Network information processing method and system based on edge computing gateway

Abstract

The invention discloses a network information processing method and system based on an edge computing gateway, and the method comprises the steps: building a dynamic data flow matrix through collecting multi-dimensional heterogeneous network flow data, inputting the dynamic data flow matrix into a recurrent neural network which is optimized by a gating mechanism and contains gradient cutting and dynamic learning rate adjustment, and generating a time sequence correlation feature vector; a space-time correlation characteristic spectrum subjected to bandwidth delay product weighting correction is constructed based on the vector, and clustering analysis is performed by using an improved DBSCAN algorithm containing dynamic threshold adjustment and adaptive radius calculation. A clustering cluster with credibility weight is generated through a dynamic credibility evaluation model fusing parameters such as energy consumption, load and communication delay, and finally a processing scheme including resource allocation, path planning and priority ranking is output by a multi-objective decision optimization engine. According to the invention, the resource utilization rate, the processing timeliness and the result reliability of the edge computing gateway in network information processing are remarkably improved, and an efficient solution is provided for an edge computing scene.

Description

Technical Field

[0001] This invention relates to the field of network information processing, and in particular to a network information processing method and system based on an edge computing gateway. Background Technology

[0002] With the rapid development of IoT and 5G technologies, network data volume is exploding. The high latency and high bandwidth pressure caused by long-distance data transmission in traditional cloud computing models are becoming increasingly prominent. Edge computing gateways, with their localized data processing capabilities, have become a key technology for solving these problems. Edge computing gateways need to process massive amounts of heterogeneous network information in real time, but existing technologies have significant bottlenecks in processing efficiency, resource utilization, and data accuracy.

[0003] The primary shortcoming of existing technologies lies in the poor compatibility between algorithms and edge computing gateway hardware parameters. Traditional recurrent neural networks, when extracting temporal features, lack awareness of dynamic parameters such as the memory capacity and computing power of the edge computing gateway. The dimension of the hidden state vector cannot adaptively adjust with the memory threshold, leading to wasted memory resources or decreased processing performance. In the DBSCAN clustering algorithm, the core point density threshold and neighborhood radius often use fixed parameters without combining real-time dynamic optimization with the edge node CPU load, network bandwidth, and other conditions. This makes it difficult for the clustering results to accurately reflect the characteristics of network data and adapt to complex and ever-changing network environments.

[0004] Secondly, existing systems lack multi-dimensional parameter collaborative optimization mechanisms. Most solutions focus on optimizing only a single metric during network information processing, such as unilaterally pursuing computational efficiency while neglecting energy consumption control, or focusing solely on data transmission latency without considering resource allocation balance. In the resource scheduling phase, parameters such as the energy consumption threshold, load rate, and communication latency of the edge computing gateway are not deeply integrated, resulting in computational resource allocation strategies and data transmission path planning schemes that fail to achieve optimal overall system performance and cannot meet the comprehensive requirements of modern networks for low latency, high reliability, and energy efficiency. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides a network information processing method and system based on edge computing gateways.

[0006] The technical solution adopted in this invention is a network information processing method based on an edge computing gateway, comprising the following steps:

[0007] Step S1: Collect multidimensional heterogeneous network traffic data through the edge computing gateway and construct a dynamic data flow matrix containing spatiotemporal characteristics. The dynamic data flow matrix is ​​adaptively adjusted in dimensions based on the edge node computing capability threshold.

[0008] Step S2: Input the dynamic data stream matrix into a recurrent neural network optimized by a gating mechanism. The recurrent neural network includes a gradient pruning strategy based on edge computing resource constraints and a dynamic learning rate adjustment module to generate a temporal correlation feature vector.

[0009] Step S3: Construct a feature similarity matrix based on the temporal correlation feature vector. The similarity matrix is ​​weighted and corrected by the bandwidth-delay product of the edge computing gateway to form a spatiotemporal correlation feature map.

[0010] Step S4: The improved DBSCAN algorithm is used to perform cluster analysis on the spatiotemporal correlation feature map. The improved algorithm includes a core point density threshold dynamic adjustment mechanism based on the processing capability of edge nodes and a neighborhood radius adaptive calculation model based on network topology.

[0011] Step S5: Construct a dynamic trust evaluation model for the clustering results. The model is based on the energy consumption threshold, computing load and communication latency of the edge computing gateway to fuse multi-dimensional parameters and generate clusters with trust weights.

[0012] Step S6: Based on the clusters with credibility weights, generate a network information processing scheme that includes computing resource allocation strategy, data transmission path planning and processing priority ranking through the multi-objective decision optimization engine of the edge computing gateway.

[0013] Furthermore, in step S2, the recurrent neural network optimized by the gating mechanism adopts the following model structure:

[0014] H t =σ g (W g ·[H t-1 X t ]+b g )⊙tanh(W h ·[H t-1 X t ]+b h )

[0015] Among them, H t Let σ be the hidden state vector at time t, and its dimension is positively correlated with the memory capacity threshold of the edge computing gateway; g The gated activation function is a piecewise linear approximation function based on the computational capabilities of edge nodes; W g and W h The weight matrix's parameter initialization strategy is determined based on the energy consumption-accuracy Pareto optimal curve of the edge computing gateway; ⊙ represents element-wise multiplication, where the cache hit rate of the edge computing gateway is used as the data prefetching decision factor during the operation; X tLet t be the input feature vector at time t, and its feature selection strategy is dynamically adjusted based on the bandwidth utilization of the edge computing gateway.

[0016] Furthermore, in step S4, the improved DBSCAN algorithm adopts the following core point density threshold calculation model: Where, ∈ t is the neighborhood radius threshold at time t, which has an exponential relationship with the real-time CPU utilization of the edge computing gateway; ∈0 is the basic neighborhood radius, determined by the initial configuration parameters of the edge computing gateway; α and β are adjustment coefficients, corresponding to the computing resource pressure factor and the network bandwidth pressure factor, respectively; C t Let C be the CPU utilization of the edge computing gateway at time t. max B represents the maximum allowed CPU utilization. t Let B be the network bandwidth utilization of the edge computing gateway at time t. max The maximum allowable utilization of network bandwidth; the update cycle of the neighborhood radius threshold is synchronized with the task scheduling cycle of the edge computing gateway.

