Data transmission method in edge computing power secure channel

Abstract

The invention provides a data transmission method in an edge computing power secure channel, and the method comprises the steps: obtaining a real-time residual energy value, node position distribution information and node historical energy consumption data of each edge node in a network, constructing a node energy prediction model through a time sequence analysis method, obtaining a predicted energy consumption rate and predicted residual energy of the node in a future period of time; the signal-to-noise ratio, the stability and the interference level of links between weak nodes and adjacent nodes in the energy depression area are obtained, the weak nodes are nodes of which the predicted residual energy is lower than a network average value preset threshold value and which are located in the energy depression area, and adaptive power control and data compression are performed on the weak nodes; and in combination with edge computing power secure channel connectivity, edge computing power secure channel coverage rate and end-to-end data transmission delay factors, network life cycles under different data transmission strategies are evaluated through a Monte Carlo simulation method, and optimal transmission parameter configuration for prolonging the network life cycles is obtained.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a data transmission method in an edge computing power secure channel. Background Art

[0002] In edge computing networks, there is a complex interaction between node energy distribution and data transmission strategies. First, the energy status of each node in the network directly affects the choice of data transmission paths. Nodes with sufficient energy are more likely to become hotspots and undertake more data forwarding tasks, while nodes with insufficient energy may become network bottlenecks. Secondly, the data transmission process will affect the energy consumption of nodes. Nodes that frequently participate in data forwarding consume energy faster, which may lead to dynamic changes in the network topology. This change will in turn affect the formulation of data transmission strategies. Therefore, how to achieve balanced distribution of network energy and extend the life cycle of the entire network while ensuring data transmission efficiency and reliability has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a data transmission method in an edge computing power secure channel, which mainly includes:

[0004] Obtain the real-time remaining energy value, node location distribution information and node historical energy consumption data of each edge node in the network, build a node energy prediction model through time series analysis method, and obtain the predicted energy consumption rate and predicted remaining energy of the node in the future period of time;

[0005] According to the node energy prediction results and network topology information, a graph theory algorithm is used to construct a network energy distribution map to identify energy anomaly areas, which include energy hotspot areas and energy depression areas. The energy hotspot area is an area where the predicted energy consumption rate exceeds a preset threshold of the average value of surrounding nodes and is located on a critical transmission path, and the energy depression area is an area where the predicted remaining energy is lower than a preset threshold of the network average value.

[0006] For the identified energy hotspot area, obtain the data transmission requirements of key edge nodes in the area, establish a parallel edge computing power secure transmission channel, obtain the transmission delay, transmission bandwidth and transmission reliability indicators of each channel, dynamically adjust the data parallel transmission strategy according to the indicator value, and optimize the low-latency and high-reliability transmission of key data;

[0007] Acquire the signal-to-noise ratio, stability and interference level of the link between the weak node and the adjacent node in the energy depression area, wherein the weak node is a node whose predicted remaining energy is lower than a preset threshold of the network average value and is located in the energy depression area, and perform adaptive power control and data compression on the weak node;

[0008] The distribution of key edge nodes, node predicted remaining energy, and node predicted energy consumption rate are used as the state input of the reinforcement learning algorithm. By establishing a reward function that combines energy consumption balance, transmission delay, and throughput, the data transmission path selection strategy is guided to optimize the energy consumption balance of the entire network and the efficient transmission of key data, and a dynamically optimized data transmission path decision is obtained.

[0009] Evaluate the stability of the data transmission path decision, obtain the path change frequency index per unit time, and if the index exceeds the preset threshold, start the path smoothing mechanism and introduce the inertia factor to control the path adjustment amplitude, thereby ensuring the convergence of the path adjustment and reducing the additional energy consumption and delay caused by frequent path switching;

[0010] Combining the connectivity of edge computing security channels, edge computing security channel coverage, and end-to-end data transmission delay factors, the network life cycle under different data transmission strategies is evaluated through the Monte Carlo simulation method to obtain the optimal transmission parameter configuration to extend the network life cycle.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a data transmission method in an edge computing power security channel. The method predicts node energy through time series analysis, constructs a network energy distribution map to identify energy abnormality areas, establishes parallel edge computing power security transmission channels for energy hotspot areas, and adopts adaptive power control and data compression strategies for weak nodes in energy depression areas. The data transmission path is optimized using a reinforcement learning algorithm, and a path smoothing mechanism is introduced to reduce the additional overhead caused by frequent switching. Finally, the network life cycle under different strategies is evaluated through Monte Carlo simulation to obtain the optimal transmission parameter configuration. The present invention realizes intelligent data transmission with energy perception in the edge network, effectively balances network energy consumption, and extends the network life cycle. At the same time, it ensures low-latency and high-reliability transmission of key data, and improves the overall performance and reliability of the edge network. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of a data transmission method in an edge computing power secure channel of the present invention.

[0014] Figure 2 A schematic diagram of a data transmission method in an edge computing power secure channel of the present invention.

[0015] Figure 3 This is another schematic diagram of a data transmission method in an edge computing power secure channel of the present invention. DETAILED DESCRIPTION

[0016] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0017] like Figure 1-3 In this embodiment, a data transmission method in an edge computing power secure channel may specifically include:

[0018] S101. Obtain the real-time remaining energy value, node location distribution information and node historical energy consumption data of each edge node in the network, build a node energy prediction model through time series analysis method, and obtain the predicted energy consumption rate and predicted remaining energy of the node in the future period of time.

[0019] The real-time remaining energy value, node location coordinates and historical energy consumption data of the network edge nodes are obtained, and the obtained data are cleaned and standardized; the moving average method is used to smooth the historical energy consumption data after cleaning and standardization; modeling is performed based on the smoothed historical energy consumption data, using the established model, combined with the current remaining energy value and location distribution information of the node, to predict the future node energy consumption rate; based on the predicted energy consumption rate, the predicted remaining energy value of the node is calculated.

