Block chain-based edge intelligent system resource management method

Abstract

The invention discloses an edge intelligent system resource management method based on a block chain. The method comprises the steps that a system controller senses artificial intelligence task information of current user equipment and current wireless environment information of a system; meanwhile, the system controller senses computing resource information of the user equipment, the edge server and the cloud server; related information is uploaded to a system controller from corresponding equipment through wireless connection; inputting the current artificial intelligence task information and the current wireless environment information of the system into a trained optimization model deployed in a system controller, and calculating user equipment transmitting power control, artificial intelligence task data volume control and user equipment artificial intelligence task unloading decisions based on a current state; calculating resource allocation conditions required by the reasoning process of the artificial intelligence task and the block generation process are transmitted to each calculation entity for execution; according to the method, the system efficiency and the user experience are improved by utilizing a task unloading and computing resource allocation algorithm.

Description

Technical Field

[0001] The present invention relates to the field of blockchain and edge intelligence technology, and specifically to a blockchain-based edge intelligence system resource management method. Background Art

[0002] Ubiquitous artificial intelligence is an important development trend for realizing the intelligence of various scenarios. It is highly resource-intensive and latency-sensitive, and has extremely high demands on communication and computing resources in the network. The traditional cloud computing model has the disadvantages of long transmission distances and concentrated computing task loads. Edge intelligence is an extension of edge computing, which deploys artificial intelligence services at the edge layer, enabling them to analyze local data without continuous reliance on cloud computing. This means that the reasoning process of artificial intelligence tasks can be carried out closer to the data source, such as sensors, smart devices, and routers, thereby reducing task transmission latency and bandwidth pressure.

[0003] Since the operation of the edge intelligent system involves the user device offloading its own artificial intelligence tasks to the edge for reasoning, some sensitive information will be transmitted outward from the user device. Therefore, the edge intelligent system will face serious privacy leakage problems. A safe and reliable edge intelligent system requires a strict protection mechanism. Blockchain is a decentralized ledger technology that connects more and more blocks together to form a chain, recording a series of encrypted and verified transactions or data. These blocks are interconnected through special encryption algorithms to form a decentralized, tamper-proof database. It can avoid the single point of failure and data tampering risks of traditional centralized databases. In blockchain technology, block generation is the computational process of verifying transactions and creating new blocks. For this reason, the concept of blockchain-assisted edge intelligent system is proposed.

[0004] In the blockchain-assisted edge intelligence system, the blockchain module is installed in the edge server located in each base station to facilitate the block generation process, and at the same time, the AI ​​model is also deployed on all edge servers for AI task reasoning. When the AI ​​task is offloaded from the user device to the edge server through the connection from the user device to the base station or further offloaded to the cloud server through the connection from the base station to the cloud server, its related information needs to be recorded in the transaction and then packaged into the block, which requires the system to have enough blocks to support the number of AI tasks. Both the block generation process and the AI ​​task reasoning process occupy the computing resources of the edge server and compete with each other. If the block generation process occupies more computing resources, the block generation speed will be faster and the AI ​​task reasoning speed will be slower. The blockchain system will generate redundant blocks, resulting in a waste of computing resources. If the AI ​​task reasoning occupies more computing resources, the system throughput can be improved, but the reduced block generation speed can hardly meet the recording requirements of the AI ​​task reasoning. Therefore, matching the computing resource allocation of the block generation process and the AI ​​task reasoning process is a prerequisite for optimizing system performance.

[0005] However, current research works regard block generation and AI task reasoning as independent processes and optimize them separately without solving the matching problem between the two. At the same time, these works have not discussed the attribute differences between AI tasks and traditional tasks, such as AI task reasoning accuracy. In particular, AI task reasoning accuracy is related to the amount of AI task data, which has not been considered in related work. Summary of the invention

[0006] 1. Technical issues to be resolved

[0007] In order to overcome the shortcomings of the above-mentioned prior art, the present invention is aimed at the tasks of the blockchain system based on mobile edge intelligence, and comprehensively considers factors such as the computing amount of artificial intelligence tasks and the network transmission rate in the system structure to determine the artificial intelligence task offloading decision, resource allocation strategy and artificial intelligence task data volume control strategy. The present invention is inherited to the system controller deployment module and resource configuration module of the edge server layer, combined with the network status perception and acquisition module, to realize the dynamic optimization of artificial intelligence task offloading decision, resource allocation strategy and artificial intelligence task data volume control strategy, and on the basis of considering practicality, accelerate the neural network training process and optimization process, and improve user experience.

