Edge computing data transmission and collaborative optimization method based on ad hoc network technology

Abstract

The invention discloses an edge computing data transmission and collaborative optimization method based on an ad-hoc network technology, which comprises the following steps: S1, constructing a communication network based on the ad-hoc network technology, and collecting a resource occupancy state and communication link bandwidth information; s2, constructing Fourier neural operator network structure parameters, obtaining optimal structure configuration by using a hunting optimization algorithm, and performing model training; s3, constructing a time sequence feature set based on the collected data, inputting the trained model, and generating an edge network state sequence; s4, scheduling and transmission optimization are carried out based on the state sequence, and an optimal combination scheme is searched by adopting a hunting optimization algorithm; s5, deploying the optimal scheduling and path to an edge ad hoc network, executing multi-hop relay transmission, and supporting path reconstruction and task migration; and S6, collecting actual operation data, comparing the actual operation data with a prediction result, updating model parameters, and executing closed-loop optimization. According to the method, efficient collaboration and dynamic self-adaption of task scheduling and data transmission in an edge computing environment are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of ad hoc networks, and particularly to an edge computing data transmission and collaborative optimization method based on ad hoc network technology. Background Art

[0002] With the rapid development of Internet of Things, artificial intelligence and big data technologies, edge computing, as an efficient computing power supplement for terminal devices, is becoming a powerful extension of cloud computing. By deploying computing nodes at the network edge close to the data source, edge computing can effectively relieve the computing and communication loads of the cloud data center and achieve low-latency and high-real-time data processing services. It is widely used in many fields such as intelligent manufacturing, vehicle networking, and smart cities. In a typical edge computing scenario, a large number of sensing devices and task requests need to be dynamically allocated to multiple edge nodes for execution, and at the same time, there are also problems such as frequent data transmission, network topology changes, and computing resource state fluctuations. How to achieve efficient task scheduling and communication path optimization has become a key technical challenge in the design of edge computing systems.

[0003] Currently, the scheduling mechanism of edge computing systems mainly relies on centralized control strategies, and a unified computing scheduling scheme is obtained by the central control node after obtaining the global state. Although this method has a certain global optimality, it depends on the single-point performance of the central node, is easily affected by system bottlenecks and failures, and at the same time, it responds slowly to the state of edge nodes and network link changes in a dynamic environment, and cannot meet the requirements of large-scale distributed deployment and high-frequency network change scenarios. In addition, existing scheduling methods mostly adopt static planning or rule-based heuristic algorithms, and it is difficult to adapt to the complex situation of dynamic coupling of multiple factors such as task types, node resources, and communication links in the edge computing environment, resulting in serious limitations in the timeliness and execution efficiency of scheduling strategies.

[0004] In terms of communication, since the connection between edge nodes is usually mainly wireless, there are problems such as unstable topological structure, large bandwidth fluctuations, and high probability of link interruption. Traditional transmission strategies based on fixed routing or centralized path planning are difficult to ensure the efficient and stable transmission of task data. Especially in multi-hop communication links, once a middle node fails or the channel degrades, it is often necessary to recalculate the path or even reschedule the task, resulting in an increase in the system response time delay. At the same time, existing solutions often ignore the differences in the collaborative capabilities between edge nodes and do not comprehensively consider key factors such as resource occupancy, bandwidth utilization, path stability, and system load balancing during the scheduling process.

[0005] To improve scheduling and transmission performance, some studies have begun to attempt to introduce artificial intelligence algorithms for optimization, such as genetic algorithms, particle swarm optimization algorithms, etc. However, the search efficiency of these algorithms in high-dimensional parameter spaces is limited and they are prone to falling into local optima; moreover, they usually lack the ability to model multi-factor constraints between edge nodes and it is difficult to effectively combine with the actual self-organizing network environment. On the other hand, some studies have proposed using deep learning models such as graph neural networks and LSTM for edge state prediction, but these models have problems such as high training costs, unstable prediction accuracy, and strong empiricism in structure design, and it is difficult to flexibly adapt to different edge environment characteristics.

[0006] Existing edge computing scheduling methods generally lack a closed-loop optimization mechanism for "state awareness - policy generation - execution feedback". Specifically, most scheduling schemes only generate a scheduling policy once at the initial state, lacking the ability to collect real-time execution feedback information and adaptively update the model, resulting in a gradual degradation of the model's prediction ability and a lack of long-term guarantee for the overall system performance.

[0007] In addition, although some systems have introduced edge network state prediction mechanisms, most use traditional regression models and it is difficult to fully extract the deep dynamic laws in time series data. The neural operator methods proposed in recent years, especially the Fourier Neural Operator (FNO), with its ability to model function mapping relationships, have strong spatio-temporal modeling capabilities and generalization, and have achieved good results in scenarios such as partial differential equations and dynamic systems. However, existing work has not fully explored its potential in edge computing state prediction and scheduling optimization, especially lacking an efficient implementation scheme for adaptively searching and training the neural operator structure in combination with optimization algorithms.

[0008] In addition, in the scheduling optimization problems of multi-tasks and multi-paths, important indicators affecting system robustness such as path stability, hop count overhead, and node load balancing have not been incorporated into the design of mainstream fitness functions, resulting in poor feasibility and unstable deployment of the optimization results in actual operation. At the same time, the system's adaptive ability to node state mutations and link failures is insufficient, lacking effective support for the self-organizing network dynamic reconstruction mechanism, which limits the system's fault tolerance and real-time response ability.

[0009] Therefore, how to provide an edge computing data transmission and collaborative optimization method based on self-organizing network technology is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0010] An object of the present invention is to propose an edge computing data transmission and collaborative optimization method based on ad hoc network technology. The present invention integrates the ad hoc network dynamic communication mechanism, the state prediction ability of the Fourier neural operator network, and an information-sharing improved cheetah optimization algorithm introducing a multi-prey mechanism, and details the steps of realizing task-node-path collaborative scheduling and intelligent optimization in a dynamic environment, with the advantages of high scheduling accuracy, strong communication stability, high resource utilization efficiency, and strong network adaptability.

