A Dynamic Optimization Method, Device, and Medium for a Centralized Mesh Network

Abstract

The present invention provides a dynamic optimization method, device and medium for a centralized Mesh network, which relates to the technical field of dynamic optimization of centralized Mesh networks. Starting from the current moment, the present invention collects the operation timing data and service request timing data of the Mesh network, including information such as the remaining energy of nodes, link quality and bandwidth utilization rate, classifies the service request data by priority, counts the request quantity and data volume of each type of service, and generates a service request feature vector; uses a time series modeling method to extract the features of the network operation data, dynamically allocates feature weights by using an attention mechanism, and generates a joint feature vector through dimensionality reduction processing to provide input features for multi-objective optimization; constructs a global optimization path planning model according to the generated joint feature vector, applies a genetic algorithm to solve the multi-objective optimization function, and finds the Pareto optimal solution, so as to output the path planning scheme with the highest fitness.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic optimization of a centralized Mesh network, and specifically to a method, device, and medium for dynamic optimization of a centralized Mesh network. Background Art

[0002] The path planning problem in the current complex network environment faces the challenge of multi-objective optimization. Especially when considering multiple constraints such as delay, packet loss rate, and remaining energy of network nodes, traditional path planning algorithms often struggle to balance globality and diversity. Most existing methods use fixed weight models or simple heuristic algorithms. Although they have high computational efficiency, they are prone to falling into local optima in a dynamic network environment, resulting in inflexible path selection and difficulty in meeting real-time and reliability requirements. In addition, the unbalanced utilization of network resources can cause some nodes to deplete energy prematurely, thereby shortening the network lifespan and reducing overall performance. This limitation is particularly prominent in energy-constrained networks such as wireless sensor networks, and a more intelligent and adaptable optimization method is needed to balance various performance metrics.

[0003] In the prior art, the path planning problem of a centralized Mesh network usually focuses on single-objective optimization through some traditional algorithms, mainly aiming at minimizing path delay. Such methods are computationally fast but ignore factors such as packet loss rate and node energy, and are less adaptable to resource-constrained network environments. Secondly, some improved heuristic algorithms can achieve multi-objective optimization to a certain extent by adjusting parameter weights to balance performance metrics such as delay and energy consumption. However, they usually require fixed weight values in advance and are difficult to adapt to the dynamically changing network environment in real time. In addition, there are also some machine learning-based methods that predict the optimal path through the training of historical data. However, in practical applications, due to high computational complexity and insufficient response to dynamics, it is difficult to be extended to scenarios with high real-time requirements.

[0004] In the actual application of a centralized Mesh network, the above prior art has obvious limitations, especially showing the following deficiencies in a dynamic network environment: First, traditional path planning methods are prone to falling into local optimal solutions and cannot find the comprehensive optimal path in multi-objective trade-offs, resulting in a decline in global performance; Second, heuristic algorithms with fixed weights lack the ability to respond to dynamic changes in network topology and service requirements during path planning and cannot achieve flexible path optimization; Third, existing machine learning or other complex algorithms have high computational overhead and are not suitable for energy-constrained or high-real-time scenarios. Therefore, the prior art is difficult to meet the actual needs in terms of how to balance delay, packet loss rate, and remaining energy in the network and provide dynamic optimization capabilities.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a dynamic optimization method, device and medium for a centralized Mesh network to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] A dynamic optimization method for a centralized Mesh network, the specific steps include:

[0009] Step 1: Set a time window, and collect the operation timing data and service request timing data of the Mesh network within the time window. The network operation timing data includes the remaining energy of each Mesh network node, the link quality between nodes, and the current bandwidth utilization rate data of each link. The service request timing data includes the data volume, priority, and timeliness index data of all services;

[0010] Step 2: Divide the collected service request timing data into high-priority, medium-priority, and low-priority services according to the service data volume, priority, and timeliness index, count the request quantity, total data volume, and service proportion of each type of service within the time window k, and combine them into a service request statistical feature vector;

[0011] Step 3: Use a time series modeling method to extract features from the collected network operation timing data to obtain node energy trend features, link quality features, and bandwidth utilization features, use an attention mechanism to assign feature weights, and generate a network dimensionality reduction feature vector through dimensionality reduction processing. At the same time, combine the service request statistical feature vector to generate a joint feature vector for the current time window;

[0012] Step 4: According to the joint feature vector of the current time window, construct a global optimization path planning model, use a genetic algorithm to solve the multi-objective optimization function, find the Pareto optimal solution, and finally output the global optimization path planning scheme with the highest fitness;

[0013] The time series data of the link quality feature has a packet loss rate P loss,i (t), which represents the change of the packet loss rate of link i over time within the time window, and the delay D i (t), which represents the end-to-end delay and delay jitter J of link i within the time window i (t), which represents the amplitude of the delay fluctuation of link i; the time series data of the bandwidth utilization rate B usa,i (t), t ∈ [t, t + k], which represents the bandwidth utilization rate of link i within the time window;

[0014] In path planning, the following core optimization objective constraints are defined first:

[0015] Minimum path delay: To ensure the real-time performance of services, the total delay D of the selected path path shall not exceed the maximum allowable end-to-end delay T of the service d , that is, the objective 1 is P path ≤T d , where D path is the total delay from the source node to the destination node, which is determined by the link delay characteristics;

[0016] Minimum link congestion: Avoid overloading the links with high bandwidth utilization in the path to improve the overall stability of the network. That is, the objective 2 is B usa,max ≤B threshold , where B usa,max is the maximum bandwidth utilization of all links in the path, and B threshold is the defined threshold of the maximum safe bandwidth utilization of the link;

[0017] Extend the network lifetime: Prioritize selecting paths with higher remaining energy to prevent premature failure of nodes. That is, the objective 3 is -E rem,min Maximize the remaining energy, where E rem,min is the lowest remaining energy of the nodes in the path, which is obtained through the node energy trend characteristics;

[0018] The multi-objective optimization function, that is, the defined fitness function is:

[0019] F(p i )=(D path (p i ),P loss (p i ),-E rem (p i ))

[0020] Use the Pareto dominance relationship to select the Pareto optimal solution set. If the path p a dominates the path p b , then p b will be excluded.

[0021] Furthermore, the logic for collecting the operation timing data and service request timing data of the Mesh network is as follows:

[0022] Taking the current moment t as the starting point of collection, define a dynamic time window k, and collect the network operation timing data and service request timing data within the time range [t, t + k];

[0023] Data collection is started in the "periodic trigger" mode: in periodic trigger, at fixed time intervals, the central node broadcasts a collection instruction to the whole network, requiring all nodes to report the latest status information;

[0024] The specific content of the collection is as follows:

[0025] The remaining energy of the node is set as E rem , and the collection method is that the central node collects the energy status of each node through a polling instruction every 1 minute, and E rem ∈[0, E total , where E total is the total energy of the node. The event trigger condition is that the energy decline rate is greater than 5% / minute, and the node needs to actively report the current remaining energy;

[0026] Link quality data measurement method: Heartbeat packets are exchanged between nodes every 10 seconds. The packet loss rate and delay are counted through ACK and RTT. The central node listens to the RSSI and SNR of the link communication, and takes the weighted average of the heartbeat packet measurement value and the central node listening value according to the weight ratio of 7:3 as the final link quality data;

[0027] The collection protocol for link bandwidth utilization rate is that the central node obtains the traffic statistics data of the switch port through SNMP, and the calculation formula is as follows:

[0028]

[0029] Among them, B usa,i is the bandwidth utilization rate of the i-th link, B used,i is the actual occupied bandwidth of the i-th link, and B total,i is the total bandwidth of the i-th link;

[0030] The service request data set for each time window contains the following information:

[0031] The service priority is set as P p , and the priority field is defined in the service request header: 0 is high, 1 is medium, 2 is low, and the digital signature of the initiating node needs to be attached; The service timeliness is set as T d , which is the maximum allowable end-to-end delay for each request and is automatically filled by the service type mapping table; The service data volume is set as R s , and when the service is initiated, the transmission data size is recorded through the request header.