[0020] Specifically, the real-time remaining energy value, node location coordinates and historical energy consumption data are obtained from the edge nodes in the network, and the collected data are cleaned and standardized. The Z-score method is used to detect outliers, and data points exceeding 3 standard deviations are marked as abnormal and removed. The linear interpolation method is used to fill missing values ​​to ensure the continuity of the data. For data of different dimensions, the minimum and maximum scaling method is used to unify them into the range of 0-1 to ensure the comparability of the data. According to the cleaned historical energy consumption data, the moving average method is used to smooth the time series data to eliminate the impact of short-term fluctuations and extract long-term change trends. The optimal moving average window size is determined by the grid search method, and the window size with the smallest root mean square error is selected by traversing in the range of 1 hour to 24 hours with a step size of 1 hour. ARIMA modeling is used for the processed time series data to capture the autocorrelation and periodic characteristics of the data. Using the established ARIMA, combined with the current remaining energy value and location distribution information of the node, the node energy consumption rate in the next 24 hours is predicted. The energy consumption coefficient is calculated according to the distance between nodes. The longer the distance, the greater the energy consumption. Using multiple linear regression, historical energy consumption data, current remaining energy value and energy consumption coefficient are used as input variables to predict the node energy consumption rate in the next 24 hours. The predicted remaining energy value of the node in 24 hours is calculated based on the predicted energy consumption rate, which provides a basis for network resource scheduling and load balancing. The real-time remaining energy value, node location coordinates and historical energy consumption data are obtained from the edge nodes in the network, and the collection frequency is once every 5 minutes. For a network of 100 nodes, 300 data points are obtained each time. The Z-score method is used to detect outliers, and data points with energy values ​​exceeding the mean value plus or minus 3 standard deviations are marked as abnormal and eliminated. For example, if the average energy value of a node is 500mAh and the standard deviation is 50mAh, data points with energy values ​​below 350mAh or above 650mAh will be eliminated. For missing data points, linear interpolation is used to fill them. The minimum and maximum scaling method is used to unify the energy value, location coordinates and energy consumption data into the range of 0-1. The moving average method is applied to the cleaned historical energy consumption data, and the optimal window size is determined by grid search. Assume that the search range is 1 hour to 24 hours, the step length is 1 hour, and select the window size with the smallest root mean square error, such as 12 hours. Use the moving average method of the 12-hour window to smooth the data and eliminate short-term fluctuations. Use ARIMA (1, 1, 1) modeling for the smoothed time series data, where the autoregressive term, the difference order, and the moving average term are all 1. Use the established ARIMA model, combined with the current remaining energy value and location distribution information of the node, to predict the node energy consumption rate in the next 24 hours. Calculate the distance between nodes as the energy consumption coefficient. For example, when the distance between two nodes is 100 meters, the energy consumption coefficient is 1, and the coefficient increases by 0.1 for every 10 meters increase in distance.Using multiple linear regression, the input variables include the average energy consumption in the past 24 hours, the current remaining energy value and the energy consumption coefficient, and predict the energy consumption rate in the next 24 hours. If the prediction results show that the average energy consumption rate of a node in the next 24 hours is 10mAh / hour and the current remaining energy is 300mAh, then the remaining energy of the node is predicted to be 60mAh after 24 hours. These prediction results can be used to optimize network resource scheduling and load balancing, such as assigning high-energy-consuming tasks to nodes with more remaining energy, thereby extending the operation time of the entire network.

[0021] S102. Based on the node energy prediction results and the network topology information, a graph theory algorithm is used to construct a network energy distribution map, and energy abnormal areas are identified. The energy abnormal areas include energy hot spots and energy depressions. The energy hot spots are areas where the predicted energy consumption rate exceeds a preset threshold of the average value of surrounding nodes and is located on a critical transmission path, and the energy depressions are areas where the predicted remaining energy is lower than a preset threshold of the network average value.

[0022] A weighted undirected graph is constructed according to the node energy prediction results and the network topology information, wherein the node represents the device in the network, the edge represents the connection relationship between the devices, and the weight of the edge is set to the physical distance between the nodes; the Kruskal algorithm is used to identify the critical transmission path on the weighted undirected graph, and the edge set with the smallest total weight is selected by comparing the weights of each edge to form a tree structure covering all nodes; the average predicted energy consumption rate and the average predicted remaining energy of all nodes in the network are calculated, and the judgment thresholds of energy hotspot areas and energy depression areas are set; the data in the moving time window is obtained and the dynamic threshold is calculated; all nodes in the network are traversed to determine whether the node is located on the critical transmission path. If the node is located on the critical transmission path and its predicted energy consumption rate exceeds a preset percentage of the average value of the surrounding one-hop neighboring nodes, the node is marked as an energy hotspot area; if the predicted remaining energy of the node is lower than the preset threshold of the network average value, the node is marked as an energy depression area.