[0011] The edge computing data transmission and collaborative optimization method based on ad hoc network technology according to an embodiment of the present invention includes the following steps:

[0012] S1. Construct a communication network between edge computing devices based on ad hoc network technology, and collect the resource occupancy status and communication link bandwidth information of each edge node;

[0013] S2. Construct the structural parameters of the Fourier neural operator network, use the cheetah optimization algorithm to globally search and optimize the structural parameters, obtain the optimal structural configuration, and train the Fourier neural operator network;

[0014] S3. Based on the resource occupancy status and communication link bandwidth information, construct a time series feature set, input it into the trained Fourier neural operator network, and generate an edge network state sequence;

[0015] S4. Based on the edge network state sequence, evaluate the resource availability and communication capabilities of each edge node in the future time period, use the cheetah optimization algorithm to search in the preset task scheduling and data transmission path combination, construct a task-edge node mapping and the corresponding transmission scheme, and determine the optimal scheduling strategy;

[0016] S5. Deploy the optimal scheduling strategy in the edge ad hoc network, and let the edge nodes receive, process, and forward task data according to the scheme, and use multi-hop relay transmission. When the node state changes or the link fails, complete path reconstruction and task migration through the update of the ad hoc network topology;

[0017] S6. After the task is completed, collect the actual operation data of each edge node, compare and analyze the actual operation data with the edge network state sequence, update the parameters of the Fourier neural operator network, and re-execute steps S2 to S6 within the set optimization period to form a prediction-driven edge computing dynamic scheduling closed-loop system.

[0018] Optionally, the S2 specifically includes:

[0019] S21. Set the structural parameter set Θ of the Fourier neural operator network;

[0020] S22, initialize the number of individuals in the cheetah optimization algorithm population to N, generate the initial structure parameter individual set {Θ1,Θ2,...,Θ N}, where each cheetah individual represents a set of structural parameter combinations;

[0021] S23. In each iteration, the fitness function value is calculated for each cheetah individual in the population, where the cheetah individual is a set of structural parameter combinations of the Fourier neural operator network, and the fitness function f(Θ i ) is defined as:

[0022]

[0023] in, Represented by the structural parameter combination Θ i The predicted output of the constructed Fourier neural operator network on the jth validation sample, y j represents the true value of the sample, n is the number of samples in the validation set, ‖‖ is the L2 norm, and according to the fitness function value f(Θ i ) sorting results, select the top k cheetah individuals with the best fitness from all cheetah individuals, and the corresponding structural parameter combination constitutes a multi-prey set

[0024]

[0025] Among them, Θ (j) represents the prey structure parameter combination of the jth ranking, j∈{1,2,...,k};

[0026] S24. For each cheetah individual in the current population, the structural parameter combination is Θ i , calculate the structural parameter combination of the current i-th cheetah individual and the multi-prey set The Euclidean distance between each prey structure parameter combination in , select the prey structure parameter combination with the closest distance Update the position of individual cheetahs:

[0027]

[0028] Among them, r1 is the tracking strength factor, t is the current iteration round, represents the structural parameter combination of the i-th cheetah individual in the t-th iteration, represents the structural parameter combination of the i-th cheetah individual at the t+1th iteration;

[0029] S25. Establish a local history structure parameter memory pool of length m for each cheetah individual, denoted as M i , used to store its structural parameter combinations in the latest m iteration rounds:

[0030]

[0031] Among them, represents the combination of structural parameters of the i-th cheetah individual at the (t - L)-th iteration, where L ∈ {1, 2,..., m};

[0032] S26. Define a neighborhood set for each cheetah individual

[0033]

[0034] Among them, Θ j represents the combination of structural parameters of the j-th cheetah individual in the current iteration, and δ is a preset distance threshold used to judge the adjacency relationship between cheetah individuals;

[0035] S27. Based on the local historical structure parameter memory pools of each cheetah individual in the neighborhood set of the i-th cheetah individual, perform an information-sharing position update operation on the structural parameter combination of the i-th cheetah individual:

[0036]

[0037] Among them, represents the position obtained after performing information-sharing fine-tuning of the i-th cheetah individual in the (t + 1)-th iteration, λ is the information-sharing intensity factor, mean(M j ) represents the mean of the historical structure parameter combinations of the j-th neighbor individual, and M j represents the historical structure parameter memory pool of the j-th neighbor individual;

[0038] S28. When the fitness improvement rate is less than the set threshold ε in consecutive G iterations, trigger the reconnaissance and patrol mechanism and perform a jump perturbation update operation on the cheetah individual:

[0039]

[0040] Among them, represents the combination of structural parameters of the i-th cheetah individual in the (t + 1)-th iteration, sign(r - 0.5) is the sign function, σ is the jump intensity factor, r is a uniformly distributed random number, and randn() represents the standard normal distribution function;

[0041] S29. Repeat steps S23 to S28 until the maximum number of iterations T is reached or the preset convergence condition is satisfied, and select the structural parameter combination Θ * with the best score as the optimal structural configuration of the Fourier neural operator network;

[0042] S210. Construct the network structure of the Fourier neural operator network according to the optimal structural parameters, and use the resource occupancy status and communication link bandwidth information as training data to complete the training of the Fourier neural operator network.

[0043] Optionally, the set of structural parameters of the Fourier neural operator network specifically includes the number of network layers, the number of Fourier projection dimensions, the number of hidden nodes in each layer, the regularization coefficient, and the learning rate, which are used to control the expression ability of the Fourier neural operator network and the stability of the training process.

[0044] Optionally, the said S3 specifically includes:

[0045] S31. Arrange the collected resource occupancy status and communication link bandwidth information in chronological order to construct multi-dimensional time series data samples;

[0046] S32. Perform data formatting and normalization processing on the constructed time series data samples to form a unified standard set of time series features;

[0047] S33. Use the processed set of time series features as input and input it into the trained Fourier neural operator network;

[0048] S34. Calculate the features of the input set of time series features through the Fourier neural operator network to obtain the corresponding change trend of the edge computing system state;

[0049] S35. Output the network state prediction results of each node in the edge computing system to form an edge network state sequence.

