Saturation throughput prediction method, device, equipment, medium and program product

Abstract

The invention discloses a saturation throughput prediction method and device, equipment, a medium and a program product, and the method comprises the steps: obtaining target parameter data of each wireless access point of a wireless local area network under to-be-tested network configuration, the target parameter data comprising an average negotiation rate, an average packet error rate and an air interface state proportion; performing data preprocessing on the target parameter data of each wireless access point to generate a first test data set; inputting the first test data set into a first target prediction model to perform saturation throughput prediction processing to obtain a saturation throughput prediction value of the wireless local area network under the to-be-tested network configuration; wherein the first target prediction model is obtained by training based on a first sample data set, and the first sample data set is generated based on sample tags of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point. The method can adapt to a complex and changeable wireless network environment, and improves the accuracy of saturation throughput prediction.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a saturation throughput prediction method, device, equipment, medium and program product. Background Art

[0002] The research on wireless performance has always been an important cornerstone of the development of wireless network technology. Saturation throughput is the core indicator for evaluating wireless network performance. Its accuracy and practicality are of vital importance for network design and optimization. However, the current saturation throughput estimation methods have many limitations and are difficult to adapt to the complex and changing wireless network environment.

[0003] Current wireless network saturation throughput prediction technologies mostly focus on theoretical model construction. These models are mainly suitable for theoretical analysis rather than real-world applications. These models are often based on idealized assumptions and ignore the complexity and variability of actual network environments, resulting in limited accuracy and practicality in actual deployment. Summary of the invention

[0004] The embodiments of the present application provide a saturation throughput prediction method, apparatus, device, medium and program product, which can adapt to complex and changeable wireless network environments, improve the accuracy of saturation throughput prediction, and achieve a more accurate and comprehensive evaluation of wireless network performance.

[0005] On the one hand, an embodiment of the present application provides a saturation throughput prediction method, the method comprising: obtaining target parameter data of each wireless access point of a wireless local area network under a network configuration to be tested, the target parameter data comprising an average negotiation rate, an average packet error rate, and an air interface state ratio; performing data preprocessing on the target parameter data of each wireless access point to generate a first test data set; inputting the first test data set into a first target prediction model to perform saturation throughput prediction processing to obtain a saturation throughput prediction value of the wireless local area network under the network configuration to be tested; wherein the first target prediction model is obtained by training based on a first sample data set, and the first sample data set is generated based on sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, and N is a natural number greater than 0.

[0006] On the other hand, an embodiment of the present application provides a saturation throughput prediction device, the device comprising:

[0007] An acquisition unit, configured to acquire target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, wherein the target parameter data includes an average negotiation rate, an average packet error rate, and a proportion of air interface status;

[0008] A first generating unit, configured to perform data preprocessing on the target parameter data of each wireless access point to generate a first test data set;

[0009] A first prediction unit, configured to input the first test data set into a first target prediction model to perform saturation throughput prediction processing, so as to obtain a saturation throughput prediction value of the wireless local area network under the network configuration to be tested;

[0010] Among them, the first target prediction model is trained based on a first sample data set, and the first sample data set is generated based on sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, where N is a natural number greater than 0.

[0011] On the other hand, an embodiment of the present application provides a computer device, comprising a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the saturation throughput prediction method as described in any of the above embodiments by calling the computer program stored in the memory.

[0012] On the other hand, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the saturation throughput prediction method as described in any of the above embodiments.

[0013] On the other hand, an embodiment of the present application provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the saturation throughput prediction method as described in any of the above embodiments.

[0014] The embodiment of the present application obtains the target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, and the target parameter data includes the average negotiation rate, the average packet error rate and the proportion of the air interface state; performs data preprocessing on the target parameter data of each wireless access point to generate a first test data set; inputs the first test data set into the first target prediction model to perform saturation throughput prediction processing to obtain the saturation throughput prediction value of the wireless local area network under the network configuration to be tested; wherein the first target prediction model is obtained based on the training of the first sample data set, and the first sample data set is generated based on the sample labels of the wireless local area network under N sample network configurations and the sample parameter data of each wireless access point, and N is a natural number greater than 0. The embodiment of the present application obtains and utilizes the target parameter data of the wireless access point under the actual network configuration, such as the average negotiation rate, the average packet error rate and the proportion of the air interface state, so as to more accurately reflect the actual network conditions, thereby improving the accuracy of the saturation throughput prediction. By performing meticulous data preprocessing on the collected target parameter data and converting it into a first test data set suitable for machine learning, high-quality input is provided for subsequent predictions. Subsequently, the first target prediction model trained based on a rich sample data set is used to predict the saturated throughput. Since the first target prediction model is trained based on a data set under a variety of sample network configurations, this enables the first target prediction model to adapt to complex and changeable wireless network environments (such as different network environments and configurations), enhancing the universality and applicability of the model, and improving the accuracy of the saturated throughput prediction, so as to achieve a more accurate and comprehensive evaluation of wireless network performance. Since the first target prediction model is trained based on a data set under a variety of sample network configurations, this enables the prediction model to adapt to different network environments and configurations, enhancing the universality and applicability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 A schematic diagram of a flow chart of a saturation throughput prediction method provided in an embodiment of the present application.

[0017] Figure 2 A schematic diagram of the internal structure of a single decision tree in the first target prediction model provided in an embodiment of the present application.

[0018] Figure 3 A schematic diagram of the internal structure of a single decision tree in the second target prediction model provided in an embodiment of the present application.

[0019] Figure 4 A schematic diagram of the structure of a saturation throughput prediction device provided in an embodiment of the present application.

[0020] Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0022] The embodiments of the present application provide a saturation throughput prediction method, apparatus, device, medium and program product. Specifically, the saturation throughput prediction method of the embodiments of the present application can be executed by a computer device, wherein the computer device can be a terminal or a server. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart TV, a smart speaker, a wearable smart device, a smart car terminal and other devices. The terminal can also include a client, which can be a financial client, a browser client or an instant messaging client. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution network services, and basic cloud computing services such as big data and artificial intelligence platforms, but is not limited to this.

[0023] It should be noted that the order of description of the following embodiments is not intended to limit the priority order of the embodiments.

[0024] See also Figures 1 to 3 , Figure 1 A schematic diagram of a flow chart of a saturation throughput prediction method provided in an embodiment of the present application. Figure 2 A schematic diagram of the internal structure of a single decision tree in the first target prediction model provided in an embodiment of the present application, Figure 3 A schematic diagram of the internal structure of a single decision tree in the second target prediction model provided in an embodiment of the present application. The method comprises the following steps:

[0025] Step 110, obtaining target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, wherein the target parameter data includes an average negotiation rate, an average packet error rate, and an air interface state ratio.

[0026] In some embodiments, the acquiring of target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested includes: acquiring an average negotiation rate between an i-th wireless access point and a client of the wireless local area network under the network configuration to be tested according to a preset period, where i is a natural number greater than 0; acquiring an average packet error rate of the i-th wireless access point according to the preset period; acquiring an air interface state proportion of the i-th wireless access point according to the preset period, where the air interface state proportion includes a transmitting state proportion, a receiving state proportion, an idle state proportion, and a busy state proportion; acquiring the target parameter data of the i-th wireless access point of the wireless local area network under the network configuration to be tested according to the average negotiation rate between the i-th wireless access point and the client, the average packet error rate of the i-th wireless access point, and the air interface state proportion; and traversing each wireless access point to acquire the target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested.

[0027] Step 110 is a detailed process for obtaining target parameter data of each wireless access point (AP) of the wireless local area network (WLAN) under the network configuration to be tested, aiming to lay a solid data foundation for subsequent network performance analysis and prediction. This process not only focuses on the comprehensiveness of the data, but also emphasizes the real-time and periodicity of data collection, ensuring the timeliness and representativeness of the parameters obtained.

[0028] A preset period (for example, every minute, every five minutes, or every hour) can be defined to ensure that data is continuously captured from the network within the period. The preset period is the time interval for periodically acquiring target parameter data. This preset period can be set according to actual needs and the frequency of changes in the network environment, so that it is neither too frequent to waste resources nor too sparse to miss key changes.

[0029] For example, for each wireless access point in the network (marked as the i-th AP, where i is a natural number sequence), the average negotiation rate between the AP and the connected client is monitored according to the above preset period. The average negotiation rate refers to the average value of the data rate negotiated and determined between the wireless access point AP and the connected client within a certain period of time. This parameter reflects the data transmission capacity of the wireless link and is one of the important factors affecting the saturation throughput.

[0030] For example, in each preset period, the average packet error rate of each wireless access point (i-th AP) is collected and calculated. The average packet error rate refers to the average value of the packet loss ratio caused by various reasons (such as signal interference, poor channel quality, etc.) during data transmission. This parameter reflects the transmission quality of the wireless link and also has a significant impact on the saturation throughput.