[0032] Furthermore, the logic for dividing the collected service request time series data into high-priority, medium-priority, and low-priority services according to service data volume, priority, and timeliness indicators is:

[0033] Service requests are divided into high, medium, and low priorities, and need to meet any of the following conditions:

[0034] For the high-priority service set D high , if the service priority P p = 0, that is, the request is clearly marked as high-priority in the service request header, then the request needs to meet any of the following conditions: a. The service timeliness T d ≤ t threshold,high , that is, the maximum allowable end-to-end delay of the request is less than or equal to the timeliness threshold of high-priority;

[0035] b. The service data volume R s ≤ r threshold,high and T d ≤ t threshold,high , that is, the request data volume is small and the timeliness requirement is high; where, t threshold,high and r threshold,high are the set timeliness and quantity thresholds;

[0036] For the medium-priority service D mid , if the service request meets all of the following conditions: a. The service priority P p = 1, that is, the request is clearly marked as medium-priority in the service request header, b. The service timeliness t threshold,high < T d ≤ t threshold,mid , that is, the delay requirement of the request is between high-priority and low-priority; c. The service data volume r threshold,high < R s ≤ r threshold,mid , that is, the request data volume is moderate; where, t threshold,mid and r threshold,mid are the set thresholds;

[0037] For the low-priority service D low , if the service request meets any of the following conditions: a. If the service priority P p = 2, that is, the request is clearly marked as low-priority in the service request header, b. The service timeliness T d > t threshold,mid , that is, the maximum allowable end-to-end delay of the request is greater than the threshold of medium-priority and the service data volume R s > r threshold,mid , that is, a large amount of data needs to be transmitted and the delay requirement is not high;

[0038] The logic for counting the request quantity, total data volume, and service proportion of each type of service within the statistical time window k and combining them into a service request statistical feature vector is as follows:

[0039] For each type of service, that is, the three types of services: high-priority, medium-priority, and low-priority, respectively count the following three key features:

[0040] Count the number of requests N belonging to this type of service within the current time window k class , and the calculation formula is:

[0041] N class = Count(D class )

[0042] where D class is a set of a certain type of service, that is, D high , D mid or D low , class represents the index. When class = 1, it refers to high-priority services. When class = 2, it refers to medium-priority services. When class = 3, it refers to low-priority services;

[0043] Accumulate the transmission data volume of all requests of this type of service within the current time window k. The calculation formula is:

[0044]

[0045] where R s,n represents the data volume of the nth request, and S class is the total data volume statistics of a certain type of service;

[0046] Calculate the proportion of the total data volume of this type of service in the total data volume of all services within the current time window. The formula is as follows:

[0047]

[0048] where S total = S high + S mid + S low is the total data volume of all services, and P class is the proportion of the data volume of a certain type of service;

[0049] If N class = 0, then S class = 0 and P class = 0; If S total = 0, then set all proportions to P class = 0;

[0050] Combine the statistical results into a structured feature vector. The definition of the service request statistical feature vector V req is as follows:

[0051] V req = {N high , N mid , N low , S high,S mid ,S low ,P high ,P mid ,P low}

[0052] Among them, N high ,N mid ,N low are the request quantities of high, medium, and low-priority services respectively, and S high ,S mid ,S low are the total data volumes of high, medium, and low-priority services respectively, and P high ,P mid ,P low are the proportions of the total data volumes of high, medium, and low-priority services respectively;

[0053] After the feature vector generation is completed, consistency verification needs to be performed on the statistical results: request quantity consistency, N high +N mid +N low =N total , where N total is the total request quantity within the current time window k; data consistency, S high +S mid +S low =S total ; proportion sum verification, P high +P mid +P low ≈100%.

[0054] Furthermore, use the time series modeling method to extract features from the network operation time series data collected in step 1. The logic for extracting node energy trend features, link quality features, and bandwidth utilization features is as follows:

[0055] The node and link time series data within the time window [t, t + k] includes: the node remaining energy time series data E rem,i (t), t ∈ [t, t + k], indicating the change in the remaining energy of the i-th node within the time window;

[0056] Use LSTM to model various time series data, and build independent time series models for each type of time series data. For the input sequence of each type of time series data, it is set as x t (t), t ∈ [t, t + k]; for each input sequence, the output of LSTM is the hidden state h t or the final hidden state h k :

[0057] h t (t) = LSTM(xt , h t-1 , c t-1 )

[0058] Among them, x t is the input of the time series feature, h t-1 , c t-1 are the hidden state and candidate state of the previous time step respectively, and h t (t) captures the dynamic representation of the time series feature, and the finally output hidden state h k represents the trend feature of the time window [t, t + k];

[0059] Through the final hidden state h k of the LSTM model, the following trend feature sets are extracted: node energy trend feature. Suppose the total number of nodes in the network is N, and T node (t) = {h k (E rem,1 ), h k (E rem,2 ), …, h k (E rem,N )}; link quality trend feature, T link (t) = {h k (P loss,i ), h k (D i ), h k (J i )}, where i represents the i-th link quality trend feature; bandwidth utilization trend feature, T bw (t) = {h k (B usa,1 ), h k (B usa,2 ), …, h k (B usa,M )}, where M is the total number of links in the network;

[0060] The logic of using the attention mechanism to assign feature weights and generating the network dimensionality reduction feature vector through dimensionality reduction processing is as follows:

[0061] For each type of feature, the weights of its various dimensions of data are dynamically calculated through the attention mechanism, and the importance of the feature is associated with the target task. The formula is as follows:

[0062]

[0063] Among them, h o is the dynamic representation of feature o, which is the feature vector obtained after forward propagation in the later layer of the neural network. Feature o represents the dimensionality feature of the later layer in the network, and W qis a trainable parameter in the attention layer and is used to dynamically represent the feature h o to perform a linear transformation is W q 's transpose matrix, and α o is the importance weight of the feature o, and α o ∈[0,1], satisfying ∑ o α o = 1;

[0064] Weighted sum of the features to obtain the aggregated representation H after the attention mechanism aat :

[0065]

[0066] where the aggregated representation H aat is obtained by weighted summing all the dynamic representations h of the features o according to their importance weights α o ; According to the aggregated representation formula, the aggregated representations of different features are calculated. By weighted summing the dynamic representations of each dimension of the node energy feature, it is calculated as follows:

[0067]

[0068] where H node is the dynamic representation of the node energy feature, is the corresponding importance weight;

[0069] Thus, the aggregated representation of the link quality feature is obtained as H link and the aggregated representation of the bandwidth utilization feature is obtained as H bw .

[0070] Furthermore, the principal component analysis method is used to perform dimensionality reduction on the feature set after attention weighting. The input feature set: H = {H node , H link , H bw};

[0071] By eigenvalue decomposition of the feature covariance matrix, the PCA projection matrix W PCA is obtained. Using this projection matrix to perform dimensionality reduction on the original feature set, the dimensionality-reduced feature vector is calculated as follows:

[0072] H PCA = H · W PCA

[0073] where H is the original feature set and H PCA is the dimensionality-reduced feature vector; The output dimensionality-reduced network state feature vector is V dim = H PCA ;

[0074] Combined with the service request statistical feature vector in Step 2, the logic for generating the joint feature vector of the current time window is as follows:

[0075] Concatenate the dimension-reduced network state features with the service request statistical features in Step 2 to generate the joint feature vector of the current time window. The formula is as follows:

[0076] V union ={V dim ,V req}

[0077] Among them, V dim represents the network operation state trend feature within the time window, V req represents the service request statistical feature within the time window, and the joint feature vector V union combines the network state features and service request features.

[0078] Furthermore, combined with the joint feature vector of the current time window generated in Step 3, the logic for constructing a global optimization path planning model and using a genetic algorithm to solve the multi-objective optimization function to find the Pareto optimal solution is as follows:

[0079] The joint feature vector V union combines the network state features and service request features. Based on these features, construct a global optimization path planning model;

[0080] The path planning must meet the following constraint conditions:

[0081] Service timeliness constraint: D path ≤T d , ensuring that the selected path can meet the timeliness requirements of the service; Link quality constraint: Among them, P loss,i is the packet loss rate of the i-th link, P threshold is the maximum allowable packet loss rate threshold, and P represents a path; Node energy constraint: Among them, E rem,i is the remaining energy of the i-th node, and E threshold is the minimum required energy threshold of the node;

[0082] Using the dimension-reduced network state feature vector V dim and the service request statistical feature vector V req generated in Step 3, define the dynamic weights of nodes and links in the network topology:

[0083] Link weight w link,i , comprehensively considering the quality characteristics and bandwidth utilization rate of the link:

[0084] w link,i = α·D i + β·P loss,i + γ·B usa,i

[0085] where D i is the delay of link i, P loss,i is the packet loss rate of link i, B usa,i is the bandwidth utilization rate of link i, and α, β, γ are all weight parameters, and α + β + γ = 1;

[0086] The node weight w node,i , is calculated based on the remaining energy of the node:

[0087]

[0088] where E rem,i is the remaining energy of node i. When E rem,i is small, the node weight increases, thus avoiding selecting nodes with high load and fast energy consumption.

[0089] Furthermore, a weighted shortest path algorithm based on feature weights is adopted. Combining the feature vector in step 3 and the service characteristics in step 2, the optimal path is dynamically planned. The specific logic is as follows:

[0090] Model the network topology as a weighted directed graph G = (A, O). Define the node set A to represent all nodes in the network, and O to represent all links in the network. The link weight is defined by w link,i . Use the improved Dijkstra algorithm to plan the path based on the following path cost function:

[0091]

[0092] where C path is the total cost of path P, reflecting the comprehensive weight of the selected path, and P node is the set of all relay nodes on the path;

[0093] Construct the network topology graph G, map the link weight w link,i and the node weight w node,i into the graph. Input the starting point, ending point and delay constraint conditions of the service request. For each candidate path P, calculate its path cost C path , and filter out the paths that do not meet the following conditions:

[0094]

[0095] Select the path with the smallest total cost C path from the remaining candidate paths as the preliminary best path;

[0096] After obtaining the preliminary optimal path, the genetic algorithm is used to further optimize the path selection to find the Pareto optimal solution:

[0097] Use the genetic algorithm to find the Pareto optimal solution of the optimized path, and randomly generate the initial path population P = {p1, p2, …, p m}, where each path p i contains a series of nodes and the links between them. For each path p i calculate: the path delay D path (p i ), the path packet loss rate P loss (p i ), and the minimum remaining energy E rem (p i );

[0098] Adopt the tournament selection method to select paths with higher fitness from the current population as the parents of the next generation; perform crossover on the selected parents to generate new path individuals; perform random mutation on the generated new individuals to increase population diversity; iterate these steps until the termination condition is met, that is, the maximum number of iterations is reached. After each iteration, update the path population and retain the path with the highest fitness;

[0099] Finally, select a path p optimal from the Pareto optimal solution set, whose performance is the best overall, and the specific description is as follows:

[0100]

[0101] where F′(p i ) is the normalized comprehensive fitness value;

[0102] Output the detailed information of the selected path p optimal , including: the path node sequence, the path link sequence, the total path delay D path , the average path packet loss rate P loss , and the lowest remaining energy E rem ;

[0103] After outputting the path planning scheme, analyze the optimization effect: compare the changes in delay, packet loss rate, and remaining energy between the initial path and the final path, and evaluate the utilization efficiency of network resources by the optimization scheme.