[0017] Furthermore, in step S5, the dynamic trust assessment model employs the following multi-dimensional parameter fusion function:

[0018]

[0019] Wherein, T(c i ) represents cluster c i The trust score ranges from [0, 1]; E(c i To process cluster c i The required power consumption of the edge computing gateway is calculated using a power-time integral model; L(c i To process cluster c i The load rate of the edge computing gateway is calculated based on the task queue length and processing latency; D(c i ) represents cluster c i The data transmission latency is dynamically estimated based on network topology and bandwidth usage; ω1, ω2, and ω3 are weighting coefficients that satisfy ω1+ω2+ω3=1 and are dynamically adjusted based on the operating mode of the edge computing gateway; λ1, λ2, and λ3 are attenuation coefficients that correspond to energy consumption sensitivity, load sensitivity, and latency sensitivity, respectively; μ is the load balancing point parameter, which corresponds to the optimal computing efficiency point of the edge computing gateway.

[0020] Furthermore, in step S6, the multi-objective decision optimization engine adopts the following resource allocation model: Constraints: Among them, R ij C represents the amount of resources allocated to the i-th cluster and the j-th edge computing node; j D represents the total available computing resources of the j-th edge computing node; ij D represents the estimated processing latency for assigning the i-th cluster to the j-th edge computing node; max E represents the maximum allowable processing latency threshold of the system. ij E represents the estimated energy consumption of assigning the i-th cluster to the j-th edge computing node; max Indicates the maximum allowable energy consumption threshold of the system; w 1i w 2i w 3i These are the resource consumption weight, latency weight, and energy consumption weight for the i-th cluster, respectively, which are dynamically adjusted based on the service quality requirements of the edge computing gateway.

[0021] Furthermore, the construction of the spatiotemporal correlation feature map in step S3 adopts the following similarity calculation model:

[0022]

[0023] Where S(x) i x j ) represents data point x i and x j The spatiotemporal similarity between them; ||x i -x j || represents the Euclidean distance in the feature space; γ is the kernel function bandwidth parameter, which is inversely proportional to the memory access speed of the edge computing gateway; T ij Represents data point x i and x j The time correlation between the two is calculated by the time series analysis module of the edge computing gateway; δ is the time correlation adjustment factor, which is dynamically adjusted based on the real-time requirements of the edge computing gateway; β is the time correlation sensitivity parameter, which is related to the clock synchronization accuracy of the edge computing gateway; τ is the time correlation threshold, which is determined by the task scheduling cycle of the edge computing gateway.

[0024] Furthermore, in step S2, the gradient pruning strategy of the recurrent neural network adopts the following adaptive threshold calculation model: Where, θ t θt is the gradient clipping threshold at time t; θ0 is the basic clipping threshold, determined based on the initial training configuration of the edge computing gateway; M t M represents the memory usage of the edge computing gateway at time t. maxV represents the maximum available memory for the edge computing gateway. t V is the norm of the gradient vector at time t; max α is the preset upper limit of the gradient norm; α and β are adjustment coefficients, corresponding to the memory pressure factor and gradient stability factor, respectively.

[0025] Furthermore, in step S4, the improved DBSCAN algorithm adopts the following neighborhood radius adaptive calculation model: Where, ∈ ij Represents data point x i Relative to cluster center c j The adaptive neighborhood radius; S ij Represents data point x i With cluster center c j Feature similarity; σ is the standard deviation of the similarity distribution, obtained statistically based on historical data from edge computing gateways; L j For cluster c j The current load is calculated using the task queue length and processing latency of the edge computing gateway; L max For cluster c j The maximum load capacity; λ and ρ are adjustment coefficients that control similarity sensitivity and load sensitivity, respectively; the parameter update frequency of the neighborhood radius adaptive calculation model is synchronized with the resource monitoring cycle of the edge computing gateway.

[0026] Furthermore, in step S6, the generation of the network information processing scheme adopts the following priority ranking model: Among them, P i For cluster c i Processing priority score; I i For cluster c i Information entropy reflects the degree of uncertainty in data; U i For cluster c i The update frequency is calculated by the time series analysis module of the edge computing gateway; D i For cluster c i The data transmission distance is determined based on the network topology of the edge computing gateway; D0 is a preset distance threshold, which is related to the communication energy consumption optimization target of the edge computing gateway; w1, w2, and w3 are weighting coefficients, which are dynamically adjusted based on the real-time operating status of the edge computing gateway; α is a distance sensitivity parameter, which is related to the wireless communication protocol characteristics of the edge computing gateway.

[0027] A network information processing system based on an edge computing gateway, the system comprising:

[0028] Multidimensional heterogeneous data acquisition unit: used to collect multidimensional heterogeneous data including network traffic, device status and user behavior through the multi-channel data interface of the edge computing gateway, and to build a dynamic data flow matrix;

[0029] Spatiotemporal feature extraction unit: connected to the multidimensional heterogeneous data acquisition unit, used to input the dynamic data stream matrix into a recurrent neural network optimized by a gating mechanism to generate a feature vector containing temporal correlation features;

[0030] Feature map construction unit: connected to the spatiotemporal feature extraction unit, used to construct a feature similarity matrix based on the temporal correlation feature vector, and perform weighted correction through the bandwidth-delay product of the edge computing gateway to generate a spatiotemporal correlation feature map;

[0031] Adaptive clustering analysis unit: connected to the feature map construction unit, used to perform clustering analysis on the spatiotemporal correlation feature map using the improved DBSCAN algorithm. The improved algorithm includes a core point density threshold dynamic adjustment mechanism based on edge node processing capability and a neighborhood radius adaptive calculation model based on network topology.

[0032] Multidimensional Trust Evaluation Unit: Connected to the adaptive clustering analysis unit, it is used to construct a multidimensional parameter fusion model based on the edge computing gateway energy consumption threshold, computing load and communication latency, and generate clusters with trust weights;

[0033] Multi-objective decision optimization unit: connected to the multi-dimensional trust evaluation unit, used to generate a network information processing scheme including computing resource allocation strategy, data transmission path planning and processing priority ranking based on the cluster with trust weight, through the multi-objective decision optimization engine of the edge computing gateway, and implement the processing scheme through the control interface of the edge computing gateway.

