Industrial Internet of Things edge computing secure communication method based on federated learning

Abstract

The invention relates to the technical field of industrial internet-of-things safety communication, in particular to an industrial internet-of-things edge computing safety communication method based on federated learning, which comprises the following steps of: constructing a unified collaboration framework of an industrial internet-of-things system; an interference signal is sent to the eavesdropping device through the edge computing server at the interference power, so that the secure transmission rate of the communication channel is a positive value in the presence of eavesdropping; calculating the total energy consumption of the edge computing server under different task unloading strategies, wherein the total energy consumption of the edge computing server does not exceed the maximum value of energy consumption which can be borne by the industrial Internet of Things system; and constructing an unloading decision model based on federated learning, and outputting an optimal unloading decision based on the unloading decision model by taking the minimum average calculation task time delay as a target. According to the method, the unified collaboration framework of the industrial Internet of Things system is constructed, federated learning and edge computing technologies are combined, data privacy protection and communication security can be remarkably improved, and the time for averaging computing tasks is minimized under the energy consumption constraint.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial Internet of Things (IIoT) secure communication, and particularly relates to a secure communication method for industrial IoT edge computing based on federated learning. Background Art

[0002] Industrial Internet of Things (IIoT), as a core component of intelligent manufacturing and Industry 4.0, is rapidly changing the operation mode of traditional industries. By connecting industrial devices, systems, and sensors to the Internet, IIoT realizes the intelligence of devices and the integration of data, thereby improving production efficiency and reducing the operation cost of enterprises. However, these advantages of IIoT also bring new challenges, especially in terms of data security and privacy protection. As more and more sensitive data flows in the IIoT network, it becomes crucial to protect this data from being stolen and misused. The existing secure communication technologies for IIoT mainly rely on traditional network security measures, such as encrypting data or access control. These measures provide a certain degree of security protection at the network layer and the transport layer.

[0003] Edge computing is a distributed computing paradigm that processes computing, storage, etc. close to the data source or the user, reducing the risk of data transmission in the network and helping to protect user privacy. At the same time, with the shortening of the transmission distance, the efficiency of the entire system is also improved. Moreover, edge computing processes data at the network edge, reducing the dependence on the central cloud, thereby saving bandwidth and storage resources. This provides significant advantages in terms of performance, cost, etc. in the IIoT scenario, and also makes edge computing technology one of the key technologies to support modern Industry 4.0 and intelligent manufacturing. With the rapid development of IIoT, the demand for data privacy, security, and edge computing is also gradually increasing, and federated learning provides a method for distributed model training without sharing raw data, which highly matches the requirements of IIoT. Therefore, federated learning has also been gradually introduced into the IIoT scenario.

[0004] Given the advantages of edge computing and federated learning in IIoT, many works have also begun to combine these two technologies with IIoT scenarios. This will also bring various problems, such as energy consumption, security, and system efficiency. Since industrial devices usually have limited resources (including computing power, bandwidth, and energy), existing solutions have failed to effectively balance the relationship among communication security, energy consumption, and task offloading efficiency, resulting in problems such as excessive energy consumption when security is improved or a decrease in task response time. Existing works have provided partial solutions in terms of federated learning, physical layer secure communication, and task scheduling optimization, but they are usually applied independently and do not form a unified framework for collaborative optimization. For the secure communication problem of edge computing in IIoT, especially in the framework of federated learning, a new secure communication method is needed to solve the problem of ignoring security in the communication process in existing technologies, and at the same time minimize the time of the average computing task under energy consumption constraints. Summary of the Invention

[0005] To solve the technical problems existing in the prior art, the present invention constructs a unified collaborative framework for the industrial Internet of Things system, combines federated learning and edge computing technologies, and can significantly improve data privacy protection and communication security, and minimize the time of the average computing task under energy consumption constraints.

[0006] The object of the present invention can be achieved by adopting the following technical solutions: A secure communication method for edge computing in an industrial Internet of Things based on federated learning, the method comprising: S1. Construct a unified collaborative framework for the industrial Internet of Things system, and divide the industrial Internet of Things system into a machine device layer, an edge layer, and a cloud server layer; The machine device layer includes a number of machine devices, and the machine devices are used to execute local computing tasks and communicate with the edge computing server, and transfer model parameters to the edge layer; The edge layer includes a number of edge computing servers, and the edge computing servers are used to collect the model parameters of the machine devices within the coverage area and update the local model using the model parameters. The edge computing servers cooperate with each other through the method of federated learning and participate in training the global model; The cloud server layer is used to coordinate the federated learning process of the entire industrial Internet of Things system and manage the aggregation and distribution of model updates among the edge computing servers; S2. The edge computing server sends interference signals to the eavesdropping device with interference power, so that the secure transmission rate of the communication channel is positive in the presence of eavesdropping; S3. Combine communication energy consumption, local training energy consumption, computing energy consumption, and interference signal energy consumption to calculate the total energy consumption of the edge computing server under different task offloading strategies, and the total energy consumption of the edge computing server does not exceed the maximum energy consumption that the industrial Internet of Things system can bear; S4. Construct an offloading decision model based on federated learning. The offloading decision model aims to minimize the average computational task latency and outputs the optimal offloading decision based on the channel state and device resources according to the offloading decision model.

[0007] Specifically, the step S2 includes: S21. Detect whether there are potential eavesdropping devices around the edge computing server by mounting cameras and synthetic aperture radars on the edge computing server; S22. When detecting that there are potential eavesdropping devices around the edge computing server, send interference signals to the eavesdropping devices with interference power so that the secure transmission rate of the communication channel is always greater than zero.

[0008] Specifically, the step S22 includes: The edge computing server captures the radio frequency energy sent by the base station and periodically transmits data. During a time period cycle, the edge computing server uses radio frequency energy harvesting technology to collect the energy sent by the base station through the energy harvesting module, and the collected energy is used to power the edge computing server; In the next time period cycle, when an edge computing server sends data to the base station and the throughput requirements of other edge computing servers do not exceed the maximum throughput that the edge computing server can achieve, the edge computing server uses interference power to send interference signals to potential eavesdropping devices so that the secure transmission rate of the edge computing server sending data to the base station is greater than zero.