[0104] The present invention also provides a dynamic optimization device for a centralized Mesh network. The device is used to implement the above-mentioned dynamic optimization method for a centralized Mesh network, and includes:

[0105] A data acquisition module, which is used to set a time window and collect the operation timing data and service request timing data of the Mesh network within the time window. The network operation timing data includes the remaining energy of each Mesh network node, the link quality between nodes, and the current bandwidth utilization rate data of each link. The service request timing data includes the data volume, priority, and timeliness index data of all services.

[0106] A data processing module, which is used to divide the collected service request timing data into high-priority, medium-priority, and low-priority services according to the service data volume, priority, and timeliness index, count the request quantity, total data volume, and service proportion of each type of service within the time window k, and combine them into a service request statistical feature vector.

[0107] A feature extraction module, which is used to extract features from the collected network operation timing data using a time series modeling method to obtain node energy trend features, link quality features, and bandwidth utilization features, allocate feature weights using an attention mechanism, and generate a network dimensionality reduction feature vector through dimensionality reduction processing. At the same time, combine with the service request statistical feature vector to generate a joint feature vector for the current time window.

[0108] An optimized path module, which is used to construct a global optimized path planning model according to the joint feature vector of the current time window, use a genetic algorithm to solve a multi-objective optimization function, find the Pareto optimal solution, and finally output the global optimized path planning scheme with the highest fitness.

[0109] The link quality feature time series data has a packet loss rate P loss,i (t), which represents the change of the packet loss rate of link i over time within the time window, and the delay D i (t), which represents the end-to-end delay and delay jitter J of link i within the time window i (t), which represents the amplitude of the delay fluctuation of link i; the bandwidth utilization time series data B usa,i (t), t ∈ [t, t + k], which represents the bandwidth utilization rate of link i within the time window.

[0110] In path planning, first define the following core optimization objective constraint conditions:

[0111] Minimum path delay: To ensure the real-time performance of services, the total delay D path of the selected path does not exceed the maximum allowable end-to-end delay T d of the service, that is, the objective 1 is D path ≤T d , where D path is the total delay from the source node to the destination node, which is determined by the link delay characteristics.

[0112] Minimum link congestion: Avoid overloading the links with high bandwidth utilization in the path to improve the overall stability of the network. That is, the target 2 is B usa,max ≤B threshold , where B usa,max is the maximum bandwidth utilization of all links in the path, and B threshold is the defined threshold of the maximum safe bandwidth utilization of the link;

[0113] Extend network lifetime: Prioritize paths with higher remaining energy to prevent premature failure of nodes. That is, the target 3 is -E rem,min Maximize the remaining energy, where E rem,min is the lowest remaining energy of nodes in the path, obtained through the node energy trend characteristics;

[0114] The multi-objective optimization function, that is, the defined fitness function is:

[0115] F(p i )=(D path (p i ),P loss (p i ),-E rem (p i ))

[0116] Use the Pareto dominance relationship to select the Pareto optimal solution set. If the path p a dominates the path p b , then p b will be excluded.

[0117] The present invention further provides a storage medium. The storage medium stores a computer program. When the computer program is executed by a processor, each step in the above-mentioned dynamic optimization method of a centralized Mesh network is implemented.

[0118] Compared with the prior art, the beneficial effects of the present invention are:

[0119] By collecting the operation timing data and service request timing data of the Mesh network, this data collection mechanism can not only reflect the operation state of the network in real time and dynamically, but also provide key input information for path planning, thus solving the problem of insufficient response ability to the dynamic changes of network topology and service requirements in the background technology; classify and statistically analyze the collected service request timing data, and generate service request statistical feature vectors. By classifying services according to priority and combining information such as the number of requests, total data volume, and service proportion within the statistical time window, the overall distribution characteristics of service requirements can be evaluated more accurately.

[0120] The present invention uses a timing modeling method to extract features from network operation timing data, and assigns feature weights through an attention mechanism. This innovative technical feature can dynamically perceive changes in node energy trends, link quality, and bandwidth utilization, and assign different weights to these features, further highlighting the impact of key performance indicators on path planning. Compared with simply processing timing data in the prior art, the present invention significantly improves the accurate modeling ability of the network dynamic environment through the attention mechanism. In addition, the introduction of the dimensionality reduction processing technology reduces the computational complexity by reducing the feature dimensions, providing a more concise and efficient input feature for global optimization path planning.

[0121] The present invention combines the generated joint feature vectors to construct a global optimization path planning model, and solves the multi-objective optimization function through a genetic algorithm to find the Pareto optimal solution. The introduction of the genetic algorithm enables the path planning model to have the global search ability, and can find the path with the highest fitness in the multi-objective trade-off, avoiding the defect that the traditional algorithm is easy to fall into the local optimum. In addition, the Pareto optimization technology enables the present solution to simultaneously optimize multiple objectives such as the delay, packet loss rate, and remaining energy of the path, rather than only focusing on a single performance, thus solving the limitation of insufficient multi-objective optimization ability in the prior art. Brief Description of the Drawings

[0122] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0123] Figure 2 It is a flowchart of the overall system module of the present invention. Detailed Embodiments

[0124] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with specific embodiments.

[0125] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meaning understood by those of ordinary skill in the field to which the present invention belongs. The "first", "second", and similar terms used in the present invention do not denote any order, quantity, or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0126] Example:

[0127] Please refer to Figure 1 , the present invention provides a technical solution:

[0128] A dynamic optimization method for a central Mesh network, the specific steps include:

[0129] Step 1: Set a time window, and collect the operation timing data and service request timing data of the Mesh network within the time window. The network operation timing data includes the remaining energy of each Mesh network node, the link quality between nodes, and the current bandwidth utilization rate data of each link. The service request timing data includes the data volume, priority, and timeliness index data of all services;

[0130] The logic for collecting the operation timing data and service request timing data of the Mesh network is:

[0131] Taking the current moment t as the starting point of collection, define a dynamic time window k, and collect the network operation timing data and service request timing data within the time range [t, t + k];

[0132] Data collection is started by the method of "periodic trigger": periodic trigger, at a fixed time interval, the central node broadcasts a collection instruction to the whole network, requiring all nodes to report the latest status information;

[0133] The specific content of the collection is as follows:

[0134] The remaining energy of the node is set as E rem , and the collection method is that the central node collects the energy status of each node through a polling instruction every 1 minute, and E rem ∈[0, E total , where E total is the total energy of the node, and the event trigger condition is that the energy decline rate is greater than 5% / minute, and the node needs to actively report the current remaining energy;

[0135] The link quality data Q link Measurement method: Nodes exchange heartbeat packets every 10 seconds, and calculate the packet loss rate and delay through ACK and RTT. The central node listens to the RSSI and SNR of the link communication, and takes the weighted average of the heartbeat packet measurement value and the central node listening value according to the weight ratio of 7:3 as the final link quality data, that is, Q link = 0.7·RSSI + 0.3·SNR, and the collection of this link quality data needs to meet the packet loss rate P loss > 10%;

[0136] Q linkIndicates the comprehensive quality of the link. The higher it is, the better the link quality and the more stable the transmission performance. RSSI reflects the signal strength. The higher the signal strength, the better the link quality. SNR represents the ratio of signal to noise. The higher it is, the less signal interference there is;

[0137] The acquisition protocol for link bandwidth utilization rate is that the central node obtains the traffic statistics data of the switch port through SNMP. The calculation formula is as follows:

[0138]

[0139] Among them, B usa,i is the bandwidth utilization rate of the i-th link, B used,i is the actual occupied bandwidth of the i-th link, B total,i is the total bandwidth of the i-th link;

[0140] When B usad,i increases and B usa,i also increases, it indicates that the burden on the network increases, and vice versa. High link utilization rate leads to network congestion and affects data transmission efficiency;

[0141] The service request data set for each time window contains the following information:

[0142] The service priority is set to P p , and the priority field is defined in the service request header: 0 is high, 1 is medium, 2 is low, and the digital signature of the initiating node needs to be attached; the service timeliness is set to T d , which is the maximum allowable end-to-end delay for each request and is automatically filled by the service type mapping table; the service data volume is set to R s , and when the service is initiated, the transmission data size is recorded through the request header;

[0143] P p affects resource allocation and scheduling. The higher the priority, the more urgent it is to be processed. T d ensures that the service request meets the delay requirements and affects the user experience. R s directly affects the network bandwidth and resource usage. The larger the data volume, the more network resources are required; the relationship between service priority, timeliness, and data volume: High-priority service requests usually require lower delays and higher bandwidth usage to ensure that the service can be completed in a timely manner. As the data volume increases, the timeliness requirements are affected, and network resources need to be dynamically adjusted, resulting in priority changes.