[0034] Beneficial Effects: This invention proposes a network information processing method and system based on an edge computing gateway. By deeply fusing an optimized recurrent neural network with an improved DBSCAN algorithm, combined with a dynamic parameter adaptive mechanism of the edge computing gateway, this invention effectively overcomes the shortcomings of existing technologies in algorithm adaptability and multi-dimensional parameter collaborative optimization. Regarding algorithm adaptability, the system constructs a neural network model structure dynamically correlated with the memory capacity and computing power of the edge computing gateway, achieving adaptive adjustment of the hidden state vector dimension, avoiding resource waste or performance bottlenecks caused by fixed parameters in traditional solutions. The improved DBSCAN algorithm dynamically adjusts the core point density threshold and neighborhood radius by real-time monitoring of CPU utilization and network bandwidth usage, making the clustering results more closely match the spatiotemporal distribution characteristics of network data, significantly improving the accuracy of information processing in complex network environments. Regarding multi-dimensional parameter collaborative optimization, the system innovatively integrates key edge computing gateway parameters such as energy consumption threshold, load rate, and communication latency into the model design. The dynamic trust assessment model achieves precise quantification of the credibility of clustering results by fusing multidimensional indicators through exponential decay functions and logistic regression. The multi-objective decision optimization engine employs a Pareto optimal strategy, balancing computational efficiency, energy consumption control, and latency sensitivity during resource allocation to generate a globally optimal processing solution. The construction of the spatiotemporal correlation feature map introduces bandwidth-delay product weighted correction, making feature similarity calculation more consistent with the data transmission characteristics of edge computing environments, further enhancing the system's adaptability to dynamic network environments. This invention significantly improves the system's processing efficiency, resource utilization, and data accuracy, providing an efficient and reliable solution for network information processing in edge computing environments. Attached Figure Description

[0035] Figure 1 This is a flowchart of the method steps of the present invention;

[0036] Figure 2 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] like Figure 1 As shown, a network information processing method based on an edge computing gateway is characterized by the following steps:

[0039] Step S1: Collect multidimensional heterogeneous network traffic data through the edge computing gateway and construct a dynamic data flow matrix containing spatiotemporal characteristics. The dynamic data flow matrix is ​​adaptively adjusted in dimensions based on the edge node computing capability threshold.

[0040] Specifically, the edge computing gateway, as the core device for data acquisition, utilizes its deployed multi-channel data interfaces to collect various types of data existing in the network, including but not limited to network traffic data, device operating status data, and user behavior data. These data exhibit heterogeneous characteristics in terms of format, structure, and source. The edge computing gateway's built-in data acquisition module converts them into a unified and processable format. During the acquisition process, the focus is on extracting the temporal dimension information (such as the time of data generation and time interval) and spatial dimension information (such as the location of the network node from which the data originates and the device deployment location) contained in the data, thereby constructing a dynamic data flow matrix containing spatiotemporal characteristics.

[0041] The dynamic data flow matrix is ​​not a fixed structure; its dimensions are adaptively adjusted based on the computing power threshold of the edge nodes. The edge computing gateway monitors its own computing resource status in real time, such as CPU processing power and memory capacity. When computing power is sufficient, the matrix dimensions can be appropriately expanded to accommodate more data features and obtain richer information. Conversely, when computing resources are strained, approaching or reaching the computing power threshold, the matrix dimensions will be reduced accordingly, removing some secondary or redundant feature dimensions. This ensures that data processing can be carried out efficiently with limited resources, avoiding situations where excessive data volume leads to excessive computing load, increased processing latency, or even system crashes.

[0042] Step S2: Input the dynamic data stream matrix into a recurrent neural network optimized by a gating mechanism. The recurrent neural network includes a gradient pruning strategy based on edge computing resource constraints and a dynamic learning rate adjustment module to generate a temporal correlation feature vector.

[0043] Specifically, the dynamic data flow matrix constructed in step S1 serves as the input data for the recurrent neural network, entering the network structure optimized by the gating mechanism. The gating mechanism plays a crucial role in filtering and controlling the flow of information within the network. By setting multiple gating units, it selectively retains or discards input data and historical information. When processing the dynamic data flow matrix, the gating units determine, based on the characteristics of the data and the current network state, which information needs to be passed to the next time step and which information can be discarded. This enhances the network's ability to process time-series data and effectively captures the temporal dependencies within the data.

[0044] This recurrent neural network also integrates a gradient pruning strategy based on edge computing resource constraints and a dynamic learning rate adjustment module. The gradient pruning strategy monitors the magnitude of gradients generated during network training in real time. Due to the limited computing resources of the edge computing gateway, excessively large gradients may lead to numerical instability or excessive resource consumption during parameter updates. By setting a gradient threshold related to edge computing resources, pruning is performed when the gradient exceeds this threshold, ensuring that parameter updates are within a reasonable range while reducing computational complexity. The dynamic learning rate adjustment module dynamically adjusts the learning rate based on the operating status of the edge computing gateway and the network training effect. In the early stages of training, a larger learning rate can be set to accelerate convergence; as training progresses and resource consumption changes, the learning rate is dynamically reduced to improve training accuracy. This ensures both training effectiveness and rational utilization of computing resources, ultimately outputting feature vectors containing temporal correlation information of the data.

[0045] Step S3: Construct a feature similarity matrix based on the temporal correlation feature vector. The similarity matrix is ​​weighted and corrected by the bandwidth-delay product of the edge computing gateway to form a spatiotemporal correlation feature map.

[0046] Specifically, after obtaining the time-series correlation feature vectors output in step S2, a feature similarity matrix is ​​constructed based on these vectors. During the construction process, similarity indices (such as Euclidean distance and cosine similarity) between different feature vectors are calculated to quantify the degree of similarity between each data feature. These similarity values ​​are then filled into the corresponding positions in the matrix, thus forming the feature similarity matrix. This matrix can intuitively reflect the similarity relationships between data features, providing a data foundation for subsequent clustering analysis.

