A Method for Evaluating the Link Quality of a Human-Robot Collaborative Network Based on Random Forest

Abstract

The present invention proposes a method for evaluating the link quality of an unmanned and manned collaborative network based on random forest, belonging to the fields of wireless communication networks and unmanned and manned collaborative ad-hoc networks. First, the present invention constructs an index system for link quality evaluation; then constructs a random forest model and balances the sample data by the stratified sampling method; next, uses the sparrow search algorithm to optimize the hyperparameters of the random forest model, improves the position update method of the discoverer in the sparrow search algorithm by introducing the butterfly optimization algorithm, and adopts an adaptive shrinking search space mechanism to limit the search space of the population in each iteration; finally, realizes the evaluation of the network link quality based on the random forest classification model optimized by the improved sparrow search algorithm. The present invention improves the accuracy and convergence speed of link quality evaluation, and enhances the stability and reliability of the interconnection and interoperability of the unmanned and manned collaborative network.

Description

Technical Field

[0001] The present invention relates to the fields of wireless communication networks and manned-unmanned cooperative ad-hoc networks, and particularly relates to a method for evaluating the link quality of a manned-unmanned cooperative network based on random forests. Background Art

[0002] With the rapid development of artificial intelligence and computer technology, unmanned equipment has joined the battlefield and undertakes important combat tasks, such as ground unmanned vehicle reconnaissance, obstacle clearance, blasting, aerial unmanned aircraft reconnaissance, strike, communication relay, etc., enriching combat styles, expanding combat spaces, and enhancing combat effectiveness. Since intelligent unmanned equipment has the ability of autonomous communication and can be used as a communication relay node, it forms a manned-unmanned cooperative ad-hoc network with manned communication nodes. Due to the numerous high-rise buildings in the city, tall buildings will cause serious occlusion and interference to the communication links in the manned-unmanned cooperative ad-hoc network, making the links intermittent and unable to achieve the ability of real-time communication between nodes. Existing ad-hoc network technologies lack the evaluation of link effectiveness and ignore the differences in link quality, thus reducing the quality of network services.

[0003] The evaluation of wireless link quality plays a fundamental role in high-level protocols such as the reliable deployment, routing strategy, and resource management of manned-unmanned cooperative ad-hoc networks. Effective link quality evaluation can assist in the design of high-level protocols such as routing protocols, select the best link as the next-hop path, reduce link loss, reduce the number of data retransmissions, improve the end-to-end reliability of data transmission and network throughput, and maximize the network lifetime.

[0004] Currently, many research works on link quality evaluation have been carried out in China. The evaluation methods mainly include the following three types: The method based on link characteristics mainly uses physical layer parameters to predict link quality. The parameters required are simple to obtain and can be directly read from nodes, but there are calibration errors in the nodes themselves; The method based on probability estimation theory needs to send a large number of probe packets and is more accurate for the long-term measurement of links, but is not sensitive to the short-term changes of links; The method based on intelligent learning can mine the potential features between data through data driving and has important research significance. Summary of the Invention

[0005] To solve the problems existing in the link quality evaluation method in the above-mentioned manned-unmanned cooperative wireless ad-hoc network, the present invention provides a link quality evaluation method based on random forests, and optimizes the hyperparameters of the model by improving the sparrow search algorithm to improve the accuracy and efficiency of the link quality evaluation of the manned-unmanned cooperative ad-hoc network.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A method for evaluating the link quality of a manned-unmanned collaborative network based on random forest, comprising the following steps:

[0008] Step 1, extract the link state data during the network operation, construct a link quality feature system, and establish a sample library through a link quality level division method;

[0009] Step 2, use the stratified sampling method to obtain the same number of samples with different link quality levels, and establish a training sample library and a test sample library;

[0010] Step 3, initialize the parameters of the sparrow population, use the recognition accuracy rate of the random forest model for the link quality level as the fitness function, input it into the sparrow search algorithm, introduce the butterfly optimization algorithm to improve the sparrow position update strategy, and limit the search space of the sparrow through the adaptive shrinking search space method, and perform iterative optimization to obtain the optimal parameters of the random forest model;

[0011] Step 4, use the obtained optimal parameters to initialize the random forest model, train the model with the training sample library, and verify it with the test sample library to obtain a link quality recognition model;

[0012] Step 5, input the data to be measured into the recognition model and output the link quality level.

