Intelligent gateway control method based on edge computing

Abstract

The invention discloses an intelligent gateway control method based on edge computing, and the method comprises the steps: S1, collecting the operation state data of network nodes through a distributed sensor, carrying out the filtering, denoising and normalization processing, sorting according to a timestamp, and generating the preprocessing data containing network flow, node load and response time delay; s2, calculating the resource demand quantity of each node based on the preprocessed data, executing a dynamic resource allocation algorithm in combination with the available resource state of the node, and generating a resource allocation result containing a calculation resource quota and a task allocation relationship; s3, inputting a resource allocation result into a gateway control model, adjusting model parameters based on a real-time network state, and generating a control strategy; and S4, encoding the control strategy into an instruction sequence, encrypting the instruction sequence through a digital signature, transmitting the encrypted instruction sequence to an execution unit through a secure channel, and executing a control instruction after verifying integrity and signature validity. Therefore, dynamic resource allocation, real-time network state adjustment and security instruction transmission are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gateway control, and in particular to an intelligent gateway control method based on edge computing. BACKGROUND

[0002] With the rapid development of Internet of Things technology, edge computing gateway is widely used in intelligent manufacturing, smart city and other fields. The existing edge gateway control technology mainly adopts static configuration method to manage network resources, and allocates computing tasks and network bandwidth through preset fixed parameters. This control method is difficult to adapt to the dynamic changes of network load, and when the network traffic surges or the node fails, the system cannot adjust the resource allocation strategy in time, resulting in overload of some nodes and idle resources of other nodes. The traditional gateway control method relies on manual experience to set control rules, lacks learning ability of historical operation data, and cannot adaptively optimize the control strategy according to the changes of network state. In addition, the existing technology lacks effective security protection mechanism in the process of control instruction transmission, and is vulnerable to malicious attacks and data tampering.

[0003] The current edge gateway system has obvious shortcomings in data acquisition and processing. Most systems adopt centralized data acquisition architecture, and single point failure will cause the whole monitoring system to be paralyzed. The raw data collected is directly used for decision without effective preprocessing, and noise and abnormal values seriously affect the control accuracy. In terms of resource allocation algorithm, the traditional method only considers a single optimization target, and cannot balance the contradictory relationship between processing delay and energy consumption. The control model mostly adopts simple linear model or rule-based decision tree, which is difficult to capture the complex nonlinear relationship between network state and control effect.

[0004] Therefore, there is an urgent need for an intelligent gateway control method based on edge computing. SUMMARY

[0005] The present application provides an intelligent gateway control method based on edge computing to solve the above problems existing in the prior art.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: An intelligent gateway control method based on edge computing, comprising: S1: collecting network node running state data through distributed sensors, sorting by time stamp after filtering, denoising and normalizing, and generating preprocessed data containing network traffic, node load and response time delay; S2: calculating the resource demand of each node based on the preprocessed data, executing a dynamic resource allocation algorithm combined with the available resource state of the node, and generating a resource allocation result containing computing resource quota and task allocation relationship; S3: input the resource allocation result into the gateway control model, adjust the model parameters based on the real-time network state, and generate a control strategy; S4: encode the control strategy into an instruction sequence, encrypt it through digital signature, and transmit it to the execution unit through a secure channel after verifying the integrity and signature validity, and then execute the control instruction.

[0007] Wherein, S2 includes: S21: weight and sum the processing delay parameters and energy consumption parameters according to the preset weight coefficients to construct the objective function of the multi-objective optimization model; S22: input the node CPU utilization and task computing complexity into a nonlinear mapping function to calculate the computing resource demand of each node and generate an initial resource allocation scheme; S23: when the network load change is detected to exceed the preset threshold, start the genetic algorithm to iteratively optimize the resource allocation scheme, and output the updated resource allocation result.

[0008] Wherein, S3 includes: S31: construct a deep network model containing convolutional layers and fully connected layers, and input the resource allocation result and real-time network state vector as input layer data; S32: store the historical running data and corresponding control effect data into the experience buffer, and pre-train the model by randomly sampling data; S33: calculate the reward value according to the weighted coefficients of network throughput and response time, and update the network weight parameters based on the error between the reward value and the predicted value through the back propagation algorithm.