[0008] (II) Technical solution

[0009] In order to solve the above technical problems, the present invention proposes a resource management method for edge computing system based on blockchain, comprising the following steps:

[0010] S1: The system controller senses the AI ​​task information generated by the current user device and the current wireless environment information of the system; at the same time, the system controller senses the computing resource information of the user device, edge server and cloud server; and uploads the relevant information from the corresponding user device to the system controller via wireless connection;

[0011] S2: Input the current artificial intelligence task information and the current wireless environment information of the system described in step S1 into the trained optimization model deployed in the system controller, and calculate the user equipment transmit power control based on the current state, the user equipment artificial intelligence task offloading decision, the user equipment artificial intelligence task data volume control, and the computing resource allocation to be used in the reasoning process and block generation process of the artificial intelligence task;

[0012] S3: extracting the artificial intelligence task offloading decision information of the user device obtained in step S2. If the information indicates that the artificial intelligence task is inferred locally, the system controller sends a corresponding artificial intelligence task offloading decision instruction to the user device via a wireless connection, and the user device directly infers the artificial intelligence task locally.

[0013] S4: Extract the artificial intelligence task unloading decision information of the user device obtained in step S2. If the information indicates that the artificial intelligence task is inferred on the edge server or cloud server, the system controller sends the corresponding artificial intelligence task unloading decision instruction, artificial intelligence task data volume control instruction, and transmission power control instruction to the user device through a wireless connection. The user device transmits the artificial intelligence task to the base station according to the corresponding transmission power, and joins the blockchain to execute the block generation process; if the artificial intelligence task is inferred on the edge server, the system controller sends a computing resource allocation instruction to the edge server through a wireless connection, and after the task reasoning process and the block generation process are all completed, the result is sent back to the user device; if the artificial intelligence task is inferred on the cloud server, the system controller sends a computing resource allocation instruction to the cloud server through a wired link, and it is necessary to further transmit the artificial intelligence task from the base station to the cloud server through a wired link for reasoning. The above instructions are used to control the parallel execution of the reasoning process and block generation process of the artificial intelligence task;

[0014] Furthermore, the step S1 is specifically as follows:

[0015] The characteristics of the network architecture of this method are as follows: represents the edge server (base station) set, One of the base stations exist In the example, edge server m (base station) covers N m User devices, represents the user equipment set under base station m, One of the user devices In addition, a system controller is placed at the edge layer to control the operation of the entire method.

[0016] The algorithm is deployed on a system controller to work. The system controller perceives the current user equipment artificial intelligence task information and the current system wireless environment information. The artificial intelligence task characteristics are composed of the amount of calculation, the amount of data, and the inference accuracy. Let c mn Indicates the computational effort of the AI ​​tasks generated by each user device, Let d mn Indicates the amount of data generated by each user device for the AI ​​task, Let μ mn represents the inference accuracy of the artificial intelligence task generated by each user device; the current wireless environment characteristics are composed of the channel gain from the user device to the base station, let g mn represents the channel gain from user equipment mn to its corresponding base station, At the same time, the system controller senses the computing resource information of user devices, edge servers, and cloud servers, and and F CS are the upper bounds of computing resources of user device mn, edge server m, and cloud server respectively.

[0017] During the working process, the system controller will upload the above information from the corresponding user equipment to the system controller through a wireless connection.

[0018] Furthermore, the step S2 is specifically as follows:

[0019] The data volume d of the current artificial intelligence task obtained in step S1 and the channel gain g between the current user equipment and the base station are input into the trained optimization model deployed on the system controller to obtain computing resource management instructions.

[0020] First, a fast numerical method is designed to obtain the transmit power control command of the user equipment in the current state, and the transmit power control command is separated from other optimization variables. Among them, p mn The total delay of the optimization target of this method decreases monotonically with the increase of the transmission power, so the transmission power control instruction can be obtained by increasing the transmission power so that the energy consumption of the user equipment reaches the energy consumption constraint.