[0013] Further, the link state data in Step 1 is the received signal strength RSSI of the physical layer parameter, and the constructed link quality feature system is F = {RSSI, RSSI AVG , RSSI STD , RSSI REP , RSSI DI , FF}, and the calculation formula is as follows:

[0014]

[0015] Among them, X i represents the i-th sample data; n represents the number of samples obtained within a certain time interval, that is, the number of samples within the moving window; represents the mean value of the RSSI values within the moving window. RSSI AVG , RSSI STD , RSSI REP and RSSI DIFF respectively represent the moving average, standard deviation, reciprocal, and first-order difference value of the RSSI data.

[0016] Further, the link quality level division method in Step 1 uses the packet reception rate PRR as the division basis, and divides the link quality into 3 levels Y = {y good , y interm , y bad}, namely the healthy link, the good link and the degraded link, which are described by the following formula:

[0017]

[0018] The constructed sample library contains sample data of three types of links: healthy links, good links and degraded links.

[0019] Furthermore, the stratified sampling method is used in step 2 to obtain the same number of samples for different link quality levels. The specific process is as follows: First, sample data is drawn from the minority class samples, and then the same amount of data as the minority class is drawn from the majority class to change the distribution of various data in the model, making the data volume of each class in the training set similar, so as to train the classification model under the condition of sample balance.

[0020] Furthermore, the process of obtaining the optimal parameters of the random forest model in step 3 is as follows:

[0021] Step 301: Initialize the parameters of the sparrow population, including the population size, the position of the sparrows, the number of iterations, and the dimension. Among them, the position (x, y) of the sparrows represents the number of decision trees and the maximum depth of the tree on which the random forest is based, and the upper and lower bounds of the two parameters are set;

[0022] Step 302: The fitness function is the change function of the recognition accuracy of the random forest for the link quality level. The butterfly optimization algorithm is used to improve the sparrow position update strategy, and the search space of the sparrows is restricted by the method of adaptively shrinking the search space to quickly find the global optimal fitness value, that is, the optimal parameters of the random forest model;

[0023] Step 303: Divide the training sample library into training data and validation data, obtain the optimal fitness value after each iteration to initialize the random forest, train the model with the training data, input the validation data into the model to obtain the recognition accuracy and return it to the next iteration until the set number of iterations is reached, stop the iteration and obtain the optimal parameters of the random forest model.

[0024] Furthermore, the specific process of improving the sparrow search algorithm in step 3 is as follows:

[0025] Initialize the positions of N = {1, 2, …, n} sparrows in the D = {1, 2, …, d} - dimensional search space:

[0026]

[0027] where, x nd The position of the nth sparrow in the dth dimension.

[0028] Individuals within the population are divided into discoverers, followers, and scouts. Among them, in each iteration process, the butterfly optimization algorithm is introduced to improve the position update formula of the discoverer as follows:

[0029]

[0030] where t represents the current iteration number; X best is the global optimal position solution of the current iteration; f n represents the fitness value of the nth sparrow; r is a random number within the range of [0, 1]; Q is a random number subject to the standard normal distribution; L represents a 1×d matrix, where each element in the matrix is all 1, and d is the dimension size; R2∈[0, 1] and ST∈[0.5, 1] represent the warning value and the safety value respectively. When R2 < ST, this means that the search environment is safe at this time, there are no predators or other dangers around foraging, and the discoverer can search widely to guide the population to obtain higher fitness; if R2≥ST, this indicates that the scouting sparrows in the population have discovered a predator and sent a danger signal to the population, and the sparrows in the population adjust the search strategy and quickly fly to other safe places for foraging.

[0031] Adopt the strategy of adaptively shrinking the search space to control the search space of the sparrows in the improved sparrow search algorithm and accelerate the convergence speed of the population. The specific description is as follows:

[0032]

[0033] where X d,lb and X d,ub are the lower and upper limits of the search for the dth dimension respectively; X d,in and X d,ax are the minimum and maximum values of the current dth dimension respectively; represents the position of the current global optimal individual in the dth dimension; r t is the space shrinkage coefficient; T is the maximum number of iterations. During the iteration process, as r t increases, the search space continuously shrinks, while controlling the expansion amplitude of the search space, it also improves the convergence speed of the algorithm.