[0031] For example, the air interface state ratio refers to the time proportion that the wireless access point AP is in different working states within a period of time. For example, different working states may include transmitting state (TX), receiving state (RX), idle state (IDLE), busy state (CCABUSY), etc.

[0032] For example, the air interface state percentage of each wireless access point (i-th AP) is counted according to a preset period, including the transmit state percentage, receive state percentage, idle state percentage, and busy state percentage. These data can reveal the working mode, resource utilization, and potential performance bottlenecks of the wireless access point.

[0033] For each AP, the average negotiation rate, average packet error rate, and air interface status ratio collected periodically are summarized to form the target parameter data of the AP under the network configuration to be tested. The target parameter data comprehensively reflects the current working status and performance of the AP.

[0034] Then, systematically traverse all wireless access points of the wireless LAN under the network configuration to be tested, repeat the above data collection and integration process to ensure that the target parameter data of all APs are fully recorded. This step ensures the comprehensiveness of the evaluation, so that the subsequent performance analysis will not miss any factors that may affect the overall network performance.

[0035] Step 120: preprocess the target parameter data of each wireless access point to generate a first test data set.

[0036] In some embodiments, the target parameter data of each wireless access point is preprocessed to generate a first test data set, including: constructing a first mathematical model corresponding to the saturation throughput prediction value according to the average negotiation rate, the average packet error rate and the statistical probability of occupying channel packets; constructing a second mathematical model corresponding to the statistical probability of occupying channel packets according to the transmission state proportion, the receiving state proportion, the idle state proportion and the busy state proportion; substituting the average negotiation rate, the average packet error rate and the second mathematical model into the first mathematical model for calculation to obtain initial feature data to be tested corresponding to the network configuration to be tested; normalizing and standardizing the initial feature data to be tested to obtain the feature data to be tested corresponding to the network configuration to be tested, and generating a first test data set based on the feature data to be tested.

[0037] According to the average negotiation rate, the average packet error rate and the statistical probability of occupying the channel packet transmission, the first mathematical model corresponding to the saturation throughput prediction value can be expressed as the following formula (1):

[0038]

[0039] Among them, Throughoutput represents the predicted value of saturation throughput; TxRate i Represents the average negotiation rate between the i-th AP and the client; PER i represents the average packet error rate of the ith AP; P i It represents the statistical probability of the ith AP occupying the channel and sending packets.

[0040] Among them, P i It can be represented by the air interface state ratio. The air interface state ratio may include the transmission state (TX) ratio, the reception state (RX) ratio, the idle state (IDLE) ratio, and the busy state (CCABUSY) ratio. For example, in the saturated state, Pi can be fully represented by the Tx ratio in the air interface state ratio. In actual scenarios, according to the transmission state ratio, the reception state ratio, the idle state ratio, and the busy state ratio, the second mathematical model corresponding to the statistical probability of occupying the channel packet can be expressed as the following formula (2):

[0041] P i = f(TX i ,RX i ,IDLE i ,CCABUSY i ) (2);

[0042] Among them, P i represents the statistical probability of the ith AP occupying the channel to send packets; TX i Indicates the transmission status ratio of the i-th AP; RX i Indicates the receiving state ratio of the i-th AP; IDLE i Indicates the idle state ratio of the i-th AP; CCABUSY i Indicates the busy status ratio of the i-th AP.

[0043] In order to effectively reduce the feature dimension, a linear expression can be used to express the second mathematical model. For example, a linear polynomial of a Taylor function can be used to express the second mathematical model, and the above formula (2) can be converted into the following formula (3):

[0044] f(TX,RX,IDLE,CCABUSY)

[0045] =Constant+w1TX+w2RX+w3IDLE+w4CCABUSY+k1TX*RX

[0046] +k2TX*IDLE+k3TX*CCABUSY+k4RX*IDLE+k5RX*CCABUSY

[0047] +k6IDLE*CCABUSY+k7TX 2 +k8RX 2 +k9IDLE 2 +k 10 CCABUSY 2 (3);

[0048] Then, the average negotiation rate, average packet error rate and the second mathematical model after conversion as shown in formula (3) are substituted into the first mathematical model as shown in formula (1) for calculation to obtain the initial characteristic data to be tested corresponding to the network configuration to be tested. Among them, the saturation throughput prediction value is approximately expressed as completely linear. Using the additivity of linear expression, the characteristics of N APs are added, and the final characteristic dimension is 14 dimensions.

[0049] Among them, due to P i Has been expanded to TX i RX i 、IDLE i 、CCABUSY i The polynomial function related to this, at this time, the saturation throughput prediction value throughoutput is completely determined by TxRate i , PER i , tTX i RX i 、IDLE i 、CCABUSY i These values ​​are actually acquired values, and the actual unknown quantity (initial feature to be measured) is the coefficient. Therefore, the final expression of the saturated throughput prediction value throughoutput can be converted into the following formula (4):

[0050] Throughoutput=a1w1+a2w2+a3w3+a4w4+b1k1+b2k2+b3k3+b4k4+b5k5+

[0051] b6k6+b7k7+b8k8+b9k9+b 10 k 10 (4);

[0052] Where a and b are constants, and the specific calculation is as follows:

[0053]

[0054]

[0055] By using the actual collected TxRate i , PER i , tTX i RX i、IDLE i 、CCABUSY i The parameters are input into a1 to a4, b1 to b 10 The corresponding expressions are calculated to obtain the initial feature data to be tested, which includes the above 14 initial features to be tested: a1, a2, a3, a4, b1, b2, b3, b4, b5, b6, b7, b8, b9, b 10 .

[0056] Then, each of the initial features to be tested in the initial feature data to be tested is normalized and standardized to obtain the feature data to be tested corresponding to the network configuration to be tested, and a first test data set is generated based on the feature data to be tested. The first test data set can be expressed as D 待估计 (X: feature to be measured), each feature to be measured is a feature obtained by normalizing and standardizing each initial feature to be measured.

[0057] Among them, feature normalization and standardization are common data preprocessing techniques, mainly to prevent excessive dimensional differences. The purpose of normalization is to scale the initial feature data to fall into a specific small interval, usually between 0 and 1. Standardization refers to scaling the data according to its characteristics (such as mean and standard deviation), for example, converting the initial feature data to be measured into a standard normal distribution with a mean of 0 and a variance of 1.

[0058] Step 130: Input the first test data set into a first target prediction model to perform saturation throughput prediction processing to obtain a saturation throughput prediction value of the wireless local area network under the network configuration to be tested.

[0059] Among them, the first target prediction model is trained based on a first sample data set, and the first sample data set is generated based on sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, where N is a natural number greater than 0.

[0060] Among them, the first target prediction model is the model described in formula (4).

[0061] For example, the training steps of the first target prediction model may include the following steps 21 to 24:

[0062] Step 21, obtain sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, the sample parameter data including the sample average negotiation rate, the sample average packet error rate and the sample air interface state ratio, and the sample label represents the true value of the saturation throughput of the wireless local area network under each sample network configuration.

[0063] In some embodiments, the acquiring of sample tags and sample parameter data of each wireless access point collected by the wireless local area network under N sample network configurations includes:

[0064] Deploy network configuration information corresponding to each sample network configuration in N sample network configurations for the wireless local area network, wherein the network configuration information includes network topology information, channel environment information, and wireless configuration parameters;

[0065] Based on the running flow test conditions of the same network topology information, the same channel environment information and the same wireless configuration parameters, a saturated running flow test and a non-saturated running flow test are respectively performed on each wireless access point under the N sample network configurations;

[0066] Based on the saturated running flow test, obtaining the true value of the saturated throughput of the wireless local area network under N kinds of sample network configurations, and obtaining the sample label;

[0067] Based on the non-saturated running flow test, sample parameter data of each wireless access point collected by the wireless local area network under N sample network configurations are obtained.

[0068] For example, sample parameter data can be collected under the experimental topology. In addition, data expansion can also be done through the simulation system.

[0069] For example, on a wireless communication simulation platform, a large-scale networking system is built, and N kinds of sample network configurations can be generated. Then, based on the same topology, in the same channel environment and with the same wireless configuration parameters, saturated flow testing and non-saturated flow testing are performed respectively. In the saturated flow testing, the actual value of the saturated throughput is recorded for the entire topology, which corresponds to the throughput in the above mathematical expression. In various types of non-saturated flow testing, all relevant sample parameter data are recorded. These sample parameter data include the sample average negotiation rate, the sample average packet error rate, and the sample air interface state ratio (sample transmission state ratio, sample receiving state ratio, sample idle state ratio, and sample busy state ratio), which correspond to TxRate in the above mathematical expression. i , PER i TX i RX i 、IDLE i 、CCABUSY i wait.