[0144] Step 2: Divide the collected service request time series data into high-priority, medium-priority, and low-priority services according to the service data volume, priority, and timeliness indicators. Count the number of requests, total data volume, and service proportion of each type of service within the time window k, and combine them into a service request statistical feature vector;

[0145] The logic for classifying the collected business request time-series data into high-priority, medium-priority, and low-priority services based on business data volume, priority, and timeliness metrics is as follows:

[0146] Business requests are classified into high, medium, and low priorities and need to meet any of the following conditions:

[0147] For the high-priority service set D high , if the service priority P p = 0, that is, the request is clearly marked as high-priority in the business request header, then the request needs to meet any of the following conditions: a. The service timeliness T d ≤ t threshold,high , that is, the maximum allowable end-to-end delay of the request is less than or equal to the delay threshold of high-priority;

[0148] b. The business data volume R s ≤ r threshold,high and T d ≤ t threshold,high , that is, the request data volume is small and the timeliness requirement is high; where t threshold,high and r threshold,high are delay and quantity threshold values dynamically set according to the service type or application scenario;

[0149] The threshold value t threshold,high is dynamically set. According to the requirements of high-priority defined by the service type, if T d is smaller, it means that the delay requirement of this service is higher and the timeliness is stronger. If T d ≤ t threshold,high , then the service meets the delay requirement of high-priority; T d represents the delay requirement of the transmission request and reflects the time sensitivity of the service;

[0150] R s represents the transmission requirement of the service. The smaller the data volume, the easier it is to transmit it quickly, thus meeting the timeliness requirement. The two threshold values jointly determine which services have high priority; R s is smaller and T d is smaller, the greater the possibility that the service meets the high-priority;

[0151] For the medium-priority service D mid , if the service request meets all of the following conditions: a. The service priority P p = 1, that is, the request is clearly marked as medium-priority in the business request header, b. The service timeliness t threshold,high < T d ≤ t threshold,mid , that is, the delay requirement of the request is between high-priority and low-priority; c. The business data volume r threshold,high < Rs ≤r threshold,mid , that is, the requested data volume is moderate; where t threshold,mid and r threshold,mid are threshold values set for medium-delay services;

[0152] T d The higher it is, the looser the delay requirement of the service is. The delay required by medium priority is between that of high priority and low priority; R s reflects the transmission requirement of the service, and a moderate data volume meets the requirements of medium priority;

[0153] For low-priority service D low , if the service request meets any of the following conditions: a. If the service priority P p = 2, that is, the request is clearly marked as low priority in the service request header, b. The service timeliness T d >t threshold,mid , that is, the maximum allowable end-to-end delay of the request is greater than the threshold value of medium priority and the service data volume R s >r threshold,mid , that is, a large amount of data needs to be transmitted and the delay requirement is not high;

[0154] Low-priority services have loose delay requirements and allow a large delay. Low-priority services allow a large data volume and usually have low requirements for transmission timeliness;

[0155] The logic for counting the number of requests, total data volume, and service proportion of each type of service within the statistical time window k and combining them into a service request statistical feature vector is:

[0156] For each type of service, that is, the three types of services: high priority, medium priority, and low priority, respectively count the following three key features:

[0157] Count the number of requests N class belonging to this type of service within the current time window k. The calculation formula is:

[0158] N class = Count(D class )

[0159] where D class is a set of a certain type of service, that is, D high , D mid or D low , class represents the index. When class = 1, it refers to high-priority services. When class = 2, it refers to medium-priority services. When class = 3, it refers to low-priority services;

[0160] N classThe number of requests for a certain type of business, which counts the total number of requests belonging to a certain type of business within the current time window k, reflecting the quantity distribution of this type of business in the total requests; D class The more requests in class , the larger N

[0161] Accumulate the transmission data volume of all requests of this type of business within the current time window k. The calculation formula is:

[0162]

[0163] Among them, R s,n represents the data volume of the nth request, and S class is the total data volume statistics of a certain type of business;

[0164] S class Statistics the total transmission data volume of a certain type of business within the current time window k, reflecting the traffic distribution of this type of business. The larger N class or R s,n , the larger S class ;

[0165] Calculate the proportion of the total data volume of this type of business in the total data volume of all businesses within the current time window. The formula is as follows:

[0166]

[0167] Among them, S total =S high +S mid +S low is the total data volume of all businesses, and P class is the proportion of the data volume of a certain type of business; P class reflects the proportion of a certain type of business in the total traffic within the current time window k, and is used to measure the traffic distribution of different priority businesses. The larger S class , the larger P class . The increase in S class results in a decrease in P class ;

[0168] If N class =0, then S class =0 and P class =0; if S total =0, then set all proportions to P class =0;

[0169] Combine the statistical results into a structured feature vector. The definition of the business request statistical feature vector V req is as follows:

[0170] V req ={N high ,Nmid , N low , S high , S mid , S low , P high , P mid , P low}

[0171] Among them, N high , N mid , N low are the request quantities of high, medium, and low-priority services respectively, S high , S mid , S low are the total data volumes of high, medium, and low-priority services respectively, P high , P mid , P low are the proportions of the total data volumes of high, medium, and low-priority services respectively;

[0172] After the feature vector generation is completed, the statistical results need to be subjected to consistency verification: request quantity consistency, N high +N mid +N low =N total , where N total is the total request quantity within the current time window k; data consistency, S high +S mid +S low =S total ; proportion sum verification, P high +P mid +P low ≈100%.

[0173] Step 3: Use the time series modeling method to extract features from the collected network operation time series data, obtain node energy trend features, link quality features, and bandwidth utilization features, allocate feature weights using the attention mechanism, and generate a network dimensionality reduction feature vector through dimensionality reduction processing. At the same time, combine the service request statistical feature vector to generate the joint feature vector of the current time window;

[0174] Use the time series modeling method to extract features from the network operation time series data collected in Step 1. The logic for extracting node energy trend features, link quality features, and bandwidth utilization features is as follows:

[0175] The node and link time series data within the time window [t, t + k] includes: the node remaining energy time series data E rem,i (t), t ∈ [t, t + k], representing the change in the remaining energy of the i-th node within the time window; the higher the energy value, the stronger the working ability and duration of the node, otherwise the node cannot continue to provide services;

[0176] The time - series data of link - quality characteristics has a packet - loss rate P loss,i (t), representing the change of the packet - loss rate of link i over time within a time window, and the delay D i (t), representing the end - to - end delay and delay jitter J of link i within a time window i (t), representing the amplitude of the delay fluctuation of link i within a time window. These characteristics describe the performance and stability of the link, and affect the reliability and efficiency of data transmission; The time - series data of bandwidth utilization B usa,i (t), t ∈ [t, t + k], representing the bandwidth usage rate of link i within a time window. A high bandwidth utilization indicates a heavy network load, which affects the quality of service;

[0177] Use LSTM to model various types of time - series data, and construct independent time - series models for each type of time - series data. For the input sequence of each type of time - series data, it is set as x t (t), t ∈ [t, t + k]; For each input sequence, the output of LSTM is the hidden state h within the time window t or the final hidden state h k :

[0178] h t (t)=LSTM(x t ,h t-1 ,c t-1 )

[0179] Among them, x t is the input of time - series features, h t-1 ,c t-1 are the hidden state and candidate state of the previous time step respectively. h t (t) captures the dynamic representation of time - series features, and the finally output hidden state h k represents the trend features of the time window [t, t + k];

[0180] h t (t) is the hidden state of the current time step, capturing the dynamic representation of time - series features. LSTM can output the hidden state h t (t) by learning the dynamic changes of historical time - series data. The higher h

[0181] Through the final hidden state h of the LSTM model k , extract the following set of trend features: Node energy trend features. Suppose there are N nodes in total in the network, T node (t)={h k (E rem,1 ),h k (E rem,2 ),…,hk (E rem,N )} Extract the energy trend features of all nodes, including the final hidden states of N nodes; the higher the h k (E rem,i ), indicating that the energy trend of node i is upward, enhancing the stability and service ability of the network;

[0182] Link quality trend feature, T link (t) = {h k (P loss,i ), h k (D i ), h k (J i )}, where i represents the i-th link quality trend feature; extract the trend features of the packet loss rate, delay, and delay jitter of the link. The higher each feature value, usually the lower the link quality, such as high packet loss rate and high delay, affecting the reliability of the network;

[0183] Bandwidth utilization trend feature, T bw (t) = {h k (B usa,1 ), h k (B usa,2 ), …, h k (B usa,M )}, where M is the total number of links in the network; extract the bandwidth usage trend features of all links, including the final hidden states of M links. A high h k (B usa,i ), value indicates that link i has a heavy load, resulting in network congestion;