[0034] The description of the follower position update is as follows:

[0035]

[0036] where, is the optimal position currently occupied by the discoverer; represents the worst position globally at the current iteration t. A represents a 1×d matrix, where each element is randomly assigned 1 or -1, and A + = A T (AAT)-1 When n > N / 2, it indicates that the nth joiner with a lower fitness value has not obtained food and is in a very hungry state. At this time, it is necessary to search for food in a larger range to obtain more energy.

[0037] Scouts in the population will be aware of danger, generally accounting for 10% - 20% of the entire population. The position update is described as follows:

[0038]

[0039] Among them, β is the step size control parameter, which is a random number following a normal distribution with a mean of 0 and a variance of 1; K is a random number in [-1, 1], representing the moving direction of the sparrow and also the step size control parameter; f g and f w are the current global best and worst fitness values respectively; ε is the minimum constant to avoid zero error. When f n > f g , it indicates that the sparrow is at the edge of the population and is extremely vulnerable to predator attacks; when f n = f g , it indicates that the sparrow is aware of the threat of predators and needs to approach other sparrows to adjust its search strategy to avoid being attacked by predators.

[0040] Furthermore, the specific process of step 4 is as follows:

[0041] Initialize the model with the optimal parameters of the random forest model obtained by the improved sparrow search algorithm, train the model using the complete training sample library, and verify it using the test sample library. Judge whether the recognition accuracy meets the set threshold requirements. If it is lower than the set threshold, return to step 3, increase the number of iterations, and correct the parameter values to obtain a link quality recognition model that meets the requirements.

[0042] Furthermore, the specific process of step 5 is as follows:

[0043] Input the newly measured link state data during the network operation into the link quality recognition model obtained in step 4, and output the link quality level of this link.

[0044] The beneficial effects of the present invention are as follows:

[0045] 1. By extracting rich and fine-grained data features and constructing a link quality evaluation feature system, the present invention can well represent the current state of the link and provide more reasonable and reliable data samples for the link quality recognition model.

[0046] 2. The present invention applies a random forest model with stronger accuracy and generalization ability to link quality identification, and uses the stratified sampling method to obtain sample data with the same link quality at three levels, reducing the impact of unbalanced data on the generalization ability of the model and improving the accuracy of model classification and identification.

[0047] 3. The present invention combines the sparrow search algorithm with the random forest model, can find the optimal parameters of the model for any link state data, introduces the butterfly optimization algorithm to optimize the sparrow position update strategy, and controls the search space of the sparrow by adapting to shrink the search space method, enabling the algorithm to quickly find the global optimal solution, further improving the accuracy of model classification and identification, and enhancing the stability and reliability of the human-unmanned collaborative network. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a calculation flow chart of a random forest link quality identification model based on an improved sparrow search algorithm in an embodiment of the present invention.

[0049] Figure 2 It is a calculation flow chart of the random forest model in the present invention.

[0050] Figure 3 It is a simulation experimental diagram of constructing an evaluation index system in an example of the present invention.

[0051] Figure 4 It is a comparison experimental diagram of typical classification models in an embodiment of the present invention.

[0052] Figure 5 It is a comparison experimental diagram of the sparrow search algorithm and the improved sparrow search algorithm in an embodiment of the present invention.

[0053] Figure 6 It is an experimental verification diagram of the method for evaluating link quality by the improved random forest model in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments, corresponding drawings, and experimental verifications.

[0055] A method for evaluating the link quality of an unmanned and manned collaborative network based on random forest. This method first constructs an evaluation index system for the link quality of the unmanned and manned collaborative network; secondly, constructs a random forest classification model, and improves the disadvantage that the random forest classification model cannot handle unbalanced data through the stratified sampling method; next, uses the newly proposed sparrow search algorithm to optimize the parameters of the random forest model, and introduces the butterfly optimization algorithm to optimize the update method of the discoverer's position in the population, designs an adaptive shrinking search space mechanism to limit the search space of the population, and while improving the global search ability of the sparrow search algorithm, speeds up the convergence speed of the algorithm; finally, selects the communication link with the highest link quality as the first choice for high-level protocol design and resource scheduling.