[0009] Specifically, the edge computing server uses radio frequency energy harvesting technology to collect the energy sent by the base station through the energy harvesting module, and the collected energy is expressed as : ; where, is the energy conversion efficiency, is the downlink channel gain coefficient of the base station sending radio frequency energy to the edge computing server, is the transmission power of the base station, is a cycle time.

[0010] Specifically, the secure transmission rate of the edge computing server sending data to the base station is equal to the data transmission throughput from the edge computing server to the base station minus the threshold of the eavesdropping throughput, and the threshold of the eavesdropping throughput is expressed as : ; where, is a cycle time, B is the channel bandwidth, is the white noise power at the base station, is the downlink channel gain coefficient for the base station to send RF energy to the edge computing server, p 1 is the transmission power of the base station, is the interference power, .

[0011] Specifically, the self-throughput requirements of the other edge computing servers satisfy the following constraint conditions: ; Among them, represents the self-throughput requirement of the other edge computing servers, is the energy collected by the edge computing server from the base station, is a cycle time, B is the channel bandwidth, is the white noise power at the base station, is the uplink channel gain coefficient for the edge computing server to send data to the base station.

[0012] Specifically, the step S3 includes: S31. Calculate the local training energy consumption, communication energy consumption, edge computing energy consumption, and interference signal energy consumption respectively. The communication energy consumption includes the edge offloading communication energy consumption and the uploading to the cloud communication energy consumption; S32. Calculate the total energy consumption of the edge computing server under different task offloading strategies according to the local training energy consumption, communication energy consumption, edge computing energy consumption, and interference signal energy consumption. The total energy consumption of the edge computing server does not exceed the maximum energy consumption that the industrial Internet of Things system can bear. The total energy consumption of the edge computing server is expressed as:

[0013] Among them, represents that the task is offloaded to the local of the machine device; represents that the task is offloaded to the edge computing server; represents that the task is offloaded to the cloud; represents the local training energy consumption, represents the interference energy consumption, represents the edge offloading communication energy consumption, represents the edge computing energy consumption, represents the uploading to the cloud communication energy consumption.

[0014] Specifically, the step S4 includes: S41. Calculate the processing time of different offloading decision tasks and the total time consumption of the machine device participating in federated learning. With the goal of minimizing the average computing task latency, construct an offloading decision model based on federated learning S42. Use federated learning to distributively train the offloading decision model, receive model updates transmitted from the edge computing server, perform weighted averaging on the model updates to obtain a new global model, and return it to the edge computing server; S43. When offloading tasks, output the optimal offloading decision based on the offloading decision model according to the channel state and device resources.

[0015] Specifically, step S41 includes: S411. Calculate the processing time of task offloading for different offloading decisions respectively: When offloading tasks locally on the machine device, the processing time of task offloading is the computing time of the task machine device; When offloading tasks on the edge computing server, the processing time of task offloading is the sum of the communication time between the machine device and the edge computing server, the computing time of the task on the edge computing server, and the transmission time when returning the result data; When offloading tasks to the cloud, the processing time of task offloading is the sum of the communication time between the machine device and the edge computing server, the transmission time required for data to be transmitted from the edge computing server to the cloud server or the transmission time for the returned data to be transmitted back from the cloud server to the edge computing server after computing, and the transmission delay for returning the result data from the edge computing server to the machine device; S412. Calculate the total time consumption of the machine device participating in federated learning, and obtain the total time consumption of all tasks based on the processing time of task offloading and the total time consumption of the machine device participating in federated learning; S413. Set the model optimization objective function with the goal of minimizing the average computing task delay according to the total time consumption of all tasks, and construct an offloading decision model based on federated learning; The model optimization objective function is expressed as: ; where ξ represents the time period, represents the reciprocal of the number of device sets, and is used to calculate the average task delay; represents the number of device sets in the industrial Internet of Things system, represents the machine device at time offloading decision. If , it means that the task is offloaded to the local of the machine device; if , the task is offloaded to the edge computing server; if , the task is offloaded to the cloud.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: The present invention proposes an edge computing secure communication method for industrial Internet of Things based on federated learning. By combining federated learning and edge computing technologies, the edge computing server sends interference signals to eavesdropping devices with interference power, enabling the secure transmission rate of the communication channel to be positive in the presence of eavesdropping, which can significantly enhance data privacy protection and communication security. By calculating the total energy consumption of the edge computing server under different task offloading strategies and making the total energy consumption of the edge computing server not exceed the maximum energy consumption that the industrial Internet of Things system can bear, a federated learning-based offloading decision model is constructed. The offloading decision model aims to minimize the average computing task delay. Based on the offloading decision model, the optimal offloading decision is output according to the channel state and device resources, which can optimize energy consumption and task offloading strategies, improve the overall performance and reliability of the system. The present invention has high practicality and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the structures shown in these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of an edge computing secure communication method for industrial Internet of Things based on federated learning in an embodiment of the present invention; Figure 2 It is a framework diagram of an industrial Internet of Things system in an embodiment of the present invention; Figure 3 It is a topological structure diagram of an industrial Internet of Things system in an embodiment of the present invention; Figure 4 It is a schematic diagram of using federated learning to distributively train an offloading decision model in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following will further describe the technical solutions of the present invention in detail in conjunction with the drawings and embodiments. Obviously, the described embodiments are some, rather than all, of the embodiments of the present invention. The embodiments of the present invention are not limited thereto. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0020] Embodiment 1: As Figure 1 shown, the edge computing secure communication method for industrial Internet of Things based on federated learning according to the present invention includes the following steps: S1. Build a unified collaboration framework for the industrial Internet of Things (IIoT) system, and divide the IIoT system into a machine device layer, an edge layer, and a cloud server layer.