[0047] To better reflect the data transmission characteristics of edge computing networks, the feature similarity matrix is ​​weighted and corrected using the bandwidth-latency product of the edge computing gateway. Bandwidth and latency are key performance parameters in edge computing networks; bandwidth reflects the data transmission rate, and latency reflects the time overhead of data transmission. By using the bandwidth-latency product as a weighting factor to weight the elements in the feature similarity matrix, the similarity between data features in high-bandwidth, low-latency network environments has a greater impact, while the impact of similarity between data features in low-bandwidth, high-latency environments is relatively reduced. After weighted correction, a spatiotemporal correlation feature map is formed. This map not only contains the similarity relationships of data features but also incorporates network transmission characteristics, more accurately depicting the spatiotemporal correlation characteristics of data.

[0048] Step S4: The improved DBSCAN algorithm is used to perform cluster analysis on the spatiotemporal correlation feature map. The improved algorithm includes a core point density threshold dynamic adjustment mechanism based on the processing capability of edge nodes and a neighborhood radius adaptive calculation model based on network topology.

[0049] Specifically, this step employs an improved DBSCAN algorithm to perform cluster analysis on the spatiotemporal correlation feature map formed in step S3. Traditional DBSCAN algorithms, when processing complex and variable network data, typically use fixed values ​​for the core point density threshold and neighborhood radius, making it difficult to adapt to different network environments and data distributions. The improved algorithm introduces a dynamic adjustment mechanism for the core point density threshold based on the processing capabilities of edge nodes. The edge computing gateway monitors its own CPU utilization, memory usage, and other processing capability-related parameters in real time. When processing capabilities are strong, the core point density threshold is appropriately lowered, enabling the algorithm to identify more relatively sparse data clusters; when processing capabilities are limited, the core point density threshold is increased to reduce the amount of clustering computation, ensuring efficient operation of the algorithm under limited resources.

[0050] Furthermore, the improved algorithm also features an adaptive neighborhood radius calculation model based on network topology. Network topology determines the data transmission path and the connections between nodes, significantly impacting data distribution characteristics. This model dynamically calculates the neighborhood radius of each data point based on network topology information, such as physical distances between nodes and link bandwidth. For data points located in areas with dense network connections and convenient data transmission, a smaller neighborhood radius is set to improve clustering accuracy; for data points in areas with sparse network connections and significant transmission delays, the neighborhood radius is increased to ensure that related data are included in the same cluster. This makes the clustering results more consistent with the actual distribution of network data, effectively improving the quality and efficiency of cluster analysis.

[0051] Step S5: Construct a dynamic trust evaluation model for the clustering results. The model is based on the energy consumption threshold, computing load and communication latency of the edge computing gateway to fuse multi-dimensional parameters and generate clusters with trust weights.

[0052] Specifically, after completing the cluster analysis in step S4, a dynamic trust evaluation model for the clustering results is constructed to assess the reliability and credibility of each cluster. This model uses three key parameters—energy consumption threshold, computational load, and communication latency—as the evaluation criteria. The energy consumption threshold reflects the maximum energy consumption range that the edge computing gateway can withstand under sustainable operating conditions; the computational load reflects the busyness of the gateway's current processing tasks, measured by indicators such as CPU utilization and task queue length; and the communication latency represents the time required for data transmission in the network. Incorporating these three parameters into the evaluation system allows for a comprehensive assessment of the credibility of the clustering results from multiple dimensions.

[0053] The model employs a multi-dimensional parameter fusion approach to quantify and weight energy consumption thresholds, computational load, and communication latency. Based on the edge computing gateway's operating mode and business requirements, each parameter is assigned a corresponding weight. These parameters are then integrated into a comprehensive trust index using specific fusion algorithms (such as weighted summation and nonlinear mapping). Clusters that perform well in terms of energy consumption, computational load, and communication latency—those that better meet system operating requirements—are assigned higher trust weights; conversely, those that perform poorly are assigned lower weights. Ultimately, the model generates evaluation results with trust weights for each cluster, providing more valuable data support for subsequent decisions and ensuring that processing solutions are based on reliable clustering results.

[0054] Step S6: Based on the clusters with credibility weights, generate a network information processing scheme that includes computing resource allocation strategy, data transmission path planning and processing priority ranking through the multi-objective decision optimization engine of the edge computing gateway.

[0055] Specifically, the clusters with credibility weights obtained in step S5 are used as the basis for decision optimization in step S6 and input into the multi-objective decision optimization engine of the edge computing gateway. This engine aims to achieve collaborative optimization of multiple objectives, including the rational allocation of computing resources, efficient planning of data transmission paths, and scientific prioritization of data processing. Regarding computing resource allocation, the engine determines the CPU time, memory space, and other resources that should be allocated to each cluster based on the data size, processing complexity, and credibility weight of each cluster, combined with the total available computing resources of the edge computing gateway, ensuring that resources are fully and rationally utilized.

[0056] In data transmission path planning, the multi-objective decision optimization engine comprehensively considers factors such as network topology, link bandwidth, and transmission latency, while also incorporating the credibility weights of clusters to plan the optimal data transmission path for each cluster. This ensures that data can quickly reach the target node while maintaining transmission quality. For processing priority ranking, the engine determines the processing order of various data types based on the importance, urgency, and credibility weights of the data contained within each cluster, prioritizing important and highly credible data to guarantee the efficient operation of critical business processes. Through comprehensive optimization decisions on these three aspects by the multi-objective decision optimization engine, a complete network information processing scheme is generated, encompassing computing resource allocation strategies, data transmission path planning, and processing priority ranking. This scheme is implemented through the control interface of the edge computing gateway, achieving efficient network information processing and overall system performance improvement.

[0057] Preferably, the recurrent neural network optimized by the gating mechanism in step S2 adopts the following model structure:

[0058] Ht =σ g (W g ·[H t-1 X t ]+b g )⊙tanh(W h ·[H t-1 X t ]+b h )

[0059] Among them, H t Let σ be the hidden state vector at time t, and its dimension is positively correlated with the memory capacity threshold of the edge computing gateway; g The gated activation function is a piecewise linear approximation function based on the computational capabilities of edge nodes; W g and W h The weight matrix's parameter initialization strategy is determined based on the energy consumption-accuracy Pareto optimal curve of the edge computing gateway; ⊙ represents element-wise multiplication, where the cache hit rate of the edge computing gateway is used as the data prefetching decision factor during the operation; X t Let t be the input feature vector at time t, and its feature selection strategy is dynamically adjusted based on the bandwidth utilization of the edge computing gateway.