[0070] The saturated traffic test is mainly used to obtain the actual saturated throughput value of the same wireless configuration in the same scenario; the non-saturated traffic test is mainly used to obtain the TxRate of the same wireless configuration in the same scenario. i , PER i TX i RXi 、IDLE i 、CCABUSY i The purpose is to obtain training data. The sample parameter data obtained through the non-saturated running flow scenario is used as X in the training data after data processing, and the actual value of the saturated throughput obtained through the saturated running flow scenario is used as Y in the training data.

[0071] In the wireless communication simulation platform, the data generation rate is set to be greater than the maximum negotiation rate * the number of terminals to ensure the saturated flow test scenario; conversely, the non-saturated flow test scenario is achieved by limiting the negotiation rate, limiting the packet sending interval and other restriction conditions.

[0072] When verifying the actual topology, use the test tool to send or receive data streams through multiple APs simultaneously without limiting the data transmission rate to obtain the performance of the network under saturation. Then, in order to simulate the daily unsaturated network traffic situation, limit the data transmission rate to 20Mbps and 50Mbps respectively, and test again to understand the performance of the device or network under this unsaturated state.

[0073] For example, network topologies of different scales, different power configurations, and different AP placements are established to obtain N sample network configurations. Some parameter examples are shown in Table 1:

[0074] Table 1

[0075]

[0076]

[0077] Then, a traffic flow test is performed on the above network topology. The traffic flow modes include a saturated traffic flow test and a non-saturated traffic flow test.

[0078] Among them, the conditions of the non-saturated flow test may include: a given total flow is sent at a uniform speed; a given total flow starts and ends randomly within a packet sending cycle; a given total throughput is sent at a random time length within 1 second (s), and stops for the rest of the time.

[0079] During the traffic flow, the negotiation rate, packet error rate, and air interface status ratio are collected regularly. The collection time is 0.1s, and the total traffic flow time is 10s. Based on the collected data, the sample average negotiation rate, sample average packet error rate, and sample air interface status ratio of each wireless access point are determined.

[0080] Step 22: preprocess the sample parameter data of each wireless access point to generate sample feature data corresponding to each sample network configuration in the N types of sample network configurations.

[0081] In some embodiments, the sample air interface state ratio includes a sample transmission state ratio, a sample reception state ratio, a sample idle state ratio and a sample busy state ratio;

[0082] The performing data preprocessing on the sample parameter data of each wireless access point to generate sample feature data corresponding to each sample network configuration in the N types of sample network configurations includes:

[0083] Constructing a first mathematical model corresponding to the statistical probability of the sample occupying a channel packet transmission according to the sample transmission state proportion, the sample receiving state proportion, the sample idle state proportion and the sample busy state proportion;

[0084] Constructing a second mathematical model corresponding to the saturation throughput prediction value according to the sample average negotiation rate, the sample average packet error rate and the sample occupied channel packet sending statistical probability;

[0085] Substituting the sample average negotiation rate, the sample average packet error rate and the first mathematical model into the second mathematical model for calculation to obtain initial sample feature data corresponding to the sample network configuration;

[0086] The initial sample feature data is normalized and standardized to generate sample feature data corresponding to each sample network configuration in the N sample network configurations, each of the sample feature data including a plurality of sample features.

[0087] Step 23: construct a first sample data set based on the sample feature data and the sample labels.

[0088] Step 24: training the initial prediction model based on the first sample data set to obtain the first target prediction model.

[0089] In some embodiments, the training of the initial prediction model based on the first sample data set to obtain the first target prediction model includes:

[0090] Inputting the sample feature data in the first sample data set into the initial prediction model to perform saturation throughput prediction processing, and obtaining a sample saturation throughput prediction value of the wireless local area network under the sample network configuration corresponding to the sample feature data;

[0091] The model parameters of the initial prediction model are trained according to the sample saturation throughput prediction value and the sample label to obtain the first target prediction model after the model parameters are updated.

[0092] When training the first target prediction model, it is necessary to predict the saturated throughput performance under a single network topology and different power schemes, and the output is a specific saturated throughput value. The sample saturated throughput prediction value of power configuration i can be expressed as the following formula (19):

[0093] Throughoutput i =a 1i w1+a 2i w2+a 3i w3+a 4i w4+b 1i k1+b 2i k2+b 3i k3+b 4i k4+

[0094] b 5i k5+b 6i k6+b 7i k7+b 8i k8+b 9i k9+b 10i k 10 (19).

[0095] Training set division:

[0096] a) Define input: initial sample features under single network topology and single power configuration; the initial sample features under single network topology and single power configuration are: a1, a2, a3, a4, b1, b2, b3, b4, b5, b6, b7, b8, b9, b 10 The expressions of the above-mentioned initial sample characteristics can refer to formula (5) to formula (18). The collected sample parameter data can be converted into a1, a2, a3, a4, b1, b2, b3, b4, b5, b6, b7, b8, b9, b10, b11, b12, b3, b4, b5, b6, b7, b8, b9 ...3, b14, b15, b6, b7, b8, b9, b15, b16, b7, b8, b9, b17, b8, b9, b18, b19, b10, b11, b12, b3, b4, b5, b6, b7, b8, b9, b18, b19, b1 10 The initial sample feature data.

[0097] b) Define output: saturated throughput performance, specifically the sample saturated throughput prediction value.

[0098] Normalization and standardization are performed on each of the initial sample features in the above initial sample feature data to generate sample feature data corresponding to each of the N sample network configurations, each of the sample feature data including a plurality of sample features.

[0099] c) Then, based on the sample feature data and sample labels, the first sample data set is constructed, and the specific division form is D train (X,Y), X represents the sample feature data, and Y represents the sample label.

[0100] For example, assume that the topology with the same number of APs, AP placement type, and AP location parameters has 20 groups of power configurations, 5 types of traffic flow patterns, and a total of 100 groups of data. The sample feature data corresponding to any one of the 100 groups of data can constitute the input X, so there are 100 output labels for the input training set, and only the saturated throughput performance recorded in the saturated traffic flow pattern (that is, the true value of the sample saturated throughput) is recorded. The sample label is obtained for the given label value, that is, there are only 20 groups of output results.

[0101] Random Forest Model Construction:

[0102] Among them, by collecting data sets under various environments, the initial prediction model is then trained using a machine learning solution to obtain the parameter values ​​of each w and k in the f() expression corresponding to formula (3). For example, the initial prediction model can be constructed based on the above formula (4) or formula (19), and the goal of the training is to ultimately determine the most appropriate model parameters w and k.

[0103] For example, a random forest model can be used to map the model parameters w and k in the initial prediction model. In actual scenarios, other models can also be used to map the model parameters w and k in the initial prediction model, such as a support vector machine (SVM).

[0104] For example, taking the random forest model as an example, the random forest model is constructed as follows:

[0105] a) For D train (X,Y) is sampled N times with replacement to generate N sub-datasets D train_bs1 , D train_bs2 ,……,D train_bsN , the sub-datasets can contain repeated entries, and the size of each sub-training set is consistent with the original data set. train The original data set of (X,Y) is shown in Table 2 below:

[0106] Table 2

[0107]

[0108] For example, the sampled sub-dataset D train_bsi (X,Y) may be as shown in Table 3 below:

[0109] Table 3

[0110]

[0111] b) Using a single sub-training set D train_bsi For data, C features are randomly selected to train a single decision tree (C<14), and finally there are N decision trees. The features used by the decision trees are not exactly the same.

[0112] If C=4, the features used by the first decision tree may be a1, b1, b3, b7, and the features used by the Nth decision tree may be a3, b2, b6, b3.

[0113] c) The final random forest regression result is determined by the weighted average of the regression of N decision trees.

[0114] For example, the saturation throughput prediction values ​​of the six regression decision trees are 100, 110, 120, 140, 100, and 90, and the final sample saturation throughput prediction value output by the random forest is (100+110+120+140+100+90) / 6=110.

[0115] Decision tree construction:

[0116] a) In step b) of the above “Random Forest Model Construction”, a single decision tree needs to be trained using the sub-training set data. The specific process is as follows. Here, we take the construction of a classification decision tree as an example. The decision tree is generated recursively.

[0117] b) Input D train_bsi (X,Y), X is the 14-dimensional feature combination, Y is the actual value of the saturation throughput, based on D train_bsi (X,Y) generates a decision tree node.

[0118] c) Determine whether the current node of the decision tree is a leaf node. The criteria for determining a leaf node are:

[0119] (1) The predicted values ​​of the current data set are the same (the mean square error (MSE) of the current data set is less than the preset threshold) and cannot be divided any further;

[0120] (2) The number of samples in the data set is less than the preset threshold;

[0121] (3) When the current node depth is greater than the maximum depth of the decision tree, the label return value of the leaf node is given according to the result with the largest label proportion in the current data set.