[0021] As Figure 2 shown, it is a framework diagram of the IIoT system. Based on the three-layer hierarchical architecture, the IIoT system is divided into a machine device layer, an edge layer, and a cloud server layer. Among them, the machine device layer includes several local machine devices. Each local machine device has limited computing resources, which are used to execute local computing tasks and communicate with the edge computing server, and transfer the model parameters to the edge layer. The data of these machine devices will not be directly shared, but the information of the machine devices is transferred to the edge computing server for model update of federated learning. The edge layer includes several edge computing servers. The edge computing servers are used to collect the model parameters of the machine devices within the coverage area and use the model parameters to update the local model. The edge computing servers work together in a federated learning manner and participate in training the global model. The cloud server layer is used to coordinate the federated learning process of the entire IIoT system and manage the aggregation and distribution of model updates between the edge computing servers. The cloud server does not directly access the data of the machine devices, but trains the global model based on the local model updates of each edge computing server. In the distributed collaboration of federated learning, in each training round, each edge computing server will collect local data from the machine devices it covers, and combine the model distributed by the central cloud server for local training. The edge computing server generates an update of the model parameters according to the results of the local training and sends the model update to the central cloud server. The cloud server will perform weighted averaging on these model updates to obtain a new global model.

[0022] Dividing the IIoT system into a machine device layer, an edge layer, and a cloud server layer based on the three-layer hierarchical architecture makes each layer of the IIoT system relatively independent, and a certain layer can be expanded independently without affecting other layers, improving the flexibility and scalability of the system. The hierarchical architecture improves the fault tolerance of the system. The failure of one layer will not cause the entire system to crash, enhancing the stability of the system. In the IIoT scenario, devices and systems may come from different manufacturers and have different technical standards. The hierarchical architecture can better integrate and manage these heterogeneous components.

[0023] S2. The edge computing server sends interference signals to the eavesdropping device at the interference power, so that the secure transmission rate of the communication channel is positive in the presence of eavesdropping, ensuring the security of the communication channel.

[0024] Currently, the security communication technologies commonly used in IIoT scenarios generally only target the protection of the network layer or the transport layer, such as the IPSec protocol, the TLS protocol, etc. The IPSec encryption protocol is a network layer security protocol that provides confidentiality, integrity, authentication, and anti-replay protection for IP communications. The IPsec protocol provides security for network communications by encrypting and authenticating IP packets at the network layer. It uses security associations (SAs) to determine the encryption and authentication parameters and can work in either transport mode or tunnel mode to ensure the confidentiality, integrity, and authenticity of data during transmission, while supporting anti-replay attacks and automatic key management. TLS (Transport Layer Security protocol) is a protocol used to provide encrypted communication and data integrity on computer networks. It protects data transmission by establishing an encrypted channel between the client and the server, ensuring the confidentiality and integrity of information and preventing data from being eavesdropped on or tampered with during transmission. TLS negotiates encryption algorithms, completes authentication, generates session keys through a handshake process, and uses these parameters to encrypt data throughout the session. However, the above solutions cannot protect the security of physical layer communications and cannot handle the situation where there are eavesdropping devices in this scenario. To address potential eavesdropping devices, this embodiment proposes a communication method for physical layer security in the IIoT scenario to resist eavesdropping devices and thus ensure secure physical layer communication.

[0025] S21. Detect whether there are potential eavesdropping devices around the edge computing server by equipping the edge computing server with devices such as cameras and synthetic aperture radars (SARs), and the location information of the eavesdropping devices can be obtained. Specifically, by equipping cameras and synthetic aperture radars, the surrounding environment can be sensed. After detecting potential eavesdropping devices, the edge computing server can take interference measures and emit interference signals to interfere with nearby eavesdroppers, thereby ensuring the security of data transmission. This detection mechanism is part of the edge computing server's secure communication model and aims to provide security guarantees for data transmission in the industrial Internet of Things environment.

[0026] S22. When it is detected that there are potential eavesdropping devices around the edge computing server, send interference signals to the eavesdropping devices at the interference power so that the secure transmission rate of the communication channel is always greater than zero. Keeping the secure transmission rate of the communication channel always greater than zero in the presence of eavesdropping, thereby ensuring the security of communication.

[0027] S221. The edge computing server captures the radio frequency energy sent by the base station and periodically transmits data. During a time period cycle, the edge computing server uses radio frequency energy harvesting technology to collect the energy sent by the base station through an energy harvesting module, and the collected energy is used to power the edge computing server.

[0028] Such as Figure 3As shown in the figure, it is the topology structure diagram of the Industrial Internet of Things (IIoT) system. The IIoT system includes a base station, machine devices, an edge computing server, and a potential eavesdropping device. The potential eavesdropping device only eavesdrops on the edge computing server. Other nodes, on the premise of ensuring their own transmission throughput requirements, send interference signals to help the edge computing server interfere with the potential eavesdropping device. At the same time, to improve the deployment flexibility, reduce the maintenance and operation costs, increase the device lifespan, and comply with the important green development concept in Industry 4.0, the edge computing server utilizes the Radio Frequency (RF) Energy Harvesting technology to collect the energy sent by the base station. The RF Energy Harvesting technology is a technology that harvests energy from radio frequency (RF) signals and can convert the wireless electromagnetic wave energy in the surrounding environment into electrical energy to provide power for the device. In the IIoT scenario, the base station is responsible for sending RF energy signals to the edge computing server. These signals propagate in the form of electromagnetic waves and carry energy that can be converted into electrical energy. The RF energy signals emitted by the base station are transmitted through the air to the edge computing server, which is equipped with an energy harvesting module. The energy harvesting module is used to capture the RF energy signals from the base station. The energy harvesting module includes an antenna, a rectifier circuit, an energy storage module, and a power management unit. The antenna is optimized for gain and directivity for specific frequency bands (such as 2.4 GHz, 5 GHz, or millimeter waves) to efficiently capture the base station signals and convert them into electrical energy. The rectifier circuit is used to convert the alternating current RF signal into direct current and boost the output voltage through a voltage multiplier (such as the Cockcroft-Walton circuit) to meet the power supply requirements of the edge computing server. The collected energy is stored through the energy storage module and the power management unit, and the energy distribution is optimized through the Maximum Power Point Tracking (MPPT) technology to ensure stable power supply. The collected electrical energy is stored in the battery for use when there is no external RF energy supply, and the stored energy is then used for the operation of the edge computing server, including local data processing, model training, communication with other devices, etc. This process is periodic, and the edge computing server needs to continuously collect energy to maintain its operation. Using this RF energy harvesting technology enables the edge computing server to be deployed in places without traditional power coverage, improving the deployment flexibility and scope. At the same time, it reduces the need for battery replacement, thereby reducing the maintenance cost and operation cost.