[0060] Specifically, the recurrent neural network structure optimized by the gating mechanism in step S2 is refined. This network constructs a specific computational model that establishes a positive correlation between the dimension of the hidden state vector and the memory capacity threshold of the edge computing gateway. When memory is sufficient, the dimension is expanded to capture more features, and when resources are scarce, the dimension is contracted to avoid overload. The gating activation function adopts a piecewise linear approximation based on the computing capabilities of edge nodes to precisely control the efficiency of information flow. The initialization strategy of the weight matrix is ​​bound to the optimal energy consumption-accuracy curve of the edge computing gateway, achieving deep adaptation of parameter settings to hardware performance. The cache hit rate is introduced as a data prefetching decision factor in the element-wise multiplication operation to reduce memory access latency. The selection strategy of the input feature vector is dynamically adjusted according to the gateway bandwidth utilization rate, ensuring data integrity while avoiding network congestion caused by excessive data volume, ultimately achieving efficient extraction of time-series correlated feature vectors.

[0061] Preferably, the improved DBSCAN algorithm in step S4 adopts the following core point density threshold calculation model: Where, ∈ t is the neighborhood radius threshold at time t, which has an exponential relationship with the real-time CPU utilization of the edge computing gateway; ∈0 is the basic neighborhood radius, determined by the initial configuration parameters of the edge computing gateway; α and β are adjustment coefficients, corresponding to the computing resource pressure factor and the network bandwidth pressure factor, respectively; C tLet C be the CPU utilization of the edge computing gateway at time t. max B represents the maximum allowed CPU utilization. t Let B be the network bandwidth utilization of the edge computing gateway at time t. max The maximum allowable utilization of network bandwidth; the update cycle of the neighborhood radius threshold is synchronized with the task scheduling cycle of the edge computing gateway.

[0062] Specifically, an innovative design was implemented for the core point density threshold calculation in step S4 of the improved DBSCAN algorithm. This model establishes an exponential correlation between the neighborhood radius threshold and the real-time CPU utilization of the edge computing gateway. When CPU load increases, the threshold is automatically increased to reduce computational load; when load decreases, the threshold is decreased to improve clustering accuracy. The basic neighborhood radius is determined by the gateway's initial configuration parameters, providing a benchmark for dynamic adjustment. Computational resource pressure factor and network bandwidth pressure factor are used as adjustment coefficients to quantify the impact of CPU utilization and bandwidth usage on the threshold, respectively. By incorporating the ratios of CPU utilization and bandwidth usage to their corresponding maximum allowable values ​​into the calculation, and combining this with product operations, the threshold is dynamically adjusted. Furthermore, the threshold update cycle is synchronized with the gateway's task scheduling cycle, ensuring that the algorithm maintains high-efficiency clustering performance under different loads.

[0063] Preferably, the dynamic trust evaluation model in step S5 adopts the following multi-dimensional parameter fusion function:

[0064]

[0065] Wherein, T(c i ) represents cluster c i The trust score ranges from [0, 1]; E(c i To process cluster c i The required power consumption of the edge computing gateway is calculated using a power-time integral model; L(c i To process cluster c i The load rate of the edge computing gateway is calculated based on the task queue length and processing latency; D(c i ) represents cluster c i The data transmission latency is dynamically estimated based on network topology and bandwidth usage; ω1, ω2, and ω3 are weighting coefficients that satisfy ω1+ω2+ω3=1 and are dynamically adjusted based on the operating mode of the edge computing gateway; λ1, λ2, and λ3 are attenuation coefficients that correspond to energy consumption sensitivity, load sensitivity, and latency sensitivity, respectively; μ is the load balancing point parameter, which corresponds to the optimal computing efficiency point of the edge computing gateway.

[0066] Specifically, the dynamic trust assessment model in step S5 quantifies the trustworthiness of clusters through a multi-dimensional parameter fusion function. The trust score comprehensively considers the gateway energy consumption, operating load rate, and data transmission latency required to process the clusters. Energy consumption is quantified using a power-time integral model, the load rate is calculated using the task queue length and processing latency, and the transmission latency is dynamically estimated based on network topology and bandwidth. The weight coefficients sum to 1 and are dynamically adjusted based on the gateway's operating mode, reflecting the differences in the importance of each parameter under different scenarios. Attenuation coefficients correspond to the sensitivity adjustment of energy consumption, load, and latency, respectively. The load balancing point parameter is associated with the gateway's optimal computational efficiency point. Through a combination of exponential and logistic functions, the multi-dimensional parameters are mapped to a trust score in the 0-1 range, providing a quantitative evaluation of the clustering results.

[0067] Preferably, the multi-objective decision optimization engine in step S6 adopts the following resource allocation model: Constraints: Among them, R ij C represents the amount of resources allocated to the i-th cluster and the j-th edge computing node; j D represents the total available computing resources of the j-th edge computing node; ij D represents the estimated processing latency for assigning the i-th cluster to the j-th edge computing node; max E represents the maximum allowable processing latency threshold of the system. ij E represents the estimated energy consumption of assigning the i-th cluster to the j-th edge computing node; max Indicates the maximum allowable energy consumption threshold of the system; w 1i w 2i w 3i These are the resource consumption weight, latency weight, and energy consumption weight for the i-th cluster, respectively, which are dynamically adjusted based on the service quality requirements of the edge computing gateway.

[0068] Specifically, regarding the multi-objective decision optimization engine resource allocation model in step S6, the model aims to minimize the weighted comprehensive cost of resource consumption, latency, and energy consumption. The model iterates through all clusters and edge computing nodes using a double summation method. The resource consumption weight, latency weight, and energy consumption weight are dynamically adjusted based on service quality requirements, reflecting the different business needs for each indicator. The ratios of resource quantity, latency, and energy consumption to the corresponding node capacity and threshold in the numerator quantify the rationality of the resource allocation scheme. Constraints ensure that the resource allocation to each node does not exceed the total available resources, and the total resource allocation for each cluster is fixed. By solving this constrained optimization problem, a computing resource allocation strategy that satisfies multi-objective balance is generated.