[0122] d) If it is not a leaf node, continue to divide and execute step e); if it is a leaf node, jump to step g).

[0123] e) Get the number M and size n of features of data Q under the current node. Select feature j and partition threshold t from M features m , the data is divided into two categories to form Q left and Q right The specific feature selection and threshold division process is:

[0124] (1) Let the current purity be impurity cur =max;

[0125] (2) Traverse M features and set the current feature to i;

[0126] (3) Obtain n data under the current feature i, sort them from small to large, traverse the n data and use them as the partition threshold, and set the current data to be t;

[0127] (4) Calculate the purity of the data after dividing it by t next ,like

[0128] impurity next <impurity cur , then set impurity cur =impurity next , j = i, t m =t;

[0129] (5) t = t + 1;

[0130] (6) Repeat steps (3) to (5) until t>n, and all the partition points of feature i have been traversed;

[0131] (7) Repeat steps (2) to (4) until i>M and all features are traversed.

[0132] f) After the division is completed, two new child nodes appear on the left and right. Repeat steps c) to e) for the new nodes.

[0133] g) Check whether there are any undivided nodes. If there are no undivided nodes, the algorithm ends; if there are any undivided nodes, jump to step c).

[0134] h) For example, the initial data input to the decision tree is as shown in Table 4 below:

[0135] Table 4

[0136]

[0137] According to the initial data in Table 4, the first node is constructed, and the data of the first feature a1 is obtained from small to large, which is 0.2, 0.2, 0.3, 0.4, 0.4, 0.6, 0.7, 0.7, 0.9. Assuming that the partition point is 0.2, a1≤0.2 is divided into a group R1={2,5}. The prediction output of the first group is Divide a1>0.2 into another group R2={1,3,4,6,7,8,9}, and the prediction output of the second group is

[0138]

[0139] Among them, because the evaluation function selected is the mean square error (MSE), the prediction output after grouping here directly uses the average value of the regression value Y after grouping. The proof is as follows. The mean square error can be expressed as the following formula (20):

[0140]

[0141] Taking the partial derivative and setting the result to 0, we get the following formula (21):

[0142]

[0143] Solved

[0144]

[0145] The same can be said

[0146]

[0147] Therefore, the prediction output that minimizes mse after grouping is the arithmetic mean of the group sum.

[0148] Among them, MSE(a1,0.2)=(110-95) 2 +(80-95) 2 +(100-101.42) 2 +

[0149] (120-101.42) 2 +(90-101.42) 2 +(100-101.42) 2 +(110-101.42) 2 +(90-101.42)

[0150] 2 +(100-101.42) 2 =1135.7148.

[0151] Similarly, the division points, prediction outputs and corresponding MSE results of feature a1 or a2 can be calculated respectively as shown in Table 5 below:

[0152] Table 5

[0153] <![CDATA[a1]]> 0.2 0.3 0.4 0.6 0.7 0.9 <![CDATA[Y R1 ]]> 95 96.67 102 103.33 100 100 <![CDATA[Y R2 ]]> 101.42 101.67 97.5 93.33 100 0 MSE 1135.71 1150.00 1155 1000.00 1200 1200 <![CDATA[a2]]> 0.05 0.3 0.4 0.6 0.7 0.8 0.9 <![CDATA[Y R1 ]]> 90 85 90 95 96 101.42 100 <![CDATA[Y R2 ]]> 101.25 104.28 105 104 105 95 0 MSE 1087.5 621.48 750 1020 1020 1135 1200

[0154] It can be seen that by dividing the data at this node with feature a2=0.3, the data can be divided into two regions, and the sum of squared output errors is minimized. The left node data set after segmentation is shown in Table 6 below:

[0155] Table 6

[0156]

[0157] At this point, you can continue to divide the left node data set, or you can stop, depending on the preset stop threshold. Here, it is assumed that the node is no longer divided, and the predicted output is directly 85.

[0158] The right node data sets after segmentation are shown in Table 7 below:

[0159] Table 7

[0160]

[0161] Obviously, the subsets shown in Table 7 can still be divided again. Repeating the above steps can obtain the threshold of feature a1 as 0.6 or the threshold of a2 as 0.8, so that the minimum MSE is 330, which is the optimal division. Similarly, the division points, prediction outputs and corresponding MSE results of feature a1 or a2 can be calculated respectively as shown in Table 8 below:

[0162] Table 8

[0163]

[0164]

[0165] For example, here is an example of how feature importance is calculated in a regression tree:

[0166]

[0167]

[0168]

[0169]

[0170] It can be seen that in this example, the importance f(a2) of feature a2 is much greater than the importance f(a1) of feature a1.

[0171] Model testing:

[0172] a) Construct the network topology of the test data, as shown in Table 9 below:

[0173] Table 9

[0174]

[0175] b) Regularly collect target parameter data such as the average negotiation rate, average packet error rate and air interface state ratio of the test network topology.

[0176] c) Construct the input D of the first test data set test (X):

[0177] The target parameter data collected in b) are converted into a1, a2, a3, a4, b1, b2, b3, b4, b5, b6, b7, b8, b9, b1 according to the above formulas (1) to (18). 10 feature data.

[0178] As shown in Table 9, there are 7 groups of configurations in total, which can generate 7 groups of feature inputs.

[0179] d) Obtain 7 sets of outputs from the random forest, and score each power configuration based on the output results. For example, the output results are shown in Table 10 below:

[0180] Table 10

[0181]

[0182]

[0183] Different output results correspond to prediction results of saturation throughput prediction values ​​of different power configurations.

[0184] For example, in the process of using the first target prediction model to perform saturation throughput prediction processing on the first test data set to obtain the saturation throughput prediction value of the wireless local area network under the network configuration to be tested, the internal structure diagram of a single decision tree in the first target prediction model can be as follows: Figure 2 The decision tree diagram shown.

[0185] In some embodiments, the method further includes: generating sample input feature data and classification labels corresponding to the sample input feature data based on the difference between any two groups of sample feature data in the first sample data set and the difference between sample labels corresponding to the any two groups of sample feature data, wherein each of the sample input feature data includes multiple sample input features; generating a second sample data set according to the sample input feature data and the classification labels corresponding to the sample input feature data; training the model parameters of the initial prediction model based on the second sample data set to obtain a second target prediction model after the model parameters are updated, wherein the second target prediction model is used to predict the saturated throughput performance score under different network configurations to be tested.

[0186] For example, any two groups of sample feature data include first sample feature data and second sample feature data, and the sample labels corresponding to any two groups of sample feature data include a first sample label corresponding to the first sample feature data and a second sample label corresponding to the second sample feature data; for example, the difference between the first sample feature data and the second sample feature data is determined as the sample input feature; the difference between the first sample label and the second sample label is determined as the classification label.

[0187] For example, establish network topologies of different scales, different power configurations, and different AP placements to obtain N kinds of sample network configurations, and some parameter examples are shown in Table 1 above. Then, a flow test is performed on the network topology shown in Table 1 above, and the flow test mode includes a saturated flow test and a non-saturated flow test. Among them, the conditions for the non-saturated flow test may include: a given total flow is sent at a uniform speed; a given total flow starts and ends randomly within a packet sending cycle; a given total throughput is sent within 1 second (s), with a random time length, and stops for the rest of the time. During the flow process, the negotiation rate, packet error rate, and air interface status ratio are collected regularly. The marking time is 0.1s, and the total flow time is 10s. Based on the collected data, the sample average negotiation rate, sample average packet error rate, and sample air interface status ratio of each wireless access point are determined.

[0188] When training the second target prediction model, since the example solution only needs to evaluate the performance relationship under different power solutions under the same network topology, such as which one has a higher saturation throughput performance between number 1 and number 2, no specific saturation throughput prediction value is required. As mentioned above, the original throughput relationship can be converted into a linear relationship, and the throughput of power configurations i and j can be expressed as the following formulas (24) and (25):

[0189] Throughoutput i =a 1i w1+a 2i w2+a 3i w3+a 4i w4+b 1i k1+b 2i k2+b 3i k3+b 4i k4+

[0190] b 5i k5+b 6i k6+b 7i k7+b 8i k8+b 9i k9+b 10i k 10 (twenty four);

[0191] Throughoutput j =a 1j w1+a 2j w2+a 3j w3+a 4j w4+b 1j k1+b 2j k2+b 3j k3+b 4j k4+

[0192] b 5jk5+b 6j k6+b 7j k7+b 8j k8+b 9j k9+b 10j k 10 (25).

[0193] The difference in saturation performance between the two configurations can be expressed as follows:

[0194] Throughoutput i-j =Throughoutput i -Throughoutput j =(a 1i -a 1j )w1+

[0195] (a 2i -a 2j )w2+(a 3i -a 3j )w3+(a 4i -a 4j )w4+(b 1i -b 1j )k1+(b 2i -b 2j )k2+(b 3i -b 3j )k3+

[0196] (b 4i -b 4j )k4+(b 5i -b 5j )k5+(b 6i -b 6j )k6+(b 7i -b 7j )k7+(b 8i -b 8j )k8+(b 9i -b 9j )k9+

[0197] (b 10i -b 10j ) 10 (26).