[0029] S222. In the next time period cycle, when an edge computing server sends data to the base station and the own throughput requirements of other edge computing servers do not exceed the maximum throughput that the edge computing server can achieve, the edge computing server sends interference signals to the potential eavesdropping device using the interference power, so that the secure transmission rate of the edge computing server sending data to the base station is greater than zero.

[0030] In this embodiment, the edge computing server captures the radio frequency energy sent by the base station and transmits data periodically. During the first time period cycle, the edge computing server first captures the radio frequency energy sent by the base station. During the next time period cycle, the edge computing server sends data to the base station. The edge computing server to respectively sends interference signals to interfere with the eavesdropping device. From to cycle, the edge computing server to uses its remaining energy to transmit data to the base station. When the edge computing server collects energy through radio frequency energy harvesting technology, considering fairness, each time period ( 、 、…、 ) is made equal because in this network environment, multiple edge computing servers share limited resources, including energy and communication bandwidth. If each edge computing server obtains different energy and communication resources in different time periods, it may lead to significantly better performance of some servers than others, resulting in unfair resource allocation. Therefore, considering fairness, assume = =…= .

[0031] The downlink channel gain coefficient of the radio frequency energy sent by the base station to the edge computing server is denoted as , and the uplink channel gain coefficient of the edge computing server sending data to the base station is denoted as . The eavesdropping channel coefficient from the edge computing server to the potential eavesdropping device is denoted as , and the interference channel coefficient from the edge computing server transmitting the interference signal to the potential eavesdropping device node is denoted as . In an actual wireless communication environment, the change of the channel is usually relatively slow, especially when the environment does not change violently. In this case, the channel gain remains relatively constant within one symbol period or several symbol periods, and changes only on a longer time scale. Therefore, we can assume that the uplink and downlink channels, the eavesdropping channel, and the interference channel are all quasi-static fading channels, that is, the channel gain (including amplitude and phase) is considered constant and does not change with time, that is 、 、 、 remain unchanged within the given working time T.

[0032] Assume that the total length of the operating cycle \(T = 1\), and the total length \(T\) is evenly divided into \(K\) time slots. So one time slot The cycle must satisfy the following inequality: (1) During the cycle, the base station broadcasts a radio frequency energy signal with transmit power . The edge computing server captures these signals through its energy harvesting module and converts them into electrical energy. The energy collected by the edge computing server can be expressed as : (2) where is the energy conversion efficiency, is the downlink channel gain coefficient for the base station to send radio frequency energy to the edge computing server, is the transmit power of the base station, is the time of one cycle.

[0033] During the cycle, the edge computing server sends information to the base station. At the same time, potential eavesdropping device nodes attempt to eavesdrop on the data of the edge computing server . On the premise of meeting its own throughput requirements, the edge computing server sends interference signals (noise data) to the eavesdropping devices with interference power to reduce the eavesdropping throughput of potential eavesdropping device nodes and improve the secure transmission rate of the edge computing server . The noise data sent by the interfering node can be effectively removed at the base station. The threshold of the eavesdropping throughput is expressed as : (3) where \(B\) is the channel bandwidth, is the white noise power at the base station, is the downlink channel gain coefficient for the base station to send radio frequency energy to the edge computing server, \(p\) 1 is the transmit power of the base station, is the interference power, .

[0034] Similarly, the data transmission throughput from the edge computing server to the base station can be expressed as (unit: bps): (4) Specifically, the secure transmission rate of the edge computing server is equal to the edge computing server The threshold of the data transmission throughput to the base station minus the eavesdropping throughput, the secure transmission rate can be expressed as: (5) Secure transmission rate is a key metric for measuring communication security. It is defined as the difference between the information reception rates of the legitimate receiver and the potential eavesdropper. If this secure transmission rate is positive, it means that the legitimate receiver can receive information at a rate higher than that of the eavesdropper, thus ensuring the security of information transmission. On the contrary, if the secure transmission rate is non-positive, it indicates that there are security risks in the channel. Therefore, to achieve secure communication, the secure transmission rate must be greater than zero.

[0035] During the period, the throughput from the edge computing server to the base station can be expressed as (unit: bps): (6) where is the transmit power of the base station, is the uplink channel gain coefficient for the edge computing server to send data to the base station.

[0036] For the edge computing server, its behavior is selfish, which means that it first needs to ensure that its own data transmission requirements are met, and then it will consider helping other nodes, that is: other edge computing servers are willing to help the edge computing server send interference signals only after meeting their own throughput requirements. The self-throughput requirements of the edge computing server satisfy the following constraint conditions: (7) That is: ; where represents the self-throughput requirement of the edge computing server, is the energy received by the edge computing server from the base station, is a period of time, B is the channel bandwidth, is the white noise power at the base station, is the uplink channel gain coefficient for the edge computing server to send data to the base station. To ensure does not exceed the maximum throughput that the edge computing server can achieve, the edge computing server uses interference power to send interference signals to potential eavesdropping device nodes, and these signals can be effectively removed at the base station.