[0069] Preferably, the spatiotemporal correlation feature map construction in step S3 adopts the following similarity calculation model:

[0070]

[0071] Where S(x) i x j ) represents data point x i and x j The spatiotemporal similarity between them; ||x i -x j || represents the Euclidean distance in the feature space; γ is the kernel function bandwidth parameter, which is inversely proportional to the memory access speed of the edge computing gateway; T ij Represents data point x i and x j The time correlation between the two is calculated by the time series analysis module of the edge computing gateway; δ is the time correlation adjustment factor, which is dynamically adjusted based on the real-time requirements of the edge computing gateway; β is the time correlation sensitivity parameter, which is related to the clock synchronization accuracy of the edge computing gateway; τ is the time correlation threshold, which is determined by the task scheduling cycle of the edge computing gateway.

[0072] Specifically, in step S3, the similarity calculation model for the spatiotemporal correlation feature map combines Euclidean distance and temporal correlation to construct a similarity metric. The kernel function bandwidth parameter is inversely proportional to the memory access speed of the edge computing gateway; faster memory access reduces bandwidth to improve similarity discriminativeness, while slower access increases bandwidth to adapt to complex features. Temporal correlation is calculated through the gateway's time series analysis module. The temporal correlation adjustment factor and sensitivity parameter are dynamically adjusted based on the gateway's real-time requirements and clock synchronization accuracy. The temporal correlation threshold is determined by the task scheduling cycle. Through the product operation of the exponential function and the logistic function, spatial distance and temporal correlation are fused into spatiotemporal similarity, providing an accurate metric for feature map construction.

[0073] Preferably, the gradient pruning strategy for the recurrent neural network in step S2 adopts the following adaptive threshold calculation model: Where, θ t θt is the gradient clipping threshold at time t; θ0 is the basic clipping threshold, determined based on the initial training configuration of the edge computing gateway; M t M represents the memory usage of the edge computing gateway at time t. max V represents the maximum available memory for the edge computing gateway. t V is the norm of the gradient vector at time t; max α is the preset upper limit of the gradient norm; α and β are adjustment coefficients, corresponding to the memory pressure factor and gradient stability factor, respectively.

[0074] Specifically, the gradient pruning strategy of the recurrent neural network in step S2 is optimized by using an adaptive threshold calculation model to balance training stability and resource consumption. The gradient pruning threshold starts from a base threshold and is dynamically adjusted based on the ratio of memory usage to maximum capacity and the ratio of the gradient vector norm to a preset upper limit. Memory pressure factor and gradient stability factor are used as adjustment coefficients to quantify the impact of memory usage and gradient fluctuations on the threshold, respectively. When memory usage increases or the gradient norm is too large, the threshold automatically increases to avoid excessive resource consumption during parameter updates; conversely, the threshold is decreased to accelerate training convergence and ensure stable and efficient training of the network under the limited resources of the edge computing gateway.

[0075] Preferably, the improved DBSCAN algorithm in step S4 adopts the following neighborhood radius adaptive calculation model: Where, ∈ ij Represents data point x i Relative to cluster center c j The adaptive neighborhood radius; S ij Represents data point x i With cluster center c j Feature similarity; σ is the standard deviation of the similarity distribution, obtained statistically based on historical data from edge computing gateways; L j For cluster c j The current load is calculated using the task queue length and processing latency of the edge computing gateway; L max For cluster c j The maximum load capacity; λ and ρ are adjustment coefficients that control similarity sensitivity and load sensitivity, respectively; the parameter update frequency of the neighborhood radius adaptive calculation model is synchronized with the resource monitoring cycle of the edge computing gateway.

[0076] Specifically, the neighborhood radius calculation in step S4 of the DBSCAN algorithm is improved by using an adaptive model to enhance clustering accuracy. The adaptive neighborhood radius is dynamically adjusted based on the base radius, taking into account the feature similarity between data points and cluster centers, and the current load of the clusters. The standard deviation of the similarity distribution is obtained statistically from historical gateway data and used as a benchmark for similarity normalization. Similarity sensitivity factors and load sensitivity factors control the degree of influence of feature similarity and load on the radius, respectively. When data point similarity is high or cluster load is low, the neighborhood radius is reduced to improve clustering accuracy; conversely, the radius is increased to ensure data integrity, and the parameter update frequency is synchronized with the gateway resource monitoring cycle.

[0077] Preferably, the generation of the network information processing scheme in step S6 adopts the following priority ranking model: Among them, P iFor cluster c i Processing priority score; I i For cluster c i Information entropy reflects the degree of uncertainty in data; U i For cluster c i The update frequency is calculated by the time series analysis module of the edge computing gateway; D i For cluster c i The data transmission distance is determined based on the network topology of the edge computing gateway; D0 is a preset distance threshold, which is related to the communication energy consumption optimization target of the edge computing gateway; w1, w2, and w3 are weighting coefficients, which are dynamically adjusted based on the real-time operating status of the edge computing gateway; α is a distance sensitivity parameter, which is related to the wireless communication protocol characteristics of the edge computing gateway.

[0078] Specifically, the network information processing schemes in step S6 are prioritized, and the processing order of clusters is determined by a comprehensive model. The priority score comprehensively considers the information entropy, update frequency, and data transmission distance of the clusters. Information entropy reflects data uncertainty, the update frequency is calculated by the time series analysis module, and the transmission distance is determined based on the network topology. The weighting coefficients are dynamically adjusted based on the real-time operating status of the gateway, reflecting the importance of each indicator in different scenarios. The distance sensitivity parameter is related to the characteristics of the gateway's wireless communication protocol, and the preset distance threshold is related to the communication energy consumption optimization target. A priority score is generated for each cluster through weighted summation and normalization calculation, guiding the optimization of the data processing order.

[0079] like Figure 2 As shown, a network information processing system based on an edge computing gateway includes:

[0080] Multidimensional heterogeneous data acquisition unit: used to collect multidimensional heterogeneous data including network traffic, device status and user behavior through the multi-channel data interface of the edge computing gateway, and to build a dynamic data flow matrix;

[0081] Spatiotemporal feature extraction unit: connected to the multidimensional heterogeneous data acquisition unit, used to input the dynamic data stream matrix into a recurrent neural network optimized by a gating mechanism to generate a feature vector containing temporal correlation features;

[0082] Feature map construction unit: connected to the spatiotemporal feature extraction unit, used to construct a feature similarity matrix based on the temporal correlation feature vector, and perform weighted correction through the bandwidth-delay product of the edge computing gateway to generate a spatiotemporal correlation feature map;

[0083] Adaptive clustering analysis unit: connected to the feature map construction unit, used to perform clustering analysis on the spatiotemporal correlation feature map using the improved DBSCAN algorithm. The improved algorithm includes a core point density threshold dynamic adjustment mechanism based on edge node processing capability and a neighborhood radius adaptive calculation model based on network topology.