[0198] Training set partitioning:

[0199] a) Define input: initial sample input features of two different configurations under the same network topology, where different configurations here not only represent power differences, but also represent running flow pattern differences;

[0200] Among them, the initial sample characteristics under the single network topology and single power configuration described above are: a1, a2, a3, a4, b1, b2, b3, b4, b5, b6, b7, b8, b9, b 10 . The expressions of the above initial sample features can refer to formula (5) to formula (18). As formula (26) describes the difference in throughput performance between the two configurations, the initial sample input features have been transformed into: a 1i -a 1j 、a 2i -a 2j 、a 3i -a 3j 、a 4i -a 4j 、b 1i -b 1j 、b 2i -b 2j 、b 3i -b 3j 、b 4i -b 4j 、b 5i -b 5j 、b 6i -b 6j 、b 7i -b 7j 、b 8i -b 8j 、b 9i -b 9j 、b 10i -b 10j For the convenience of expression, in the subsequent expressions, they are still expressed as a1, a2, a3, a4, b1, b2, b3, b4, b5, b6, b7, b8, b9, b 10 .

[0201] b) Define output: the magnitude relationship of two saturated throughputs. There are only three magnitude relationships, and the classification labels are defined as -1, 0, and 1. The specific meanings are:

[0202] -1 means that the saturated throughput of power configuration i is less than the saturated throughput of power configuration j;

[0203] 0 means that the saturated throughput of power configuration i is similar to the saturated throughput of power configuration j, for example, the difference in saturated throughput is within 5%;

[0204] 1 means that the saturated throughput of power configuration i is greater than the saturated throughput of power configuration j.

[0205] Each initial sample input feature in the above-mentioned initial sample input feature data is normalized and standardized to generate sample input feature data corresponding to each sample network configuration in N types of sample network configurations, and each sample feature data includes multiple sample input features.

[0206] c) Generate a second sample data set based on the sample input feature data and the classification label corresponding to the sample input feature data. The specific division form is D train (X,Y), X represents the sample input feature data, and Y represents the classification label.

[0207] For example, suppose that a topology with the same number of APs, AP placement type, and AP location parameters has 20 power configurations, 5 traffic flow modes, and a total of 100 sets of data. The feature difference between any two sets of data in the 100 sets of data can constitute the input X, so there are 100*99 / 2 input combinations that together constitute X. For example, the input of the first set is i=1, j=2, and the input of the last set is i=99, j=100. However, the output label of the training set is only based on the throughput performance difference recorded in the two saturated traffic flow modes, and the classification label value is given.

[0208] Random Forest Model Construction:

[0209] a) For D train (X,Y) is sampled N times with replacement to generate N sub-datasets D train_bs1 , D train_bs2 ,……,D train_bsN , the sub-datasets can contain repeated entries, and the size of each sub-training set is consistent with the original data set. train The original data set of (X,Y) is shown in Table 11 below:

[0210] Table 11

[0211]

[0212]

[0213] For example, the sampled sub-dataset D train_bsi (X,Y) may be as shown in Table 12 below:

[0214] Table 12

[0215]

[0216] b) Using a single sub-training set D train_bsi For data, C features are randomly selected to train a single decision tree (C<14), and finally there are N decision trees. The features used by the decision trees are not exactly the same.

[0217] If C=4, the features used by the first decision tree may be a1, b1, b3, b7, and the features used by the Nth decision tree may be a3, b2, b6, b3.

[0218] c) The final random forest regression result is determined by the weighted average of the regression of N decision trees.

[0219] For example, among 6 decision trees, 0 has a classification result of -1, 1 has a classification result of 0, and 5 have a classification result of 1, then the final classification result of the decision tree is 1.

[0220] Decision tree construction:

[0221] a) In step b) of the above “Random Forest Model Construction”, a single decision tree needs to be trained using the sub-training set data. The specific process is as follows. Here, we take the construction of a classification decision tree as an example. The decision tree is generated recursively.

[0222] b) Input D train_bsi (X,Y), X is a 14-dimensional feature combination, and Y is a classification label. train_bsi (X,Y) generates a decision tree node.

[0223] c) Determine whether the current node of the decision tree is a leaf node. The criteria for determining a leaf node are:

[0224] (1) The labels of the current data set are the same and cannot be divided further;

[0225] (2) When the current node depth is greater than the maximum depth of the decision tree, the label return value of the leaf node is given according to the result with the largest label proportion in the current data set.

[0226] d) If it is not a leaf node, continue to divide and execute step e); if it is a leaf node, jump to step g).

[0227] e) Get the number M and size n of features of data Q under the current node. Select feature j and partition threshold t from M features m , the data is divided into two categories to form Q left and Q right The specific feature selection and threshold division process is:

[0228] (1) Let the current purity be impurity cur =max;

[0229] (2) Traverse M features and set the current feature to i;

[0230] (3) Obtain n data under the current feature i, sort them from small to large, traverse the n data and use them as the partition threshold, and set the current data to be t;

[0231] (4) Calculate the purity of the data after dividing it by t next ,like

[0232] impurity next <impurity cur , then set impurity cur =impurity next , j = i, t m =t;

[0233] (5) t = t + 1;

[0234] (6) Repeat steps (3) to (5) until t>n, and all the partition points of feature i have been traversed;

[0235] (7) Repeat steps (2) to (4) until i>M and all features are traversed.

[0236] f) After the division is completed, two new child nodes appear on the left and right. Repeat steps c) to e) for the new nodes.

[0237] g) Check whether there are any undivided nodes. If there are no undivided nodes, the algorithm ends; if there are any undivided nodes, jump to step c).

[0238] h) For example, the initial data input to the decision tree is as shown in Table 13 below:

[0239] Table 13

[0240]

[0241] According to the initial data in Table 13, the first node is constructed, and the data of the first feature a1 is obtained from small to large, which are 0.2, 0.2, 0.3, 0.4, 0.4, 0.6, 0.7, 0.7, 0.9. Assuming that the partition point is 0.2, a1≤0.2 is divided into one group, and a1>0.2 is divided into another group. The purity after partition is calculated as:

[0242] impurity next (a1,0.2) = Gini(a1,0.2);

[0243] impurity cur =Gini(D)=1-(3 / 9) 2 -(3 / 9) 2 -(3 / 9) 2 =2 / 3;

[0244] Gini(a1,0.2)=Gini(D left )+Gini(D right )=(1-(0 / 2)2 -(1 / 2) 2 -(1 / 2) 2 )*(2 / 9)+(1-(3 / 7) 2 -(2 / 7) 2 -(2 / 7) 2 )*(2 / 7)=0.5*(2 / 9)+0.653*(7 / 9)=0.619;

[0245] Less than the current purity, update the optimal purity to 0.619, the optimal feature to a1, and the optimal partition threshold to 0.2;

[0246] Gini(a1,0.3)=Gini(D left )+Gini(D right )=(1-(0 / 3) 2 -(2 / 3) 2 -(1 / 3) 2 )*(3 / 9)+(1-(3 / 6) 2 -(1 / 6) 2 -(2 / 6) 2 )*(6 / 9)=0.556*(3 / 9)+0.611*(6 / 9)=0.593;

[0247] Less than the current purity, update the optimal purity to 0.593, the optimal feature to a1, and the optimal partition threshold to 0.3;

[0248] Gini(a1,0.4)=Gini(D left )+Gini(D right )

[0249] =(1-(0 / 5) 2 -(3 / 5) 2 -(2 / 5) 2 )*(5 / 9)+(1-(3 / 4) 2 -(0 / 4) 2 -(1 / 4) 2 )*(4 / 9)=0.48*(5 / 9)+0.375*(4 / 9)=0.433;

[0250] Less than the current purity, update the optimal purity to 0.433, the optimal feature to a1, and the optimal partition threshold to 0.4;

[0251] Gini(a1,0.6)=Gini(D left )+Gini(D right )=(1-(1 / 6) 2 -(3 / 6) 2 -(2 / 6)2 )*(6 / 9)+(1-2 / 3) 2 -0 / 3) 2 -(1 / 3) 2 )*(3 / 9)=0.611*(6 / 9)+0.444*(3 / 9)=0.555;

[0252] Greater than the current purity, no update;

[0253] Gini(a1,0.7)=Gini(D left )+Gini(D right )

[0254] =(1-(2 / 8) 2 -(3 / 8) 2 -(3 / 8) 2 )*(8 / 9)+(1-(1 / 1) 2 -(0 / 1) 2 -(0 / 1) 2 )*(1 / 9)=0.65625*(8 / 9)+0*(1 / 9)=0.583;

[0255] Greater than the current purity, no update;

[0256] Gini(a1,0.9)=Gini(D);

[0257] If it is greater than the current purity, it will not be updated. All the partition thresholds of feature a1 have been traversed, so the optimal threshold of a1 is 0.4.