[0050] Optionally, the time series data samples specifically include the resource occupancy rate of edge nodes at each time point and the bandwidth values of the corresponding communication links, which are used to reflect the operating state characteristics of the edge computing environment changing over time.

[0051] Optionally, the calculating the features of the input set of time series features through the Fourier neural operator network specifically means inputting the information including the resource occupancy status and communication link bandwidth into the Fourier neural operator network, and through frequency domain mapping and non-linear mapping operations, extracting deep features changing over time and generating prediction results of future states, which are used to extract potential change patterns in the time series features of edge nodes.

[0052] Optionally, the said S4 specifically includes:

[0053] S41. Construct a set of scheduling tasks A set of edge nodes And a set of transmission paths between each pair of scheduling tasks and edge nodes where, T jDenote the j-th scheduling task, N i Denote the i-th edge node, j ∈ {1, 2,..., a}, i ∈ {1, 2,..., b}, a represents the number of scheduling tasks, b represents the number of edge nodes, P ij Denote the scheduling task T j To the edge node N i The allocated path scheme;

[0054] S42. Construct the solution space of the task scheduling and transmission path combination As the search individual in the cheetah optimization algorithm, each individual corresponds to a set of scheduling and transmission combination schemes;

[0055] S43. Define the fitness function

[0056]

[0057] Among them, α, β, γ, δ, λ are adjustable control factors, D j Is the task execution delay, C j Is the communication consumption, H j Is the number of path hops, S ij Is the path stability, Var(L) is the variance of the node resource utilization rates of all edge nodes;

[0058] S44. Initialize the population using the cheetah optimization algorithm with the multi-prey mechanism introduced, and select the top k combination schemes with the best fitness as the multi-prey set Each cheetah individual determines the target prey according to the principle of the minimum Euclidean distance and performs the tracking update operation;

[0059] S45. Based on the information sharing mechanism, define the neighborhood set Extract the mean of the neighbor historical structure parameters and perform the fine-tuning update;

[0060] S46. When the fitness improvement is less than the threshold ∈ in G consecutive iterations, trigger the detection mechanism, perform perturbation jumps on some cheetah individuals, and finally select the combination scheme with the best fitness As the optimal scheduling strategy.

[0061] Optionally, the specific steps of S5 include:

[0062] S51. Send the obtained optimal scheduling strategy to each relevant edge node in the edge ad hoc network;

[0063] S52. The edge node receiving the scheduling instruction receives and locally buffers the task data according to the assigned task execution role, and prepares to perform the calculation or forwarding operation;

[0064] S53. The edge node undertaking the task execution role completes the data processing and resource scheduling process locally according to the assigned task content;

[0065] S54. The edge node undertaking the relay role forwards the task data from the source node to the target node in a multi-hop relay manner according to the optimal scheduling strategy;

[0066] S55. During the task execution and data transmission process, the system monitors the running state information of each edge node and the connection stability of the communication link in real time;

[0067] S56. When it is detected that the resources of the edge node are abnormal or the communication link is disconnected, the self-organizing network topology update mechanism is triggered to dynamically adjust the network structure and reconstruct the data transmission path, and at the same time, the task migration operation is executed.

[0068] The beneficial effects of the present invention are as follows:

[0069] By introducing the self-organizing network technology, the Fourier neural operator network, and the improved cheetah optimization algorithm, the present invention solves the key problems in the existing edge computing system, such as slow response of the scheduling strategy, poor path stability, weak model prediction ability, and lack of an adaptive adjustment mechanism in the system, and achieves beneficial technical effects in many aspects. Compared with the traditional centralized scheduling method, the present invention uses the self-organizing network technology to construct a decentralized and dynamically reconstructable communication network among edge nodes, enabling edge nodes to complete distributed task scheduling and multi-hop relay transmission without the support of fixed infrastructure, effectively improving the fault tolerance and topology adaptation ability of the system. At the same time, the self-organizing network mechanism can automatically complete path reconstruction and task migration when a node fails or a link is interrupted, ensuring the continuity and transmission reachability of tasks in a dynamic environment.

[0070] In terms of edge state modeling, the present invention constructs a multi-dimensional time series feature set centered on the resource occupancy state and communication link bandwidth information, and introduces the Fourier neural operator network to learn and predict this feature set, significantly improving the modeling accuracy of the future state change trend of edge nodes. Compared with traditional deep networks, the Fourier neural operator network has stronger function fitting ability and spatio-temporal generalization ability, can adapt to the changing and non-linear system state changes in the edge computing environment, and improves the forward-looking and accuracy of the generation of scheduling strategies.

[0071] At the optimization strategy level, the present invention constructs a scheduling solution space based on the task-node-path combination, and on this basis, introduces an improved cheetah optimization algorithm. This algorithm integrates the multi-prey target mechanism, neighborhood information sharing mechanism and detection perturbation mechanism, overcoming the problems of traditional intelligent optimization algorithms being prone to falling into local optima and having a slow convergence speed. The multi-prey mechanism enhances the global exploration ability, the neighborhood sharing mechanism improves the co-evolution efficiency of the solution space, and the detection perturbation mechanism ensures the diversity and robustness of the optimization process, thus effectively searching for the optimal scheduling and transmission strategy with high resource utilization, low communication overhead, reasonable path hops and balanced load distribution.