[0009] Wherein, the digital signature processing in S4 includes: S41: generate a private key and public key pair using elliptic curve algorithm, and use the private key to perform digital signature operation on the hash value of the instruction sequence; S42: encapsulate the digital signature data and instruction sequence data into data blocks, and write the data blocks into the distributed ledger maintained by the edge node.

[0010] Wherein, the deployment of distributed sensors in S1 includes: S51: calculate the path length between nodes according to the network topology structure, and select the sink node with smaller path length sum as the sensor deployment location; S52: group the network nodes based on the communication delay constraint and bandwidth capacity limit conditions through clustering algorithm, and determine the number of sensor deployments; S53: deploy the sensor unit at the selected sink node location, and access the existing network architecture through standard network protocol for data collection.

[0011] Wherein, it also includes: S61: Collect gateway CPU utilization, memory occupancy and network throughput parameters in real time; S62: Input the collected parameters into the preset evaluation function to calculate the performance evaluation index value, compare the evaluation index value with the historical data to generate a performance evaluation report; S63: Extract the performance deviation value in the evaluation report, and adjust the mutation probability and crossover probability parameters of the genetic algorithm in S2 according to the size of the deviation value.

[0012] Wherein, the execution trigger linkage mechanism of the control strategy includes: S71: Adjust the processor working frequency and power consumption management state of the target network node according to the control instruction; S72: The node state change data is transmitted to the distributed sensor through the network management protocol; S73: After receiving the state change data, the distributed sensor triggers the resource allocation algorithm to re-execute the calculation process and generates an updated resource allocation result.

[0013] Wherein, the verification method of the execution unit for the instruction includes: S81: Extract the digital signature data of the corresponding instruction block from the distributed ledger; S82: Use the pre-stored elliptic curve public key to perform verification operation on the extracted digital signature data to obtain the verification result; S83: Calculate the hash value of the received instruction based on the verification result, compare the calculated hash value with the ledger record, and output the integrity verification result.

[0014] Wherein, the parallel processing method of the gateway control model in S3 includes: S91: Distribute the genetic algorithm resource allocation operation and control logic operation to multiple general processing cores for parallel execution; S92: Distribute the deep network inference operation and elliptic curve encryption and decryption operation to the special neural processing unit for execution; S93: Distribute subsequent operation tasks to the corresponding processing units according to the operation type and priority through the hardware task scheduler to achieve load balancing.

[0015] Wherein, the construction method of the historical running data in S32 includes: S101: Establish a mapping relationship between the resource allocation result and the corresponding network delay and throughput parameters, and store it as relational data; S102: Record the time sequence change process of gateway CPU utilization and memory occupancy after different control actions are executed, and generate time sequence data; S103: Collect the parameter correction process of the control strategy under network congestion state and node fault state, and generate correction parameter data.

[0016] Compared with the prior art, the present application has the following advantages: An intelligent gateway control method based on edge computing, comprising: S1: collecting network node running state data through a distributed sensor, sorting the data after filtering, denoising and normalization processing according to time stamp, and generating preprocessed data containing network traffic, node load and response time delay; S2: calculating the resource demand of each node based on the preprocessed data, executing a dynamic resource allocation algorithm combined with the available resource state of the node, and generating a resource allocation result containing the calculation resource quota and the task allocation relationship; S3: inputting the resource allocation result into a gateway control model, adjusting the model parameters based on the real-time network state, and generating a control strategy; S4: encoding the control strategy into an instruction sequence, encrypting it through digital signature, and transmitting it to the execution unit through a secure channel after verifying the integrity and signature validity, and executing the control instruction. The present application collects network node running state data through a distributed sensor, calculates the resource demand of each node after preprocessing, executes a dynamic resource allocation algorithm to generate a resource allocation result. The resource allocation result is input into a gateway control model, the model parameters are adjusted to generate a control strategy, and the control strategy is encoded into an instruction sequence, which is encrypted through digital signature and transmitted to the execution unit through a secure channel. After verifying the integrity and signature validity, the control instruction is executed, thereby realizing dynamic resource allocation, real-time network state adjustment and secure instruction transmission.

[0017] Other features and advantages of the present application will be described in the following description, and some will become apparent from the description, or will be understood from the practice of the present application.

