A computing power resource security scheduling method in a computing power network

Abstract

The application relates to a computing power resource security scheduling method in a computing power network, which comprises the following steps: 1) forming a computing power resource pool; 2) a user requesting a computing power resource; 3) computing power resource security scheduling; and 4) obtaining a computing task result. The application considers the case that multiple data center computing power resource nodes exist in the whole network, integrates the computing power resources of multiple data centers, and simultaneously realizes unified security optimization scheduling of the computing power resources. Under the constraint of meeting the computing time delay requirement, the computing time delay and energy consumption cost are jointly optimized, and the data privacy and integrity are guaranteed in the process of computing power scheduling.

Description

Technical Field

[0001] This invention belongs to the field of computing resource security scheduling technology, specifically relating to a computing resource security scheduling method in a computing network. Background Technology

[0002] With the launch and development of the East-West Computing Project, my country has begun large-scale construction of computing power hub nodes and national data center clusters. This has two significant implications: First, it expands the overall computing power of the country and coordinates regional development through a nationwide integrated planning and layout. Second, deploying data centers in the western region, where green energy is abundant, is conducive to promoting green, low-carbon, and circular development.

[0003] Computing power networks have provided a powerful boost to the application and development of East-West computing. A computing power network is a new type of network that integrates and links computing resources, including those from various data centers, to provide users with ubiquitous computing power and meet their computing needs. However, in its development and application, it still faces significant challenges, such as how to jointly schedule the computing resources of numerous data centers to improve the utilization rate of computing resources; and how to consider energy consumption costs and ensure the latency requirements and data security of computing tasks during the scheduling process.

[0004] Currently, most existing methods only consider models where the entire network has only a single node with supercomputing power resources, similar to a data center. However, this model suffers from single points of failure and difficulties in cross-domain use, making it unusable. Furthermore, it doesn't fully realize the goal of computing power networks to integrate computing resources from various domains and data centers. In addition, existing methods lack security considerations during computing power scheduling. However, data used for computing power scheduling is vulnerable to malicious tampering and privacy breaches during transmission through the computing power network. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a secure scheduling method for computing resources in a computing network. It introduces multi-access edge computing (MEC), which can be regarded as a data center running at the network edge such as base stations or gateways, located between the data center and the user, and is highly flexible. The network control plane is deployed in the multi-access edge computing, which serves as an interface to receive computing tasks offloaded by users, and aggregates computing resources from the data center, thereby realizing secure scheduling of computing resources and meeting the needs of user computing tasks.

[0006] The technical problem solved by this invention is achieved through the following technical solution:

[0007] A method for secure scheduling of computing resources in a computing power network, characterized in that the method comprises the following steps:

[0008] 1) Forming a computing power resource pool: Aggregating the computing power resources of the data center to form a computing power resource pool;

[0009] 101) Multi-access edge computing (MEC) establishes a communication connection with the data center, and at the same time, MEC and the data center generate their respective public keys m. e = (n,e) and private key m d = (n,d), and keep the private key and broadcast the public key;

[0010] The generation of public and private keys satisfies:

[0011] gcd((p-1)(q-1),e)=1

[0012] ed mod m = 1

[0013] Where p and q are two large prime numbers, and n = pq;

[0014] 102) The MEC queries the data center with which it has established a connection for information on its computing resources, and at the same time probes the network status information;

[0015] 103) Gather computing resource information and network status information to prepare for the secure scheduling of computing resources;

[0016] 2) User requests computing resources: When a user's own computing power is insufficient to meet the computing task requirements, it will request computing resources from the nearest MEC to meet the computing task requirements.