[0072] In addition, the present invention constructs a dynamic feedback mechanism during the execution of the scheduling strategy, uses the real-time operation data of the edge nodes to incrementally update the Fourier neural operator network, forms a closed-loop system that couples state prediction and strategy optimization, and realizes the continuous optimization of the model performance and the adaptive adjustment of the scheduling strategy. The overall system has good real-time performance, scalability and deployment flexibility, and can be widely applied to edge computing systems in scenarios such as the Internet of Things, smart cities, and industrial Internet, significantly improving the intelligent scheduling level and communication cooperation efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the accompanying drawings:

[0074] Figure 1 is a flowchart of the edge computing data transmission and collaborative optimization method based on the ad hoc network technology proposed by the present invention;

[0075] Figure 2 is a flowchart of the scheduling-path combination search and optimization based on the improved cheetah optimization algorithm for the edge computing data transmission and collaborative optimization method proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0077] Refer to Figure 1 and Figure 2 , the edge computing data transmission and collaborative optimization method based on the ad hoc network technology includes the following steps:

[0078] S1. Construct a communication network between edge computing devices based on the ad hoc network technology, and collect the resource occupancy status and communication link bandwidth information of each edge node;

[0079] S2. Construct the structural parameters of the Fourier neural operator network, use the cheetah optimization algorithm to globally search and optimize the structural parameters to obtain the optimal structural configuration, and train the Fourier neural operator network;

[0080] S3. Based on the resource occupancy status and communication link bandwidth information, construct a time series feature set, input it into the trained Fourier neural operator network, and generate an edge network state sequence;

[0081] S4. Based on the edge network state sequence, evaluate the resource availability and communication capabilities of each edge node in the future time period, use the cheetah optimization algorithm to search in the preset task scheduling and data transmission path combinations, construct a mapping between tasks and edge nodes and the corresponding transmission scheme, and determine the optimal scheduling strategy;

[0082] S5. Deploy the optimal scheduling strategy in the edge ad hoc network. The edge nodes receive, process, and forward task data according to the scheme, and use multi-hop relay transmission. When the node status changes or the link fails, the path reconstruction and task migration are completed through the update of the ad hoc network topology;

[0083] S6. After the task is completed, collect the actual operation data of each edge node, compare and analyze the actual operation data with the edge network state sequence, update the parameters of the Fourier neural operator network, and re-execute steps S2 to S6 within the set optimization period to form a prediction-driven dynamic scheduling closed-loop system for edge computing.

[0084] Through the construction of an edge communication network based on ad hoc network technology, combined with the Fourier neural operator network and the improved cheetah optimization algorithm, the present invention realizes the efficient collaborative optimization of task scheduling and data transmission in the edge computing environment, and has significant technical advantages. First, the ad hoc network mechanism improves the connection stability and path reconstruction ability between edge nodes, and can dynamically complete network self-recovery in case of node failures or link interruptions, ensuring the continuity of data transmission. Second, the Fourier neural operator network can accurately predict the resource status and communication capabilities of each node, providing forward-looking support for scheduling. Through the multi-objective search of the cheetah optimization algorithm for the scheduling and transmission scheme, comprehensively considering task latency, communication load, and resource balance, the global optimality and execution efficiency of the scheduling scheme are improved. In addition, the system establishes a closed-loop optimization mechanism based on feedback data, realizes the adaptive update of the model and the dynamic adjustment of the scheduling strategy, and enhances the system's response ability to environmental changes and long-term stability. The present invention has the advantages of high scheduling accuracy, low communication overhead, strong robustness, and strong self-evolution ability, and is applicable to a variety of complex distributed edge computing scenarios.

[0085] In this embodiment, the S2 specifically includes:

[0086] S21. Set the structural parameter set Θ of the Fourier neural operator network;

[0087] S22. Initialize the population size of the cheetah optimization algorithm as N, and generate an initial set of individual structural parameter sets {Θ1, Θ2,..., Θ N}, where each cheetah individual represents a set of structural parameter combinations;

[0088] S23. In each generation of iteration, calculate the fitness function value for each cheetah individual in the population. The cheetah individual is a set of structural parameter combinations of the Fourier neural operator network. The fitness function f(Θ i ) is defined as:

[0089]

[0090] where, represents the predicted output of the Fourier neural operator network constructed by the structural parameter combination Θ i on the j-th validation sample, y j represents the true value of the sample, n is the number of samples in the validation set, ‖‖ is the L2 norm. According to the sorting result of the fitness function value f(Θ i ), select the top k cheetah individuals with the best fitness from all cheetah individuals, and the corresponding structural parameter combinations form a multi-prey set

[0091]

[0092] where, Θ (j) represents the structural parameter combination of the j-th ranked prey, j ∈ {1, 2,..., k};

[0093] S24. For each cheetah individual in the current population, with the structural parameter combination being Θ i , calculate the Euclidean distance between the structural parameter combination of the current i-th cheetah individual and each prey structural parameter combination in the multi-prey set , and select the prey structural parameter combination with the closest distance to update the position of the cheetah individual:

[0094]

[0095] where, r1 is the tracking intensity factor, t is the current iteration round, represents the structural parameter combination of the i-th cheetah individual at the t-th iteration, represents the structural parameter combination of the i-th cheetah individual at the (t + 1)-th iteration;

[0096] S25. Establish a local historical structural parameter memory pool with a length of m for each cheetah individual, denoted as M i, for storing the combination of its structural parameters in the most recent m iteration rounds:

[0097]

[0098] where, represents the combination of structural parameters of the i-th cheetah individual at the (t - L)-th iteration, where L ∈ {1, 2,..., m};

[0099] S26. Define a neighborhood set for each cheetah individual

[0100]

[0101] where, Θ j represents the combination of structural parameters of the j-th cheetah individual in the current iteration, and δ is a preset distance threshold for judging the adjacency relationship between cheetah individuals;

[0102] S27. Based on the local historical structure parameter memory pools of each cheetah individual in the neighborhood set of the i-th cheetah individual, perform an information-sharing based position update operation on the structural parameter combination of the i-th cheetah individual:

[0103]

[0104] where, represents the position obtained after performing information-sharing fine-tuning of the i-th cheetah individual in the (t + 1)-th iteration, λ is the information-sharing intensity factor, mean(M j ) represents the mean of the historical structure parameter combinations of the j-th neighbor individual, and M j represents the historical structure parameter memory pool of the j-th neighbor individual;

[0105] S28. When the fitness improvement rate is less than the set threshold ε in G consecutive iterations, trigger the reconnaissance and patrol mechanism to perform a jump perturbation update operation on the cheetah individual:

[0106]

[0107] where, represents the combination of structural parameters of the i-th cheetah individual in the (t + 1)-th iteration, sign(r - 0.5) is the sign function, σ is the jump intensity factor, r is a uniformly distributed random number, and randn() represents the standard normal distribution function;

[0108] S29. Repeat steps S23 to S28 until the maximum number of iterations T is reached or the preset convergence condition is satisfied, and select the structural parameter combination Θ * with the best score as the optimal structural configuration of the Fourier neural operator network;

[0109] S210. Construct the network structure of the Fourier neural operator network according to the optimal structure parameters, and use the resource occupancy status and communication link bandwidth information as training data to complete the training of the Fourier neural operator network.