[0023] Specifically, a weighted undirected graph is constructed based on the node energy prediction results and network topology information, where nodes represent devices in the network, edges represent the connection relationship between devices, and the weight of the edge is set to the physical distance between nodes. The predicted energy consumption rate is mapped to the color depth of the node, and the predicted remaining energy is mapped to the size of the node to achieve an intuitive representation of node attributes. The Kruskal algorithm is used to identify the key transmission path on the constructed weighted undirected graph, and the edge set with the smallest total weight is selected by comparing the weights of each edge to form a tree structure covering all nodes. The specific implementation process includes sorting all edges in ascending order of weight, selecting the minimum weight edge in turn, judging whether a loop is formed with the selected edge, and adding the minimum spanning tree if a loop is not formed until all nodes are connected. The average predicted energy consumption rate and average predicted remaining energy of all nodes in the network are calculated, and the judgment thresholds of energy hotspot areas and energy depression areas are set. The dynamic threshold is calculated using a 24-hour moving time window and updated every hour. The energy consumption rate threshold is set to 1.5 times the average value in the window, and the remaining energy threshold is set to 0.5 times the average value in the window. All nodes in the network are traversed, and the surrounding nodes are defined as one-hop neighbors. If a node is located on a critical transmission path and its predicted energy consumption rate exceeds the preset threshold of the average value of the surrounding one-hop neighbor nodes, the node is marked as an energy hotspot area; if the node's predicted remaining energy is lower than the preset threshold of the network average value, the node is marked as an energy depression area. For nodes on critical transmission paths, a weight of 1.2 times is given when judging energy hotspot areas to increase their importance. In a wireless sensor network containing 100 nodes, a weighted undirected graph is constructed based on the prediction results. The predicted energy consumption rate of each node is in the range of 0-10mW / h, and the predicted remaining energy is in the range of 0-1000mAh. The energy consumption rate is mapped to the node color, 0mW / h is blue, 10mW / h is red, and linear interpolation is used to obtain the intermediate color. The remaining energy is mapped to the node size, 0mAh is 1 pixel, 1000mAh is 10 pixels, and linear interpolation is used to obtain the intermediate size. The weight of the edge is set to the physical distance between nodes, ranging from 1 to 100 meters. Use the Kruskal algorithm to identify the critical transmission path. First, sort the 500 edges from small to large weights, and then select the minimum weight edges in turn. If the two nodes connected by the selected edge are already in the same connected component, skip the edge; otherwise, add the edge to the minimum spanning tree. Repeat this process until 99 edges are selected to form a tree structure covering all nodes. Use a 24-hour moving time window to calculate the dynamic threshold, which is updated every hour. Assuming that the average energy consumption rate in the current time window is 5mW / h and the average remaining energy is 500mAh, the energy consumption rate threshold is set to 7.5mW / h (5*1.5) and the remaining energy threshold is set to 250mAh (500*0.5). Traverse all nodes and define one-hop neighbors as directly connected nodes.For node A, if it is located on a critical transmission path, the predicted energy consumption rate is 9mW / h, and the average energy consumption rate of the surrounding one-hop neighbors is 6mW / h, then 9mW / h>6mW / h*1.2 (given 1.2 times weight), and A is marked as an energy hotspot. For node B, if its predicted remaining energy is 200mAh, which is lower than the preset threshold of 250mAh of the network average, B is marked as an energy depression area. In this way, multiple energy abnormal areas can be identified in the network, providing a basis for subsequent load balancing and routing optimization.

[0024] S103. For the identified energy hotspot area, obtain the data transmission requirements of the key edge nodes in the area, establish a parallel edge computing power secure transmission channel, obtain the transmission delay, transmission bandwidth and transmission reliability indicators of each channel, dynamically adjust the data parallel transmission strategy according to the indicator value, and optimize the low-latency and high-reliability transmission of key data.

[0025] The data transmission demand information of the edge nodes is obtained, and the data transmission demand information is generated when the depth-first search algorithm traverses the edge nodes in the energy hotspot area; a three-dimensional data transmission demand matrix is ​​constructed according to the data transmission demand information, and the dimensions of the three-dimensional data transmission demand matrix include source nodes, target nodes and transmission demand attributes; a Dijkstra algorithm is used to calculate multiple paths from the source node to the target node, and the Dijkstra algorithm uses the node residual energy and link quality as calculation factors of the edge weight; a weighted polling load balancing algorithm is applied to the multiple paths, and the weighted polling load balancing algorithm is assigned weights according to the transmission bandwidth of the path, the node residual energy and the current load situation; an adaptive weight adjustment method is used for the multiple paths, and the adaptive weight adjustment method uses the exponential moving average method to calculate the change trend of the performance indicator, and adjusts the path weight according to the change trend.

[0026] Specifically, for the identified energy hotspot area, a depth-first search algorithm is used to traverse the edge nodes in the area to obtain the data transmission requirements of each node, including data volume, transmission priority and deadline, and construct a three-dimensional data transmission requirement matrix. The three dimensions of the matrix represent the source node, the target node and the transmission requirement attributes, which are used to guide the subsequent multi-path selection. According to the data transmission requirement matrix, the improved Dijkstra algorithm is used to calculate multiple paths from the source node to the target node. The algorithm takes the node residual energy and link quality as the calculation factors of the edge weight, and comprehensively considers the energy efficiency and transmission quality of the path. The first N paths that meet the transmission requirements are selected as candidate transmission channels, where the N value is dynamically adjusted according to the degree of network congestion. When the congestion is severe, the N value is increased, and vice versa. A weighted polling load balancing algorithm is applied to each candidate transmission channel, and a weight is assigned according to the transmission bandwidth of the channel, the node residual energy and the current load situation. The specific implementation process includes: calculating the comprehensive performance index of each channel, normalizing the index value as the weight, and allocating the data volume according to the weight ratio. In this way, the data allocation ratio of each channel is calculated, and a parallel edge computing power secure transmission channel is established. The real-time transmission delay, transmission bandwidth and transmission reliability indicators of each transmission channel are obtained through regular detection, and the adaptive weight adjustment algorithm is used to dynamically optimize data allocation. The algorithm uses the exponential moving average method to calculate the trend of performance indicators and adjust the channel weight according to the trend. At the same time, AES encryption technology is applied to the transmission channel to ensure the security of data transmission and achieve low-latency and high-reliability transmission of key data. In an energy hotspot area containing 50 edge nodes, the depth-first search algorithm is used to traverse all nodes to collect data transmission requirements. A 50x50x3 three-dimensional matrix is ​​constructed, in which the third dimension contains the data volume (unit: MB), transmission priority (1-10, 10 is the highest) and deadline (unit: minute). For example, the transmission demand from node A to node B is [100,8,30], which means that 100MB of data needs to be transmitted, the priority is 8, and the deadline is 30 minutes. When calculating the path, the Dijkstra algorithm defines the edge weight as 1 / (residual energy*link quality), where the residual energy range is 0-100% and the link quality range is 0-1. The algorithm selects the top 5 optimal paths as candidate transmission channels. When the average network load exceeds 80%, it increases to the top 8 paths. The weighted round-robin load balancing algorithm calculates the comprehensive performance index of each channel, index = 0.4*bandwidth+0.3*remaining energy+0.3*(1-current load rate). The index value is normalized as the weight. For example, if the weights of the four channels are [0.3, 0.25, 0.2, 0.25], 100MB of data is distributed as [30MB, 25MB, 20MB, 25MB].The adaptive weight adjustment method is used to detect the transmission performance every 5 seconds, and the exponential moving average method with a smoothing factor of α = 0.2 is used to calculate the trend as EMA_new = α * current value + (1-α) * EMA_old, where EMA_old is the exponential moving average calculated in the previous cycle and EMA_new is the new exponential moving average. If the EMA changes by more than 10% from the last time, the weight is adjusted according to the change ratio. At the same time, the AES-GCM mode with a 256-bit key is used to encrypt data to ensure transmission security, achieving low-latency and high-reliability data transmission in a dynamic network environment.