[0017] 201) Users within the MEC service area will randomly generate different computing tasks within each time slot t;

[0018] 202) Determine whether the user will be unable to complete the computing task i according to the computing task requirements due to limited computing power;

[0019] 203) If the user can complete the computation task i according to the computation task requirements, then the user will process the computation task locally;

[0020] 204) If the user is unable to complete the task, they will send a request for computing power resource scheduling to the MEC;

[0021] 205) After the MEC accepts the request, the user offloads the computing task to the MEC;

[0022] Where: the computation task i can be represented as {D} i C i ,T i ,S i}, D i Represents the data size, C iT represents the number of CPU cycles required to complete computation task i. i S represents the latency requirement for MEC to complete this computational task. i Data security requirements for computation task i;

[0023] 3) Secure scheduling of computing resources: Based on the information of the computing resource pool and the needs of computing tasks, MEC achieves joint optimization of computing latency T and energy consumption cost E in computing resource scheduling under the constraints of computing latency requirements and data security.

[0024] 301) Formalize the computing power security scheduling process by combining computing tasks with computing resource pools and network state information;

[0025] 302) Determine if the network state has been updated; if it has, then the model parameters based on the action constraint reinforcement learning method need to be retrained according to the formalization of the computing power security scheduling process; otherwise, directly apply action constraint reinforcement learning to obtain the optimal computing power resource security scheduling scheme.

[0026] 303) According to the optimal computing resource security scheduling scheme, the computing resources are scheduled to complete the user's computing tasks;

[0027] 4) Obtain the results of the computation task: Complete the computation task using the scheduled computing resources and return the final computation results to the user;

[0028] 401) Perform computing power scheduling and data transmission in accordance with the computing power resource security scheduling plan;

[0029] 402) The computing task is completed by the scheduled computing resources;

[0030] 403) After obtaining the calculation results of the calculation task, return the calculation results to the user.

[0031] Furthermore, the specific steps of the formalized computing power security scheduling process in step 301) are as follows:

[0032] 30101) Receive information about computation task i;

[0033] 30102) If scheduling only MEC computing resources can meet the requirements of computing task i, then schedule MEC computing resources to process the computing task, and update the information of the computing resource pool at the same time. In this case, the computing latency is expressed as:

[0034]

[0035] Where: j = 0 represents that the computing resources of the data center do not need to be scheduled;

[0036] f m This represents the computing speed of MEC, i.e., the CPU cycle frequency;

[0037] The cost of energy consumption is expressed as:

[0038]

[0039] Where: μ m A coefficient representing the energy consumed per CPU cycle;

[0040] k m Represents the cost per unit of energy;

[0041] 30103) If the MEC computing resources are still insufficient to meet the computing needs, then the computing resources of the data center will be scheduled to assist in the computing. At this time, it is necessary to pre-occupy network resources and data center computing resources, and update network status information and computing resource pool information.

[0042] 30104) Based on the requirements of computing task i, determine whether it needs security guarantees when scheduling data center computing resources, that is, based on the S of computing task i. i To determine whether or not safety measures are in place;

[0043] 30105) If the security requirements of computation task i are low, then the corresponding data center CC can be called directly. j The computing resources are used to directly transmit the data of computing task i. The data transmission rate is expressed as:

[0044] r ij =Blog(1+pgd) j -v / σ)

[0045] Where: B represents from MEC to CC j Transmission link channel bandwidth;

[0046] σ represents MEC local noise;

[0047] P represents the transmission power of the MEC;

[0048] gd j -v Representing MEC to CC j Channel gain;

[0049] d j Representing MEC to CC j The distance between them, where v is the channel loss factor;

[0050] Therefore, the data upload time is represented as:

[0051]

[0052] Energy consumption is expressed as:

[0053]

[0054] 30106) Through scheduling data center CC j The computing resources are used to process computing task i. The computational delay is expressed as:

[0055]

[0056] in: Represents CC j In terms of computational speed, CC generally has a stronger computational capability than MEC, i.e., f C >f m ;

[0057] CC j The energy consumption cost is expressed as:

[0058]

[0059] in: Represents CC j The coefficient of energy consumed per CPU cycle;

[0060] Represents CC j Cost per unit of energy;