[0258] Similarly, the data of the second feature a2 are 0.05, 0.3, 0.4, 0.6, 0.7, 0.8, 0.8, 0.9, and 0.9 from small to large, and the calculated purities are shown in Table 14 below:

[0259] Table 14

[0260] Threshold 0.05 0.3 0.4 0.6 0.7 0.8 0.9 purity 0.583 0.476 0.333 0.433 0.433 0.653 0.667

[0261] It is concluded that the first node has the highest benefit when divided according to feature a2 and threshold 0.4. The left node data set after division is shown in Table 15 below:

[0262] Table 15

[0263]

[0264] Obviously, the subset shown in Table 15 meets the leaf node standard and all data labels are the same.

[0265] The right node data sets after segmentation are shown in Table 16 below:

[0266] Table 16

[0267]

[0268] Obviously, the subsets shown in Table 16 can still be divided again. Repeating the above steps can obtain the threshold value of feature a1 of 0.4 so that the purity of the left and right subsets is 0, which is the optimal division. For example, the following example illustrates the calculation process of the importance of two features:

[0269]

[0270]

[0271]

[0272]

[0273] At this time, the importance of the two features f(a2) and f(a1) is the same, both 0.5, indicating that the two features are equally important.

[0274] In some embodiments, if the network configuration to be tested includes multiple different network configurations to be tested, the first test data set is generated by data preprocessing based on the target parameter data of each wireless access point under each network configuration to be tested in the multiple different network configurations to be tested, and the first test data set includes the feature data to be tested corresponding to each network configuration to be tested; the method also includes: generating input feature data based on the difference between any two groups of feature data to be tested in the first test data set, and generating a second test data set based on the input features to be tested, each of the input feature data including multiple input features; inputting the second test data set into a second target prediction model for saturated throughput performance score prediction processing to obtain the saturated throughput performance score of the wireless local area network under different network configurations to be tested.

[0275] For example, a common application scenario is that for the same wireless network, it is often necessary to compare whether the network has better wireless performance under different wireless configuration parameters, that is, to evaluate the pros and cons of the deployment scheme. In this scenario, the mapping outputs (saturated throughput performance scores) of the two deployment schemes can be evaluated separately as the evaluation comparison results of the two deployment schemes.

[0276] For example, the model test is as follows:

[0277] a) Construct the network topology of the test data, including a variety of different network configurations to be tested, as shown in Table 17 below:

[0278] Table 17

[0279] serial number AP placement type Positional parameters Number of APs AP Power 1 square AP distance 12m 25 23 2 square AP distance 12m 25 20 3 square AP distance 12m 25 17 4 square AP distance 12m 25 14 5 square AP distance 12m 25 11 6 square AP distance 12m 25 8 7 square AP distance 12m 25 5

[0280] b) Regularly collect target parameter data such as the average negotiation rate, average packet error rate and air interface status ratio of the test network topology.

[0281] c) Construct the input D of the second test data set test (X):

[0282] The target parameter data collected in b) are converted into a1, a2, a3, a4, b1, b2, b3, b4, b5, b6, b7, b8, b9, b10, b11, b12, b3, b4, b5, b6, b7, b8, b9, b12, b13, b14, b15, b16, b17, b18, b19, b20, b21, b22, 10 feature data.

[0283] As shown in Table 17, there are 7 groups of configurations in total, which can generate 7*6 / 2=21 groups of feature inputs.

[0284] d) 21 sets of outputs are obtained from the random forest. Based on the output results, the saturation throughput performance of each power configuration is scored. For example, the output results are shown in Table 18 below:

[0285] Table 18

[0286]

[0287]

[0288] The saturated throughput performance score of each configuration is calculated. The scoring rule is to add up the output results of each configuration. The output results of the same configuration are 0. For example, the saturated throughput performance score is shown in Table 19 below:

[0289] Table 19

[0290] Configuration Saturated Throughput Performance Rating 1 (0)+(-1)+(-1)+(-1)+(-1)+(-1)+(-1)＝-6 2 (1)+(0)+(-1)+(-1)+(-1)+(-1)+(-1)＝-4 3 (1)+(1)+(0)+(-1)+(-1)+(-1)+(-1)＝-2 4 (1)+(1)+(1)+(0)+(0)+(-1)+(-1)＝1 5 (1)+(1)+(1)+(0)+(0)+(-1)+(-1)＝1 6 (1)+(1)+(1)+(1)+(1)+(0)+(1)＝6 7 (1)+(1)+(1)+(1)+(1)+(-1)+(0)＝4

[0291] For example, in the process of using the second target prediction model to predict the saturation throughput performance scores under different network configurations to be tested, the internal structure diagram of a single decision tree in the second target prediction model can be as follows: Figure 3 The decision tree diagram shown.

[0292] The embodiment of the present application provides a mathematical model for predicting the saturated throughput of a wireless local area network, and provides a parameter training scheme based on machine learning to support the mathematical model. The machine learning training model can be trained offline, and the actual prediction can be obtained in real time.

[0293] Compared with the traditional prediction method, the embodiment of the present application has the following characteristics:

[0294] Without strong a priori assumptions, all parameters can be used in various wireless scenarios after being trained and acquired during the R&D phase; therefore, it is more adaptable to complex and changing network environments.

[0295] Applicable to non-ideal channel models, that is, transmission errors are not limited to those caused by transmission collisions, but can also be caused by the channel quality of the physical layer; the saturation throughput prediction method provided in the embodiment of the present application can take these factors into account, and thus can more accurately predict the saturation throughput of the network.

[0296] Applicable to non-saturated scenarios and user scenarios, even under non-saturated network usage conditions, the saturated throughput prediction method provided by the embodiment of the present application can predict the throughput in saturated service scenarios by inference. This is of great significance for network planning and optimization.

[0297] It is applicable to non-single channel models, and each communication node may work on a different channel, and may have a working state of being detectable, partially detectable, or completely undetectable; in the network, different communication nodes may work on different channels. The saturation throughput prediction method provided in the embodiment of the present application can handle this situation and is applicable to multi-channel models.

[0298] Applicable to multi-node operation, and each node does not need symmetry or other constraints. In a wireless local area network, there are usually multiple nodes working at the same time. The saturation throughput prediction method provided in the embodiment of the present application does not require symmetry or other constraints between each node, so it is more adaptable to complex network environments.

[0299] The embodiment of the present application effectively solves the problem that it is difficult to obtain the maximum bearing capacity of the network in a large-scale wireless local area network. The embodiment of the present application can more accurately reflect the actual network conditions by obtaining and using the target parameter data of the wireless access point under the actual network configuration, such as the average negotiation rate, the average packet error rate and the proportion of the air interface state, thereby improving the accuracy of the saturation throughput prediction. By performing meticulous data preprocessing on the collected target parameter data, it is converted into a first test data set suitable for machine learning, which provides high-quality input for subsequent predictions. Subsequently, the saturation throughput prediction is performed using the first target prediction model trained based on a rich sample data set. Since the first target prediction model is trained based on a data set under a variety of sample network configurations, this enables the first target prediction model to adapt to complex and changeable wireless network environments (such as different network environments and configurations), enhances the universality and applicability of the model, and improves the accuracy of the saturation throughput prediction, so as to achieve a more accurate and comprehensive evaluation of the wireless network performance. Since the first target prediction model is trained based on a data set under a variety of sample network configurations, this enables the prediction model to adapt to different network environments and configurations, enhancing the universality and applicability of the model.

[0300] In addition, through the embodiments of the present application, the saturated throughput performance evaluation can be further performed on the current state of the network to provide reference indicators for the wireless configuration strategy under the wireless networking system. The administrator can understand the performance of the current network under different configurations based on the evaluation results, and thus select the optimal configuration scheme. This can not only improve the throughput and performance of the network, but also reduce the risk of network failures and improve the stability and reliability of the network. When the network configuration to be tested includes a variety of different configurations, this embodiment can predict the saturated throughput performance scores under different configurations by generating input feature data and constructing a second test data set. This makes the evaluation results more comprehensive and meets the evaluation needs under different complex scenarios.

[0301] All of the above technical solutions can be arbitrarily combined to form optional embodiments of the present application, which will not be described in detail here.