[0018] The technical solutions of the present application will be further described in detail below with the aid of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with embodiments of the present application, and do not constitute a limitation on the present application. In the drawings: Figure 1 A flowchart of an intelligent gateway control method based on edge computing in an embodiment of the present application; Figure 2 A flowchart of generating a resource allocation result containing a calculation resource quota and a task allocation relationship in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The preferred embodiments of the present application will be described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to explain and illustrate the present application, and do not limit the present application.

[0021] The embodiment of the present application provides an intelligent gateway control method based on edge computing as shown in Figure 1 The embodiment of the present application provides an intelligent gateway control method based on edge computing as shown in S1: Collect network node running state data through distributed sensors, sort by timestamp after filtering, denoising and normalization processing, generate preprocessed data containing network traffic, node load and response delay; S2: Calculate the resource demand of each node based on the preprocessed data, execute the dynamic resource allocation algorithm combined with the available resource state of the node, generate the resource allocation result containing the calculation resource quota and task allocation relationship; S3: Input the resource allocation result into the gateway control model, adjust the model parameters based on the real-time network state, generate the control strategy; S4: Encode the control strategy into instruction sequence, encrypt through digital signature after verification of integrity and signature validity, transmit to the execution unit through the secure channel, and execute the control instruction after verification of integrity and signature validity.

[0022] The working principle and beneficial effects of the above technical solution are as follows: Step S1, deploy distributed sensor units in the edge network environment, each sensor unit monitors the network node running condition in real time. The technical personnel install temperature sensors, current sensors and flow sensors in the key positions of each network node. The original signal collected by the sensor is first filtered by the analog filter circuit to eliminate noise interference, and then converted into a digital signal by the analog-to-digital converter. The digital signal enters the normalization processing unit to map the data of different dimensions to the standard interval. The processed data is stored in the cache queue in timestamp order to form a data set containing network traffic records, node load values and response delay measurements.

[0023] Step S2, after the gateway system receives the preprocessed data, the resource scheduling module starts to calculate the resource demand of each node. The system reads the current CPU occupation, memory usage state and storage capacity information of each node, combines the calculation complexity requirement of the to-be-processed task, and generates an initial resource allocation scheme through a load balancing algorithm. The dynamic adjustment mechanism continuously monitors the network load change, and when the load fluctuation exceeds the set range, triggers the resource reallocation process. Finally, the complete scheme containing the calculation resource quota allocation and task scheduling relationship is output.

[0024] Step S3, the gateway control module constructs an intelligent decision-making model, taking the resource allocation scheme as input data. The control model contains multiple neural network structures, which can dynamically adjust internal parameters according to the current network running state. The model predicts the optimal control strategy by analyzing historical running data and current network state. The system compares the prediction result with the actual running effect, continuously optimizes the model parameters, and finally generates a control strategy that adapts to the current network environment.

[0025] Step S4, the control policy encoding module converts the policy instruction into a standard instruction sequence format. The instruction sequence is encrypted by a digital signature algorithm to ensure the security and integrity of the instruction during transmission. The encrypted instruction is transmitted to each execution unit through a dedicated secure channel. After receiving the instruction, the execution unit first verifies the validity of the digital signature and the integrity of the instruction data, and executes the corresponding control operation according to the instruction after verification.

[0026] In another embodiment, as shown in Figure 2 S2 includes: S21: weight the processing delay parameter and the energy consumption parameter by a preset weight coefficient to obtain a weighted sum, and construct a target function of a multi-objective optimization model; S22: input the node CPU utilization and the task calculation complexity into a nonlinear mapping function to calculate the calculation resource demand of each node, and generate an initial resource allocation scheme; S23: when detecting that the network load change exceeds a preset threshold, starting a genetic algorithm to iteratively optimize the resource allocation scheme, and outputting an updated resource allocation result.

[0027] The working principle and beneficial effects of the above technical solution are as follows: Step S21, the system administrator pre-sets the values of the processing delay weight parameter and the energy consumption weight parameter. The multi-objective optimization module multiplies the two parameters by the corresponding weight coefficients, and then adds the results to form the target function. The target function serves as the evaluation standard for subsequent resource allocation optimization, ensuring that resource allocation considers both processing efficiency and energy consumption control.