[0061] 30107) If the security requirements for computation task i are high, the data x of computation task i will be generated into a 128-bit key using h(x) = x mod 128 to facilitate subsequent integrity verification; at the same time, the data of computation task i will be transmitted through the data center CC. j public key m e Encryption is performed to generate ciphertext y to protect data privacy, where the generation process of y is represented as:

[0062] y = x e mod n

[0063] The additional computational latency caused by data security protection is:

[0064]

[0065] The additional energy costs incurred due to data security protection are:

[0066]

[0067] in: The computation task i is transferred to CC. j The total number of CPU cycles required to perform privacy protection;

[0068] ωij The total number of CPU cycles required to perform integrity verification for a computational task;

[0069] 30108) transmits data that has undergone security protection mechanisms to CC. j At this point, the data size D i This may vary, therefore the transmission time can be expressed as:

[0070]

[0071] Where: τ represents the coefficient of data size change caused by the security protection mechanism;

[0072] Energy consumption is expressed as:

[0073]

[0074] 30109) When CC j After receiving data that has passed through a security protection mechanism, it will first use the private key m d Decryption yields the original data x, i.e., x = y d mod n; then generate keywords again using h(x) = x mod 128, and compare them with the keywords generated by MEC. If they are the same, the data is intact; otherwise, the data is corrupted.

[0075] By CC j The additional computational delay caused by decryption is represented as follows:

[0076]

[0077] The additional energy consumption cost incurred by integrity verification is expressed as:

[0078]

[0079] in: Representing CC j Total number of CPU cycles required for decryption;

[0080] 30110) via scheduled CC j The computing resources are used to perform computational processing for computational task i, where the computational latency and energy consumption costs are represented in the same way as in step 30106).

[0081] Furthermore, the specific steps of the computational resource security scheduling scheme based on action constraints in step 302) are as follows:

[0082] 30201) If the network state is updated, the model parameters of the method need to be retrained. First, the optimization problem is formalized as a Markov Decision Process (MDP), which can be represented by M = {S, A, P, R, γ}, where S represents the state space, A represents the action space, P represents the state transition probability, R represents the reward, and γ is the discount factor. Specifically:

[0083] State space: The state specifically refers to the user's requirements, that is, the computational tasks generated by the user. It can be known that the state space constituted by the computational task i in time slot t is s. t ={D i,t C i,t ,T i,t ,S i};

[0084] Action Space: The action space refers to the security scheduling strategy for computing resources, denoted as A. t ={a i,t ,a i,1,t ,…,a i,j,t ,…,a i,M,t};

[0085] Where: a i,t This indicates whether MEC computing resources are scheduled for computation task i within time slot t;

[0086] a i,j,t This indicates whether task i needs to be scheduled (CC) within time slot t. j Computing resources;

[0087] Reward: The objective is to minimize the total cost of the system, which can be expressed as:

[0088]

[0089] in:

[0090] α = {α1, α2} represents the weights of computation delay and energy consumption cost, and satisfies ∑ i＝1,2 α i =1, while ensuring that the computation latency of each computation task meets the latency requirements, that is Otherwise, C will be given t A certain punishment;

[0091] 30202) Create a predictive neural network and a target neural network model. The model parameters of the predictive neural network are represented as θ, and the parameters of the target neural network are represented as... The input is the reduced state space {D}. i,t C i,tThe output is the reward value Q(θ) that can be obtained for each action in the action space;

[0092] 30203) By setting the penalty value for unavailable actions through action constraints, the penalty vector A for unavailable actions is obtained. unavl This allows the computation latency requirements of the computation task to be met, while avoiding the over-selection of unavailable actions during model parameter training, thus improving the performance of the method.

[0093] 30204) The output value of the neural network is combined with the penalty value of the unavailable action, and then action a is selected according to the ε-greedy rule, i.e.