[0056] The operation process of this method is as Figure 1 shown, and specifically includes the following steps:

[0057] A. Obtain network link status data, construct a link quality feature system, and establish a sample library through a link quality level division method;

[0058] The present invention obtains the received signal strength RSSI of the physical layer parameter as a sample of the link status data, and extracts the moving average RSSI AVG , standard deviation RSSI STD , reciprocal RSSI REP and first-order difference value RSSI DIFF and other eigenvalue to construct a link quality feature system F = {RSSI, RSSI AVG , RSSI STD , RSSI REP , RSSI DIFF}, and the calculation formula is as follows:

[0059]

[0060] Among them, X i represents the i-th sample data; n represents the number of samples obtained within a certain time interval, that is, the number of samples within the moving window; represents the mean value of RSSI values within the moving window.

[0061] Adopt the packet reception rate PRR as the basis for dividing the link quality level, and divide the link quality into 3 levels Y = {y good , y interm , y bad}, which are healthy link, good link and deteriorated link respectively, and are described by the formula as follows:

[0062]

[0063] The constructed sample library contains sample data of 3 types of links: healthy link, good link and deteriorated link.

[0064] B. Use the stratified sampling method to obtain the same number of samples for different link quality levels, and establish a training sample library and a test sample library;

[0065] The specific process is as follows: First, sample data is drawn from the minority class samples, and then the same amount of data as the minority class is drawn from the majority class to change the distribution of various data in the model, making the amount of data for each class in the training set similar, so as to train the classification model under the condition of sample balance.

[0066] C. The specific process of using the improved sparrow search algorithm to obtain the optimal parameters of the random forest model is as follows:

[0067] Step1: Initialize the population, the number of iterations, and parameters such as the proportion of discoverers and followers;

[0068] Initialize the sparrow population parameters, including setting the population size to 50, the number of iterations to 30, and the dimension to 2. The position of the sparrow is initialized as follows:

[0069]

[0070] where, x nd The position of the nth sparrow in the dth dimension.

[0071] The position of the sparrow (x n1 , x n2 ) represents the number of decision trees and the maximum depth of the tree on which the random forest model is based, respectively. Set the upper and lower bounds of the two parameters to (1, 1000) and (1, 50), respectively.

[0072] Step2: Calculate the fitness value of each sparrow individual and sort them to find the best and worst fitness values;

[0073] The fitness function is the change function f n of the recognition accuracy rate of the random forest for the link quality level. Calculate the fitness value according to the initial position of the sparrow, and determine the best fitness value f g and the worst fitness value f w .

[0074] Step3: Select a part of the sparrows as discoverers according to the proportion, and introduce the butterfly optimization algorithm to update the position of the discoverers;

[0075] Select sparrows with better fitness values in the population as discoverers according to a certain proportion (20%), and the rest are followers. The discoverers are responsible for finding food for the entire sparrow population and providing foraging directions for all followers. At the same time, any sparrow in the population can become a discoverer as long as it can search for a better food location. Since the sparrow search algorithm will prematurely converge and fall into a local optimum, the position update strategy of the global search stage in the butterfly optimization algorithm is introduced to optimize the position update strategy of the discoverers in the sparrow search algorithm. The formula is described as follows:

[0076]

[0077] Among them, t represents the current iteration number; X best is the global optimal position solution of the current iteration; f n represents the fitness value of the nth sparrow; r is a random number in the range of [0, 1]; Q is a random number subject to the standard normal distribution; L represents a 1×d matrix, where each element in the matrix is all 1, and d is the dimension size; R2∈[0, 1] and ST∈[0.5, 1] represent the warning value and the safety value respectively. When R2 < ST, it means that the search environment is safe at this time, there are no predators or other dangers around foraging, and the discoverers can search widely to guide the population to obtain higher fitness; if R2≥ST, it means that the scout sparrows in the population have discovered predators and sent a danger signal to the population, and the sparrows in the population adjust the search strategy and quickly fly to other safe places for foraging.