[0302] The embodiment of the present application obtains the target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, and the target parameter data includes the average negotiation rate, the average packet error rate and the proportion of the air interface state; performs data preprocessing on the target parameter data of each wireless access point to generate a first test data set; inputs the first test data set into the first target prediction model to perform saturation throughput prediction processing to obtain the saturation throughput prediction value of the wireless local area network under the network configuration to be tested; wherein the first target prediction model is obtained based on the training of the first sample data set, and the first sample data set is generated based on the sample labels of the wireless local area network under N sample network configurations and the sample parameter data of each wireless access point, and N is a natural number greater than 0. The embodiment of the present application obtains and utilizes the target parameter data of the wireless access point under the actual network configuration, such as the average negotiation rate, the average packet error rate and the proportion of the air interface state, so as to more accurately reflect the actual network conditions, thereby improving the accuracy of the saturation throughput prediction. By performing meticulous data preprocessing on the collected target parameter data and converting it into a first test data set suitable for machine learning, high-quality input is provided for subsequent predictions. Subsequently, the first target prediction model trained based on a rich sample data set is used to predict the saturated throughput. Since the first target prediction model is trained based on a data set under a variety of sample network configurations, this enables the first target prediction model to adapt to complex and changeable wireless network environments (such as different network environments and configurations), enhancing the universality and applicability of the model, and improving the accuracy of the saturated throughput prediction, so as to achieve a more accurate and comprehensive evaluation of wireless network performance. Since the first target prediction model is trained based on a data set under a variety of sample network configurations, this enables the prediction model to adapt to different network environments and configurations, enhancing the universality and applicability of the model.

[0303] In order to better implement the saturation throughput prediction method of the embodiment of the present application, the embodiment of the present application also provides a saturation throughput prediction device. Figure 4 , Figure 4 A schematic diagram of the structure of a saturation throughput prediction device provided in an embodiment of the present application. The saturation throughput prediction device 200 may include:

[0304] An acquisition unit 210 is used to acquire target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, wherein the target parameter data includes an average negotiation rate, an average packet error rate, and an air interface state ratio;

[0305] A first generating unit 220, configured to perform data preprocessing on the target parameter data of each wireless access point to generate a first test data set;

[0306] A first prediction unit 230 is used to input the first test data set into a first target prediction model to perform saturation throughput prediction processing to obtain a saturation throughput prediction value of the wireless local area network under the network configuration to be tested;

[0307] Among them, the first target prediction model is trained based on a first sample data set, and the first sample data set is generated based on sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, where N is a natural number greater than 0.

[0308] In some embodiments, the acquisition unit 210 is specifically used to: acquire an average negotiation rate between an i-th wireless access point and a client in the wireless local area network under the network configuration to be tested according to a preset period, where i is a natural number greater than 0; acquire an average packet error rate of the i-th wireless access point according to the preset period; acquire an air interface state ratio of the i-th wireless access point according to the preset period, where the air interface state ratio includes a transmitting state ratio, a receiving state ratio, an idle state ratio, and a busy state ratio; obtain target parameter data of the i-th wireless access point in the wireless local area network under the network configuration to be tested according to the average negotiation rate between the i-th wireless access point and the client, the average packet error rate of the i-th wireless access point, and the air interface state ratio; and traverse each wireless access point to acquire the target parameter data of each wireless access point in the wireless local area network under the network configuration to be tested.

[0309] In some embodiments, the first generating unit 220 is specifically used to: construct a first mathematical model corresponding to the statistical probability of occupying a channel packet according to the transmitting state proportion, the receiving state proportion, the idle state proportion and the busy state proportion; construct a second mathematical model corresponding to the saturation throughput prediction value according to the average negotiation rate, the average packet error rate and the statistical probability of occupying a channel packet; substitute the average negotiation rate, the average packet error rate and the first mathematical model into the second mathematical model for calculation to obtain initial feature data to be tested corresponding to the network configuration to be tested; perform normalization and standardization on the initial feature data to be tested to obtain the feature data to be tested corresponding to the network configuration to be tested, and generate a first test data set based on the feature data to be tested.

[0310] In some embodiments, the saturated throughput prediction device 200 also includes a first training unit, which is used to: obtain sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, the sample parameter data including a sample average negotiation rate, a sample average packet error rate and a sample air interface state ratio, and the sample label represents the true value of the saturated throughput of the wireless local area network under each sample network configuration; perform data preprocessing on the sample parameter data of each wireless access point to generate sample feature data corresponding to each sample network configuration in the N sample network configurations; construct a first sample data set based on the sample feature data and the sample label; and train the initial prediction model based on the first sample data set to obtain the first target prediction model.

[0311] In some embodiments, when the first training unit obtains the sample labels and sample parameter data of each wireless access point collected by the wireless local area network under N sample network configurations, it can be used to: deploy network configuration information corresponding to each sample network configuration in N sample network configurations for the wireless local area network, the network configuration information including network topology information, channel environment information and wireless configuration parameters; based on the running flow test conditions of the same network topology information, the same channel environment information and the same wireless configuration parameters, perform saturated running flow test and non-saturated running flow test on each wireless access point under the N sample network configurations; based on the saturated running flow test, obtain the true value of the saturated throughput of the wireless local area network under the N sample network configurations to obtain the sample label; based on the non-saturated running flow test, obtain the sample parameter data of each wireless access point collected by the wireless local area network under the N sample network configurations.

[0312] In some embodiments, the sample air interface state ratio includes a sample transmission state ratio, a sample reception state ratio, a sample idle state ratio and a sample busy state ratio;

[0313] When the first training unit performs data preprocessing on the sample parameter data of each wireless access point to generate sample feature data corresponding to each sample network configuration in the N types of sample network configurations, it can be used to: construct a first mathematical model corresponding to the statistical probability of sample occupying channel packets according to the sample transmission state proportion, the sample receiving state proportion, the sample idle state proportion and the sample busy state proportion; construct a second mathematical model corresponding to the saturation throughput prediction value according to the sample average negotiation rate, the sample average packet error rate and the sample occupying channel packet statistical probability; substitute the sample average negotiation rate, the sample average packet error rate and the first mathematical model into the second mathematical model for calculation to obtain the initial sample feature data corresponding to the sample network configuration; perform normalization and standardization on the initial sample feature data to generate sample feature data corresponding to each sample network configuration in the N types of sample network configurations, each of the sample feature data includes multiple sample features.

[0314] In some embodiments, when the first training unit trains the initial prediction model based on the first sample data set to obtain the first target prediction model, it can be used to: input the sample feature data in the first sample data set into the initial prediction model to perform saturation throughput prediction processing to obtain the sample saturation throughput prediction value of the wireless local area network under the sample network configuration corresponding to the sample feature data; train the model parameters of the initial prediction model according to the sample saturation throughput prediction value and the sample label to obtain the first target prediction model after the model parameters are updated.

[0315] In some embodiments, the saturated throughput prediction device 200 further includes a second training unit, which is used to: generate sample input feature data and classification labels corresponding to the sample input feature data based on the difference between any two groups of sample feature data in the first sample data set and the difference between sample labels corresponding to the any two groups of sample feature data, wherein each of the sample input feature data includes multiple sample input features; generate a second sample data set according to the sample input feature data and the classification labels corresponding to the sample input feature data; train the model parameters of the initial prediction model based on the second sample data set to obtain a second target prediction model after the model parameters are updated, and the second target prediction model is used to predict the saturated throughput performance score under different network configurations to be tested.

[0316] In some embodiments, if the network configuration to be tested includes multiple different network configurations to be tested, the first test data set is generated by data preprocessing based on the target parameter data of each wireless access point under each network configuration to be tested in the multiple different network configurations to be tested, and the first test data set includes the feature data to be tested corresponding to each network configuration to be tested; the saturated throughput prediction device 200 also includes a second prediction unit, which is used to: generate input feature data based on the difference between any two groups of feature data to be tested in the first test data set, and generate a second test data set based on the input features to be tested, each of the input feature data includes multiple input features; the second test data set is input into a second target prediction model to perform saturated throughput performance score prediction processing to obtain the saturated throughput performance score of the wireless local area network under different network configurations to be tested.

[0317] All of the above technical solutions can be arbitrarily combined to form optional embodiments of the present application, which will not be described in detail here.

[0318] It should be understood that the saturated throughput prediction device embodiment and the method embodiment may correspond to each other, and similar descriptions may refer to the method embodiment. To avoid repetition, no further description is given here. Specifically, the saturated throughput prediction device may execute the above-mentioned saturated throughput prediction method embodiment, and the aforementioned and other operations and / or functions of each unit in the saturated throughput prediction device respectively implement the corresponding processes of the above-mentioned method embodiment, which will not be described here for the sake of brevity.

[0319] Optionally, the present application further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0320] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present application, and the computer device may be a terminal or a server. Figure 5 As shown, the computer device 300 may include: a communication interface 301, a memory 302, a processor 303 and a communication bus 304. The communication interface 301, the memory 302, and the processor 303 communicate with each other through the communication bus 304. The communication interface 301 is used for the computer device 300 to communicate data with external devices. The memory 302 can be used to store software programs and modules, and the processor 303 runs the software programs and modules stored in the memory 302, such as the software programs of the corresponding operations in the aforementioned method embodiments.