[0094]

[0095] ε is determined after each training iteration based on the factor ε decay It decays exponentially until it reaches a minimum value ε. min ,Right now

[0096] ε=ε·ε decay ;

[0097] 30205) When selecting action a t Afterwards, performing the selected action will earn a reward r. t and the next state space s generated by the next computational task t+1 Then the tuple {s t ,a t ,r t ,s t+1 Stored in the replay buffer experience pool;

[0098] During training (30206), data {s} are randomly selected from the replay buffer experience pool. t ,a t ,r t ,s t+1}, and generate the target Q-value based on the target network, represented as

[0099]

[0100] 30207) Update the network parameters, i.e., reduce the loss function, which is expressed as:

[0101] L(θ)=E[(y t -Q(θ)) 2 ];

[0102] After training (30208), the optimal computing resource security scheduling scheme can be obtained by directly inputting the computing task into the state space.

[0103] The advantages and beneficial effects of this invention are as follows:

[0104] This invention considers scenarios with multiple data center computing resource nodes across the network, integrating the computing resources of multiple data centers and achieving unified, secure, and optimized scheduling of computing resources. Under the constraint of meeting computational latency requirements, it jointly optimizes computational latency and energy consumption costs, while ensuring data privacy and integrity during computing resource scheduling. Attached Figure Description

[0105] Figure 1 The flowchart of this invention

[0106] Figure 2 This is a flowchart illustrating a user request for computing resources according to the present invention.

[0107] Figure 3 This is a flowchart illustrating the formalized computing power security scheduling process of this invention. Detailed Implementation

[0108] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0109] This invention discloses a secure scheduling method for computing resources in a computing network. It introduces multi-access edge computing (MEC), which operates at the network edge (e.g., base stations or gateways), positioned between the data center and users, offering greater flexibility. Therefore, the network control plane is deployed within the MEC. On one hand, it serves as an interface to receive computing tasks offloaded by users; on the other hand, it aggregates computing resources from the data center, thereby achieving secure scheduling of computing resources to meet the computing needs of users.

[0110] like Figure 1 As shown, a method for secure scheduling of computing resources in a computing power network is innovative in that the steps of the method are as follows:

[0111] 1) Forming a computing power resource pool: Aggregating the computing power resources of the data center to form a computing power resource pool;

[0112] 101) The MEC establishes a communication connection with the data center, and at the same time, the MEC and the data center generate their respective public keys m. e = (n,e) and private key m d = (n,d), and keep the private key and broadcast the public key;

[0113] The generation of public and private keys satisfies:

[0114] gcd((p-1)(q-1),e)=1

[0115] ed mod m = 1

[0116] Where p and q are two large prime numbers, and n = pq;

[0117] 102) The MEC queries the data center with which it has established a connection for information on its computing resources, and at the same time probes the network status information;

[0118] 103) Gather computing resource information and network status information to prepare for the secure scheduling of computing resources;

[0119] 2) User requests computing resources: When a user's own computing power is insufficient to meet the demands of a computing task, it will request computing resources from the nearest MEC to fulfill the task's requirements; the specific steps are as follows... Figure 2 As shown,

[0120] 201) Users within the MEC service area will randomly generate different computing tasks within each time slot t;

[0121] 202) Determine whether the user will be unable to complete the computing task i according to the computing task requirements due to limited computing power;

[0122] 203) If the user can complete the computation task i according to the computation task requirements, then the user will process the computation task locally;

[0123] 204) If the user is unable to complete the task, they will send a request for computing power resource scheduling to the MEC;

[0124] 205) After the MEC accepts the request, the user offloads the computing task to the MEC;

[0125] Where: the computation task i can be represented as {D} i C i ,T i ,S i}, D i Represents the data size, C i T represents the number of CPU cycles required to complete computation task i. i S represents the latency requirement for MEC to complete this computational task. i Data security requirements for computation task i;

[0126] 3) Secure scheduling of computing resources: Based on the information of the computing resource pool and the needs of computing tasks, MEC achieves joint optimization of computing latency T and energy consumption cost E in computing resource scheduling under the constraints of computing latency requirements and data security.