[0078] In the improved position update formula, on the one hand, at each iteration, the sparrow individuals will communicate information with the optimal individual to make full use of the information of the current optimal solution and improve the defect of the lack of information exchange between individuals in the original algorithm; on the other hand, the introduction of the butterfly optimization algorithm expands the search space of the population to a certain extent. However, the expansion of the search space does not guarantee the improvement of the global optimization ability of the algorithm. Therefore, a strategy of adaptively shrinking the search space is proposed. By controlling the upper and lower limits of the search space, the search space is restricted within a certain range to accelerate the convergence speed of the population. The specific description is as follows:

[0079]

[0080] Among them, X d,lb and X d,ub are the lower and upper limits of the dth dimension search respectively; X d,in and X d,ax are the minimum and maximum values of the current dth dimension respectively; represents the position of the current global optimal individual in the dth dimension; r t is the space reduction coefficient; T is the maximum number of iterations. During the iteration process, as rt With the increase of

[0081] Step4: Update the positions of the followers;

[0082] Except for the discoverer, the remaining sparrows are all followers. They will always observe the behavior of the discoverer and adjust their positions according to the discoverer's behavior. Once they notice that the discoverer has found better food, they will immediately leave their current positions to compete for the food. The position update of the followers is described as follows:

[0083]

[0084] where is the optimal position currently occupied by the discoverer; represents the globally worst position at the current iteration t. A is a 1×d matrix, where each element is randomly assigned a value of 1 or -1, and A + = A T (AA T ) -1 ; when n > N / 2, it indicates that the nth joiner with a lower fitness value has not obtained food and is in a very hungry state. At this time, it is necessary to search for food in a larger range to obtain more energy.

[0085] Step5: Randomly select some sparrows with a scouting and warning mechanism from the entire population as scouts and update their positions;

[0086] The scouts in the population will be aware of the danger and thus adjust their overall positions in the sparrow population. This part of the sparrows is randomly generated in the entire population, generally accounting for 10% - 20% of the entire population. The position update is described as follows:

[0087]

[0088] where β, as the step size control parameter, is a random number obeying a normal distribution with a mean of 0 and a variance of 1; K is a random number in [-1, 1], representing the moving direction of the sparrow and also the step size control parameter; f g and f w are the current global optimal and worst fitness values respectively; ε is the smallest constant to avoid zero error. When f n > f g , it indicates that the sparrow is at the edge of the population and is extremely vulnerable to predator attacks; when f n = f g , it indicates that the sparrow is aware of the threat of predators and needs to approach other sparrows to adjust its search strategy to avoid being attacked by predators.

[0089] Step6: Calculate the fitness values of the entire updated sparrow population and update the sparrow positions;

[0090] Step7: Check whether the stopping condition is satisfied. If it is satisfied, exit and output the result. Otherwise, repeat Steps 2 - 6.

[0091] Divide the training sample library into training data and validation data with a ratio of 7:3. Obtain the optimal fitness value after each iteration to initialize the random forest. Train the model with the training data, input the validation data into the model to obtain the recognition accuracy rate and return it to the next iteration until the set number of iterations is reached. Then stop the iteration and obtain the optimal parameters of the random forest model.

[0092] D. Construct a random forest recognition model:

[0093] Initialize the model with the optimal parameters of the random forest model obtained by using the improved sparrow search algorithm. Train the model with the complete training sample library and validate it with the test sample library. Determine whether the recognition accuracy rate meets the set threshold requirement. If it is lower than the set threshold, return to Step 3, increase the number of iterations, and correct the parameter values to obtain a link quality recognition model that meets the requirements.

[0094] The random forest model has the advantages of fast training speed, not being easily overfitted, and being applicable to feature - missing and high - dimensional data. Its basic principle is as Figure 2 shown:

[0095] Step1: Randomly sample n training data sets from the original data samples by the Bootstrap sampling - with - replacement method;

[0096] Step2: Randomly select k features (k is less than the total number of features in the original data) from each training data set;

[0097] Step3: Build n decision trees (weak classifiers) based on these k features;

[0098] Step4: Apply each decision tree to predict the result and save all the predicted results;

[0099] Step5: Vote on the classification model, calculate the number of votes for each predicted result, and select the model with the highest number of votes as the final decision.