[0027] S104, obtaining the signal-to-noise ratio, stability and interference level of the link between the weak node and the adjacent node in the energy depression area, wherein the weak node is a node whose predicted remaining energy is lower than a preset threshold of the network average value and is located in the energy depression area; and performing adaptive power control and data compression on the weak node.

[0028] The predicted remaining energy values ​​of all nodes in the energy depression area are obtained, and the average remaining energy of the network is calculated by using the sliding window averaging method, wherein the size of the sliding window is a preset time period; a preset threshold is determined according to the average remaining energy of the network, and the preset threshold is a preset percentage of the average remaining energy of the network; it is determined whether the predicted remaining energy value of the node is lower than the preset threshold, and if the predicted remaining energy value of the node is lower than the preset threshold, the node is identified as a weak node; the signal-to-noise ratio, stability and interference level of the link between the weak node and the adjacent nodes are obtained; a link quality assessment matrix is ​​constructed according to the signal-to-noise ratio, the stability and the interference level; according to the link quality assessment matrix, a transmission power adjustment value is determined through fuzzification, reasoning and defuzzification processes; the data to be transmitted by the weak node is adaptively compressed; a compression ratio is determined according to the link quality assessment matrix, and the compression ratio is inversely proportional to the link quality assessment matrix.

[0029] Specifically, the predicted remaining energy values ​​of all nodes in the energy depression area are obtained through periodic detection, and the average remaining energy of the network is calculated by the sliding window average method, with a window size of 24 hours. The threshold is adaptively adjusted according to the overall energy status of the network, and the preset threshold is set to 70% of the sliding window average value. The nodes with predicted remaining energy lower than the threshold are identified as weak nodes. For the identified weak nodes, the channel state estimation algorithm is used to obtain the signal-to-noise ratio, stability and interference level of the link between it and the adjacent nodes. The pilot signal method is used to estimate the signal-to-noise ratio, and the received signal to noise power ratio is calculated by sending a known signal sequence. The link quality index is used to evaluate stability, and the received signal strength indicator and packet error rate are measured. The interference level is measured, and the signal energy is sampled on the idle channel. A link quality evaluation matrix is ​​constructed, and the matrix elements include the signal-to-noise ratio, stability index and interference level. According to the link quality evaluation matrix, adaptive power control is implemented, and the input variables are the signal-to-noise ratio, stability and interference level, and the output variable is the transmission power adjustment value. Using the fuzzification, reasoning and defuzzification process, the input variables are converted into fuzzy sets, and the preset fuzzy rules are applied for reasoning, such as "if the signal-to-noise ratio is lower than 10dB and the interference level is higher than -70dBm, increase the transmission power by 2dBm". The fuzzy output is converted into a specific power adjustment value by the centroid method. An adaptive compression method is applied to the data to be transmitted by the weak node, and the compression method is selected according to the data type. Text data uses Huffman encoding, image data uses JPEG compression, and audio data uses MP3 encoding. The compression ratio is dynamically adjusted according to the link quality. The worse the link quality, the higher the compression ratio is used. Combined with the adaptive power control strategy, the node transmission power consumption is optimized to reduce energy consumption while ensuring link stability. In a wireless sensor network containing 100 nodes, the energy depression area occupies 20% of the network area. The node residual energy is detected every 5 minutes, and the average residual energy of the network is calculated using the 24-hour sliding window average method. The current sliding window average value is 500mAh, and the preset threshold is set to 350mAh (70%). 15 weak nodes are identified with residual energy below 350mAh. For these weak nodes, the pilot signal method is used to estimate the signal-to-noise ratio. 100 known bit sequences are sent, and the calculated signal-to-noise ratio range is 5-15dB. The link quality index evaluates stability, and the received signal strength indication is measured to be -75 to -60dBm, and the packet error rate is 0.1% to 1%. The energy detection method samples 1000 times within a 20MHz bandwidth, and the interference level is measured to be -90 to -70dBm. A 15x3 link quality evaluation matrix is ​​constructed, each row represents a weak node, and the columns are signal-to-noise ratio, stability index, and interference level. In adaptive power control, the signal-to-noise ratio of 5-15dB is mapped to three fuzzy sets of low, medium, and high, the stability of 0.9-1 is mapped to stable, and below 1 is mapped to unstable, and the interference level of -90 to -70dBm is mapped to low, medium, and high.Fuzzy rules such as "if the signal-to-noise ratio is low and the interference is high, increase the power significantly" are applied, and the power adjustment value is obtained by the centroid method, ranging from -2 to +5dBm. For data compression, Huffman coding is used to achieve a 50% compression rate for text, the JPEG image quality factor is set to 70%, and the MP3 audio bit rate is set to 128kbps. For every 10% decrease in link quality, the compression rate increases by 5%. The average transmission power consumption of weak nodes is reduced from 50mW to 35mW, extending the expected survival time.

[0030] S105. The distribution of key edge nodes, the predicted remaining energy of nodes, and the predicted energy consumption rate of nodes are used as the state inputs of the reinforcement learning algorithm. By establishing a reward function that combines energy consumption balance, transmission delay, and throughput, the data transmission path selection strategy is guided to optimize towards the direction of energy consumption balance of the entire network and efficient transmission of key data, and a dynamically optimized data transmission path decision is obtained.

[0031] Obtain the distribution of key edge nodes, the predicted remaining energy of nodes and the predicted energy consumption rate of nodes, and map the obtained information into a state space for constructing a reinforcement learning environment; according to the state space, calculate the energy balance, transmission delay and throughput, and obtain the reward function R, where R = energy balance weight * energy balance + transmission delay weight * (1 / transmission delay) + throughput weight * throughput; adopt a deep Q network algorithm to implement the reinforcement learning process, and the input layer of the deep Q network includes the node position coordinates, predicted remaining energy and predicted energy consumption rate; set a playback buffer to store the state, action, reward and next state of each time step, randomly sample training samples from the playback buffer, and the training samples are used for model training; minimize the loss function by the gradient descent method, update the target network parameters every preset time step, and obtain a dynamically optimized data transmission path decision.