[0184] The logic of using the attention mechanism to assign feature weights and generating the network dimensionality reduction feature vector through dimensionality reduction processing is as follows:

[0185] For each type of feature, the weights of its various dimensions of data are dynamically calculated through the attention mechanism, associating the importance of the feature with the target task. The formula is as follows:

[0186]

[0187] Among them, h o is the dynamic representation of feature o, which is the feature vector obtained after forward propagation in the subsequent layer of the neural network, reflecting the state of the feature in the current context. Feature o represents the dimensionality feature of the subsequent layer in the network. W q is the trainable parameter in the attention layer, used to perform a linear transformation on the feature dynamic representation h o , is the transpose matrix of W q , α ois the importance weight of feature o, reflecting the importance of this feature, and α o ∈ [0, 1], satisfying ∑ o α o = 1; The higher the weight α o is, the more important this feature is in the current task. The sum of the weights is 1, ensuring that the relative importance of features can be normalized;

[0188] In the attention mechanism, the main role of W q is to convert the feature representation h o into a query vector, enabling the model to calculate the correlation between features and the target task. After the feature passes through the transformation of the W q matrix, the formed vector directly participates in the subsequent calculation of attention weights, thereby determining the contribution size of each feature in the aggregated representation;

[0189] In the formula, is part of the query vector and is used to calculate the importance of each feature's dynamic representation h o . By calculating the dot product , a scalar value can be obtained, representing the similarity or correlation between feature o and the query. This scalar value will then be exponentiated and normalized to generate the weight α o . Therefore, 's transpose operation ensures the dimensional compatibility of matrix operations and plays a key role in the attention mechanism;

[0190] Performs a weighted sum of the features to obtain the aggregated representation H aat after the attention mechanism:

[0191]

[0192] where the aggregated representation H aat is obtained by performing a weighted sum of all feature dynamic representations h o according to their importance weights α o ; According to the aggregated representation formula, the aggregated representations of different features are calculated. By performing a weighted sum of the dynamic representations of each dimension of the node energy feature, it is calculated as:

[0193]

[0194] where H node is the dynamic representation of the node energy feature, is the corresponding importance weight;

[0195] Thus, the aggregated representation of the link quality feature is H link and the aggregated representation of the bandwidth utilization feature is H bw; These aggregated representations not only retain the important information related to the target task in the original multi-dimensional features but also reduce the data dimension, simplifying the input complexity of the subsequent multi-objective optimization model;

[0196] Use the principal component analysis method to perform dimensionality reduction on the feature set after attention weighting. The input feature set: H = {H node , H link , H bw};

[0197] Through the eigenvalue decomposition of the feature covariance matrix, obtain the PCA projection matrix W PCA , and use this projection matrix to perform dimensionality reduction on the original feature set. Calculate the reduced feature vector:

[0198] H PCA = H · W PCA

[0199] where H is the original feature set, and H PCA is the reduced feature vector; the output reduced network state feature vector is V dim = H PCA ;

[0200] W PCA is the projection matrix obtained through the eigenvalue decomposition of the feature covariance matrix. After dimensionality reduction, the most important feature information is retained, and reducing the feature dimension helps the subsequent training and computational efficiency of the model;

[0201] Meanwhile, the logic of generating the joint feature vector of the current time window by combining the business request statistical feature vector in step 2 is as follows:

[0202] Concatenate the reduced network state feature with the business request statistical feature in step 2 to generate the joint feature vector of the current time window. The formula is:

[0203] V union = {V dim , V req}

[0204] where V dim represents the network operation state trend feature within the time window, V req represents the business request statistical feature within the time window, and the joint feature vector V union combines the network state feature and the business request feature;

[0205] The different parts of the joint feature vector V union together provide a comprehensive perspective on the network state and business requests, which helps subsequent decision-making and optimization.

[0206] Step 4: Based on the joint feature vector of the current time window, construct a global optimization path planning model, use the genetic algorithm to solve the multi-objective optimization function, find the Pareto optimal solution, and finally output the global optimization path planning scheme with the highest fitness;

[0207] Combined with the joint feature vector of the current time window generated in Step 3, construct a global optimization path planning model. The logic of using the genetic algorithm to solve the multi-objective optimization function and find the Pareto optimal solution is as follows:

[0208] Joint feature vector V union Integrates network state features and service request features. Based on these features, construct a global optimization path planning model. In path planning, first define the following core optimization objective constraints:

[0209] Minimum path delay: To ensure the real-time performance of the service, the total delay D of the selected path path Does not exceed the maximum allowable end-to-end delay T of the service d , that is, the objective 1 is D path ≤T d , where D path Is the total delay from the source node to the destination node, determined by the link delay characteristics;

[0210] D path That is, the total delay of the selected path, reflecting the total delay from the source node to the destination node, describing the real-time performance and response speed of the network. The smaller it is, the faster the link transmission speed on the path, which is beneficial to meeting the timeliness requirements; T is the maximum allowable end-to-end delay of the service request, determined by the characteristics of the service request, restricting the upper limit of the path delay;

[0211] If D path >T d , the path does not meet the service timeliness requirements and will be filtered out. The larger T d , the looser the network allows the path delay limit;

[0212] Minimum link congestion: Avoid overloading links with high bandwidth utilization in the path to improve the overall stability of the network. That is, the objective 2 is B usa,max ≤B threshold , where B usa,max Is the maximum bandwidth utilization rate of all links in the path, and B threshold Is the defined threshold of the maximum safe bandwidth utilization rate of the link, defining the load upper limit of the link in the path to avoid network overload;

[0213] B usa,max , the maximum bandwidth utilization rate of all links in the path, reflecting the load situation of the most congested link on the path. The larger its value, the closer the load of a certain link in the path is to its bandwidth limit, resulting in congestion; If Busa,max >B threshold If the path does not meet the link congestion requirement, it will be filtered out, B threshold The larger it is, the higher the load allowed for the link;

[0214] Extend network lifetime: Prioritize paths with higher remaining energy to prevent premature node failure, i.e., objective 3 is -E rem,min Maximize the remaining energy, where E rem,min is the lowest remaining energy of the nodes in the path, obtained through the node energy trend feature, reflecting the remaining energy state of the weakest node in the path. The larger its value, the more sufficient the energy of all nodes in the path, and the network lifetime and node usage time can be extended;

[0215] The optimization objective is to maximize -E rem,min That is, select paths with higher remaining energy to avoid premature node failure. If the remaining energy of a certain node in the path is lower than the threshold, the path will be filtered;

[0216] Path planning must meet the following constraint conditions:

[0217] Business timeliness constraint: D path ≤T d , ensuring that the selected path can meet the business timeliness requirements; Link quality constraint: Among them, P loss,i is the packet loss rate of the i-th link, P threshold is the maximum allowable packet loss rate threshold. The upper limit of the packet loss rate is determined by the network service requirements. P represents a path; P loss,i reflects the reliability of link transmission. The larger its value, the more serious the packet loss situation of the link, resulting in data loss or retransmission; If P loss,i >P threshold , the path does not meet the link quality requirements and will be filtered out;

[0218] Node energy constraint: Among them, E rem,i is the remaining energy of the i-th node, E threshold is the minimum required energy threshold of the node, determining the energy lower limit of all nodes in the path; E rem,i reflects the current energy state of the node. The smaller its value, the more energy the node consumes, affecting the service life of the path; If E rem,i <E threshold , the path will be filtered out;

[0219] Utilize the dimension-reduced network state feature vector V dim generated in step 3 and the service request statistical feature vector V req , and define the dynamic weights of nodes and links in the network topology:

[0220] Link weight w link,i , comprehensively considering the quality characteristics and bandwidth utilization rate of the link:

[0221] w link,i = α·D i + β·P loss,i + γ·B usa,i

[0222] Among them, D i is the delay of link i, P loss,i is the packet loss rate of link i, B usa,i is the bandwidth utilization rate of link i, α, β, γ are all weight parameters, and α + β + γ = 1;

[0223] w link,i reflects the comprehensive impact of the delay, packet loss rate and bandwidth utilization rate of the link. D i is the delay of link i, the larger the value, the higher the weight. P loss,i is the packet loss rate of link i, the larger the value, the higher the weight. B usa,i is the bandwidth utilization rate of link i, the larger the value, the higher the weight; α is the weight of the link delay D i reflecting the importance of the delay to the link weight, β is the weight of the link packet loss rate P loss,i reflecting the importance of the packet loss to the link weight, γ is the weight of the bandwidth utilization rate B usa,i reflecting the importance of the bandwidth utilization to the link weight;

[0224] When setting parameters in different network environments, if the sensitivity to delay is high, such as applications with strong real-time requirements, set α larger. The possible relationship is: α > β, γ; if the packet loss rate has a significant impact on the application, such as voice calls or video streams, β can be set larger. The possible relationship is: β > α, γ; if the bandwidth utilization rate is the main consideration factor, such as big data transmission, γ can be set larger. The possible relationship is: γ > α, β;

[0225] Node weight w node,i , calculated based on the remaining energy of the node:

[0226]

[0227] Among them, E rem,i is the remaining energy of node i. When E rem,i is small, the node weight increases, thus avoiding selecting nodes with high load and fast energy consumption; w node,i reflects the inverse ratio of the remaining energy of the node. The larger its value, the lower the remaining energy of the node and the higher the weight, thus reducing the possibility of selecting this node; E rem,i The larger it is, the node weight w node,iThe smaller it is, the more inclined to select nodes with sufficient remaining energy;

[0228] Adopt a weighted shortest path algorithm based on feature weights, combine the feature vectors in step 3 and the service characteristics in step 2, and dynamically program the optimal path. The specific logic is as follows:

[0229] Model the network topology as a weighted directed graph G=(A, O), define the node set A to represent all nodes in the network, and O to represent all links in the network. The link weight is determined by w link,i Define and use an improved Dijkstra algorithm to plan the path based on the following path cost function:

[0230]

[0231] Among them, C path is the total cost of path P, reflecting the comprehensive weight of the selected path. P node is the set of all relay nodes on the path;

[0232] C path reflects the total weight of the links and nodes in the path. The smaller its value, the better the comprehensive performance of the path; an increase in the weight of any link or node will cause the total path cost C path to increase, thereby reducing the priority of selecting this path;

[0233] Construct the network topology graph G, map the link weight w link,i and the node weight w node,i to the graph, input the starting point, ending point and delay constraint conditions of the service request, and calculate the path cost C path for each candidate path P, and filter out the paths that do not meet the following conditions:

[0234]

[0235] Select the path with the smallest total cost C path from the remaining candidate paths as the preliminary best path;

[0236] Select paths with high fitness:

[0237] In each generation, first select paths with higher fitness from the current population, which is achieved through the tournament selection method. The specific process is as follows:

[0238] Adopt the tournament selection method to select paths with higher fitness from the current population as the parents of the next generation; perform crossover on the selected parents to generate new path individuals; perform random mutation on the generated new individuals to increase population diversity; iterate these steps until the termination condition is met, that is, the maximum number of iterations is reached. After each iteration, update the path population and retain the path with the highest fitness;

[0239] After obtaining the preliminary best path, use the genetic algorithm to further optimize the path selection to find the Pareto optimal solution:

[0240] Use the genetic algorithm to find the Pareto optimal solution of the optimized path. Randomly generate an initial path population P = {p1, p2,..., p m}, where each path p i contains a series of nodes and the links between them. For each path p i calculate: the path delay D path (p i ), the path packet loss rate P loss (p i ), and the minimum remaining energy E rem (p i );

[0241] Fitness definition: The fitness of each path is determined by its performance in terms of delay, packet loss rate, and remaining energy. Usually, the fitness function takes these factors into comprehensive consideration to ensure that the selected path performs well in multiple dimensions;

[0242] The fitness function is for multi-objective optimization, reflecting the performance of each path in terms of delay, packet loss rate, and remaining energy. The multi-objective optimization function defines the fitness function as:

[0243] F(p i ) = (D path (p i ), P loss (p i ), -E rem (p i ))

[0244] Use the Pareto dominance relationship to select the Pareto optimal solution set. If path p a dominates path p b , then p b will be excluded;

[0245] Fitness optimization direction: F(p i ) reflects the comprehensive performance of the path in terms of delay, packet loss rate, and remaining energy; D path (p i ) and P loss (pi ) The smaller it is, the higher the fitness; E rem (p i ) The larger it is, that is, -E rem (p i ) The smaller it is, the higher the fitness;

[0246] Crossover operation. The purpose of the crossover operation is to combine two parent paths into a new path individual, retain the advantages of the parent paths, and explore new possibilities. Crossover is usually done by swapping a part of the nodes or links in the parent paths, and the specific method can vary according to the way of path encoding;

[0247] Mutation operation is to randomly adjust the new path individual after crossover, which involves changing some nodes or links in the path, or randomly replacing some parts. The purpose of mutation is to prevent the algorithm from falling into a local optimal solution and at the same time explore a new solution space;

[0248] The Pareto optimal solution set means that there is no other path that can perform better in all dimensions in terms of these three dimensions: delay, packet loss rate, and remaining energy. Mathematically, if path p a dominates path p b , it means that:

[0249] D path (p a ) ≤ D path (p b ), P loss (p a ) ≤ P loss (p b ), -E rem (p a ) ≥ -E rem (p b )

[0250] That is, path p a is either better than, or equal to, path p b in all dimensions, and is better than path p b in at least one dimension;

[0251] Finally, select a path p optimal from the Pareto optimal solution set, which has the best comprehensive performance, that is, determine the optimal path according to the normalized comprehensive fitness value F(p i ), and the specific description is as follows:

[0252]

[0253] Among them, here F′(p i) is the normalized comprehensive fitness value, which reflects the comprehensive performance of the path under multiple objectives. Finally, the path with the optimal fitness is selected as the optimal solution;

[0254] Output the selected path p optimal Details, including: path node sequence: a list of nodes from the source node to the destination node and path link sequence: details of each link on the path, which may include link delay, bandwidth utilization, etc., and the total delay D of the path path That is, the sum of the delays of all links in the path, and the average packet loss rate P of the path loss That is, the average value of the packet loss rates of all links on the path, and the lowest remaining energy E of the path rem That is, the remaining energy of the weakest node in the path;

[0255] After outputting the path planning scheme, analyze the optimization effect: compare the changes in delay, packet loss rate, and remaining energy between the initial path and the final path, and evaluate the utilization efficiency of network resources by the optimization scheme;

[0256] Delay change, by comparing the delays of the initial path and the optimized path, evaluate the improvement of the optimization scheme on service timeliness; packet loss rate change, by comparing the changes in packet loss rate, evaluate the improvement of the reliability of the optimized path; remaining energy change, by comparing the lowest remaining energy of the path, evaluate the utilization efficiency of the optimized path for network resources, especially node energy;

[0257] By comparing the performance of the initial path and the optimized path, analyze whether the optimization scheme improves the utilization efficiency of network resources and whether it allocates network resources more reasonably to improve the overall stability and lifespan of the network.

[0258] Please refer to Figure 2 , the present invention also provides a dynamic optimization device for a centralized Mesh network, and the device is used to implement the above-mentioned dynamic optimization method for a centralized Mesh network, including:

[0259] A data acquisition module, which is used to set a time window and collect the operation timing data and service request timing data of the Mesh network within the time window. The network operation timing data includes the remaining energy of each Mesh network node, the link quality between nodes, and the current bandwidth usage data of each link. The service request timing data includes the data volume, priority, and timeliness index data of all services;

[0260] A data processing module, which is used to divide the collected service request timing data into high-priority, medium-priority, and low-priority services according to service data volume, priority, and timeliness index, count the request quantity, total data volume, and service proportion of each type of service within the time window k, and combine them into a service request statistical feature vector;

[0261] A feature extraction module, which is used to extract features from the collected network operation time series data by using a time series modeling method, obtain node energy trend features, link quality features, and bandwidth utilization features, allocate feature weights by using an attention mechanism, generate a network dimensionality reduction feature vector through dimensionality reduction processing, and combine it with a service request statistical feature vector to generate a joint feature vector for the current time window;

[0262] An optimization path module, which is used to construct a global optimization path planning model according to the joint feature vector of the current time window, solve a multi-objective optimization function by using a genetic algorithm, find the Pareto optimal solution, and finally output the global optimization path planning scheme with the highest fitness;

[0263] The time series data of the link quality feature has a packet loss rate P loss,i (t), which represents the change of the packet loss rate of link i within the time window over time, and the delay D i (t), which represents the end-to-end delay and delay jitter J of link i within the time window i (t), which represents the amplitude of the delay fluctuation of link i within the time window; the time series data of the bandwidth utilization rate B usa,i (t), t ∈ [t, t + k], which represents the bandwidth usage rate of link i within the time window;

[0264] In path planning, first define the following core optimization objective constraints:

[0265] Minimum path delay: To ensure the real-time performance of services, the total delay D path of the selected path does not exceed the maximum allowable end-to-end delay T d of the service, that is, the objective 1 is D path ≤T d , where D path is the total delay from the source node to the destination node, which is determined by the link delay feature;

[0266] Minimum link congestion: Avoid overloading links with high bandwidth utilization rates in the path to improve the overall stability of the network, that is, the objective 2 is B usa,max ≤B threshold , where B usa,max is the maximum bandwidth utilization rate of all links in the path, and B threshold is the defined link maximum safe bandwidth utilization rate threshold;

[0267] Prolong the network life: Prioritize selecting paths with higher remaining energy to prevent premature failure of nodes, that is, the objective 3 is -E rem,min to maximize the remaining energy, where E rem,min is the lowest remaining energy of the nodes in the path, which is obtained through the node energy trend feature;

[0268] The multi-objective optimization function, that is, the fitness function is defined as:

[0269] F(p i )=(D path (p i ),P loss (p i ),-E rem (p i ))

[0270] Use the Pareto dominance relationship to select the Pareto optimal solution set. If the path p a dominates the path p b , then p b will be excluded.

[0271] The present invention also provides a storage medium storing a computer program, which when executed by a processor, implements each step in the above-mentioned dynamic optimization method for a centralized Mesh network.