[0021] The characteristics of artificial intelligence task scheduling in this method are as follows: regarding the artificial intelligence task offloading decision of the user device, let α mn =1 means that the AI ​​task mn is offloaded to the edge server for inference, otherwise, α mn =0; similarly, βmn =1 means that the AI ​​task mn is offloaded to the cloud server for reasoning, otherwise, β mn =0; obviously, 1-α mn -β mn =1 means that the AI ​​task mn is inferred locally on the user device mn, At the same time, regarding the allocation of computing resources used in the AI ​​task reasoning process and block generation process, The computing resources allocated to the user device mn for the reasoning process of the artificial intelligence task mn, make and are the computing resources allocated by edge server m to the block generation process and the reasoning process of artificial intelligence task mn, respectively. make The computing resources allocated by the cloud server to the reasoning process of the artificial intelligence task mn, The above optimization variables are designed based on the deep reinforcement learning technology to design an optimization algorithm and establish a Markov decision process, where the state space of this method is set to s = {c, g} and the action space is set to a = {α, β, f MINE ,f ES ,f CS ,f U ,d}. Finally, the optimization result is obtained according to the algorithm output.

[0022] Furthermore, the step S3 is specifically as follows:

[0023] Extract the offloading decision information obtained in step S2. If the information indicates that the artificial intelligence task is inferred locally on the user device, the network controller sends the artificial intelligence task offloading decision instructions α and β, the artificial intelligence task data volume control instruction d, and the user device computing resource allocation instruction f to the user device via a wireless connection. U , execute the instruction in the manner described in step S3 of claim 1, and directly return the result to the user device.

[0024] Furthermore, the step S4 is specifically as follows:

[0025] Extract the offloading decision information obtained in step S2. If the information indicates that the artificial intelligence task is executed on the edge server or cloud server, the network controller sends the artificial intelligence task offloading decision instructions α and β, the transmission power control instruction p, and the artificial intelligence task data volume control instruction d to the user device through the wireless connection, and sends the computing resource allocation instruction f to the edge server through the wireless connection MINE and f ES , or send computing resource allocation instructions to the cloud server through a wired link fC , and execute the instruction in the manner described in step S4 of claim 1, and after the artificial intelligence task reasoning process and the block generation process are all completed, the result is returned to the user device. After the corresponding instruction execution is completed, the total delay is obtained, that is, the sum of the larger of the artificial intelligence task reasoning delay and the block generation delay and the artificial intelligence task transmission delay, and the formula is:

[0026] t TOTAL =t TRAN +max(t INF ,t BC )

[0027] Among them, t TOTAL is the total processing delay of the AI ​​task, t TRAN is the transmission delay of the AI ​​task (if the AI ​​task is executed on the user device, the value is 0), t INF is the inference latency of the AI ​​task, t BC Delay for block generation.

[0028] (III) Beneficial effects

[0029] Compared with the prior art, the present invention adopts the above technical solution and has the following beneficial effects:

[0030] 1. When making AI task offloading decisions, resource allocation, and AI task data volume control, the task reasoning delay and block generation delay are jointly optimized through analysis of actual scenarios, making the technology more practical;

[0031] 2. In the decision-making of AI task offloading, resource allocation and AI task data volume control, a Markov decision process is constructed. Based on deep reinforcement learning technology, complex optimization problems are solved through neural networks, which is more practical;

[0032] 3. By analyzing and optimizing the target structure, the original problem is decomposed into two sub-problems for solution, which greatly reduces the learning difficulty of deep reinforcement learning and greatly improves the calculation speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a schematic diagram of a system model of an embodiment;

[0034] Figure 2 It is a schematic diagram of the training process of AI task offloading decision, resource allocation, and AI task data volume control;

[0035] Figure 3 It is a schematic diagram of the application process of AI task offloading decision, resource allocation, and AI task data volume control;

[0036] Figure 4 This is a schematic diagram comparing the rewards of this scheme with other schemes as the amount of computation of AI tasks increases;

[0037] Figure 5 This is a schematic diagram comparing the rewards of this scheme with other schemes as the bandwidth increases;

[0038] Figure 6 This is a schematic diagram comparing the rewards of this scheme with other schemes as the transmission rate between base stations increases. DETAILED DESCRIPTION

[0039] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0040] The present invention proposes a task scheduling method combining an edge intelligence system with a blockchain system, and the embodiment includes the following steps:

[0041] Step 1: The system controller senses the AI ​​task information generated by the current user device and the current wireless environment information of the system; at the same time, the system controller senses the computing resource information of the user device, edge server and cloud server; and uploads the relevant information from the corresponding user device to the system controller via wireless connection;

[0042] Step 2: Input the current artificial intelligence task information and the current wireless environment information of the system described in step 1 into the trained optimization model deployed in the system controller, and calculate the user equipment transmit power control based on the current state, the user equipment artificial intelligence task offloading decision, the user equipment artificial intelligence task data volume control, and the computing resource allocation to be used in the reasoning process and block generation process of the artificial intelligence task;

[0043] Step 3: extracting the artificial intelligence task offloading decision information of the user device obtained in step 2. If the information indicates that the artificial intelligence task is inferred locally, the system controller sends a corresponding artificial intelligence task offloading decision instruction to the user device through a wireless connection, and the user device directly infers the artificial intelligence task locally.