[0037] For the eavesdropping devices during the communication process, the interference strategy in the secure communication design is adopted for interference, and the secure communication rate is used to reflect the security state of the current channel, ensuring that the channel can always be higher than the minimum security standard. Calculate the energy consumed for each task to ensure that the energy consumption is less than the maximum energy consumption requirement and will not cause damage to the device, enabling it to operate normally for a long time. This method can effectively deal with the interference of physical-layer eavesdropping devices on industrial systems. By sending interference signals to the eavesdropping devices through the edge computing server, the communication rate of the eavesdropping devices can be reduced, and the security status of the communication channel can be reflected through the secure transmission rate to ensure the channel security during the communication process.

[0038] S3. Combine the local training energy consumption, communication energy consumption, edge computing energy consumption, and interference signal energy consumption to calculate the total energy consumption of the edge computing server under different task offloading strategies. The total energy consumption of the edge computing server does not exceed the maximum energy consumption that the industrial Internet of Things system can bear.

[0039] S31. Calculate the local training energy consumption, communication energy consumption, edge computing energy consumption, and interference signal energy consumption respectively.

[0040] In each round of training, each edge computing server collects local data from the machine devices it covers and conducts local model training, and updates by generating a local model according to the model parameters. Calculate the local training energy consumption. The local training energy consumption is equal to the product of the local training power and the local training time. The local training energy consumption can be expressed as: (8) where represents the local training power; represents the local training time.

[0041] Calculate the communication energy consumption. The communication energy consumption includes the edge offloading communication energy consumption and the communication energy consumption for uploading to the cloud. The edge offloading communication energy consumption is equal to the product of the edge offloading communication power and the edge offloading communication time. The communication energy consumption for uploading to the cloud (uploading to the cloud server) is equal to the product of the communication power for uploading to the cloud and the communication time for uploading to the cloud. The edge offloading communication energy consumption and the communication energy consumption for uploading to the cloud are respectively expressed as: (9) (10) where: represents the edge offloading communication power; represents the edge offloading communication time; represents the communication power for uploading to the cloud; represents the communication time for uploading to the cloud.

[0042] Calculate the edge computing energy consumption. When the task is offloaded to the edge computing server, corresponding edge computing energy consumption will also be generated. The edge computing energy consumption is calculated according to the following formula: (11) Where, is the power consumption coefficient, The value of depends on the CPU chip structure of the edge computing server; is the computing rate when the edge computing server performs computing tasks; represents the computing time.

[0043] Calculate the interference energy consumption. When the j-th edge computing server is eavesdropped, the interference energy consumption generated by the remaining devices : (12) Where, represents the interference signal power transmitted by the edge computing server; represents the interference time.

[0044] S32. According to the local training energy consumption, communication energy consumption, edge computing energy consumption, and interference signal energy consumption, calculate the total energy consumption of the edge computing server under different task offloading strategies. The total energy consumption of the edge computing server does not exceed the maximum energy consumption that the industrial Internet of Things system can bear. The total energy consumption is expressed as: (13) Where, , indicates that the task is offloaded to the local of the machine device; , indicates that the task is offloaded to the edge computing server; , indicates that the task is offloaded to the cloud.

[0045] By comprehensively considering the training energy consumption, communication energy consumption, edge computing energy consumption, and interference energy consumption of sending interference signals to eavesdropping devices, and there are also different energy consumptions for different task offloading schemes. By establishing an energy consumption model, it is ensured that the energy consumption of the system will not exceed the maximum value that the entire industrial system can bear, ensuring the long-term stable operation of the device and also ensuring the stability of the entire industrial system.

[0046] S4. Construct an offloading decision model based on federated learning. The offloading decision model aims to minimize the average computing task delay, and based on the offloading decision model, output the optimal offloading decision according to the channel state and device resources.

[0047] Due to the limitations of computing resources and energy on local devices, it is impossible to complete huge computing tasks in a short time. Therefore, local devices need to offload some computing tasks to edge computing servers for processing. Edge computing servers have more computing, storage, and communication resources. An edge computing server is a computing platform deployed at the edge of the network, aiming to sink cloud computing capabilities to nodes closer to users or data sources to reduce latency, improve efficiency, and optimize resource utilization. The present invention aims to minimize the average computing task latency, constructs an offloading decision model based on federated learning, and outputs an optimal offloading decision based on the offloading decision model according to the channel state and device resources under the premise of satisfying the channel security constraint condition and the energy consumption constraint condition.

[0048] S41. Calculate the processing time of task offloading for different offloading decisions and the total time consumption of the machine device participating in federated learning. Aiming to minimize the average computing task latency, construct an offloading decision model based on federated learning.

[0049] S411. Calculate the processing time of task offloading for different offloading decisions respectively.

[0050] When task offloading is performed locally on the machine device, the processing time of task offloading is the computing time of the task machine device, expressed as: = ; Among them, represents the computing time of the task machine device.

[0051] When task offloading is performed on the edge computing server, the processing time is the sum of the communication time between the machine device and the edge computing server, the computing time of the task on the edge computing server, and the transmission time when returning the result data, expressed as: (14) Among them, represents the communication time between the machine device and the edge computing server; represents the computing time of the task on the edge computing server; represents the transmission time when returning the result data; When task offloading is performed in the cloud, the processing time is the sum of the communication time between the machine device and the edge computing server, the transmission time required for data to be transmitted from the edge computing server to the cloud server or the transmission time for the returned data to be transmitted back from the cloud server to the edge computing server after computing, and the transmission latency for returning the result data from the edge computing server to the machine device, expressed as: (15) Among them, represents the communication time between the machine device and the edge computing server; It represents the transmission time required for data to be transmitted from the edge computing server to the cloud server, or the transmission time for the returned data to be transmitted back from the cloud server to the edge computing server after the calculation is completed; It represents the transmission delay for returning the result data from the edge computing server to the machine device.

[0052] Therefore, the processing time of task offloading can be expressed as: (16) Among them, represents the offloading decision of the machine device at time . If , it means that the task is offloaded to the local of the machine device; if , the task is offloaded to the edge computing server; if , the task is offloaded to the cloud.