[0100] E. Input the newly measured link - state data during the network operation process into the recognition model and output the link quality level of this link.

[0101] F. Simulation experiment

[0102] (1) Experimental data set

[0103] The effectiveness of the method proposed in the present invention is verified using the dataset of RSSI measured on the ORBIT test bed by Rutgers University. This dataset contains information about the locations of physical nodes, timestamps of transmitted data packets, and hardware information of the devices used. The dataset for each physical node contains the original RSSI values and serial numbers, and various levels of noise are injected into the ORBIT test bed, and the RSSI of each correctly received frame at each physical node is recorded. Since this dataset has a large data scale, a simple and complete structure, and does not require complex data preprocessing, it has been proven to be the most suitable dataset for data-driven machine learning link quality assessment models.

[0104] (2) Feature selection experiment

[0105] Through feature transformation and feature selection, the data features are divided into 13 groups, as shown in Table 1. The training data sample sets of each feature group are the same, and the link quality assessment and classification are all performed through a random forest model. The classification results obtained by taking the mean of multiple experiments are as Figure 3 shown. It can be seen from the figure that the feature system constructed in the present invention has the highest final recognition accuracy, and it is not that the more features the better. Too many features may form feature redundancy and instead reduce the classification accuracy of the model.

[0106] Table 1 Feature grouping experiment table

[0107]

[0108]

[0109] (3) Classification model comparison experiment

[0110] To verify the superiority of the performance of the random forest classification model on the Rutgers dataset, a comparative analysis experiment is conducted on the random forest RF classification model and other classification models such as logistic regression LG, decision tree DT, fully connected neural network MLP, and support vector machine SVM. The experimental results are as Figure 4 shown. The left vertical axis represents the recognition accuracy of the model, the right vertical axis represents the running time of the model, the horizontal axis represents each classification model, and good, interm, and bad respectively represent healthy links, good links, and deteriorated links of the link quality level.

[0111] As can be seen from the figure, the LG classification model performs poorly in all aspects on the Rutgers dataset; the DT classification model has the least running time, but its classification accuracy is poor, and it is prone to overfitting and has poor generalization ability; the MLP classification model has a relatively high classification accuracy, but its running time is too long; the SVM classification model performs well on many datasets, but it performs mediocrely on the Rutgers dataset with a large number of data samples, has a long training time, and the classification accuracy of various link qualities varies greatly; the RF classification model performs well in terms of running time and comprehensive recognition accuracy and is capable of performing the link quality assessment task.

[0112] (4) Comparative experiment on model improvement

[0113] The Rutgers dataset is an imbalanced dataset, where the proportion of healthy links (good) is approximately 63%, the proportion of good links (interm) is approximately 32.6%, and the proportion of deteriorated links (bad) is 4.4%. However, the dataset has a large number of sample data, and the least proportion of bad links has approximately 50,000 sample data. Therefore, the improved data sample sampling method proposed in the present invention will not reduce the richness of data features.

[0114] By setting a comparative experiment between the Sparrow Search Algorithm (SSA) and the Improved Sparrow Search Algorithm (ISSA), the effectiveness of introducing the Butterfly Optimization Algorithm and designing an adaptive shrinking search space method in the present invention is proven. As Figure 5 shown, the ISSA algorithm can not only find the global optimal order, enabling the classification model to obtain a higher accuracy, but also improve the convergence speed of the algorithm. Therefore, the ISSA algorithm effectively improves the classification accuracy and running speed of the model.

[0115] By setting a link quality classification comparative experiment between the RF model, the improved RF model (BRF), and the RF model optimized based on the ISSA algorithm (ISSA-BRF), the effectiveness of the improvement measures proposed in the present invention is proven. As Figure 6 shown, the traditional RF model performs poorly in imbalanced data. Although the recognition accuracy of good links with a large number of samples is relatively high, the recognition effect of interm links with a small number of samples is very poor, resulting in a low overall recognition accuracy of the model; the BRF classification model with the improved sampling method has a greatly improved ability to recognize interm links, but due to the lack of an optimization algorithm, the hyperparameters of the model cannot be set to the optimal, so the overall recognition accuracy is poor; the ISSA-BRF model proposed in the present invention shows advantages in both interm links with a small number of samples and overall recognition accuracy, thus proving the effectiveness of the method proposed in the present invention.