[0044] Step 4: Extract the artificial intelligence task offloading decision information of the user device obtained in step 2. If the information indicates that the artificial intelligence task is inferred on the edge server or cloud server, the system controller sends the corresponding artificial intelligence task offloading decision instruction, artificial intelligence task data volume control instruction, and transmission power control instruction to the user device through a wireless connection. The user device transmits the artificial intelligence task to the base station according to the corresponding transmission power and joins the blockchain to execute the block generation process. If the artificial intelligence task is inferred on the edge server, the system controller sends a computing resource allocation instruction to the edge server through a wireless connection. After the task reasoning process and the block generation process are all completed, the result is sent back to the user device. If the artificial intelligence task is inferred on the cloud server, the system controller sends a computing resource allocation instruction to the cloud server through a wired link, and the artificial intelligence task needs to be further transmitted from the base station to the cloud server through a wired link for reasoning. The above instructions are used to control the parallel execution of the reasoning process and block generation process of the artificial intelligence task;

[0045] Furthermore, the step 1 includes:

[0046] The characteristics of the network architecture of this method are as follows: represents the edge server (base station) set, One of the base stations exist In the example, edge server m (base station) covers N m User devices, represents the user equipment set under base station m, One of the user devices In addition, a system controller is placed at the edge layer to control the operation of the entire method.

[0047] The algorithm is deployed on a system controller to work. The system controller perceives the current user equipment artificial intelligence task information and the current system wireless environment information. The artificial intelligence task characteristics are composed of the amount of calculation, the amount of data, and the inference accuracy. Let c mn Indicates the computational effort of the AI ​​tasks generated by each user device, Let d mn Indicates the amount of data generated by each user device for the AI ​​task, Let μ mn represents the inference accuracy of the artificial intelligence task generated by each user device; the current wireless environment characteristics are composed of the channel gain from the user device to the base station, let g mn represents the channel gain from user equipment mn to its corresponding base station, At the same time, the system controller senses the computing resource information of user devices, edge servers, and cloud servers, and and FCS are the upper bounds of computing resources of user device mn, edge server m, and cloud server respectively.

[0048] During the working process, the system controller will upload the above information from the corresponding user equipment to the system controller through a wireless connection.

[0049] Furthermore, the step 2 includes:

[0050] The data volume d of the current artificial intelligence task obtained in step 1 and the channel gain g between the current user equipment and the base station are input into the trained optimization model deployed on the system controller to obtain computing resource management instructions.

[0051] First, a fast numerical method is designed to obtain the transmit power control command of the user equipment in the current state, and the transmit power control command is separated from other optimization variables. Among them, p mn The transmission power allocated to the user equipment mn. The transmission delay from the user equipment to the base station And transmission energy consumption for:

[0052]

[0053] Among them, the wireless transmission rate from user equipment to base station for:

[0054]

[0055] Where W is the bandwidth, is the noise power, and χ is the inter-channel interference. The total delay of the optimization target of this method decreases monotonically with the increase of the transmission power. Therefore, the transmission power control instruction can be obtained by increasing the transmission power so that the energy consumption of the user equipment reaches the energy consumption constraint.

[0056] The characteristics of artificial intelligence task scheduling in this method are as follows: regarding the artificial intelligence task offloading decision of the user device, let α mn =1 means that the AI ​​task mn is offloaded to the edge server for inference, otherwise, α mn =0; similarly, β mn =1 means that the AI ​​task mn is offloaded to the cloud server for reasoning, otherwise, β mn =0; obviously, 1-α mn -β mn =1 means that the AI ​​task mn is inferred locally on the user device mn, At the same time, regarding the allocation of computing resources used in the AI ​​task reasoning process and block generation process, The computing resources allocated to the user device mn for the reasoning process of the artificial intelligence task mn, make and are the computing resources allocated by edge server m to the block generation process and the reasoning process of artificial intelligence task mn, respectively. make The computing resources allocated by the cloud server to the reasoning process of the artificial intelligence task mn, The above optimization variables are designed based on the deep reinforcement learning technology to design an optimization algorithm and establish a Markov decision process, where the state space of this method is set to s = {c, g} and the action space is set to a = {α, β, f MINE ,f ES ,f CS ,f U ,d}. Finally, the optimization result is obtained according to the algorithm output.