[0032] Specifically, a reinforcement learning environment is constructed, the distribution of key edge nodes, the predicted remaining energy of nodes, and the predicted energy consumption rate of nodes are mapped into the state space, and the action space is defined as a set of selectable data transmission paths. The reward function R=0.4*energy balance+0.3*(1 / transmission delay)+0.3*throughput is designed, where the energy balance is reflected by calculating the variance of the remaining energy between nodes, and the transmission delay and throughput are obtained through network simulation. The deep Q network algorithm is used to implement the reinforcement learning process. The input layer contains the node position coordinates, predicted remaining energy, and predicted energy consumption rate. The hidden layer uses 3 fully connected layers, and the number of neurons in each layer is 64, 32, and 16 respectively. The activation function uses ReLU, and the output layer corresponds to the Q value of each optional transmission path. The action selection is performed using the ε-greedy strategy. The transmission path is randomly selected with a probability of ε, and the path with the largest Q value is selected with a probability of 1-ε. The ε value adopts a decay strategy,

[0033] ε=max(0.01,0.5*0.995^episode), where 0.01 is the minimum exploration rate, 0.5 is the initial exploration rate, 0.995 is the decay rate, and episode is the current number of training rounds. Model training is performed based on the experience replay mechanism, the replay buffer size is set to 10000, and 256 samples are randomly sampled for training each time. The state, action, reward, and next state of each time step are stored, and the loss function is minimized by the gradient descent method. The target network parameters are updated every 100 time steps. The model performance is tested regularly on the validation set. If the performance of the validation set does not improve for 5 consecutive times, the learning rate is halved until the preset minimum learning rate is reached, and a dynamically optimized data transmission path decision is obtained. A reinforcement learning environment is constructed in a network containing 100 edge nodes. The state space includes the two-dimensional coordinates of the node (range 0-1000 meters), the predicted remaining energy 0-100%, and the predicted energy consumption rate 0-10mW / h. The action space contains all possible paths from the source node to the target node, totaling about 1000. In the reward function, the energy balance is obtained by calculating the standard deviation of the residual energy of all nodes. The smaller the standard deviation, the higher the balance; the transmission delay and throughput are obtained through the network simulation tool NS-3. The input layer of the deep Q network contains 7 neurons, including 2 coordinate values, 1 residual energy value, 1 energy consumption rate value, 3 one-hot encodings to represent the node type, 3 hidden layers with 64, 32, and 16 neurons respectively, and the output layer has 1000 neurons corresponding to each possible path. The initial ε value is set to 0.5, and multiplied by 0.995 after each round of training until the minimum value is 0.01. The replay buffer stores the most recent 10,000 experience tuples (state, action, reward, next state), and randomly samples every 256 time steps for training. The Adam optimizer is used, and the initial learning rate is 0.001. The model performance was evaluated every 1,000 rounds of training on a validation set consisting of 1,000 node pairs. If the performance did not improve for five consecutive times, the learning rate was halved to a minimum of 0.0001. After 100,000 rounds of training, the model achieved an average reward value on the test set from the initial -5 to 8.5, an increase in energy consumption balance by 30%, a reduction in average transmission delay by 25%, and an increase in network throughput by 20%. The final decision model was able to select the optimal data transmission path for any node pair within 50 milliseconds based on the real-time network status.

[0034] S106, evaluating the stability of the data transmission path decision, obtaining a path change frequency index per unit time, and if the index exceeds a preset threshold, starting a path smoothing mechanism, introducing an inertia factor to control the path adjustment amplitude, thereby ensuring the convergence of the path adjustment and reducing the additional energy consumption and delay caused by frequent path switching.

[0035] The sliding window method is used to obtain the path change frequency index per unit time, and the path change frequency index is obtained by dividing the number of path changes in the calculation window by the window size; the path change frequency index is compared with the preset frequency threshold, and if the path change frequency index exceeds the frequency threshold, the path smoothing mechanism is triggered; the exponential moving average method is used to achieve path smoothing, and the path smoothing includes determining a new path node sequence, wherein the new path node sequence is obtained by multiplying the current path node sequence by α plus (1-α) multiplied by the decision path node sequence, and α is the inertia factor; an adaptive α value adjustment strategy is adopted, and the adaptive α value The adjustment strategy includes: obtaining a network stability index, wherein the network stability index is obtained by calculating the variance of node residual energy and link quality in a recent period of time; dynamically adjusting the α value according to the network stability index; obtaining the convergence speed of the path adjustment and the network performance change, wherein the convergence speed is obtained by calculating the average similarity of multiple consecutive path updates, and the network performance change includes the change rate of end-to-end delay, throughput and energy consumption indicators; according to the convergence speed and the network performance change, a fuzzy controller is used to adjust the α value, and the fuzzy controller is used to minimize the additional energy consumption and delay caused by path switching while ensuring convergence.

[0036] Specifically, the sliding window method is used to calculate the path change frequency index per unit time. The window size is set to 5 minutes and updated every 30 seconds. The number of path changes in the window is recorded and divided by the window size to obtain the path change frequency. The judgment standard for path change is that the node overlap of two paths is less than 80%. The path change frequency threshold is set as an adjustable parameter, and the initial value is set to 1 time per minute, which is dynamically adjusted according to network performance feedback. If the calculated path change frequency exceeds the threshold, the path smoothing mechanism is triggered and the inertia factor α is introduced, with a value range of 0 to 1. The exponential moving average method is used to achieve path smoothing, and the new path node sequence = α*current path node sequence + (1-α)*decision path node sequence, where the initial value of α is set to 0.7. The adaptive α value adjustment strategy is adopted to dynamically adjust the α value according to the network stability index. The network stability index is obtained by calculating the variance of the node residual energy and link quality in the last 10 minutes. The inertia factor α is dynamically optimized by the adaptive adjustment method, and the convergence speed of the path adjustment is calculated, which is defined as the average similarity of 5 consecutive path updates. The higher the similarity, the faster the convergence speed. At the same time, the network performance changes are monitored, including the change rate of end-to-end delay, throughput and energy consumption indicators. According to the convergence speed and network performance changes, the fuzzy controller is used to adjust the α value to minimize the additional energy consumption and delay caused by path switching while ensuring convergence. The effect of the path smoothing mechanism is regularly evaluated to monitor the change trend of average energy consumption, delay and throughput. If the performance degrades for three consecutive evaluation cycles, the fine-tuning of the path smoothing strategy is triggered, including adjusting the change range and update frequency of the α value. In a wireless sensor network containing 100 nodes, a 5-minute sliding window is used to calculate the path change frequency. It is updated every 30 seconds and the number of path changes in the window is recorded. For example, if 8 path changes are detected within 5 minutes, the path change frequency is 1.6 times / minute. The path change judgment standard is set to a node overlap of less than 80%. For example, if the original path is [1,2,3,4,5] and the new path is [1,2,6,4,5], the overlap is 80%, which is not counted as a change. The initial threshold is set to 1 time / minute. When the network throughput drops by 10%, the threshold is adjusted to 0.8 times / minute. After the path smoothing mechanism is triggered, the inertia factor α is introduced with an initial value of 0.7. Assuming the current path is [1,3,5,7,9] and the decision path is [1,4,6,7,9], the new path after smoothing is