[0110] In the optimization process of the Fourier neural operator network structure of the present invention, an improved cheetah optimization algorithm is introduced, which solves the problems of traditional neural network structure design relying on experience, low search efficiency, and easy to fall into local optimum. By setting a set of structure parameters and representing each cheetah individual as a combination of a set of structure parameters, the multi-prey mechanism is used to enhance the global exploration ability to ensure that the algorithm can quickly locate multiple potential optimal solution regions in the high-dimensional search space. By constructing a neighborhood set and a local historical memory pool, an information-sharing position update operation is performed to effectively improve the local development accuracy and convergence speed. At the same time, a detection perturbation mechanism is introduced to enhance the ability of the algorithm to jump out of the local optimum when the convergence stagnates, further improving the overall search robustness. Finally, the optimal structure configuration obtained through the above multi-stage collaborative optimization process can significantly improve the modeling ability and prediction accuracy of the Fourier neural operator network for the resources and communication states of edge nodes. Experimental results show that the present invention can reduce the model error and improve the training stability, achieving the purpose of automatically constructing a high-performance prediction model and enhancing the state perception ability, providing a more reliable model basis for the scheduling optimization of edge computing systems.

[0111] In this embodiment, the set of structure parameters of the Fourier neural operator network specifically includes the number of network layers, the number of Fourier projection dimensions, the number of hidden nodes in each layer, the regularization coefficient, and the learning rate, which are used to control the expression ability of the Fourier neural operator network and the stability of the training process.

[0112] In this embodiment, the S3 specifically includes:

[0113] S31. Arrange the collected resource occupancy status and communication link bandwidth information in chronological order to construct multi-dimensional time series data samples;

[0114] S32. Perform data formatting and normalization processing on the constructed time series data samples to form a unified standard time series feature set;

[0115] S33. Use the processed time series feature set as input and input it into the trained Fourier neural operator network;

[0116] S34. Calculate the features of the input time series feature set through the Fourier neural operator network to obtain the corresponding change trend of the edge computing system state;

[0117] S35. Output the network state prediction results of each node in the edge computing system to form an edge network state sequence.

[0118] In the process of predicting the edge network state, the present invention proposes a modeling method based on standardized time series features and Fourier neural operator network, which effectively solves the problems of inconsistent data features, insufficient prediction accuracy, and weak ability to model time series dynamics in existing methods. By constructing time series of the resource occupancy status of edge nodes and the bandwidth information of communication links, the integrity and timeliness of data are ensured; further, the format is unified and normalized to improve the computability of data and the stability of model training. The processed high-quality feature set is input into the trained Fourier neural operator network, and by using its advantages in frequency domain mapping and spatial pattern modeling, the deep feature relationships affecting network state changes are effectively extracted. Finally, the model can accurately predict the resource and communication state trends of each edge node in the future time period and output a stable edge network state sequence. Experimental results show that, on the premise of maintaining low computational overhead, this method significantly improves the accuracy and response speed of state prediction, provides reliable and forward-looking support for generating scheduling strategies, and achieves the purpose of enhancing the system's prediction ability and supporting subsequent intelligent scheduling optimization.

[0119] In this embodiment, the time series data sample specifically includes the resource occupancy rate of the edge node at each time point and the bandwidth value of the corresponding communication link, which is used to reflect the operating state characteristics of the edge computing environment changing with time.

[0120] In this embodiment, the feature calculation of the input time series feature set by the Fourier neural operator network specifically refers to inputting the information including the resource occupancy status and the communication link bandwidth into the Fourier neural operator network, and through frequency domain mapping and non-linear mapping operations, extracting the deep features changing with time and generating the prediction results of the future state, which is used to extract the potential change patterns in the time series features of the edge nodes.

[0121] In this embodiment, S4 specifically includes:

[0122] S41. Construct a set of scheduling tasks A set of edge nodes And a set of transmission paths between each pair of scheduling tasks and edge nodes Where, T j represents the jth scheduling task, N i represents the ith edge node, j ∈ {1, 2,..., a}, i ∈ {1, 2,..., b}, a represents the number of scheduling tasks, b represents the number of edge nodes, and P ij represents the path scheme assigned by the scheduling task T j to the edge node N i ;

[0123] S42. Construct the solution space of the task scheduling and transmission path combination As a search individual in the cheetah optimization algorithm, each individual corresponds to a set of scheduling and transmission combination schemes;

[0124] S43. Define the fitness function

[0125]

[0126] where α, β, γ, δ, λ are adjustable control factors, D j is the task execution delay, C j is the communication consumption, H j is the number of path hops, S ij is the path stability, and Var(L) is the variance of the resource utilization rates of all edge nodes;

[0127] S44. Initialize the population using the cheetah optimization algorithm with the multi-prey mechanism introduced, and select the top k combination schemes with the best fitness as the multi-prey set Each cheetah individual determines the target prey based on the principle of the minimum Euclidean distance and performs the tracking update operation;

[0128] S45. Based on the information sharing mechanism, define the neighborhood set Extract the mean of the neighbor historical structure parameters and perform the fine-tuning update;

[0129] S46. When the fitness improvement is less than the threshold ∈ in G consecutive iterations, trigger the detection mechanism, perform perturbation jumps on some cheetah individuals, and finally select the combination scheme with the best fitness as the optimal scheduling strategy.