[0127] 301) Formalize the computing power security scheduling process by combining computing tasks with computing resource pools and network state information;

[0128] 302) Determine if the network state has been updated; if it has, then the model parameters based on the action constraint reinforcement learning method need to be retrained according to the formalization of the computing power security scheduling process; otherwise, directly apply action constraint reinforcement learning to obtain the optimal computing power resource security scheduling scheme.

[0129] 303) According to the optimal computing resource security scheduling scheme, the computing resources are scheduled to complete the user's computing tasks;

[0130] 4) Obtain the results of the computation task: Complete the computation task using the scheduled computing resources and return the final computation results to the user;

[0131] 401) Perform computing power scheduling and data transmission in accordance with the computing power resource security scheduling plan;

[0132] 402) The computing task is completed by the scheduled computing resources;

[0133] 403) After obtaining the calculation results of the calculation task, return the calculation results to the user.

[0134] like Figure 3 As shown, the specific steps of the formalized computing power security scheduling process in step 301) are as follows:

[0135] 30101) Receive information about computation task i;

[0136] 30102) If scheduling only MEC computing resources can meet the requirements of computing task i, then schedule MEC computing resources to process the computing task, and update the information of the computing resource pool at the same time. In this case, the computing latency is expressed as:

[0137]

[0138] Where: j = 0 represents that the computing resources of the data center do not need to be scheduled;

[0139] f m This represents the computing speed of MEC, i.e., the CPU cycle frequency;

[0140] The cost of energy consumption is expressed as:

[0141]

[0142] Where: μ m A coefficient representing the energy consumed per CPU cycle;

[0143] k m Represents the cost per unit of energy;

[0144] 30103) If the MEC computing resources are still insufficient to meet the computing needs, then the computing resources of the data center will be scheduled to assist in the computing. At this time, it is necessary to pre-occupy network resources and data center computing resources, and update network status information and computing resource pool information.

[0145] 30104) Based on the requirements of computing task i, determine whether it needs security guarantees when scheduling data center computing resources, that is, based on the S of computing task i. i To determine whether or not safety measures are in place;

[0146] 30105) If the security requirements of computation task i are low, then the corresponding data center CC is called directly. j The computing resources are used to directly transmit the data of computing task i. The data transmission rate is expressed as:

[0147] r ij =Blog(1+pgd) j -v / σ)

[0148] Where: B represents from MEC to CC j Transmission link channel bandwidth;

[0149] σ represents MEC local noise;

[0150] P represents the transmission power of the MEC;

[0151] gd j -v Representing MEC to CC j Channel gain;

[0152] d j Representing MEC to CC j The distance between them, where v is the channel loss factor;

[0153] Therefore, the data upload time is represented as:

[0154]

[0155] Energy consumption is expressed as:

[0156]

[0157] 30106) Through scheduling data center CC j The computing resources are used to process computing task i. The computational delay is expressed as:

[0158]

[0159] in: Represents CCj In terms of computational speed, CC generally has a stronger computational capability than MEC, i.e., f C >f m ;

[0160] CC j The energy consumption cost is expressed as:

[0161]

[0162] in: Represents CC j The coefficient of energy consumed per CPU cycle;

[0163] Represents CC j Cost per unit of energy;

[0164] 30107) If the security requirements for computation task i are high, the data x of computation task i will be generated into a 128-bit key using h(x) = x mod 128 to facilitate subsequent integrity verification; at the same time, the data of computation task i will be transmitted through the data center CC. j public key m e Encryption is performed to generate ciphertext y to protect data privacy, where the generation process of y is represented as:

[0165] y = x e mod m

[0166] The additional computational latency caused by data security protection is:

[0167]

[0168] The additional energy costs incurred due to data security protection are:

[0169]

[0170] in: The computation task i is transferred to CC. j The total number of CPU cycles required to perform privacy protection;

[0171] ω ij The total number of CPU cycles required to perform integrity verification for a computational task;

[0172] 30108) transmits data that has undergone security protection mechanisms to CC. j At this point, the data size D i This may vary, therefore the transmission time can be expressed as:

[0173]

[0174] Where: τ represents the coefficient of data size change caused by the security protection mechanism;

[0175] Energy consumption is expressed as:

[0176]

[0177] 30109) When CC j After receiving data that has passed through a security protection mechanism, it will first use the private key m d Decryption yields the original data x, i.e., x = y d mod n; then generate keywords again using h(x) = x mod 128, and compare them with the keywords generated by MEC. If they are the same, the data is intact; otherwise, the data is corrupted.