[0037] [1,3.3,5.3,7,9] (rounded to [1,3,5,7,9]). The network stability index is obtained by calculating the variance of the residual energy of 100 nodes in the last 10 minutes. The smaller the variance, the more stable the network. If the variance decreases from 0.1 to 0.05, the α value is increased by 0.1. The path adjustment convergence speed is measured by calculating the average similarity of 5 consecutive path updates. The similarity uses the Jaccard coefficient with a value range of 0-1. If the average similarity increases from 0.6 to 0.8, the α value is increased by 0.05. At the same time, the network performance is monitored. For example, if the end-to-end delay increases from 50ms to 60ms, the throughput decreases from 10Mbps to 9Mbps, and the average energy consumption per node increases from 5mW / h to 5.5mW / h, the α value is reduced by 0.03. A fuzzy controller is used to comprehensively consider these factors and dynamically adjust the α value. The effectiveness of the path smoothing mechanism is evaluated every hour. If the network performance decreases by more than 5% for three consecutive hours, the step size of the α value is adjusted from 0.05 to 0.03, and the update frequency is changed from once every 10 minutes to once every 5 minutes to respond more sensitively to network changes.

[0038] S107. Combined with the edge computing power security channel connectivity, edge computing power security channel coverage, and end-to-end data transmission delay factors, the network life cycle under different data transmission strategies is evaluated through the Monte Carlo simulation method to obtain the optimal transmission parameter configuration to extend the network life cycle.

[0039] Define the security channel connectivity index of the edge computing power security channel network model, the security channel connectivity index is the end-to-end connectivity probability of the channel; calculate the security channel connectivity index; obtain the security channel coverage index of the edge computing power security channel network model, the security channel coverage index is the proportion of network nodes covered by at least one security channel; establish a parameter space sampling scheme according to the security channel connectivity index and the security channel coverage index; discretize the node transmission power, data compression ratio, data encryption strength and routing weight to obtain a parameter combination set; use the Monte Carlo method to perform network life cycle simulation, and the network life cycle simulation repeats the experiment for each set of parameters; simulate the network operation process, the simulation includes data packet generation, routing selection, transmission and energy consumption calculation; record the network life cycle length, and the recording is performed when the energy of the first node is exhausted; use a non-dominated sorting genetic algorithm for multi-objective optimization, and the multi-objective optimization is based on the network life cycle simulation results; obtain a Pareto optimal solution set, and determine the optimal transmission parameter configuration from the Pareto optimal solution set.

[0040] Specifically, a secure channel network model for edge computing power is constructed, and the connectivity indicator of the secure channel is defined as the end-to-end connectivity probability of the channel. The maximum flow algorithm is used to calculate the connectivity. The secure channel coverage indicator is the proportion of network nodes covered by at least one secure channel. The greedy algorithm of the set covering problem is used to calculate the coverage. The end-to-end data transmission delay threshold is set to 100ms, which is determined based on application requirements. According to the defined indicators, a parameter space sampling scheme is designed to discretize the node transmission power, data compression ratio, data encryption strength, and routing weight. The node transmission power range is 0-20dBm, with a step size of 2dBm; the data compression ratio range is 0-0.9, with a step size of 0.1; the data encryption strength level is {low, medium, high}; the routing weight range is 0-1, with a step size of 0.1. A set of parameter combinations is generated, and each set of parameters corresponds to a data transmission strategy. The Monte Carlo method is used to simulate the network life cycle, and each set of parameters is repeated 1000 times to ensure statistical significance. The network operation process is simulated, including data packet generation, routing selection, transmission, and energy consumption calculation. Packet generation follows Poisson distribution, routing selection is based on Dijkstra algorithm, channel fading and interference are considered in the transmission process, and energy consumption calculation includes transmission, reception and computing energy consumption. The network is simulated to run until the energy of the first node is exhausted, and the length of the network life cycle is recorded. Based on the simulation results, the non-dominated sorting genetic algorithm II (NSGA-II) is used for multi-objective optimization, taking into account the network life cycle, secure channel connectivity, coverage and transmission delay. The algorithm implementation includes population initialization, crossover, mutation and non-dominated sorting process. From the Pareto optimal solution set obtained, the analytic hierarchy process (AHP) is used to determine the weight of each objective, and the solution with the highest comprehensive score is calculated as the optimal transmission parameter configuration. In a network containing 100 edge nodes, an edge computing power secure channel model is constructed. The Ford-Fulkerson algorithm is used to calculate the maximum flow, and the secure channel connectivity is obtained as 0.85. The greedy algorithm is used to calculate the coverage, and the result is 92%. The end-to-end delay threshold is set to 100ms. The parameter space is designed as node transmission power range 0-20dBm, step size 2dBm, a total of 11 values; data compression ratio 0-0.9, step size 0.1, a total of 10 values; data encryption strength is {low, medium, high}; routing weight 0-1, step size 0.1, a total of 11 values. A total of 11*10*3*11=3630 parameter combinations are generated. 1000 Monte Carlo simulations are performed on each set of parameters to simulate the network operation process. The data packet generation rate λ=10 / second, which obeys the Poisson distribution. The Dijkstra algorithm is used for routing selection, considering the free space path loss model and -100dBm background noise. The energy consumption model includes the transmission power consumption

