privacy reward function and the usage pattern privacy reward function according to the calculation model to obtain the privacy protection level expression includes: obtaining an average calculation speed expression according to the local maximum calculation speed expression and the maximum task offloading speed expression; obtaining the location privacy reward function according to the average calculation speed expression and the maximum task offloading speed expression; acquiring the data transmission ratio without privacy protection and the data transmission ratio under privacy protection to obtain a data transmission ratio difference; obtaining the usage pattern privacy reward function according to the data transmission ratio difference and the maximum task offloading speed expression; obtaining the privacy protection level expression according to the location privacy reward function and the usage pattern privacy reward function; The obtaining the privacy protection level expression according to the location privacy reward function and the usage pattern privacy reward function includes: setting a location privacy weight and a usage pattern privacy weight; adjusting the privacy protection level expression according to the location privacy weight and the usage pattern privacy weight; The obtaining the comprehensive model according to the maximum calculation speed expression and the privacy protection level expression is: obtaining the comprehensive model based on the maximum calculation speed expression and the privacy protection level expression: ； ， ， ， ； Among them , , and , x m represents an indication variable, the location privacy weight, the usage pattern privacy weight, M represents the set of terminal devices, i.e., , aT (0 < a ≤ 1) represents the charging duration within a time slot, T is the length of the time slot, h m represents the channel energy gain between the current terminal device m and the MEC server, is the energy efficiency factor, where B represents the communication bandwidth between device m and the MEC server, represents the noise power, represents the data transmission ratio, E is the average computing speed expression, represents the CPU cycles required to process 1 bit of data, b m T represents the time required for offloading tasks; μ represents the charging efficiency factor, P represents the energy transmission power of wireless charging; represents the ratio between the data transmitted without the usage pattern privacy protection strategy and the original data.

2. The security protection method based on the MEC system of an active distribution network according to claim 1, characterized in that, The establishing the calculation model according to the system model to obtain the maximum calculation speed expression includes: acquiring terminal device information and establishing a local calculation model; calculating the local maximum calculation speed expression according to the local calculation model; acquiring the interaction information between the device and the server to obtain a mobile edge computing model; calculating the local maximum task offloading speed expression according to the mobile edge computing model; obtaining the total maximum calculation speed expression of the terminal device according to the local maximum calculation speed expression and the maximum task offloading speed expression.

3. The security protection method based on the MEC system of an active distribution network according to claim 2, characterized in that, The acquiring the terminal device information and establishing the local calculation model includes: acquiring the charging power, consumed energy, and data processing speed of the terminal device in each preset time slot and establishing a local calculation model.

4. A security protection method based on an active distribution network MEC system according to claim 2, characterized in that, the obtaining of the interaction information between the terminal device and the server to obtain the mobile edge computing model includes: Obtaining the communication bandwidth between the terminal device and the server, the transmission power of the terminal device, the noise power, and the data transmission ratio in each preset time slot to obtain the mobile edge computing model.

5. A security protection method based on an active distribution network MEC system according to claim 1, characterized in that, the training of the neural network according to the comprehensive model and the preset wireless channel energy gain to obtain the training model includes: Obtaining the preset additional redundant information amount, the offloading decision, and the wireless channel energy gain and training the neural network according to the comprehensive model to obtain the first training set; Performing secondary learning on the neural network according to the first training set and the wireless channel energy gain to obtain the training model.

6. A security protection method based on an active distribution network MEC system according to claim 5, characterized in that, the obtaining of the preset additional redundant information amount, the offloading decision, and the wireless channel energy gain and training the neural network according to the comprehensive model to obtain the first training set includes: Performing a Lagrangian transformation on the comprehensive model to obtain a Lagrangian dual model; Solving the Lagrangian dual model according to the preset additional redundant information amount, the offloading decision, and the wireless channel energy gain to obtain the first training set.

7. A security protection terminal based on an active distribution network MEC system, including a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements each step in a security protection method based on an active distribution network MEC system according to any one of claims 1-6.

Images