[0028] Step S22, the resource demand calculation unit obtains the CPU utilization data of each node and the calculation complexity parameter of the to-be-executed task from the system monitoring module. The two groups of data are input into a nonlinear mapping function for processing. The mapping function uses a polynomial transformation method to convert the input parameters into resource demand quantity values. According to the calculation result, the system allocates corresponding calculation resources to each node to generate an initial allocation scheme.

[0029] Step S23, the network load monitor continuously detects the changes of network traffic and node load. When the monitor finds that the load change exceeds the pre-set threshold value, the genetic algorithm optimizer is automatically started. The optimizer takes the current resource allocation scheme as the initial population, and iteratively evolves through selection, crossover and mutation operations. After multiple generations of evolution, the optimized resource allocation result is output.

[0030] In another embodiment, S3 includes: S31: construct a deep network model containing convolutional layers and fully connected layers, and input the resource allocation result and the real-time network state vector as input layer data; S32: store the historical running data and the corresponding control effect data into an experience buffer, and pre-train the model by randomly sampling data; S33: calculate the reward value according to the network throughput and the response time by a weighting coefficient, and update the network weight parameters by a back propagation algorithm based on the error between the reward value and the predicted value.

[0031] The working principle and beneficial effects of the above technical solution are as follows: in step S31, a network architecture engineer designs a deep network model including a convolution layer and a full connection layer. The convolution layer is responsible for extracting feature patterns in resource allocation data, and the full connection layer comprehensively analyzes the extracted features. The model input layer receives two types of data, resource allocation matrix and real-time network state vector, and outputs control decisions after processing by multiple neurons.

[0032] In step S32, a historical data management module stores past network running records and corresponding control effect data in an experience buffer. In the model training stage, data samples are randomly selected from the buffer, and the sample data is input into the neural network for pre-training. The pre-training process adjusts the network connection weight, so that the model can learn the effective control mode in the historical experience.

[0033] In step S33, a reward value calculation module calculates the network throughput and the response time according to a preset weighting coefficient, and obtains the reward value of the system running. The model compares the reward value with the predicted value, calculates the error between them, and adjusts the weight parameters in the network according to the error value by a back propagation algorithm, so that the model prediction result gradually approaches the actual running effect.

[0034] In another embodiment, the digital signature processing in S4 includes: S41: generate a private key and public key pair using an elliptic curve algorithm, and perform a digital signature operation on the hash value of the instruction sequence using the private key; S42: encapsulate the digital signature data and the instruction sequence data into a data block, and write the data block into a distributed ledger maintained by the edge node.

[0035] The working principle and beneficial effects of the above technical solution are as follows: in step S41, a key generation module creates a private key and public key pair using an elliptic curve cryptography algorithm. The private key is generated by a random number generator, and the public key is derived from the private key by elliptic curve operation. The instruction sequence first calculates the digest value by a hash algorithm, and then performs a digital signature operation on the digest value using the private key to generate signature data.

[0036] Step S42, the data encapsulation module combines the digital signature data and the original instruction sequence into a data block structure. Each data block contains three basic components: timestamp, data content and signature information. The encapsulated data block is written into the distributed ledger maintained by each edge node through the network interface, realizing the secure storage and synchronous update of the instruction data.

[0037] In another embodiment, the deployment of distributed sensors in S1 includes: S51: Calculate the path length between nodes according to the network topology structure, and select the sink node with smaller path length sum as the sensor deployment position; S52: Based on the communication delay constraint and bandwidth capacity limit condition, the network nodes are grouped by clustering algorithm, and the number of sensor deployment is determined; S53: Deploy the sensor unit at the selected sink node position, and access the existing network architecture through the standard network protocol for data collection.

[0038] The working principle and beneficial effects of the above technical solution are: step S51, the network topology analysis module scans the entire network structure and calculates the shortest path length between nodes. The analysis algorithm traverses all possible node connection paths and counts the path length sum of each node to other nodes. The system selects nodes with smaller path length sum as sink points, which are convenient for data collection and transmission.

[0039] Step S52, the node grouping module executes the clustering algorithm according to the communication delay limit and bandwidth capacity constraint condition. The algorithm groups the nodes in the network according to the communication distance and connection bandwidth, and the nodes in each group have similar network characteristics. According to the grouping result, the number of sensors to be deployed is determined to ensure that at least one monitoring sensor is configured in each group.