[0116] In summary, the present invention first constructs a link quality evaluation index system; then constructs a random forest model and balances the sample data by the stratified sampling method; next, the sparrow search algorithm is used to optimize the hyperparameters of the random forest model, and the position update method of the discoverer in the sparrow search algorithm is improved by introducing the butterfly optimization algorithm, and an adaptive shrinking search space mechanism is designed to limit the search space of the population in each iteration, enhance the global search ability of the sparrow population, and improve the convergence speed of the algorithm; finally, the random forest classification model optimized by the improved sparrow search algorithm is used to realize the evaluation of the network link quality, laying a foundation for high-level protocols such as the reliable deployment, routing strategy, and resource management of the unmanned and manned collaborative self-organizing network. The present invention solves the problems of the traditional link quality evaluation algorithm, such as the complex construction of the evaluation index system, difficult to give an accurate link level, unable to guarantee the algorithm convergence speed, and unable to meet the requirements of future unmanned and manned collaborative networking, etc., improves the accuracy and convergence speed of the link quality evaluation, and enhances the stability and reliability of the interconnection of the unmanned and manned collaborative network.

Claims

1. A method for evaluating the link quality of an unmanned and manned collaborative network based on random forest, characterized in that, It includes the following steps: Step 1: Extract the link state data during the network operation process, construct a link quality feature system, and establish a sample library through a link quality level division method; The link state data is the received signal strength RSSI of the physical layer parameters, and the constructed link quality feature system is F = {RSSI, RSSI AVG , RSSI STD , RSSI REP , RSSI DIFF}, and the calculation formula is as follows: Among them, X i represents the i-th sample data; n represents the number of samples obtained within a certain time interval, that is, the number of samples within the moving window; represents the mean value of RSSI values within the moving window, RSSI AVG , RSSI STD , RSSI REP and RSSI DIFF respectively represent the moving average, standard deviation, reciprocal and first-order difference value of RSSI data; Step 2: Use the stratified sampling method to obtain the same number of samples of different link quality levels, and establish a training sample library and a test sample library; Step 3: Initialize the sparrow population parameters, use the recognition accuracy of the random forest model for the link quality level as the fitness function, input it into the sparrow search algorithm, introduce the butterfly optimization algorithm to improve the sparrow position update strategy, and limit the search space of the sparrow through the adaptive shrinking search space method, and perform iterative optimization to obtain the optimal parameters of the random forest model; Step 4: Initialize the random forest model with the obtained optimal parameters, train the model using the training sample library, and verify it with the test sample library to obtain a link quality recognition model; Step 5: Input the data to be measured into the link quality recognition model and output the link quality level.

2. The method for evaluating the quality of the manned and unmanned collaborative network link based on the random forest according to claim 1, wherein In the above step 1, the link quality level division method uses the packet reception rate (PRR) as the division basis, and divides the link quality into three levels \(Y = \{y\) good , y interm , y bad \}, namely healthy link, good link and deteriorated link, which are described by the following formula: The constructed sample library contains sample data of three types of links: healthy links, good links, and deteriorated links.

3. The method for evaluating the quality of an unmanned and manned collaborative network link based on random forest according to claim 1, characterized in that, In step 2, the stratified sampling method is used to obtain the same number of samples of different link quality levels. The specific process is as follows: First, sample data is extracted from the minority class samples, and then the same amount of data as the minority class is extracted from the majority class, changing the distribution of various data in the model to make the data volume of each category in the training set similar, so as to train the classification model under the condition of sample balance.

4. The method for evaluating the quality of an unmanned and manned collaborative network link based on a random forest according to claim 1, wherein The process of obtaining the optimal parameters of the random forest model in step 3 is as follows: Step 301: Initialize the sparrow population parameters, including the population size, the position of the sparrows, the number of iterations, and the dimension. Among them, the position (x, y) of the sparrows represents the number of decision trees and the maximum depth of the tree on which the random forest is based, and the upper and lower bounds of the two parameters are set; Step 302: The fitness function is the change function of the recognition accuracy of the random forest for the link quality level. The butterfly optimization algorithm is used to improve the sparrow position update strategy, and the search space of the sparrows in the improved sparrow search algorithm is limited through the adaptive shrinking search space method to quickly find the global optimal fitness value, that is, the optimal parameters of the random forest model; Step 303: Divide the training sample library into training data and validation data, obtain the optimal fitness value after each iteration to initialize the random forest, train the model with the training data, input the validation data into the model to obtain the recognition accuracy and return it to the next iteration until the set number of iterations is reached, stop the iteration and obtain the optimal parameters of the random forest model.