[0053] S412. Calculate the total time consumption of the machine device participating in federated learning, and obtain the total time consumption of all tasks based on the processing time of task offloading and the total time consumption of the machine device participating in federated learning.

[0054] In the optimization process of federated learning, the total time of each round includes the local training time of the machine device, the time for the machine device to transmit model updates to the cloud, the time for the cloud to aggregate model updates, and the time for the machine device to receive the global model. Specifically, the machine device The total time consumption of participating in federated learning is: (17) Among them, represents the local training time of device ; represents the time for device to transmit model updates to the cloud; represents the time for the cloud to aggregate model updates (the same for all participating devices); represents the time for device to receive the global model.

[0055] Obtain the total time consumption of all tasks based on the processing time of task offloading and the total time consumption of the machine device participating in federated learning. The total time consumption refers to the total time required to complete all tasks, including the processing time, data transmission time, waiting time, etc. of a single task. The total time consumption of all tasks is equal to the processing time of task offloading plus the total time consumption of the machine device participating in federated learning, and can be expressed as: (18) S413. Set the model optimization objective function with the goal of minimizing the average computing task delay based on the total time consumption of all tasks, and construct an offloading decision model based on federated learning.

[0056] The goal of the industrial Internet of Things system is to minimize the time of the average computing task while considering the time overhead in the federated learning process. Each device , set the model optimization objective function with the goal of minimizing the average computing task delay. Minimizing the average computing task delay means reducing the average time from the start to the completion of a single task. By minimizing the average computing task delay, the total time required to complete all tasks can be effectively reduced. That is, if the average delay of each task is reduced, the total time required to complete all tasks will also be reduced accordingly. Reducing the average time required for each task from start to completion includes reducing the task processing time: the time required for the task to execute on the local device, edge server, or cloud server; the data transmission time: the time required for data to be transmitted between devices, including uploading to the edge server or cloud server, and downloading the processing results from these servers; the waiting time: the time the task waits in the queue to be processed, which may be due to busy computing resources or network congestion. The model optimization objective function can be expressed as: (19) Among them, ξ represents the time period. As ξ approaches infinity, it can ensure that the optimization objective reflects the average behavior of the system in the long-term operation rather than short-term fluctuations; Represents the reciprocal of the number of device sets, used to calculate the average task delay; Represents the number of device sets in the industrial Internet of Things system, Represents the machine device At time The offloading decision of, if , it means the task is offloaded to the local of the machine device; if , the task is offloaded to the edge computing server; if , the task is offloaded to the cloud.

[0057] The present invention jointly considers the security of the communication channel of the edge computing server and the energy consumption of task processing, specifically including: each machine device at time The total energy consumption does not exceed the maximum energy consumption that the edge computing server can bear. To ensure communication security, the guaranteed secure transmission rate should always be positive. The edge computing server will only assist the eavesdropped device in interference when it meets its own throughput requirements. The value of the edge computing server meeting its own throughput cannot exceed the maximum throughput that the edge computing server can achieve. The bandwidth and computing resources allocated to the machine device cannot exceed the allocable bandwidth and computing resources. Under the conditions of secure communication and energy consumption constraints of the edge computing server, minimize the time of the average computing task and improve the overall efficiency of the system.

[0058] S42. Use federated learning to distributively train the offloading decision model, receive the model updates transmitted from the edge computing server, and perform weighted averaging on them to obtain a new global model, and then return it to the edge computing server; As Figure 4 shown, it is a schematic diagram of using federated learning to distributively train the offloading decision model. Using federated learning to distributively train the offloading decision model, in the current IIoT scenario, the system is divided into three-layer architecture, namely the machine device layer, the edge layer, and the cloud service layer. The system uses federated learning to distributively train the offloading decision model. The machine device no longer needs to upload the original data to the cloud server, thus ensuring data security. The cloud server only needs to perform weighted averaging on the model updates transmitted from the edge computing server to generate the global model. This distributed training reduces the data transmission in the network, reduces the communication cost, and reduces the computing burden on the cloud server. At the same time, the transmitted data changes from the original data to model updates, and the data volume also becomes smaller, thus optimizing the use of network resources and reducing energy consumption. At the same time, after each round of federated learning model aggregation, the global model update is obtained, so as to more effectively allocate computing and storage resources.

[0059] S421. For each edge computing server, the edge computing server receives the data of the machine devices within its coverage area for local training and generates updates of the local model parameters.

[0060] The edge computing server receives data from local machine devices within its coverage area, including the amount of input parameter data required for a computing task, the amount of computing power required to execute the computing task, and the amount of result data that needs to be returned to the industrial device after the computing task is completed. The amount of input parameter data required for a computing task, that is, the total amount of data that the device needs to obtain from the outside when executing the task. This data is real-time monitoring data from sensors or intermediate results obtained from other devices, used to support the execution of the task. The amount of computing power required to execute the computing task reflects the complexity of the task, helping the edge computing server evaluate the computing resources required for the task, so as to reasonably allocate resources and determine the execution location of the task; the amount of result data that needs to be returned to the industrial device after the computing task is completed. These returned data usually contain the final result or decision information of the task, used to guide the subsequent operations of the device. Through these three types of data, the edge computing server can comprehensively understand the requirements of each computing task, so as to achieve efficient and secure scheduling decisions. And use these data to train the offloading decision model and update the model parameters. In each training process, the edge computing server will optimize according to the objective function and constraint conditions to improve the accuracy of the model S422. The edge computing server sends the update of the local model parameters to the cloud server. The cloud server performs weighted aggregation on all model updates to obtain the global model parameter update. The cloud server sends the global model update back to each edge computing server for optimizing the next round of offloading decisions.

[0061] The edge computing server sends the local model update to the cloud server. The cloud server performs weighted averaging on these updates to obtain the global model: (20) Where: represents the global model parameters at time step (a time step is a cycle of global model updates) ; represents the local model parameters of the i th client at time step ; represents the i th client's sample quantity, N represents the total sample quantity of all clients; represents the weight of the i th client, indicating its contribution degree to the global model. Using federated learning to distributively train the offloading decision model, using federated learning not only protects the security of the data, but also continuously updates the device model, and at the same time considers the channel state, security, and energy consumption, so as to optimize its next offloading decision, making the entire system operate safely, stably, and efficiently.