[0040] Step S53, the equipment installer installs the sensor hardware equipment at the selected sink node position. The sensor unit is connected to the existing network switching equipment through the network cable and establishes communication connection with the gateway system using the standard Ethernet protocol. After installation, each sensor starts collecting node operation data according to the preset collection period and transmits the data to the gateway control system in real time.

[0041] In another embodiment, it also includes: S61: Real-time acquisition of gateway CPU utilization, memory occupancy and network throughput parameters; S62: Input the collected parameters into the preset evaluation function to calculate the performance evaluation index value, and compare the evaluation index value with the historical data to generate a performance evaluation report; S63: Extract the performance deviation value in the evaluation report, and adjust the mutation probability and crossover probability parameters of the genetic algorithm in S2 according to the size of the deviation value.

[0042] The working principle and beneficial effects of the above technical solution are as follows: step S61, during the operation of the edge gateway, the operation and maintenance technician deploys a performance collection module on the gateway mainboard. The collection module obtains core occupancy data by reading the processor status register. The memory controller interface provides the number of currently used memory pages. The network card driver program counts the total number of data frames passing through the network interface. These raw data are collected once every fixed period and temporarily stored in a ring buffer for subsequent processing.

[0043] Step S62, based on the collected raw data, the system substitutes each parameter into the evaluation function constructed in advance for operation. The evaluation function adopts a weighted summation method, and different parameters are assigned different weight coefficients according to their importance. The comprehensive performance value calculated is compared with the benchmark value saved in the database. The technician draws a performance change curve according to the comparison result, marks the abnormal period, and forms a visual evaluation report.

[0044] Step S63, based on the generated report, the program automatically identifies the gap between the current performance and the expected value. When the gap exceeds the allowable range, the parameter adjustment mechanism is triggered. Larger deviation prompts the system to increase the mutation rate of the genetic algorithm and expand the search space. Smaller deviation reduces the crossover rate to protect the existing good solution. The adjusted algorithm parameters take effect immediately, affecting the next round of resource allocation calculation.

[0045] In another embodiment, the execution of the control strategy triggers a linkage mechanism, which includes: S71: The control instruction adjusts the processor working frequency and power consumption management state of the target network node; S72: The node state change data is transmitted to the distributed sensor through the network management protocol; S73: After receiving the state change data, the distributed sensor triggers the resource allocation algorithm to re-execute the calculation process and generates an updated resource allocation result.

[0046] The working principle and beneficial effects of the above technical solution are as follows: step S71, the control system issues a frequency adjustment command to a specific network node. The node processor receives the command and modifies the frequency division ratio of the internal clock generator. The voltage regulator sets the corresponding core voltage according to the new frequency. The power consumption controller selects the appropriate power management mode according to the task load. The entire adjustment process is completed within milliseconds, ensuring business continuity.

[0047] Step S72, after the node completes the internal adjustment, a data packet containing the new state information is generated. The packet is encapsulated according to the network management protocol format, including node address, time mark and parameter value fields. The packet is sent to each sensor node distributed in the network through the Ethernet link. The transmission adopts an acknowledgement retransmission mechanism to ensure that each sensor receives complete information.

[0048] Step S73, the sensor node parses the received state packet and identifies the change in network topology or node capability. This change immediately triggers the resource allocation program to be re-run. The program re-calculates the amount of resources to be allocated to each node according to the latest network state. The calculation result forms a new allocation table, which is executed through the control channel to complete the entire linkage process.

[0049] In another embodiment, the verification method of the execution unit for the instruction includes: S81: Extracting the digital signature data corresponding to the instruction block from the distributed ledger; S82: Using the pre-stored elliptic curve public key to perform verification operation on the extracted digital signature data to obtain a verification result; S83: Based on the verification result, calculating the hash value of the received instruction, comparing the calculated hash value with the ledger record, and outputting an integrity verification result.

[0050] The working principle and beneficial effects of the above technical solution are as follows: Step S81, after receiving the instruction, the execution unit searches for the block storing the instruction in the blockchain ledger. The target block is quickly located through the hash link. The signature field is read from the block data structure, which contains the coordinate point on the elliptic curve and the random number. These data are identity credentials generated by the private key of the instruction sender.