[0178] By CC j The additional computational delay caused by decryption is represented as follows:

[0179]

[0180] The additional energy consumption cost incurred by integrity verification is expressed as:

[0181]

[0182] in: Representing CC j Total number of CPU cycles required for decryption;

[0183] 30110) via scheduled CC j The computing resources are used to perform computational processing for computational task i, where the computational latency and energy consumption costs are represented in the same way as in step 30106).

[0184] The specific steps of the computational resource security scheduling scheme based on action constraint reinforcement learning in step 302) are as follows:

[0185] 30201) If the network state is updated, the model parameters of the method need to be retrained. First, the optimization problem is formalized as a Markov Decision Process (MDP), which can be represented by M = {S, A, P, R, γ}, where S represents the state space, A represents the action space, P represents the state transition probability, R represents the reward, and γ is the discount factor. Specifically:

[0186] State space: The state specifically refers to the user's requirements, that is, the computational tasks generated by the user. It can be known that the state space constituted by the computational task i in time slot t is s. t ={D i,t C i,t ,T i,t ,Si};

[0187] Action Space: The action space refers to the security scheduling strategy for computing resources, denoted as A. t ={a i,t ,a i,1,t ,…,a i,j,t ,…,a i,M,t};

[0188] Where: a i,t This indicates whether MEC computing resources are scheduled for computation task i within time slot t;

[0189] a i,j,t This indicates whether task i needs to be scheduled (CC) within time slot t. j Computing resources;

[0190] Reward: The objective is to minimize the total cost of the system, which can be expressed as:

[0191]

[0192] in:

[0193] α = {α1, α2} represents the weights of computation delay and energy consumption cost, and satisfies ∑ i＝1,2 α i =1, while ensuring that the computation latency of each computation task meets the latency requirements, that is Otherwise, C will be given t A certain punishment;

[0194] 30202) Create a predictive neural network and a target neural network model. The model parameters of the predictive neural network are represented as θ, and the parameters of the target neural network are represented as... The input is the reduced state space {D}. i,t C i,t The output is the reward value Q(θ) that can be obtained for each action in the action space;

[0195] 30203) By setting the penalty value for unavailable actions through action constraints, the penalty vector A for unavailable actions is obtained. unavl This allows the computation latency requirements of the computation task to be met, while avoiding the over-selection of unavailable actions during model parameter training, thus improving the performance of the method.

[0196] 30204) The output value of the neural network is combined with the penalty value of the unavailable action, and then action a is selected according to the ε-greedy rule, i.e.

[0197]

[0198] ε is determined after each training iteration based on the factor ε decay It decays exponentially until it reaches a minimum value ε. min ,Right now

[0199] ε=ε·ε decay ;

[0200] 30205) When selecting action a t Afterwards, performing the selected action will earn a reward r. t and the next state space s generated by the next computational task t+1 Then the tuple {s t ,a t ,r t ,s t+1 Stored in the replay buffer experience pool;

[0201] During training (30206), data {s} are randomly selected from the replay buffer experience pool. t ,a t ,r t ,s t+1}, and generate the target Q-value based on the target network, represented as

[0202]

[0203] 30207) Update the network parameters, i.e., reduce the loss function, which is expressed as:

[0204] L(θ)=E[(y t -Q(θ)) 2 ];

[0205] After training (30208), the optimal computing resource security scheduling scheme can be obtained by directly inputting the computing task into the state space.