[0062] S43. When task offloading occurs, based on the edge computing network state of the current industrial Internet of Things system and the machine device model parameters, an optimal offloading decision is output based on the offloading decision model.

[0063] When task offloading occurs, with the goal of minimizing the average computing task latency based on the offloading decision model, an optimal offloading decision is output according to the current edge computing network state of the industrial Internet of Things and the machine device model parameters. The network state information includes the link transmission rate between the machine device and the edge computing server, which reflects the current network transmission efficiency.

[0064] The machine device model parameters include machine device information, edge computing server information, and eavesdropping device information. The machine device information includes: Computing rate: The computing rate when the machine device executes computing tasks locally, which is used to evaluate the efficiency of local task execution. Signal transmission power: The signal transmission power when the machine device communicates with the edge computing server, which affects the reliability and rate of data transmission. Channel gain: The channel gain between the machine device and the edge computing server, which reflects the quality and transmission efficiency of the communication link.

[0065] The edge computing server information includes: Computing rate: The computing rate when the edge computing server executes computing tasks, which is used to evaluate the processing capacity of the edge computing server. Interference power: The power of the edge computing server to emit electromagnetic waves to interfere with the eavesdropping device, which is used to ensure the security of data transmission. Receiving power: The power of the edge computing server to receive the offloaded data of the machine device, which affects the efficiency of data reception. Sending power: The sending power when the edge computing server returns the computing result to the machine device, which affects the efficiency of data transmission. Energy consumption limit: The energy consumption limit allocated by the edge computing server to a single task, which is used to control energy consumption. Number of sub-channels: The maximum number of sub-channels that the edge computing server can allocate, which is used to optimize data transmission. Number of computing resources: All the computing resources owned by the edge computing server, which is used to reasonably allocate computing tasks.

[0066] Eavesdropping device information: Presence or absence of the eavesdropping device: Detect the presence of the eavesdropping device through sensors (such as cameras or synthetic aperture radars). Location information of the eavesdropping device: If the eavesdropping device is detected, obtain its location information in order to take corresponding interference measures.

[0067] The specifically output offloading decisions include the task offloading location and resource allocation. Among them, the task offloading location includes: local execution, which decides whether to execute the computing task locally on the machine device. If the computational volume of the task is small and the local computing resources are sufficient, local execution can be selected. Offloading to the edge computing server, which decides whether to offload the task to the edge computing server for execution. If the computing resources of the edge computing server are sufficient and the communication link quality is good, offloading to the edge computing server can be selected. Offloading to the cloud server, which decides whether to offload the task to the cloud server for execution. If the computational volume of the task is large and the resources of the edge computing server are insufficient, offloading to the cloud server can be selected. Resource allocation includes: sub-channel allocation, which decides how many sub-channels to allocate to each task to optimize the data transmission efficiency. Computational resource allocation, which decides how much computational resource to allocate to each task to ensure that the task can be executed efficiently. Energy consumption control, which ensures that the energy consumption of the task does not exceed the preset upper limit to avoid excessive consumption of the energy of the edge computing server.

[0068] Combining task offloading with the federated learning optimization process, comprehensively considering the data security, channel security, task energy consumption, task processing latency, and channel state of industrial devices, an offloading decision is output. Each industrial device calculates the task energy consumption and latency under different offloading decisions based on the current channel state, local computing resources, task load and other information, combined with the model obtained from federated learning training. The device selects the optimal offloading decision on the premise of meeting the constraints such as data security, channel security, energy consumption, and latency. When offloading tasks, combining the channel state and device resources, an optimal offloading decision is output to achieve the multi-objective balance of communication, computing, and energy consumption, and improve the efficiency of task processing in the industrial Internet of Things environment. The edge computing server feeds back the task processing results to the local device, and the device updates and adjusts the local model according to the feedback information. The system regularly updates its own channel state, computing resources and other information, and re-performs offloading decisions to adapt to the dynamically changing industrial network environment. As the tasks continue, the models of industrial devices are also constantly updated, while avoiding a large amount of raw data from being transmitted from edge devices to the cloud, reducing the risk of being eavesdropped or leaked during data transmission. At the same time, considering the combination of channel security, energy consumption, and channel state, the offloading decisions of tasks are also continuously optimized, thus ensuring the safe, stable, and efficient operation of the system and achieving the optimization of performance and decision-making.

[0069] In summary, this embodiment proposes an industrial Internet of Things edge computing security communication method based on federated learning. By combining federated learning and edge computing technologies, it can significantly enhance data privacy protection and communication security. By adopting a distributed federated learning mechanism, the privacy risks of centralized data storage and transmission are avoided, and the edge computing server emits interference signals to counter eavesdropping devices, effectively ensuring the security of the communication channel. By optimizing the energy consumption and task offloading strategies, the overall performance and reliability of the system are improved. The present invention has high practicality and broad application prospects.

[0070] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. A secure communication method for industrial Internet of Things edge computing based on federated learning, characterized in that: The following steps are involved: S1. Build a unified collaborative framework for the industrial Internet of Things system, dividing the industrial Internet of Things system into the machine equipment layer, edge layer and cloud server layer; The machine device layer includes several machine devices, which are used to perform local computing tasks and communicate with the edge computing server to pass the model parameters to the edge layer; The edge layer includes several edge computing servers, which are used to collect model parameters of machines and devices in the coverage area and use the model parameters to update the local model. The edge computing servers work together through federated learning to participate in training the global model. The cloud server layer is used to coordinate the federated learning process of the entire industrial IoT system and manage the aggregation and distribution of model updates between edge computing servers; S2, sending an interference signal to the eavesdropping device with interference power through the edge computing server, so that the secure transmission rate of the communication channel is positive in the presence of eavesdropping; S3. Combine communication energy consumption, local training energy consumption, computing energy consumption and interference signal energy consumption to calculate the total energy consumption of the edge computing server under different task offloading strategies. The total energy consumption of the edge computing server does not exceed the maximum energy consumption that the industrial Internet of Things system can bear. S4. Construct an offloading decision model based on federated learning. The offloading decision model aims to minimize the average computing task delay. The offloading decision model outputs the optimal offloading decision based on the channel status and device resources.

2. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 1 is characterized in that: The step S2 comprises: S21, using cameras and synthetic aperture radars installed on edge computing servers to detect whether there are potential eavesdropping devices around the edge computing servers; S22. When a potential eavesdropping device is detected around the edge computing server, an interference signal is sent to the eavesdropping device with an interference power so that the secure transmission rate of the communication channel is always greater than zero.

3. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 2 is characterized in that: The step S22 comprises: The edge computing server captures the radio frequency energy sent by the base station and transmits data periodically. During a period of time, the edge computing server uses radio frequency energy collection technology to collect the energy sent by the base station through the energy collection module. The collected energy is used to power the edge computing server. In the next time period, when an edge computing server sends data to the base station, and the throughput requirements of other edge computing servers do not exceed the maximum throughput that the edge computing servers can achieve, the edge computing servers use interference power to send interference signals to potential eavesdropping devices, so that the secure transmission rate of the edge computing servers sending data to the base station is greater than zero.

4. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 3 is characterized in that: The edge computing server uses radio frequency energy collection technology to collect energy sent by the base station through an energy collection module, where the collected energy is expressed as : ； in, is the energy conversion efficiency, The downlink channel gain coefficient for the base station to send RF energy to the edge computing server. is the transmission power of the base station, For one cycle time.

5. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 3 is characterized in that: The secure transmission rate of the edge computing server sending data to the base station is equal to the data transmission throughput from the edge computing server to the base station minus the threshold of the eavesdropping throughput. The threshold of the eavesdropping throughput is expressed as : ； in, is a cycle time, B is the channel bandwidth, is the white noise power at the base station, is the downlink channel gain coefficient for the base station to send RF energy to the edge computing server, p1 is the transmission power of the base station, is the interference power, .

6. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 4 is characterized in that: The throughput of the other edge computing servers is required to meet the following constraints: ； in, Indicates the throughput requirements of other edge computing servers. The energy sent by the base station is collected by the edge computing server. is a cycle time, B is the channel bandwidth, is the white noise power at the base station, The uplink channel gain coefficient for the edge computing server to send data to the base station.

7. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 1, characterized in that: The step S3 comprises: S31, respectively calculating local training energy consumption, communication energy consumption, edge computing energy consumption and interference signal energy consumption, where the communication energy consumption includes edge offloading communication energy consumption and uploading cloud communication energy consumption; S32. According to the local training energy consumption, communication energy consumption, edge computing energy consumption and interference signal energy consumption, the total energy consumption of the edge computing server under different task offloading strategies is calculated. The total energy consumption of the edge computing server does not exceed the maximum energy consumption that the industrial Internet of Things system can bear. The total energy consumption of the edge computing server is expressed as: ； in, , indicating that the task is offloaded to the local machine device; , indicating that the task is offloaded to the edge computing server; , indicating that the task is offloaded to the cloud; represents the local training energy consumption, represents the interference energy consumption, represents the edge offloading communication energy consumption, represents the edge computing energy consumption, Indicates the energy consumption of uploading to the cloud.

8. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 1, characterized in that: The step S4 comprises: S41. Calculate the processing time of different offloading decision tasks and the total time consumption of machine equipment participating in federated learning, and build an offloading decision model based on federated learning with the goal of minimizing the average computing task delay. S42, using federated learning to distribute the training of the offloading decision model, receiving the model update from the edge computing server and performing weighted average on the model update, obtaining a new global model and returning it to the edge computing server; S43: When the task is offloaded, an optimal offloading decision is output based on the offloading decision model according to the channel status and device resources.

9. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 8, characterized in that: The step S41 comprises: S411, respectively calculating the task unloading processing time of different unloading decisions: When the task is offloaded locally on the machine device, the processing time of the task offload is the calculation time of the task machine device; When offloading tasks to an edge computing server, the processing time for offloading tasks is the sum of the communication time between the machine device and the edge computing server, the computing time of the task on the edge computing server, and the transmission time when the result data is returned; When offloading tasks in the cloud, the processing time of task offloading is the sum of the communication time between the machine device and the edge computing server, the transmission time required for data to be transmitted from the edge computing server to the cloud server or the transmission time for the return data after the calculation is completed to be transmitted from the cloud server to the edge computing server, and the transmission delay of the result data from the edge computing server back to the machine device; S412, calculating the total time consumption of the machine equipment participating in the federated learning, and obtaining the total time consumption of all tasks according to the processing time of the task offloading and the total time consumption of the machine equipment participating in the federated learning; S413, setting a model optimization objective function according to the total time consumption of all tasks and minimizing the average computing task delay as the goal, and constructing an offloading decision model based on federated learning; The model optimization objective function is expressed as: ； Where ξ represents the time period, Indicates the reciprocal of the number of device sets, used to calculate the average task latency; Indicates the number of devices in the industrial Internet of Things system. Indicates machine equipment At the moment The uninstall decision if , it means that the task is offloaded to the local machine device; if , then the task is offloaded to the edge computing server; if , the task is offloaded to the cloud.

10. The method for secure communication of industrial Internet of Things edge computing based on federated learning according to claim 8, characterized in that: The step S42 comprises: S421. For each edge computing server, the edge computing server receives data from machine devices in its coverage area for local training and generates updates to local model parameters; S422. The edge computing server sends the update of the local model parameters to the cloud server. The cloud server performs weighted aggregation on all model updates to obtain a global model parameter update. The cloud server sends the global model update back to each edge computing server.

Images