[0051] Step S82, based on the read signature information, the execution unit calls the corresponding public key data from the key storage area. Point operation and scalar multiplication are performed using the mathematical properties of the elliptic curve. The operation verifies whether the signature is generated by the paired private key. If the verification is passed, it means that the instruction source is reliable, and if it is rejected, it means that there may be a forgery.

[0052] Step S83, after identity verification, the instruction text is calculated for digest. The selected hash algorithm compresses the instruction of any length into a fixed length digest value. The newly calculated digest is compared byte by byte with the original digest saved in the ledger. A perfect match proves that the instruction has not been tampered with and can be safely executed. If it does not match, it is rejected and an exception is recorded.

[0053] In another embodiment, the parallel processing method of the gateway control model in S3 includes: S91: Distribute the genetic algorithm resource allocation operation and control logic operation to multiple general processing cores for parallel execution; S92: distribute the deep network inference operation and the elliptic curve encryption and decryption operation to the dedicated neural processing unit for execution; S93: distribute the subsequent operation tasks to the corresponding processing units according to the operation type and priority through the hardware task scheduler, and achieve load balancing.

[0054] The working principle and beneficial effects of the above technical solution are as follows: in step S91, the task decomposer identifies the independent calculation units in the genetic algorithm. The fitness calculation of each individual in the population is independent of each other and can be distributed to different processor cores. Genetic operations such as selection, crossover, and mutation are also split into multiple parallel tasks. The condition judgment and state conversion in the control logic are also parallelized. Multiple cores work simultaneously, greatly shortening the algorithm execution time.

[0055] In step S92, hardware matching is performed according to the operation characteristics. The matrix multiplication and convolution operation of the deep network is given to the neural processing unit equipped with a large number of multiply-accumulators. The large number of modular operations and point operations of the elliptic curve are undertaken by the cryptographic processor with built-in special circuits. These special hardware has much higher efficiency than general-purpose processors in specific operations.

[0056] In step S93, the task scheduler continuously monitors the busy state of each processing unit. When a new task arrives, the appropriate processor is selected according to the task type. High-priority tasks are executed in queue, and low-priority tasks are queued. When a processor accumulates tasks, the scheduler directs new tasks to idle processors. The dynamic scheduling strategy keeps the entire system running efficiently.

[0057] In another embodiment, the method for constructing historical running data in S32 includes: S101: establish a mapping relationship between the resource allocation result and the corresponding network delay and throughput parameters, and store it as relational data; S102: record the time sequence change process of gateway CPU utilization and memory occupancy after different control actions are executed, and generate time sequence data; S103: collect the parameter correction process of the control strategy under network congestion state and node fault state, and generate correction parameter data.

[0058] The working principle and beneficial effects of the above technical solution are as follows: in step S101, after each resource allocation is completed, the system records the detailed parameters of the allocation scheme. Including the bandwidth quota and processing capacity share of each node. At the same time, the actual effect after the execution of the scheme is measured, such as data packet transmission delay and network throughput rate. The configuration parameters and effect indicators are paired and stored to establish a causal relationship database.

[0059] Step S102, the system state is continuously recorded before and after the control action. The trajectory of the processor occupancy over time is completely saved. The dynamic fluctuation process of the memory usage is also recorded in detail. These time-labeled data points are connected into curves, reflecting the response characteristics of the system to the control action.

[0060] Step S103, when an abnormal situation occurs, the system enters a special recording mode. The increase in packet loss rate caused by network congestion is immediately detected. The routing changes caused by node failure are also accurately captured. The control strategy adjusts the parameters according to the type of abnormality, and the entire adjustment process, including the trigger condition, modification action and final effect, is completely recorded, forming valuable emergency handling experience.

[0061] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application.

Claims

1. A smart gateway control method based on edge computing, characterized in that, include: S1: Collect network node operating status data through distributed sensors, and after filtering, denoising and normalization, sort the data by timestamp to generate preprocessed data containing network traffic, node load and response latency. S2: Calculate the resource requirements of each node based on the preprocessed data, and execute a dynamic resource allocation algorithm in combination with the available resource status of the nodes to generate resource allocation results that include computing resource quotas and task allocation relationships; S3: Input the resource allocation results into the gateway control model, adjust the model parameters based on the real-time network status, and generate a control strategy; S4: The control policy is encoded into a sequence of instructions, encrypted with a digital signature, and transmitted to the execution unit through a secure channel. After verifying the integrity and validity of the signature, the control instructions are executed.