[0041] (50nJ / bit+100pJ / bit / m^2*d^2), receiving power consumption (50nJ / bit) and computing power consumption (5nJ / bit). The simulation runs until the energy of the first node is lower than 1% of the initial energy, and the network life cycle is recorded. The NSGA-II algorithm is applied for multi-objective optimization, the population size is set to 100, and the evolution is 500 generations. The crossover probability is 0.9 and the mutation probability is 0.1. 200 non-dominated solutions are obtained to form the Pareto frontier. The target weights are determined by the AHP method as 0.4 for network life cycle, 0.3 for connectivity, 0.2 for coverage, and 0.1 for latency. The solution with the highest comprehensive score is calculated as 14dBm for transmission power, 0.7 for compression ratio, medium for encryption strength, and 0.6 for routing weight. Under this configuration, the network life cycle reaches 3650 hours, connectivity is 0.92, coverage is 98%, and average latency is 85ms.

[0042] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations. In addition, the various different embodiments of the present invention can also be combined arbitrarily, as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A data transmission method in an edge computing power secure channel, characterized in that: The method comprises: Obtain the real-time remaining energy value, node location distribution information and node historical energy consumption data of each edge node in the network, build a node energy prediction model through time series analysis method, and obtain the predicted energy consumption rate and predicted remaining energy of the node in the future period of time; According to the node energy prediction results and network topology information, a graph theory algorithm is used to construct a network energy distribution map to identify energy anomaly areas, which include energy hotspot areas and energy depression areas. The energy hotspot area is an area where the predicted energy consumption rate exceeds a preset threshold of the average value of surrounding nodes and is located on a critical transmission path, and the energy depression area is an area where the predicted remaining energy is lower than a preset threshold of the network average value. For the identified energy hotspot area, obtain the data transmission requirements of key edge nodes in the area, establish a parallel edge computing power secure transmission channel, obtain the transmission delay, transmission bandwidth and transmission reliability indicators of each channel, dynamically adjust the data parallel transmission strategy according to the indicator value, and optimize the low-latency and high-reliability transmission of key data; Acquire the signal-to-noise ratio, stability and interference level of the link between the weak node and the adjacent node in the energy depression area, wherein the weak node is a node whose predicted remaining energy is lower than a preset threshold of the network average value and is located in the energy depression area, and perform adaptive power control and data compression on the weak node; The distribution of key edge nodes, node predicted remaining energy, and node predicted energy consumption rate are used as the state input of the reinforcement learning algorithm. By establishing a reward function that combines energy consumption balance, transmission delay, and throughput, the data transmission path selection strategy is guided to optimize the energy consumption balance of the entire network and the efficient transmission of key data, and a dynamically optimized data transmission path decision is obtained. Evaluate the stability of the data transmission path decision, obtain the path change frequency index per unit time, and if the index exceeds the preset threshold, start the path smoothing mechanism and introduce the inertia factor to control the path adjustment amplitude, thereby ensuring the convergence of the path adjustment and reducing the additional energy consumption and delay caused by frequent path switching; Combining the connectivity of edge computing security channels, edge computing security channel coverage, and end-to-end data transmission delay factors, the network life cycle under different data transmission strategies is evaluated through the Monte Carlo simulation method to obtain the optimal transmission parameter configuration to extend the network life cycle.

2. The method according to claim 1, characterized in that: The method of obtaining the real-time remaining energy value, node location distribution information and node historical energy consumption data of each edge node in the network, building a node energy prediction model through a time series analysis method, and obtaining the predicted energy consumption rate and predicted remaining energy of the node in the future period of time includes: Obtain the real-time remaining energy value, node location coordinates and historical energy consumption data of network edge nodes, and clean and standardize the acquired data; According to the historical energy consumption data after cleaning and standardization, the moving average method is used for smoothing; Based on the smoothed historical energy consumption data modeling, the established model is used to combine the current remaining energy value and location distribution information of the node to predict the future node energy consumption rate; According to the predicted energy consumption rate, the predicted remaining energy value of the node is calculated.

3. The method according to claim 1, characterized in that According to the node energy prediction results and network topology information, a graph theory algorithm is used to construct a network energy distribution map, and energy abnormal areas are identified. The energy abnormal areas include energy hot spots and energy depression areas. The energy hot spots are areas where the predicted energy consumption rate exceeds a preset threshold of the average value of surrounding nodes and is located on a critical transmission path, and the energy depression areas are areas where the predicted remaining energy is lower than a preset threshold of the network average value, including: A weighted undirected graph is constructed based on the node energy prediction results and network topology information, where nodes represent devices in the network, edges represent the connection relationship between devices, and the weight of the edge is set as the physical distance between nodes; The Kruskal algorithm is used to identify the key transmission path on the weighted undirected graph, and the edge set with the smallest total weight is selected by comparing the weights of each edge to form a tree structure covering all nodes; Calculate the average predicted energy consumption rate and average predicted remaining energy of all nodes in the network, and set the judgment thresholds for energy hot spots and energy depressions; Get the data in the moving time window and calculate the dynamic threshold; Traverse all nodes in the network to determine whether the node is located on a critical transmission path. If the node is on the critical transmission path and its predicted energy consumption rate exceeds a preset percentage of the average value of the surrounding one-hop neighboring nodes, mark the node as an energy hotspot; If the predicted remaining energy of a node is lower than a preset threshold of the network average value, the node is marked as an energy depression area.