[0206] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for secure scheduling of computing resources in a computing power network, characterized in that: The steps of the method are as follows: 1) Forming a computing power resource pool: Aggregating the computing power resources of the data center to form a computing power resource pool; 101) Multi-access edge computing (MEC) establishes a communication connection with the data center, and at the same time, MEC and the data center generate their respective public keys. With private key The private key is kept secret, while the public key is broadcast and distributed. The generation of public and private keys satisfies: ； ； Where p and q are two large prime numbers, and 102) The MEC queries the data center with which it has established a connection for information on its computing resources, and at the same time probes the network status information; 103) Gather computing resource information and network status information to prepare for the secure scheduling of computing resources; 2) User requests computing resources: When a user's own computing power is insufficient to meet the computing task requirements, it will request computing resources from the nearest MEC to meet the computing task requirements; 201) Users within the MEC service area will randomly generate different computing tasks within each time slot t; 202) Determine whether the user will be unable to complete the computing task i according to the computing task requirements due to limited computing power; 203) If the user can complete the computation task i according to the computation task requirements, then the user will process the computation task locally; 204) If the user is unable to complete the task, they will send a request for computing power resource scheduling to the MEC. 205) After the MEC accepts the request, the user offloads the computing task to the MEC; Where: computation task i can be represented as , Represents the size of the data. This represents the number of CPU cycles required to complete computation task i. This represents the latency requirement for MEC to complete the computation task. Data security requirements for computation task i; 3) Secure scheduling of computing resources: Based on the information of the computing resource pool and the needs of computing tasks, MEC achieves joint optimization of computing latency T and energy consumption cost E in computing resource scheduling under the constraints of computing latency requirements and data security. 301) Formalize the computing power security scheduling process by combining computing tasks with computing resource pools and network state information; 302) Determine if the network state has been updated; if it has, then the model parameters based on the action constraint reinforcement learning method need to be retrained according to the formalization of the computing power security scheduling process; otherwise, directly apply action constraint reinforcement learning to obtain the optimal computing power resource security scheduling scheme. 303) According to the optimal computing resource security scheduling scheme, the computing resources are scheduled to complete the user's computing tasks; 4) Obtain the results of the computation task: Complete the computation task using the scheduled computing resources and return the final computation results to the user; 401) Perform computing power scheduling and data transmission in accordance with the computing power resource security scheduling plan; 402) The computing task is completed by the scheduled computing resources; 403) After obtaining the calculation results of the calculation task, return the calculation results to the user.