[0130] In the process of task scheduling and data transmission path optimization, the present invention introduces an improved cheetah optimization algorithm, which solves the problems in traditional scheduling methods such as single optimization objective, blind path selection, weak global search ability, and easy to fall into local optimum. By constructing a combined solution space of tasks, nodes and paths, and taking each combination as a search individual, a multi-dimensional fitness function is introduced, comprehensively considering multiple key indicators such as task execution delay, communication overhead, path hop count, path stability and node load balancing, which can comprehensively reflect the overall advantages and disadvantages of the scheduling scheme. The algorithm adopts a multi-prey mechanism to guide multiple search directions simultaneously, enhancing the global exploration ability; the neighborhood information sharing mechanism realizes the knowledge transfer between individuals and improves the local convergence speed; when the search gets stuck, the detection mechanism can improve the ability to jump out of the local optimum through perturbation jumps, ensuring the stable evolution of the algorithm. The finally output optimal scheduling-path combination scheme can reduce the network load while meeting the real-time requirements of tasks, improving the robustness and resource utilization efficiency of the system. Experiments prove that the method of the present invention can effectively improve the task completion rate and shorten the overall processing delay of the system, achieving the purpose of realizing efficient, stable and adaptive scheduling optimization in complex edge environments.

[0131] In this embodiment, the S5 specifically includes:

[0132] S51. Send the obtained optimal scheduling strategy to each relevant edge node in the edge ad hoc network;

[0133] S52. The edge node receiving the scheduling instruction receives and locally buffers the task data according to the assigned task execution role, and prepares to perform calculation or forwarding operations;

[0134] S53. The edge node undertaking the task execution role completes the data processing and resource scheduling process locally according to the assigned task content;

[0135] S54. The edge node undertaking the relay role forwards the task data from the source node to the target node in a multi-hop relay manner according to the optimal scheduling strategy;

[0136] S55. During the task execution and data transmission process, the system monitors the running state information of each edge node and the connection stability of the communication link in real time;

[0137] S56. When it is detected that the resources of the edge node are abnormal or the communication link is disconnected, trigger the ad hoc network topology update mechanism, dynamically adjust the network structure and reconstruct the data transmission path, and at the same time perform the task migration operation.

[0138] In the scheduling policy deployment and execution phase of the present invention, the flexible communication capabilities and dynamic topology management advantages of the ad-hoc network technology are fully utilized, solving problems such as task deployment lag, path rigidity, and slow node failure response in traditional edge computing systems. By real-time sending the optimized scheduling policy to each edge node, the system can automatically distinguish the execution nodes and relay nodes according to the task roles, and quickly enter the task execution and data forwarding states. Adopting a multi-hop relay transmission mechanism, it can achieve efficient task data transfer under the condition of no fixed infrastructure, improving the communication coverage and extension capabilities of the system. During operation, the system continuously monitors the resource status and link stability of each node. Once a node overload or link interruption is detected, it immediately triggers the ad-hoc network topology update mechanism, automatically reconstructs the optimal communication path and executes task migration to ensure that tasks are not interrupted and data is not lost. This mechanism significantly improves the dynamic adaptability and fault self-healing capabilities of the system. Practical applications show that the present invention can effectively improve the task processing continuity and communication reliability in complex network environments, reduce the scheduling interruption rate, enhance the real-time performance, stability, and intelligence level of the edge computing system, achieving the collaborative optimization goal of flexible deployment, stable operation, and efficient scheduling.

[0139] Embodiment 1:

[0140] To verify the feasibility of the present invention in implementation, the present invention is applied to the intelligent patrol inspection system of a certain coastal port. The management unit has deployed a large number of image monitoring devices and environmental sensors for real-time collection of the operation status and potential safety hazard information of the dock operation area. To reduce the data backhaul delay and improve the local response efficiency, multiple edge computing nodes are arranged in different operation areas. These nodes have basic image recognition and sensing processing capabilities, can complete preliminary analysis and alarm judgment locally, and send key information back to the background system.

[0141] However, during the actual operation process, the system faces three main problems: First, the node deployment positions are scattered and the environment is complex. The nodes only rely on wireless networks to form a temporary communication structure, and the network topology changes frequently, making the data transmission path easily interrupted; second, the resource capabilities of the edge nodes are uneven, and some nodes are prone to overload, resulting in task delays or even failures; third, the existing task scheduling mechanism cannot make predictive judgments based on the future states of the nodes, nor can it achieve real-time dynamic migration of tasks.

[0142] In this typical application scenario, the edge computing data transmission and collaborative optimization method based on the ad-hoc network technology proposed by the present invention effectively solves the above problems. First, the system automatically constructs an edge communication network based on the ad-hoc network protocol, and each node can dynamically establish a communication relationship according to the signal strength and connection quality. Each node periodically collects its current CPU usage rate, memory occupancy, and communication bandwidth with neighboring nodes, and uses these data as inputs to construct a standardized time-series feature set. This set is input into the trained Fourier neural operator network to generate the state prediction results of each node in the future time period, including the change trend of resource availability and link stability.

[0143] According to the prediction results, the scheduling optimization module constructs the combined space of all tasks, nodes, and paths, and uses the improved cheetah optimization algorithm in the present invention for searching. This algorithm introduces a multi-prey mechanism for parallel target guidance, introduces a neighborhood information sharing mechanism to improve the local search accuracy, and automatically performs reconnaissance perturbations when the convergence stagnates to avoid falling into a local optimum. The finally determined scheduling strategy can comprehensively balance the execution efficiency, communication overhead, path hop count, and node load distribution.

[0144] In a 7-day field test after the system deployment, about 200 video analysis tasks are assigned to 12 edge nodes every day. The average size of the task data is 8.5MB, and it is required to complete the local analysis and result upload within 20 seconds. Under the condition of maintaining the same number of tasks and data scale, the system uses the traditional static scheduling strategy, the standard genetic algorithm scheduling scheme, and the scheduling optimization method proposed by the present invention for comparison, and collects a number of key performance indicators, including task latency, path hop count, task completion rate, link reconstruction times, etc.