[0084] Multidimensional Trust Evaluation Unit: Connected to the adaptive clustering analysis unit, it is used to construct a multidimensional parameter fusion model based on the edge computing gateway energy consumption threshold, computing load and communication latency, and generate clusters with trust weights;

[0085] Multi-objective decision optimization unit: connected to the multi-dimensional trust evaluation unit, used to generate a network information processing scheme including computing resource allocation strategy, data transmission path planning and processing priority ranking based on the cluster with trust weight, through the multi-objective decision optimization engine of the edge computing gateway, and implement the processing scheme through the control interface of the edge computing gateway.

[0086] This invention effectively overcomes the limitations of traditional technologies in adaptive processing and resource balancing through deep coupling design of algorithms and hardware parameters and multi-dimensional parameter collaborative optimization mechanism.

[0087] To address the issue of poor compatibility between traditional algorithms and edge computing gateway hardware, this invention constructs a dynamic correlation model to achieve precise adaptation. In the recurrent neural network structure, a positive correlation is established between the dimension of the hidden state vector and the memory capacity threshold of the edge computing gateway. The gating activation function adopts a piecewise linear approximation based on node computing power, avoiding memory resource waste or performance bottlenecks from the perspectives of data processing scale and computational complexity. The improved DBSCAN algorithm exponentially correlates the neighborhood radius threshold with CPU utilization, dynamically adjusting the core point density according to network bandwidth usage. This enables clustering analysis to adapt to network load fluctuations in real time, significantly improving the accuracy of data feature recognition in complex and ever-changing network environments compared to traditional fixed-parameter methods.

[0088] To address the performance imbalance caused by optimizing a single metric, this invention employs a multi-dimensional parameter fusion and global decision-making strategy. The dynamic trust assessment model quantifies the credibility of clustering results by fusing exponential decay functions and logistic regressions of parameters such as energy consumption, load, and latency. The multi-objective decision optimization engine determines resource allocation weights based on Pareto optimal curves, simultaneously optimizing energy consumption, latency, and resource utilization during computational resource scheduling and data transmission path planning. For example, the resource allocation model constrains the resource requirements of clusters with available resources at edge nodes, processing latency, and energy consumption limits, ensuring balanced performance across all stages and preventing a decline in overall system efficiency due to excessive pursuit of a single objective.

[0089] Furthermore, this invention achieves dynamic optimization across the entire process through real-time parameter feedback from the edge computing gateway. From the adaptive adjustment of the dynamic data flow matrix dimensions during the data acquisition phase to the similarity-weighted correction based on the bandwidth-delay product during feature map construction, each step is closely related to the gateway's operational status parameters. This design, which deeply embeds hardware parameters into the algorithm logic, forms a closed-loop optimization system from data processing to decision execution. Ultimately, this significantly improves the system's core indicators such as resource utilization, processing timeliness, and result reliability, providing a more efficient and stable solution for network information processing in edge computing scenarios.

[0090] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A network information processing method based on an edge computing gateway, characterized in that, Includes the following steps: Step S1: Collect multidimensional heterogeneous network traffic data through the edge computing gateway and construct a dynamic data flow matrix containing spatiotemporal characteristics. The dynamic data flow matrix is ​​adaptively adjusted in dimensions based on the edge node computing capability threshold. Step S2: Input the dynamic data stream matrix into a recurrent neural network optimized by a gating mechanism. The recurrent neural network includes a gradient pruning strategy based on edge computing resource constraints and a dynamic learning rate adjustment module to generate a temporal correlation feature vector. Step S3: Construct a feature similarity matrix based on the temporal correlation feature vector. The similarity matrix is ​​weighted and corrected by the bandwidth-delay product of the edge computing gateway to form a spatiotemporal correlation feature map. Step S4: The improved DBSCAN algorithm is used to perform cluster analysis on the spatiotemporal correlation feature map. The improved algorithm includes a core point density threshold dynamic adjustment mechanism based on the processing capability of edge nodes and a neighborhood radius adaptive calculation model based on network topology. Step S5: Construct a dynamic trust evaluation model for the clustering results. The model is based on the energy consumption threshold, computing load and communication latency of the edge computing gateway to fuse multi-dimensional parameters and generate clusters with trust weights. Step S6: Based on the clusters with credibility weights, generate a network information processing scheme that includes computing resource allocation strategy, data transmission path planning and processing priority ranking through the multi-objective decision optimization engine of the edge computing gateway.

2. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, In step S2, the recurrent neural network optimized by the gating mechanism adopts the following model structure: H t =σ g (W g ·[H t-1 ,X t ]+b g )⊙tanh(W h ·[H t-1 ,X t ]+b h ) Among them, H t Let σ be the hidden state vector at time t, and its dimension is positively correlated with the memory capacity threshold of the edge computing gateway; g The gated activation function is a piecewise linear approximation function based on the computational capabilities of edge nodes; W g and W h The weight matrix's parameter initialization strategy is determined based on the energy consumption-accuracy Pareto optimal curve of the edge computing gateway; ⊙ represents element-wise multiplication, where the cache hit rate of the edge computing gateway is used as the data prefetching decision factor during the operation; X t Let t be the input feature vector at time t, and its feature selection strategy is dynamically adjusted based on the bandwidth utilization of the edge computing gateway.

3. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, In step S4, the improved DBSCAN algorithm adopts the following core point density threshold calculation model: Where, ∈ t is the neighborhood radius threshold at time t, which has an exponential relationship with the real-time CPU utilization of the edge computing gateway; ∈0 is the basic neighborhood radius, determined by the initial configuration parameters of the edge computing gateway; α and β are adjustment coefficients, corresponding to the computing resource pressure factor and the network bandwidth pressure factor, respectively; C t Let C be the CPU utilization of the edge computing gateway at time t. max B represents the maximum allowed CPU utilization. t Let B be the network bandwidth utilization of the edge computing gateway at time t. max The maximum allowable utilization of network bandwidth; the update cycle of the neighborhood radius threshold is synchronized with the task scheduling cycle of the edge computing gateway.

4. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, In step S5, the dynamic trust assessment model uses the following multi-dimensional parameter fusion function: Wherein, T(c i ) represents cluster c i The trust score ranges from [0, 1]; E(c i To process cluster c i The required power consumption of the edge computing gateway is calculated using a power-time integral model; L(c i To process cluster c i The load rate of the edge computing gateway is calculated based on the task queue length and processing latency; D(c i ) represents cluster c i The data transmission latency is dynamically estimated based on network topology and bandwidth usage; ω1, ω2, and ω3 are weighting coefficients that satisfy ω1+ω2+ω3=1 and are dynamically adjusted based on the operating mode of the edge computing gateway; λ1, λ2, and λ3 are attenuation coefficients that correspond to energy consumption sensitivity, load sensitivity, and latency sensitivity, respectively; μ is the load balancing point parameter, which corresponds to the optimal computing efficiency point of the edge computing gateway.

5. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, In step S6, the multi-objective decision optimization engine adopts the following resource allocation model: Constraints: Among them, R ij C represents the amount of resources allocated to the i-th cluster and the j-th edge computing node; j D represents the total available computing resources of the j-th edge computing node; ij D represents the estimated processing latency for assigning the i-th cluster to the j-th edge computing node; max E represents the maximum allowable processing latency threshold of the system. ij E represents the estimated energy consumption of assigning the i-th cluster to the j-th edge computing node; max Indicates the maximum allowable energy consumption threshold of the system; w 1i w 2i w 3i These are the resource consumption weight, latency weight, and energy consumption weight for the i-th cluster, respectively, which are dynamically adjusted based on the service quality requirements of the edge computing gateway.

6. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, The spatiotemporal correlation feature map construction in step S3 adopts the following similarity calculation model: Where S(x) i x j ) represents data point x i and x j The spatiotemporal similarity between them; ||x i -x j || represents the Euclidean distance in the feature space; γ is the kernel function bandwidth parameter, which is inversely proportional to the memory access speed of the edge computing gateway; T ij Represents data point x i and x j The time correlation between the two is calculated by the time series analysis module of the edge computing gateway; δ is the time correlation adjustment factor, which is dynamically adjusted based on the real-time requirements of the edge computing gateway; β is the time correlation sensitivity parameter, which is related to the clock synchronization accuracy of the edge computing gateway; τ is the time correlation threshold, which is determined by the task scheduling cycle of the edge computing gateway.

7. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, In step S2, the gradient pruning strategy of the recurrent neural network adopts the following adaptive threshold calculation model: Where, θ t θt is the gradient clipping threshold at time t; θ0 is the basic clipping threshold, determined based on the initial training configuration of the edge computing gateway; M t M represents the memory usage of the edge computing gateway at time t. max V represents the maximum available memory for the edge computing gateway. t V is the norm of the gradient vector at time t; max α is the preset upper limit of the gradient norm; α and β are adjustment coefficients, corresponding to the memory pressure factor and gradient stability factor, respectively.

8. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, In step S4, the improved DBSCAN algorithm adopts the following neighborhood radius adaptive calculation model: Where, ∈ ij Represents data point x i Relative to cluster center c j The adaptive neighborhood radius; S ij Represents data point x i With cluster center c j Feature similarity; σ is the standard deviation of the similarity distribution, obtained statistically based on historical data from edge computing gateways; L j For cluster c j The current load is calculated using the task queue length and processing latency of the edge computing gateway; L max For cluster c j The maximum load capacity; λ and ρ are adjustment coefficients that control similarity sensitivity and load sensitivity, respectively; the parameter update frequency of the neighborhood radius adaptive calculation model is synchronized with the resource monitoring cycle of the edge computing gateway.

9. The network information processing method based on an edge computing gateway according to claim 1, characterized in that, In step S6, the network information processing scheme is generated using the following priority ranking model: Among them, P i For cluster c i Processing priority score; I i For cluster c i Information entropy reflects the degree of uncertainty in data; U i For cluster c i The update frequency is calculated by the time series analysis module of the edge computing gateway; D i For cluster c i The data transmission distance is determined based on the network topology of the edge computing gateway; D0 is a preset distance threshold, which is related to the communication energy consumption optimization target of the edge computing gateway; w1, w2, and w3 are weighting coefficients, which are dynamically adjusted based on the real-time operating status of the edge computing gateway; α is a distance sensitivity parameter, which is related to the wireless communication protocol characteristics of the edge computing gateway.

10. A network information processing system based on an edge computing gateway, characterized in that, include: Multidimensional heterogeneous data acquisition unit: used to collect multidimensional heterogeneous data including network traffic, device status and user behavior through the multi-channel data interface of the edge computing gateway, and to build a dynamic data flow matrix; Spatiotemporal feature extraction unit: connected to the multidimensional heterogeneous data acquisition unit, used to input the dynamic data stream matrix into a recurrent neural network optimized by a gating mechanism to generate a feature vector containing temporal correlation features; Feature map construction unit: connected to the spatiotemporal feature extraction unit, used to construct a feature similarity matrix based on the temporal correlation feature vector, and perform weighted correction through the bandwidth-delay product of the edge computing gateway to generate a spatiotemporal correlation feature map; Adaptive clustering analysis unit: connected to the feature map construction unit, used to perform clustering analysis on the spatiotemporal correlation feature map using the improved DBSCAN algorithm. The improved algorithm includes a core point density threshold dynamic adjustment mechanism based on edge node processing capability and a neighborhood radius adaptive calculation model based on network topology. Multidimensional Trust Evaluation Unit: Connected to the adaptive clustering analysis unit, it is used to construct a multidimensional parameter fusion model based on the edge computing gateway energy consumption threshold, computing load and communication latency, and generate clusters with trust weights; Multi-objective decision optimization unit: connected to the multi-dimensional trust evaluation unit, used to generate a network information processing scheme including computing resource allocation strategy, data transmission path planning and processing priority ranking based on the cluster with trust weight, through the multi-objective decision optimization engine of the edge computing gateway, and implement the processing scheme through the control interface of the edge computing gateway.

Images