[0057] Furthermore, the step three includes:

[0058] Extract the offloading decision information obtained in step 2. If the information indicates that the AI ​​task is inferred locally on the user device, the network controller sends AI task offloading decision instructions α and β, AI task data volume control instruction d, and user device computing resource allocation instruction f to the user device via a wireless connection. U , execute the instruction in the manner described in step S3 of claim 1, and directly return the result to the user device.

[0059] During the local reasoning process, the reasoning delay for:

[0060]

[0061] in, The amount of computing required to infer AI tasks on user devices, The computing resources of the user device. The computing energy consumption of local inference for:

[0062]

[0063] where κ is the computational efficiency of the GPU.

[0064] Furthermore, the step 4 includes:

[0065] Extract the offloading decision information obtained in step 2. If the information indicates that the AI ​​task is inferred on the edge server or cloud server, the network controller sends the AI ​​task offloading decision instructions α and β, the transmission power control instruction p, and the AI ​​task data volume control instruction d to the user device through the wireless connection, and sends the computing resource allocation instruction f to the edge server through the wireless connection MINE and f ES , or send computing resource allocation instructions to the cloud server through a wired link f C , and execute the instructions in the manner described in step S4 of claim 1, and after the artificial intelligence task reasoning process and the block generation process are all completed, the results are returned to the user device.

[0066] Among them, during the edge server reasoning process, the reasoning delay for:

[0067]

[0068] in, The amount of computing required for AI tasks to be inferred on edge servers or cloud servers. During the inference process on cloud servers, the inference latency for:

[0069]

[0070] At the same time, during cloud server inference, the wired transmission delay from the base station to the cloud server is:

[0071]

[0072] in, It is the wired transmission rate from the base station to the cloud server.

[0073] At the same time, based on the proof of work, the block generation delay is calculated Find the total latency of AI task processing:

[0074]

[0075] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the scope of protection of the claims of the present invention.

Claims

1. A resource management method for edge intelligent system based on blockchain, characterized in that: The following steps are involved: S1: The system controller senses the AI ​​task information generated by the current user device and the current wireless environment information of the system; at the same time, the system controller senses the computing resource information of the user device, edge server and cloud server; and uploads the relevant information from the corresponding user device to the system controller via wireless connection; S2: Input the current artificial intelligence task information and the current wireless environment information of the system described in step S1 into the trained optimization model deployed in the system controller, and calculate the user equipment transmit power control based on the current state, the user equipment artificial intelligence task offloading decision, the user equipment artificial intelligence task data volume control, and the computing resource allocation to be used in the reasoning process and block generation process of the artificial intelligence task; S3: extracting the artificial intelligence task offloading decision information of the user device obtained in step S2. If the information indicates that the artificial intelligence task is inferred locally, the system controller sends a corresponding artificial intelligence task offloading decision instruction to the user device via a wireless connection, and the user device directly infers the artificial intelligence task locally. S4: extracting the user equipment artificial intelligence task offloading decision information obtained in step S2. If the information indicates that the artificial intelligence task is inferred on the edge server or cloud server, the system controller sends the corresponding artificial intelligence task offloading decision instruction, artificial intelligence task data volume control instruction, and transmission power control instruction to the user equipment through a wireless connection. The user equipment transmits the artificial intelligence task to the base station according to the corresponding transmission power, and joins the blockchain to execute the block generation process. If the AI ​​task is inferred on the edge server, the system controller sends a computing resource allocation instruction to the edge server through a wireless connection, and after the task inference process and the block generation process are all completed, the result is sent back to the user device; if the AI ​​task is inferred on the cloud server, the system controller sends a computing resource allocation instruction to the cloud server through a wired link, and the AI ​​task needs to be further transmitted from the base station to the cloud server through a wired link for inference. The above instructions control the parallel execution of the inference process and block generation process of the AI ​​task.