[0272] All the above formulas are dimensionless and take their numerical calculations. The formula is obtained by collecting a large amount of data for software simulation to obtain a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0273] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0274] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0275] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered within the protection scope of the present application.

Claims

1. A dynamic optimization method for a centralized Mesh network, characterized in that: The specific steps include: Step 1: Set a time window and collect the operation sequence data and service request sequence data of the Mesh network within the time window. The network operation sequence data includes the remaining energy of each Mesh network node, the link quality between nodes and the current bandwidth utilization rate of each link. The service request sequence data includes the data volume, priority and timeliness index data of all services. Step 2: Divide the collected service request time series data into high priority, medium priority and low priority services according to the service data volume, priority and timeliness indicators, and count the number of requests, total data volume and service proportion of each type of service in the time window k, and combine them into a service request statistical feature vector; Step 3: Use the time series modeling method to extract features from the collected network runtime time series data to obtain node energy trend features, link quality features, and bandwidth utilization features. Use the attention mechanism to assign feature weights, and generate network dimension reduction feature vectors through dimensionality reduction processing. Combined with the service request statistical feature vector, generate a joint feature vector for the current time window. Step 4: Based on the joint feature vector of the current time window, a global optimization path planning model is constructed, and a genetic algorithm is used to solve the multi-objective optimization function, find the Pareto optimal solution, and finally output the global optimization path planning solution with the highest fitness; The link quality characteristic time series data has packet loss rate P loss,i (t), represents the change of packet loss rate of link i over time in the time window, and the delay D i (t), represents the end-to-end delay and delay jitter J of link i in the time window i (t), represents the amplitude of the delay fluctuation of link i in the time window; bandwidth utilization time series data B usa,i (t),t∈[t,t+k], represents the bandwidth utilization of link i in the time window; In path planning, the following core optimization objective constraints are defined first: Minimum path delay: To ensure the real-time performance of the service, the total delay D of the selected path path Does not exceed the maximum allowable end-to-end delay T of the service d , that is, target 1 is P path ≤T d , where D path is the total delay from the source node to the destination node, which is determined by the link delay characteristics; Minimum link congestion: Avoid overloading of links with high bandwidth utilization in the path to improve the overall stability of the network, that is, goal 2 is B usa,max ≤B threshold , where B usa,max is the maximum bandwidth utilization of all links in the path, B threshold The maximum safe bandwidth utilization threshold of the defined link; Prolong network life: Prioritize paths with higher residual energy to prevent premature node failure, that is, goal 3 is -E rem,min Maximize the remaining energy, where E rem,min is the minimum residual energy of the nodes in the path, obtained through the node energy trend characteristics; The multi-objective optimization function defines the fitness function as: F(p i )=(D path (p i ),P loss (p i ),-E rem (p i )) Use the Pareto dominance relationship to select the Pareto optimal solution set. If the path p a Dominant Path p b , then p b Excluded.

2. The method for dynamic optimization of a centralized Mesh network according to claim 1, characterized in that: The logic for collecting the Mesh network's runtime data and service request data is as follows: Take the current time t as the starting point of collection, define the time window k, and collect network operation time series data and business request time series data within the time range of [t, t+k]; Data collection is started through "periodic triggering": periodic triggering, at fixed time intervals, the central node broadcasts collection instructions to the entire network, requiring all nodes to report the latest status information; The specific contents of the collection are as follows: The node remaining energy is set to E rem The collection method is that the central node collects the energy status of each node through polling instructions every 1 minute, and E rem ∈[0,E total ], where E total The total energy of the node. When the energy drop rate is greater than 5% / minute, the node needs to actively report the current remaining energy; Link quality data measurement method: Nodes exchange heartbeat packets every 10 seconds, and use ACK and RTT to count packet loss rate and delay. The central node monitors the RSSI and SNR of link communication, and calculates the weighted average of the heartbeat packet measurement value and the central node monitoring value in a weight ratio of 7:3 as the final link quality data; The link bandwidth usage collection protocol is that the central node obtains the traffic statistics of the switch port through SNMP. The calculation formula is as follows: Among them, B usa,i is the bandwidth utilization rate of the ith link, B used,i is the actual occupied bandwidth of the ith link, B total,i is the total bandwidth of the ith link; The business request dataset for each time window contains the following information: The business priority is set to P p , define the priority field in the service request header: 0 for high, 1 for medium, 2 for low, and the initiating node digital signature is required; the service timeliness is set to T d , the maximum allowed end-to-end delay for each request, which is automatically filled in by the service type mapping table; the service data volume is set to R s ,When the service is initiated, the transmission data size is recorded through the request header.

3. The method for dynamic optimization of a centralized Mesh network according to claim 2, characterized in that: The logic of dividing the collected business request time series data into high priority, medium priority and low priority businesses according to the business data volume, priority and timeliness indicators is as follows: Business requests are classified as high, medium, or low priority, and must meet any of the following conditions: For high priority service set D high , if the service priority P p =0, that is, the request is clearly marked as high priority in the service request header, then the request needs to meet any of the following conditions: a. Service timeliness T d ≤t threshold,high , that is, the maximum allowed end-to-end delay of the request is less than or equal to the delay threshold of the high priority; b. Business data volume R s ≤r threshold,high And T d ≤t threshold,high , that is, the amount of requested data is small and the timeliness requirement is high; among them, t threshold,high and r threshold,high are the set delay and quantity thresholds; For medium priority service D mid , if the service request meets all the following conditions: a. Service priority P p =1, that is, the request is clearly marked as medium priority in the service request header, b. Service timeliness t threshold,high <T d ≤t hreshold,mid , that is, the latency requirement of the request is between high priority and low priority; c. Business data volume r threshold,high <R s ≤r threshold,mid , that is, the amount of requested data is moderate; among them, t threshold,mid and r threshold,mid is the set threshold value; For low priority services D low , if the service request meets any of the following conditions: a. If the service priority P p =2, that is, the request is clearly marked as low priority in the service request header, b. Service timeliness T d >t threshold,mid , that is, the maximum allowed end-to-end delay of the request is greater than the threshold value of the medium priority, and the service data volume R s >r threshold,mid , that is, large data needs to be transmitted, and the latency requirement is not high; The logic of counting the number of requests, total data volume, and business proportion of each type of business within the statistical time window k and combining them into a business request statistical feature vector is as follows: For each type of business, namely high priority, medium priority and low priority, the following three key characteristics are counted respectively: Count the number of requests N belonging to this type of business in the current time window k class , the calculation formula is: N class =Count(D class ) Among them, D class is a collection of certain types of services, namely D high , D mid or D low , class represents the index. When class=1, it refers to high-priority services, when class=2, it refers to medium-priority services, and when class=3, it refers to low-priority services; The transmission data volume of all services of this type in the current time window k is accumulated, and the calculation formula is: Among them, R s,n Indicates the data volume of the nth request, S class Statistics of total data volume for a certain type of business; Calculate the ratio of the total data volume of this type of business to the total data volume of all businesses in the current time window. The formula is as follows: Among them, S total =S high +S mid +S low is the total data volume of all services, P class The proportion of data volume for a certain type of business; If N class =0, then S class = 0 and P class =0; if S total =0, then all proportions are set to P class =0; Combine the statistical results into a structured feature vector, the business request statistical feature vector V req is defined as follows: V req ={N high ,N mid ,N low ,S high ,S mid ,S low ,P high ,P mid ,P low } Among them, N high ,N mid ,N low are the number of requests for high, medium and low priority services, respectively. high ,S mid ,S low are the total data volume of high, medium and low priority services respectively, P high ,P mid ,P low They are the total data volume proportions of high, medium and low priority services respectively; After the feature vector is generated, the statistical results need to be checked for consistency: request quantity consistency, N high +N mid +N low =N total , where N total is the total number of requests in the current time window k; data consistency, S high +S mid +S low =S total ; Total percentage check, P high +P mid +P low ≈100%.