5. The method for evaluating the quality of an unmanned collaborative network link based on random forest according to claim 4, wherein The specific process of using the butterfly optimization algorithm to improve the sparrow search algorithm in step 3 is as follows: Initialize the positions of N = {1, 2, …, n} sparrows in the D = {1, 2, …, d} - dimensional search space: where x nd is the position of the n-th sparrow in the d-th dimension; Divide the individuals in the population into discoverers, followers, and scouts. Among them, in each iteration process, introduce the butterfly optimization algorithm to improve the position update formula of the discoverers as: where t represents the current iteration number; X best is the global optimal position solution of the current iteration; f n represents the fitness value of the nth sparrow; r is a random number in the range of [0, 1]; Q is a random number obeying the standard normal distribution; L represents a 1×d matrix, where each element in the matrix is all 1, and d is the dimension size; R2 ∈ [0, 1] and ST ∈ [0.5, 1] represent the warning value and the safety value respectively; when R2 < ST, it means that the search environment is safe at this time, there are no predators or other dangers around foraging, and the discoverer can search widely to guide the population to obtain higher fitness; if R2 ≥ ST, it means that the scouting sparrows in the population have discovered predators and sent a danger signal to the population, and the sparrows in the population adjust the search strategy and quickly fly to other safe places to forage; Adopt the strategy of adaptive shrinking search space to control the search space of the sparrows in the improved sparrow search algorithm and accelerate the convergence speed of the population. The specific method is as follows: Among them, X d,lb and X d,ub are respectively the lower and upper bounds of the search in the d-th dimension; X d,min and X d,max are respectively the minimum and maximum values in the current d-th dimension; represents the position of the current globally optimal individual in the d-th dimension; r t is the space reduction coefficient; T is the maximum number of iterations; during the iteration process, as r t increases, the search space continuously shrinks, while controlling the expansion amplitude of the search space, the convergence speed of the algorithm is improved; The position update method of the followers is as follows: Among them, is the optimal position occupied by the current discoverer; represents the position of the globally worst at the current iteration t; A represents a 1×d matrix, where each element is randomly assigned 1 or -1, and A + = A T (AA T ) -1 ; when n > N / 2, it indicates that the nth joiner with a lower fitness value has not obtained food and is in a very hungry state. At this time, it is necessary to search for food in a larger range to obtain more energy; The position update method of the scouts is as follows: Among them, β is a step size control parameter and is a random number obeying the normal distribution with a mean of 0 and a variance of 1; K is a random number in [-1, 1], representing the moving direction of the sparrow and also being a step size control parameter; f g and f w are the current global best and worst fitness values respectively; ε is the smallest constant to avoid zero error; when f n > f g , it indicates that the sparrow is at the edge of the population and is extremely vulnerable to predators; when f n = f g , it indicates that the sparrow realizes the threat of predators and needs to approach other sparrows to adjust its search strategy so as to avoid being attacked by predators.

6. The method for evaluating the quality of an attended-unattended collaborative network link based on a random forest according to claim 1, wherein The specific process of step 4 is as follows: Initialize the model with the optimal parameters of the random forest model obtained by using the improved sparrow search algorithm, train the model using the complete training sample library, and verify it using the test sample library. Determine whether the recognition accuracy meets the set threshold requirement. If it is lower than the set threshold, return to step 3, increase the number of iterations, and correct the parameter values to obtain a link quality recognition model that meets the requirements.

7. The method for evaluating the quality of the manned and unmanned collaborative network link based on random forest according to claim 1, wherein The specific process in step 5 is as follows: Input the newly measured link state data during the network operation into the link quality recognition model obtained in step 4, and output the link quality level of this link.

Images