2. According to the blockchain-based edge intelligence system resource management method of claim 1, it is characterized in that: The step S1 includes: the network architecture of the method is characterized by: represents the edge server (base station) set, One of the base stations exist In the example, edge server m (base station) covers N m User devices, represents the user equipment set under base station m, One of the user devices In addition, a system controller is placed at the edge layer to control the operation of the entire method. The algorithm is deployed on a system controller to work. The system controller perceives the current user equipment artificial intelligence task information and the current system wireless environment information. The artificial intelligence task characteristics are composed of the amount of calculation, the amount of data, and the inference accuracy. Let c mn Indicates the computational effort of the AI ​​tasks generated by each user device, Let d mn Indicates the amount of data generated by each user device for the AI ​​task, Let μ mn represents the inference accuracy of the artificial intelligence task generated by each user device; the current wireless environment characteristics are composed of the channel gain from the user device to the base station, let g mn represents the channel gain from user equipment mn to its corresponding base station, At the same time, the system controller senses the computing resource information of user devices, edge servers, and cloud servers, and and F CS are the upper bounds of computing resources of user device mn, edge server m, and cloud server respectively. During the working process, the system controller will upload the above information from the corresponding user equipment to the system controller through a wireless connection.

3. According to a blockchain-based edge intelligence system resource management method according to claim 1, it is characterized in that: The step S2 includes: inputting the data volume d of the current artificial intelligence task obtained in step S1 and the channel gain g between the current user equipment and the base station into the trained optimization model deployed on the system controller to obtain computing resource management instructions. First, a fast numerical method is designed to obtain the transmit power control command of the user equipment in the current state, and the transmit power control command is separated from other optimization variables. Among them, p mn The total delay of the optimization target of this method decreases monotonically with the increase of the transmission power, so the transmission power control instruction can be obtained by increasing the transmission power so that the energy consumption of the user equipment reaches the energy consumption constraint. The characteristics of artificial intelligence task scheduling in this method are as follows: regarding the artificial intelligence task offloading decision of the user device, let α mn =1 means that the AI ​​task mn is offloaded to the edge server for inference, otherwise, α mn =0; similarly, β mn =1 means that the AI ​​task mn is offloaded to the cloud server for reasoning, otherwise, β mn =0; obviously, 1-α mn -β mn =1 means that the AI ​​task mn is inferred locally on the user device mn, At the same time, regarding the allocation of computing resources used in the AI ​​task reasoning process and block generation process, The computing resources allocated to the user device mn for the reasoning process of the artificial intelligence task mn, make and are the computing resources allocated by edge server m to the block generation process and the reasoning process of artificial intelligence task mn, respectively. make The computing resources allocated by the cloud server to the reasoning process of the artificial intelligence task mn, The above optimization variables are designed based on the deep reinforcement learning technology to design an optimization algorithm and establish a Markov decision process, where the state space of this method is set to s = {c, g} and the action space is set to a = {α, β, f MINE ,f ES ,f CS ,f U ,d}. Finally, the optimization result is obtained according to the algorithm output.

4. According to a blockchain-based edge intelligence system resource management method according to claim 1, it is characterized in that: The step S3 includes: extracting the offloading decision information obtained in step S2, if the information indicates that the artificial intelligence task is inferred locally on the user device, the network controller sends the artificial intelligence task offloading decision instructions α and β, the artificial intelligence task data volume control instruction d, and the user device computing resource allocation instruction f to the user device via a wireless connection U , execute the instruction in the manner described in step S3 of claim 1, and directly return the result to the user device.

5. According to a blockchain-based edge intelligence system resource management method according to claim 1, it is characterized in that: The step S4 includes: extracting the offloading decision information obtained in step S2, if the information indicates that the artificial intelligence task is inferred on the edge server or cloud server, the network controller sends the artificial intelligence task offloading decision instructions α and β, the transmission power control instruction p, the artificial intelligence task data volume control instruction d to the user device through the wireless connection, and sends the computing resource allocation instruction f to the edge server through the wireless connection MINE and f ES , or send computing resource allocation instructions to the cloud server through a wired link f C , and execute the instruction in the manner described in step S4 of claim 1, and after the artificial intelligence task reasoning process and the block generation process are all completed, the result is returned to the user device. After the corresponding instruction execution is completed, the total delay is obtained, that is, the sum of the larger of the artificial intelligence task reasoning delay and the block generation delay and the artificial intelligence task transmission delay, and the formula is: t TOTAL =t TRAN +max(t INF ,t BC ) Among them, t TOTAL is the total processing delay of the AI ​​task, t TRAN is the transmission delay of the AI ​​task (if the AI ​​task is inferred on the user device, the value is 0), t INF is the inference latency of the AI ​​task, t BC Delay for block generation.

Images