4. The method for dynamic optimization of a centralized Mesh network according to claim 2, characterized in that: The time series modeling method is used to extract features from the network runtime time series data collected in step 1. The logic for extracting node energy trend features, link quality features, and bandwidth utilization features is as follows: The node and link timing data in the time window [t, t+k], including: the node remaining energy timing data E rem,i (t),t∈[t,t+k], represents the residual energy change of the i-th node in the time window; Use LSTM to model various types of time series data, build independent time series models for each type of time series data, and set the input sequence of each type of time series data to x t (t), t∈[t,t+k]; for each input sequence, the output of LSTM is the hidden state h in the time window t Or the final hidden state h k : h t (t)=LSTM(x t ,h t-1 ,c t-1 ) Among them, x t is the input of the time series feature, h t-1 ,c t-1 are the hidden state and candidate state of the previous time step, respectively, h t (t) captures the dynamic representation of the temporal features, and the final output hidden state is h k , represents the trend characteristics of the time window [t, t+k]; The final hidden state h of the LSTM model k , extract the following trend feature set: node energy trend feature, assuming that the total number of nodes in the network is N, T node (t) = {h k (E rem,1 ),h k (F rem,2 ),…,h k (E rem,N )}; Link quality trend characteristics, T link (t) = {h k (P loss,i ),h k (D i ),h k (J i )}, where i represents the trend characteristic of the i-th link quality; bandwidth utilization trend characteristic, T bw (t) = {h k (B usa,1 ),h k (B usa,2 ),…,h k (B usa,M )}, where M is the total number of links in the network; The logic of using the attention mechanism to assign feature weights and generating network dimension reduction feature vectors through dimensionality reduction processing is: For each type of feature, the weight of each dimension of data is dynamically calculated through the attention mechanism to associate the importance of the feature with the target task. The formula is as follows: Among them, h o is the dynamic representation of feature o, which is the feature vector obtained after forward propagation in the latter layer of the neural network. Feature o represents the dimensional feature of the latter layer in the network. W q is a trainable parameter in the attention layer, used to dynamically represent the feature h o Perform a linear transformation, It is W q The transposed matrix of o is the importance weight of feature o, and α o ∈[0,1], satisfying ∑ o α o =1; Perform weighted summation on the features to obtain the aggregate representation H after the attention mechanism aat : Wherein, polymerization means H aat is achieved by dynamically representing all features h o According to their importance weights α o The weighted sum is obtained; according to the aggregate representation formula, the aggregate representation of different features is calculated, and the weighted sum of the dynamic representation of each dimension of the node energy feature is calculated: Among them, H node is a dynamic representation of the node energy characteristics, is the corresponding importance weight; The link quality feature aggregation is expressed as H Link and bandwidth utilization feature aggregation is denoted as H bw .

5. The method for dynamic optimization of a centralized Mesh network according to claim 4, characterized in that: Use the principal component analysis method to reduce the dimension of the attention-weighted feature set, and input the feature set: H = {H node ,H link ,H bw }; By decomposing the eigenvalues ​​of the feature covariance matrix, we get the PCA projection matrix W PCA , use the projection matrix to reduce the dimension of the original feature set, and calculate the feature vector after dimension reduction: H PCA =H·W PCA Among them, H is the original feature set, H PCA is the feature vector after dimensionality reduction; the output network state feature vector after dimensionality reduction is V dim =H PCA ; At the same time, combined with the business request statistical feature vector in step 2, the logic for generating the joint feature vector of the current time window is: The network status features after dimensionality reduction are concatenated with the business request statistical features in step 2 to generate the joint feature vector of the current time window. The formula is: V union ={V dim ,V req } Among them, V dim Represents the trend characteristics of network operation status within the time window, V req Represents the statistical characteristics of business requests within the time window, and the joint feature vector V union It integrates network status characteristics and business request characteristics.

6. The method for dynamic optimization of a centralized Mesh network according to claim 5, characterized in that: Combined with the joint feature vector of the current time window generated in step 3, a global optimization path planning model is constructed, and the genetic algorithm is used to solve the multi-objective optimization function. The logic of finding the Pareto optimal solution is: Joint eigenvector V union The network status characteristics and business request characteristics are integrated, and based on these characteristics, a global optimization path planning model is constructed; Path planning must meet the following constraints: Business timeliness constraints: D path ≤T d , ensuring that the selected path can meet the timeliness requirements of the business; link quality constraint: P loss,i ≤P threshold , Among them, P loss,i is the packet loss rate of the ith link, P threshold is the maximum allowed packet loss rate threshold, P represents a path; node energy constraint: E rem,i ≥E threshold , Among them, E rem,i is the residual energy of the ith node, E threshold is the minimum required threshold of node energy; Using the reduced-dimensional network state feature vector V generated in step 3 dim and business request statistical feature vector V req , defining the dynamic weights of nodes and links in the network topology: Link weight w link,i , comprehensively considering the link quality characteristics and bandwidth utilization: w link,i =α·D i +β·P loss,i +γ·B usa,i Among them, D i is the delay of link i, P loss,i is the packet loss rate of link i, B usa,i is the bandwidth utilization of link i, α, β, γ are all weight parameters, and α+β+γ=1; Node weight w node,i , calculated based on the node residual energy: Among them, E rem,i is the residual energy of node i, when E rem,i When it is smaller, the node weight increases, thus avoiding the selection of nodes with high load and fast energy consumption.

7. The method for dynamic optimization of a centralized Mesh network according to claim 6, characterized in that: The weighted shortest path algorithm based on feature weights is used to dynamically plan the optimal path by combining the feature vector in step 3 and the business characteristics in step 2. The specific logic is as follows: The network topology is modeled as a weighted directed graph G = (A, O), where the node set A represents all nodes in the network, O represents all links in the network, and the link weight is represented by w link,i Definition, using the improved Dijkstra algorithm, planning the path based on the following path cost function: Among them, C path is the total cost of path P, reflecting the comprehensive weight of the selected path, P node is the set of all relay nodes on the path; Construct the network topology graph G and transform the link weight w link,i and node weight w node,i Map it to the graph, input the starting point, end point and delay constraint of the service request, and calculate the path cost C for each candidate path P path , filter out paths that do not meet the following conditions: Select the total cost C from the remaining candidate paths path The smallest path is taken as the preliminary optimal path; After obtaining the initial optimal path, the genetic algorithm is used to further optimize the path selection to find the Pareto optimal solution: Use genetic algorithm to find the Pareto optimal solution of the optimization path and randomly generate the initial path population P = {p1, p2, ..., p m }, where each path p i Contains a series of nodes and links between them. For each path p i Calculation: Path delay D path (p i ), path packet loss rate P loss (p i ), the minimum residual energy E of the path rem (p i ); The tournament selection method is used to select paths with higher fitness from the current population as the parents of the next generation; the selected parents are crossovered to generate new path individuals; the generated new individuals are randomly mutated to increase population diversity; these steps are iterated until the termination condition is met, that is, the maximum number of iterations is reached, and the path population is updated after each iteration, and the path with the highest fitness is retained; Finally select a path p from the Pareto optimal solution set optimal , which has the best overall performance, as described below: Here, F′(p i ) is the normalized comprehensive fitness value; Output selected path p optimal Detailed information, including: path node sequence, path link sequence, and total path delay D path , the average packet loss rate of the path P loss , the minimum residual energy E of the path rem ; After outputting the path planning solution, analyze the optimization effect: compare the delay, packet loss rate and remaining energy changes between the initial path and the final path, and evaluate the optimization solution's utilization efficiency of network resources.

8. A dynamic optimization device for a centralized Mesh network, characterized in that: The device is used to execute a dynamic optimization method of a centralized Mesh network according to any one of claims 1 to 7, comprising: The data collection module is used to set a time window and collect the operation sequence data and service request sequence data of the Mesh network within the time window. The network operation sequence data includes the remaining energy of each Mesh network node, the link quality between nodes and the current bandwidth usage rate of each link. The service request sequence data includes the data volume, priority and timeliness index data of all services. The data processing module is used to divide the collected service request time series data into high priority, medium priority and low priority services according to the service data volume, priority and timeliness indicators, and to count the number of requests, total data volume and service proportion of each type of service in the time window k, and combine them into a service request statistical feature vector; The feature extraction module is used to extract features from the collected network runtime data using a time series modeling method, obtain node energy trend features, link quality features, and bandwidth utilization features, use an attention mechanism to assign feature weights, and generate a network dimension reduction feature vector through dimensionality reduction processing. At the same time, it combines the service request statistical feature vector to generate a joint feature vector for the current time window; The optimization path module is used to build a global optimization path planning model based on the joint feature vector of the current time window, use genetic algorithms to solve multi-objective optimization functions, find the Pareto optimal solution, and finally output the global optimization path planning solution with the highest fitness; The link quality characteristic time series data has packet loss rate P loss,i (t), represents the change of packet loss rate of link i over time in the time window, and the delay D i (t), represents the end-to-end delay and delay jitter J of link i in the time window i (t), represents the amplitude of the delay fluctuation of link i in the time window; bandwidth utilization time series data B usa,i (t),t∈[t,t+k], represents the bandwidth utilization of link i in the time window; In path planning, the following core optimization objective constraints are defined first: Minimum path delay: To ensure the real-time performance of the service, the total delay D of the selected path path Does not exceed the maximum allowable end-to-end delay T of the service d , that is, target 1 is D path ≤T d , where D path is the total delay from the source node to the destination node, which is determined by the link delay characteristics; Minimum link congestion: Avoid overloading of links with high bandwidth utilization in the path to improve the overall stability of the network, that is, goal 2 is B usa,max ≤B threshold , where B usa,max is the maximum bandwidth utilization of all links in the path, B threshold The maximum safe bandwidth utilization threshold of the defined link; Prolong network life: Prioritize paths with higher residual energy to prevent premature node failure, that is, goal 3 is -E rem,min Maximize the remaining energy, where E rem,min is the minimum residual energy of the nodes in the path, obtained through the node energy trend characteristics; The multi-objective optimization function defines the fitness function as: F(p i )=(D path (p i ),P loss (p i ),-E rem (p i )) Use the Pareto dominance relationship to select the Pareto optimal solution set. If the path p a Dominant Path p b , then p b Excluded.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, a dynamic optimization method for a centralized Mesh network as described in any one of claims 1 to 7 is implemented.

Images