4. The method according to claim 1, characterized in that The method of obtaining the data transmission requirements of key edge nodes in the identified energy hotspot area, establishing a parallel edge computing power secure transmission channel, obtaining the transmission delay, transmission bandwidth and transmission reliability indicators of each channel, dynamically adjusting the data parallel transmission strategy according to the indicator value, and optimizing the low-delay and high-reliability transmission of key data includes: Acquire data transmission demand information of edge nodes, where the data transmission demand information is generated when a depth-first search algorithm traverses edge nodes in an energy hotspot area; Constructing a three-dimensional data transmission requirement matrix according to the data transmission requirement information, wherein the dimensions of the three-dimensional data transmission requirement matrix include a source node, a target node, and a transmission requirement attribute; The Dijkstra algorithm is used to calculate multiple paths from the source node to the target node, wherein the Dijkstra algorithm uses the node residual energy and the link quality as the calculation factors of the edge weight; Applying a weighted round-robin load balancing algorithm to multiple paths, wherein the weighted round-robin load balancing algorithm assigns weights according to the transmission bandwidth of the paths, the remaining energy of the nodes, and the current load conditions; An adaptive weight adjustment method is adopted for the multiple paths. The adaptive weight adjustment method uses an exponential moving average method to calculate the change trend of the performance indicator and adjusts the path weight according to the change trend.

5. The method according to claim 1, characterized in that The obtaining of the signal-to-noise ratio, stability and interference level of the link between the weak node and the adjacent node in the energy depression area, wherein the weak node is a node whose predicted remaining energy is lower than a preset threshold of the network average value and is located in the energy depression area, and performing adaptive power control and data compression on the weak node, includes: Obtain the predicted remaining energy values ​​of all nodes in the energy depression area, and use the sliding window averaging method to calculate the average remaining energy of the network, where the sliding window size is a preset time period; Determine a preset threshold value according to the average remaining energy of the network, wherein the preset threshold value is a preset percentage of the average remaining energy of the network; Determine whether the predicted remaining energy value of the node is lower than the preset threshold, and if the predicted remaining energy value of the node is lower than the preset threshold, identify the node as a weak node; Obtaining the signal-to-noise ratio, stability and interference level of the link between the weak node and the adjacent nodes; constructing a link quality assessment matrix according to the signal-to-noise ratio, the stability and the interference level; Determining a transmit power adjustment value according to the link quality assessment matrix through a fuzzification, reasoning and defuzzification process; Adaptively compressing the data to be transmitted by the weak node; A compression ratio is determined according to the link quality evaluation matrix, and the compression ratio is in inverse proportion to the link quality evaluation matrix.

6. The method according to claim 1, characterized in that The distribution of key edge nodes, node predicted remaining energy, and node predicted energy consumption rate are used as state inputs of the reinforcement learning algorithm. By establishing a reward function combining energy consumption balance, transmission delay, and throughput, the data transmission path selection strategy is guided to optimize the energy consumption balance of the entire network and the efficient transmission of key data, and a dynamically optimized data transmission path decision is obtained, including: Obtain the distribution of key edge nodes, node predicted remaining energy, and node predicted energy consumption rate, and map the acquired information into state space for building a reinforcement learning environment; According to the state space, the energy balance, transmission delay and throughput are calculated to obtain a reward function R, where R = energy balance weight * energy balance + transmission delay weight * (1 / transmission delay) + throughput weight * throughput; A deep Q network algorithm is used to implement the reinforcement learning process, wherein the input layer of the deep Q network includes node position coordinates, predicted remaining energy, and predicted energy consumption rate; Setting a replay buffer to store the state, action, reward and next state of each time step, and randomly sampling training samples from the replay buffer, where the training samples are used for model training; The loss function is minimized by the gradient descent method, and the target network parameters are updated every preset time step to obtain a dynamically optimized data transmission path decision.

7. The method according to claim 1, characterized in that The evaluation of the stability of the data transmission path decision is to obtain a path change frequency index per unit time. If the index exceeds a preset threshold, a path smoothing mechanism is started, and an inertia factor is introduced to control the path adjustment amplitude, thereby ensuring the convergence of the path adjustment and reducing the additional energy consumption and delay caused by frequent path switching, including: A sliding window method is used to obtain a path change frequency index within a unit time, wherein the path change frequency index is obtained by dividing the number of path changes within a calculation window by the window size; Comparing the path change frequency index with a preset frequency threshold, if the path change frequency index exceeds the frequency threshold, triggering a path smoothing mechanism; Path smoothing is achieved using an exponential moving average method, wherein the path smoothing includes determining a new path node sequence, wherein the new path node sequence is obtained by multiplying the current path node sequence by α plus (1-α) multiplied by the decision path node sequence, where α is an inertia factor; Adopting an adaptive α value adjustment strategy, the adaptive α value adjustment strategy includes: obtaining a network stability index, the network stability index is obtained by calculating the variance of node residual energy and link quality in a recent period of time; Dynamically adjust the α value according to the network stability indicator; Obtaining the convergence speed of the path adjustment and the network performance change, wherein the convergence speed is obtained by calculating the average similarity of multiple consecutive path updates, and the network performance change includes the change rate of end-to-end delay, throughput and energy consumption indicators; According to the convergence speed and the network performance change, a fuzzy controller is used to adjust the α value, and the fuzzy controller is used to minimize the additional energy consumption and delay caused by path switching while ensuring convergence.

8. The method according to claim 1, characterized in that: The network life cycle under different data transmission strategies is evaluated by Monte Carlo simulation method based on the edge computing power security channel connectivity, edge computing power security channel coverage, and end-to-end data transmission delay factors to obtain the optimal transmission parameter configuration for extending the network life cycle, including: Define the security channel connectivity index of the edge computing power security channel network model, where the security channel connectivity index is the end-to-end connectivity probability of the channel; Calculating the security channel connectivity index; Obtain a security channel coverage index of the edge computing power security channel network model, where the security channel coverage index is a proportion of network nodes covered by at least one security channel; Establishing a parameter space sampling scheme according to the security channel connectivity index and the security channel coverage index; Discretize the node transmission power, data compression ratio, data encryption strength and routing selection weight to obtain a parameter combination set; A Monte Carlo method is used to perform a network life cycle simulation, wherein the network life cycle simulation performs repeated experiments for each set of parameters; Simulating a network operation process, wherein the simulation includes packet generation, routing selection, transmission, and energy consumption calculation; Recording the length of the network life cycle, wherein the recording is performed when the energy of the first node is exhausted; A non-dominated sorting genetic algorithm is used for multi-objective optimization, wherein the multi-objective optimization is based on a network life cycle simulation result; A Pareto optimal solution set is obtained, and an optimal transmission parameter configuration is determined from the Pareto optimal solution set.

Images