2. The method for secure scheduling of computing resources in a computing network according to claim 1, characterized in that: The specific steps of the formalized computing power security scheduling process in step 301) are as follows: 30101) Receive information about computation task i; 30102) If scheduling only MEC computing resources can meet the requirements of computing task i, then schedule MEC computing resources to process the computing task, and update the information of the computing resource pool at the same time. In this case, the computing latency is expressed as: ； in: This means that there is no need to schedule data center computing resources; This represents the computing speed of MEC, i.e., the CPU cycle frequency; The cost of energy consumption is expressed as: ； in: A coefficient representing the energy consumed per CPU cycle; Represents the cost per unit of energy; 30103) If the MEC computing resources are still insufficient to meet the computing needs, then the computing resources of the data center will be scheduled to assist in the computing. At this time, it is necessary to pre-occupy network resources and data center computing resources, and update network status information and computing resource pool information. 30104) Based on the requirements of computing task i, determine whether it needs security guarantees when scheduling data center computing resources, that is, based on the requirements of computing task i. To determine whether or not safety measures are in place; 30105) If the security requirements of computation task i are low, then the corresponding data center can be directly invoked. The computing resources are used to directly transmit the data of computing task i. The data transmission rate is expressed as: ； Where: B represents from MEC to Transmission link channel bandwidth; Represents MEC local noise; p Represents the transmission power of the MEC; Representing MEC to Channel gain; Representing MEC to The distance between them It is the channel loss factor; Therefore, the data upload time is represented as: ； Energy consumption is expressed as: ； 30106) By scheduling data centers The computing resources are used to process computing task i. The computational delay is expressed as: ； in: represent The calculation speed, i.e. ; The energy consumption cost is expressed as: ； in: represent The coefficient of energy consumed per CPU cycle; represent Cost per unit of energy; 30107) If the security requirements for computation task i are high, the data x of computation task i will be transferred through... Generate a 128-bit key to facilitate subsequent integrity verification; simultaneously, transmit the data from computation task i through the data center. public key Encryption is performed to generate ciphertext y to protect data privacy, where the generation process of y is represented as: ； The additional computational latency caused by data security protection is: ； The additional energy costs incurred due to data security protection are: ； in: The computation task i is transferred to The total number of CPU cycles required to perform privacy protection; The total number of CPU cycles required to perform integrity verification for a computational task; 30108) transmits data that has undergone security protection mechanisms to At this time, the data size This may vary, therefore the transmission time can be expressed as: ； in: This represents the coefficient of data size variation caused by security protection mechanisms; Energy consumption is expressed as: ； 30109) When After receiving data that has passed through a security protection mechanism, it will first use its private key. Decryption is performed to obtain the original data x, i.e. Then through again Generate keywords and compare them with the keywords generated by MEC. If they are the same, the data is intact; otherwise, the data has been corrupted. Depend on The additional computational delay caused by decryption is represented as follows: ； The additional energy consumption cost incurred by integrity verification is expressed as: ； in: Represent Total number of CPU cycles required for decryption; 30110) through scheduling The computing resources are used to perform computational processing for computational task i, where the computational latency and energy consumption costs are represented in the same way as in step 30106.

3. The method for secure scheduling of computing resources in a computing network according to claim 1, characterized in that: The specific steps of the computational resource security scheduling scheme based on action constraints in step 302) are as follows: 30201) If the network state is updated, the model parameters of the method need to be retrained; first, the optimization problem is formalized as a Markov Decision Process (MDP), which can be used... Let S represent the state space, A represent the action space, P represent the state transition probability, and R represent the reward. It is a discount factor, specifically: State space: Specifically, "state" refers to user requirements, i.e., the computational tasks generated by the user. Therefore, the state space constituted by computational task i in time slot t is: ; Action Space: The action space refers to the security scheduling strategy for computing resources, represented as... ; in: This indicates whether MEC computing resources are scheduled for computation task i within time slot t; This indicates whether task i needs to be scheduled within time slot t. Computing resources; Reward: The objective is to minimize the total cost of the system, which can be expressed as: in: , ; Represents the respective weights of computation latency and energy consumption cost, and satisfies At the same time, it ensures that the computation latency of each computation task meets the latency requirements, that is... Otherwise, will be given A certain punishment; 30202) Create predictive neural network and target neural network models. The model parameters of the predictive neural network are expressed as follows: The parameters of the target neural network are expressed as follows: The input is the reduced state space { The output is the reward value that can be obtained for each action in the action space. ; 30203) By setting penalty values ​​for unavailable actions through action constraints, the penalty vector for unavailable actions is obtained. This allows the computation latency requirements of the computation task to be met, while avoiding the over-selection of unavailable actions during model parameter training, thus improving the performance of the method. 30204) combines the output value of the neural network with the penalty value for unavailable actions, and then according to -greedy rule selects actions ,Right now ； Based on factors after each training session It decays exponentially until it reaches a minimum. ,Right now ； 30205) When selecting an action Afterwards, performing the selected action will earn you a reward. and the next state space generated by the next computational task Then the tuple Stored in the replay buffer experience pool; During training (30206), data is randomly selected from the replay buffer experience pool. And generate the target Q value based on the target network, represented as ； 30207) Update the network parameters, i.e., reduce the loss function, which is expressed as: ； (30208) After training is completed, the optimal computing resource security scheduling scheme can be obtained by directly inputting the computing task into the state space.

Images