[0145] Table 1 Performance comparison of different scheduling methods in the edge intelligent patrol system

[0146]

[0147] As can be seen from Table 1, the scheduling optimization method proposed by the present invention is superior to the traditional static scheduling strategy and the genetic algorithm optimization strategy in multiple key performance indicators, fully demonstrating its comprehensive advantages in the edge computing environment. First, in terms of the average task completion latency, the method of the present invention achieves an excellent result of 11.2 seconds, shortening by 8.5 seconds compared with the static scheduling and reducing by 4.4 seconds compared with the genetic algorithm scheme, which fully shows that it has a significant improvement in task response speed and scheduling efficiency.

[0148] In terms of the path hop count, the optimized solution of the present invention only requires an average of 2.6 hops, significantly less than the 4.3 hops of the static scheduling and the 3.5 hops of the genetic algorithm. This reflects that the system is more intelligent in the selection of communication paths, the transmission link is shorter and more stable, effectively reducing the transmission delay and failure risk brought by relay nodes.

[0149] In terms of the node resource load balance, the standard deviation of the method of the present invention is 8.1%, significantly lower than 14.3% of the genetic algorithm and 18.9% of the static strategy, indicating that the task allocation is more uniform among edge nodes, effectively avoiding resource bottlenecks and node overload situations, and improving the overall operation stability of the system.

[0150] In terms of the task completion rate, the solution of the present invention reaches 98.7%, far higher than 89.5% of the static strategy and 94.2% of the genetic algorithm. This shows that this method has higher scheduling success rate and reliability in multi-task high-concurrency scenarios, and is suitable for edge computing applications with high requirements for timeliness and stability.

[0151] The number of link reconstructions is also significantly reduced, from 12 times per day under static scheduling to 2 times under the method of the present invention. This benefits from the introduction of the ad-hoc network topology awareness and dynamic update mechanism. The system can quickly respond to emergencies such as node failures or channel interferences, and achieve efficient path recovery and task migration.

[0152] Finally, in terms of the response time of the scheduling strategy, the solution of the present invention is controlled within 2.7 seconds, shortening by 3.1 seconds and 1.9 seconds compared with the static solution and the genetic algorithm respectively. This shows that this method has higher computational efficiency when generating scheduling strategies and is more suitable for deployment in practical applications with high real-time requirements.

[0153] In summary, through the integration of state prediction, intelligent optimization and ad-hoc network communication mechanism, the present invention significantly improves the comprehensive performance of the edge computing system in task scheduling and data transmission. Especially, it shows obvious advantages in processing efficiency, load balance, system stability and response ability, and has good engineering practicability and promotion value.

[0154] The above is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent replacements or changes, should be covered by the protection scope of the present invention.

Claims

1. An edge computing data transmission and collaborative optimization method based on ad hoc network technology, characterized in that, It includes the following steps: S1. Build a communication network among edge computing devices based on ad-hoc network technology, and collect the resource occupancy status and communication link bandwidth information of each edge node; S2. Build the structural parameters of the Fourier neural operator network, use the cheetah optimization algorithm to globally search and optimize the structural parameters, obtain the optimal structure configuration, and train the Fourier neural operator network; S3. Based on the resource occupancy status and communication link bandwidth information, build a time series feature set, input it into the trained Fourier neural operator network, and generate an edge network state sequence; S4. Based on the edge network state sequence, evaluate the resource availability and communication capabilities of each edge node in the future time period, use the cheetah optimization algorithm to search in the preset task scheduling and data transmission path combinations, build a mapping between tasks and edge nodes and the corresponding transmission scheme, and determine the optimal scheduling strategy; S5. Deploy the optimal scheduling strategy in the edge ad-hoc network, and let the edge nodes receive, process, and forward task data according to the scheme, and use multi-hop relay transmission. When the node status changes or the link fails, complete path reconstruction and task migration through self-organizing network topology update; S6. After the task is completed, collect the actual operation data of each edge node, compare and analyze the actual operation data with the edge network state sequence, update the parameters of the Fourier neural operator network, and re-execute steps S2 to S6 within the set optimization period to form a prediction-driven dynamic scheduling closed-loop system for edge computing.

2. The edge computing data transmission and collaborative optimization method based on the ad hoc network technology according to claim 1, wherein The specific content of S2 includes: S21. Set the structural parameter set Θ of the Fourier neural operator network; S22. Initialize the population size of the cheetah optimization algorithm as N, and generate an initial set of individual structural parameters {Θ1, Θ2,..., Θ N}, where each cheetah individual represents a set of structural parameter combinations; S23. In each generation of iteration, calculate the fitness function value for each cheetah individual in the population. The cheetah individual is a set of structural parameter combinations of the Fourier neural operator network, and the fitness function f(Θ i ) is defined as: Among them, represents the predicted output of the Fourier neural operator network constructed by the structural parameter combination Θ i on the j-th validation sample, and y j represents the true value of the sample. n is the number of samples in the validation set, and ‖‖ is the L2 norm. According to the sorting result of the fitness function value f(Θ i ), the top k cheetah individuals with the best fitness are selected from all cheetah individuals, and the corresponding structural parameter combinations form a multi-prey set Among them, Θ (j) represents the combination of prey structure parameters ranked at the j-th position, where j ∈ {1, 2,..., k}; S24. For each cheetah individual in the current population, with the structural parameter combination being Θ i , calculate the Euclidean distance between the structural parameter combination of the current i-th cheetah individual and the structural parameter combination of each prey in the multi-prey set , and select the structural parameter combination of the prey with the closest distance Update the position of the cheetah individual: where r1 is the tracking intensity factor, and t is the current iteration round. represents the combination of structural parameters of the i-th cheetah individual at the t-th iteration round, represents the combination of structural parameters of the i-th cheetah individual at the (t + 1)-th iteration round; S25. Establish a local historical structure parameter memory pool with a length of m for each cheetah individual, denoted as M i , which is used to store the combination of its structure parameters in the most recent m iteration rounds: Among them, represents the combination of structural parameters of the i-th cheetah individual at the (t - L)-th iteration, where L ∈ {1, 2,..., m}; S26. Define a neighborhood set for each cheetah individual Among them, Θ j represents the combination of structural parameters of the j-th cheetah individual in the current iteration, and δ is a preset distance threshold used to judge the adjacency relationship between cheetah individuals; S27. Based on the neighborhood set of the i-th cheetah individual in the local historical structure parameter memory pool of each cheetah individual, perform an information-sharing position update operation on the structure parameter combination of the i-th cheetah individual: Among them, represents the position obtained after the information sharing fine-tuning of the i-th cheetah individual in the (t + 1)-th iteration. λ is the information sharing intensity factor, and mean(M j ) represents the mean of the historical structure parameter combinations of the j-th neighbor individual, and M j represents the historical structure parameter memory pool of the j-th neighbor individual; S28. When the fitness improvement rate is less than the set threshold ε in continuous G iterations, trigger the reconnaissance and patrol mechanism, and perform a jump perturbation update operation on the cheetah individuals: Among them, represents the combination of structural parameters of the i-th cheetah individual in the (t + 1)-th iteration, sign(r - 0.5) is the sign function, σ is the jump intensity factor, r is a uniformly distributed random number, and randn() represents the standard normal distribution function; S29. Repeat steps S23 to S28 until the maximum number of iterations T is reached or the preset convergence condition is satisfied, and select the structural parameter combination Θ with the best score. * As the optimal structural configuration of the Fourier neural operator network; S210. According to the optimal structural parameters, build the network structure of the Fourier neural operator network, and use the resource occupancy status and communication link bandwidth information as training data to complete the training of the Fourier neural operator network.