2. The intelligent gateway control method based on edge computing according to claim 1, characterized in that, S2 include: S21: The processing delay parameter and energy consumption parameter are weighted and summed according to preset weight coefficients to construct the objective function of the multi-objective optimization model; S22: Input the node CPU utilization and task computation complexity into a nonlinear mapping function to calculate the computational resource requirements of each node and generate an initial resource allocation scheme. S23: When the network load change is detected to exceed the preset threshold, the genetic algorithm is started to iteratively optimize the resource allocation scheme and output the updated resource allocation result.

3. The intelligent gateway control method based on edge computing according to claim 1, characterized in that, S3 include: S31: Construct a deep network model containing convolutional layers and fully connected layers, and use the resource allocation results and real-time network state vector as input layer data; S32: Store historical operating data and corresponding control effect data in an experience buffer, and randomly sample data to pre-train the model; S33: Calculate the reward value by weighting network throughput and response time, and update the network weight parameters based on the error between the reward value and the predicted value through the backpropagation algorithm.

4. The intelligent gateway control method based on edge computing according to claim 1, characterized in that, Digital signature processing in S4 includes: S41: Use the elliptic curve algorithm to generate a private key and public key pair, and use the hash value of the instruction sequence using the private key pair to perform digital signature calculation; S42: Encapsulate the digital signature data and instruction sequence data into a data block, and write the data block into the distributed ledger maintained by the edge node.

5. The intelligent gateway control method based on edge computing according to claim 1, characterized in that, The deployment of distributed sensors in S1 includes: S51: Calculate the path length between each node based on the network topology, and select the convergence node with the smaller total path length as the sensor deployment location; S52: Based on communication delay constraints and bandwidth capacity limitations, network nodes are grouped using a clustering algorithm to determine the number of sensors to be deployed. S53: Deploy sensor units at selected aggregation node locations and connect the sensor units to the existing network architecture for data acquisition via standard network protocols.

6. The intelligent gateway control method based on edge computing according to claim 1, characterized in that, Also includes: S61: Real-time collection of gateway CPU utilization, memory usage, and network throughput parameters; S62: Input the collected parameters into the preset evaluation function to calculate the performance evaluation index value, compare the evaluation index value with historical data to generate a performance evaluation report; S63: Extract the performance deviation value from the evaluation report, and adjust the mutation probability and crossover probability parameters of the genetic algorithm in S2 according to the magnitude of the deviation value.

7. The intelligent gateway control method based on edge computing according to claim 1, characterized in that, The execution triggering mechanism of the control strategy includes: S71: Control instructions adjust the processor operating frequency and power management status of the target network node; S72: Node status change data is transmitted to the distributed sensors via the network management protocol; S73: After receiving state change data, the distributed sensor triggers the resource allocation algorithm to re-execute the calculation process and generate an updated resource allocation result.

8. The intelligent gateway control method based on edge computing according to claim 4, characterized in that, The methods for verifying instructions by the execution unit include: S81: Extract the digital signature data of the corresponding instruction block from the distributed ledger; S82: Use the pre-stored elliptic curve public key to perform verification operations on the extracted digital signature data and obtain the verification result; S83: Calculate the hash value of the received instruction based on the verification result, compare the calculated hash value with the ledger record, and output the integrity verification result.

9. The intelligent gateway control method based on edge computing according to claim 1, characterized in that, Parallel processing methods in the gateway control model of S3 include: S91: Distribute the genetic algorithm resource allocation operation and control logic operation to multiple general-purpose processing cores for parallel execution; S92: Distribute deep network inference operations and elliptic curve encryption / decryption operations to dedicated neural processing units for execution; S93: The hardware task scheduler distributes subsequent computing tasks to the corresponding processing units according to the type and priority of the operation, thereby achieving load balancing.

10. The intelligent gateway control method based on edge computing according to claim 3, characterized in that, The methods for constructing historical operational data in S32 include: S101: Establish a mapping relationship between the resource allocation results and the corresponding network latency and throughput parameters, and store it as relational data; S102: Record the timing changes of gateway CPU utilization and memory usage after different control actions are executed, and generate timing data; S103: Collect the parameter correction process of the control strategy under network congestion and node failure states, and generate correction parameter data.

Images