3. The edge computing data transmission and collaborative optimization method based on the ad hoc network technology according to claim 2, wherein The structural parameter set of the Fourier neural operator network specifically includes the number of network layers, the number of Fourier projection dimensions, the number of hidden nodes in each layer, the regularization coefficient, and the learning rate, which are used to control the expression ability of the Fourier neural operator network and the stability of the training process.

4. The edge computing data transmission and collaborative optimization method based on the ad hoc network technology according to claim 1, wherein The specific content of S3 includes: S31. Arrange the collected resource occupancy status and communication link bandwidth information in chronological order to build multi-dimensional time series data samples; S32. Perform data formatting and normalization processing on the constructed time series data samples to form a unified standard time series feature set; S33. Use the processed time series feature set as the input and input it into the trained Fourier neural operator network; S34. Calculate the features of the input time series feature set through the Fourier neural operator network to obtain the corresponding change trend of the edge computing system state; S35. Output the network state prediction results of each node in the edge computing system to form an edge network state sequence.

5. The edge computing data transmission and collaborative optimization method based on the ad hoc network technology according to claim 3, wherein The time series data sample specifically includes the resource occupancy rate of the edge node at each time point and the bandwidth value of the corresponding communication link, which is used to reflect the operation state characteristics of the edge computing environment changing with time.

6. The edge computing data transmission and collaborative optimization method based on the ad hoc network technology according to claim 3, wherein The feature calculation of the input time series feature set through the Fourier neural operator network specifically refers to inputting the set containing resource occupancy status and communication link bandwidth information into the Fourier neural operator network, extracting deep features that change over time through frequency domain mapping and non-linear mapping operations, and generating prediction results of future states, which are used to extract potential change patterns in the time series features of edge nodes.

7. The edge computing data transmission and collaborative optimization method based on the ad hoc network technology according to claim 1, wherein The specific steps of S4 include: S41. Construct a set of scheduling tasks Set of edge nodes And a set of transmission paths between each pair of scheduling tasks and edge nodes Among them, T j Represents the j-th scheduling task, N i Represents the i-th edge node, j ∈ {1, 2,..., a}, i ∈ {1, 2,..., b}, a represents the number of scheduling tasks, b represents the number of edge nodes, P ij Represents the path scheme assigned by the scheduling task T j To the edge node N i ; S42. Construct the solution space of the task scheduling and transmission path combination As a search individual in the cheetah optimization algorithm, each individual corresponds to a set of scheduling and transmission combination schemes; S43. Define a fitness function Among them, α, β, γ, δ, and λ are adjustable control factors, D j is the task execution delay, C j is the communication consumption, H j is the number of path hops, S ij is the path stability, and Var(L) is the variance of the resource utilization rates of all edge nodes; S44. Initialize the population using the cheetah optimization algorithm with the mechanism of introducing multiple prey, and select the top k combination schemes with the best fitness as the multiple prey set Each cheetah individual determines the target prey according to the principle of the minimum Euclidean distance and performs the tracking update operation; S45. Define the neighborhood set based on the information sharing mechanism Extract the mean value of the neighbor historical structure parameters and perform fine-tuning updates; S46. When the fitness improvement is less than the threshold ∈ in G consecutive iterations, trigger the detection mechanism, perform perturbation jumps on some cheetah individuals, and finally select the combination plan with the optimal fitness. As the optimal scheduling strategy.

8. The edge computing data transmission and collaborative optimization method based on the ad hoc network technology according to claim 1, characterized in that The specific steps of S5 include: S51. Send the obtained optimal scheduling policy to each relevant edge node in the edge ad hoc network; S52. The edge nodes that receive the scheduling instructions receive and locally buffer the task data according to the assigned task execution roles, and prepare to perform calculation or forwarding operations; S53. The edge nodes undertaking the task execution role complete the data processing and resource scheduling processes locally according to the assigned task content; S54. The edge nodes undertaking the relay role forward the task data from the source node to the target node in a multi-hop relay manner according to the optimal scheduling policy; S55. During the task execution and data transmission processes, the system monitors the running state information of each edge node and the connection stability of the communication links in real time; S56. When resource anomalies or communication link disconnections are detected in the edge nodes, trigger the ad hoc network topology update mechanism, dynamically adjust the network structure and reconstruct the data transmission path, and at the same time perform task migration operations.

Images