[0321] Optionally, the processor 303 can call the software program and module stored in the memory 302 to perform the following operations: obtain target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, the target parameter data including the average negotiation rate, the average packet error rate and the air interface state ratio; perform data preprocessing on the target parameter data of each wireless access point to generate a first test data set; input the first test data set into a first target prediction model to perform saturation throughput prediction processing to obtain a saturation throughput prediction value of the wireless local area network under the network configuration to be tested; wherein the first target prediction model is obtained by training based on a first sample data set, and the first sample data set is generated based on sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, and N is a natural number greater than 0.

[0322] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0323] To this end, the embodiment of the present application provides a computer-readable storage medium, in which multiple computer programs are stored, and the computer program can be loaded by a processor to execute the steps in any saturation throughput prediction method provided in the embodiment of the present application. The specific implementation of each of the above operations can be referred to the previous embodiment, and will not be repeated here.

[0324] The storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0325] Since the computer program stored in the storage medium can execute the steps in any one of the saturation throughput prediction methods provided in the embodiments of the present application, the beneficial effects that can be achieved by any one of the saturation throughput prediction methods provided in the embodiments of the present application can be achieved. For details, please refer to the previous embodiments and will not be repeated here.

[0326] The embodiment of the present application also provides a computer program product, which includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the corresponding process in any saturation throughput prediction method in the embodiment of the present application, which will not be described here for brevity.

[0327] The embodiment of the present application also provides a computer program, which includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the corresponding process in any saturation throughput prediction method in the embodiment of the present application, which will not be described here for the sake of brevity.

[0328] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A saturation throughput prediction method, characterized in that: The method comprises: Obtain target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, wherein the target parameter data includes an average negotiation rate, an average packet error rate, and an air interface state ratio; Performing data preprocessing on the target parameter data of each wireless access point to generate a first test data set; Inputting the first test data set into a first target prediction model to perform saturation throughput prediction processing to obtain a saturation throughput prediction value of the wireless local area network under the network configuration to be tested; Among them, the first target prediction model is trained based on a first sample data set, and the first sample data set is generated based on sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, where N is a natural number greater than 0.

2. The saturation throughput prediction method according to claim 1, characterized in that: The step of obtaining target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested includes: Obtaining an average negotiation rate between an i-th wireless access point and a client in the wireless local area network under the network configuration to be tested according to a preset period, where i is a natural number greater than 0; Obtaining an average packet error rate of the i-th wireless access point according to the preset period; Acquire the air interface state ratio of the i-th wireless access point according to the preset period, where the air interface state ratio includes a transmitting state ratio, a receiving state ratio, an idle state ratio and a busy state ratio; Obtaining target parameter data of the i-th wireless access point in the wireless local area network under the network configuration to be tested according to the average negotiation rate between the i-th wireless access point and the client, the average packet error rate of the i-th wireless access point, and the air interface state ratio; Each wireless access point is traversed to obtain target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested.

3. The saturation throughput prediction method according to claim 2, characterized in that: The performing data preprocessing on the target parameter data of each wireless access point to generate a first test data set includes: Constructing a first mathematical model corresponding to the saturation throughput prediction value according to the average negotiation rate, the average packet error rate and the statistical probability of occupying a channel for sending packets; Constructing a second mathematical model corresponding to the statistical probability of occupying a channel for sending packets according to the transmit state proportion, the receive state proportion, the idle state proportion and the busy state proportion; Substituting the average negotiation rate, the average packet error rate and the second mathematical model into the first mathematical model for calculation, to obtain initial characteristic data to be tested corresponding to the network configuration to be tested; The initial feature data to be tested is normalized and standardized, the feature data to be tested corresponding to the network configuration to be tested is obtained, and a first test data set is generated based on the feature data to be tested.

4. The saturation throughput prediction method according to claim 1, characterized in that: The training step of the first target prediction model includes: Obtaining sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, the sample parameter data including a sample average negotiation rate, a sample average packet error rate, and a sample air interface state ratio, the sample label representing a true value of a saturated throughput of the wireless local area network under each sample network configuration; Performing data preprocessing on the sample parameter data of each wireless access point to generate sample feature data corresponding to each sample network configuration in the N types of sample network configurations; Constructing a first sample data set based on the sample feature data and the sample label; The initial prediction model is trained based on the first sample data set to obtain the first target prediction model.

5. The saturation throughput prediction method according to claim 4, characterized in that: The obtaining of sample labels collected by the wireless local area network under N sample network configurations and sample parameter data of each wireless access point includes: Deploy network configuration information corresponding to each sample network configuration in N sample network configurations for the wireless local area network, wherein the network configuration information includes network topology information, channel environment information, and wireless configuration parameters; Based on the running flow test conditions of the same network topology information, the same channel environment information and the same wireless configuration parameters, a saturated running flow test and a non-saturated running flow test are respectively performed on each wireless access point under the N sample network configurations; Based on the saturated running flow test, obtaining the true value of the saturated throughput of the wireless local area network under N kinds of sample network configurations, and obtaining the sample label; Based on the non-saturated running flow test, sample parameter data of each wireless access point collected by the wireless local area network under N sample network configurations are obtained.

6. The saturation throughput prediction method according to claim 4, characterized in that: The sample air interface state ratio includes the sample transmission state ratio, the sample reception state ratio, the sample idle state ratio and the sample busy state ratio; The performing data preprocessing on the sample parameter data of each wireless access point to generate sample feature data corresponding to each sample network configuration in the N types of sample network configurations includes: Constructing a first mathematical model corresponding to the saturation throughput prediction value according to the sample average negotiation rate, the sample average packet error rate and the sample occupied channel packet sending statistical probability; Constructing a second mathematical model corresponding to the statistical probability of the sample occupying a channel for sending packets according to the sample transmission state proportion, the sample receiving state proportion, the sample idle state proportion and the sample busy state proportion; Substituting the sample average negotiation rate, the sample average packet error rate and the second mathematical model into the first mathematical model for calculation to obtain initial sample characteristic data corresponding to the sample network configuration; The initial sample feature data is normalized and standardized to generate sample feature data corresponding to each sample network configuration in the N sample network configurations, each of the sample feature data including a plurality of sample features.

7. The saturation throughput prediction method according to claim 4, characterized in that: The training of the initial prediction model based on the first sample data set to obtain the first target prediction model includes: Inputting the sample feature data in the first sample data set into the initial prediction model to perform saturation throughput prediction processing, and obtaining a sample saturation throughput prediction value of the wireless local area network under the sample network configuration corresponding to the sample feature data; The model parameters of the initial prediction model are trained according to the sample saturation throughput prediction value and the sample label to obtain the first target prediction model after the model parameters are updated.

8. The saturation throughput prediction method according to any one of claims 4 to 6, characterized in that: The method further comprises: Based on the difference between any two groups of sample feature data in the first sample data set and the difference between the sample labels corresponding to the any two groups of sample feature data, generate sample input feature data and classification labels corresponding to the sample input feature data, each of the sample input feature data includes a plurality of sample input features; Generate a second sample data set according to the sample input feature data and the classification label corresponding to the sample input feature data; The model parameters of the initial prediction model are trained based on the second sample data set to obtain a second target prediction model after the model parameters are updated, and the second target prediction model is used to predict the saturation throughput performance score under different network configurations to be tested.

9. The saturation throughput prediction method according to claim 8, characterized in that: If the network configuration to be tested includes multiple different network configurations to be tested, the first test data set is generated by data preprocessing based on the target parameter data of each wireless access point under each network configuration to be tested in the multiple different network configurations to be tested, and the first test data set includes the feature data to be tested corresponding to each network configuration to be tested; The method further comprises: Generate input feature data based on the difference between any two sets of feature data to be tested in the first test data set, and generate a second test data set based on the input features to be tested, each of the input feature data includes a plurality of input features; The second test data set is input into a second target prediction model to perform saturation throughput performance score prediction processing to obtain a saturation throughput performance score of the wireless local area network under different network configurations to be tested.

10. A saturation throughput prediction device, characterized in that: The device comprises: An acquisition unit, configured to acquire target parameter data of each wireless access point of the wireless local area network under the network configuration to be tested, wherein the target parameter data includes an average negotiation rate, an average packet error rate, and a proportion of air interface status; A first generating unit, configured to perform data preprocessing on the target parameter data of each wireless access point to generate a first test data set; A first prediction unit, configured to input the first test data set into a first target prediction model to perform saturation throughput prediction processing, so as to obtain a saturation throughput prediction value of the wireless local area network under the network configuration to be tested; Among them, the first target prediction model is trained based on a first sample data set, and the first sample data set is generated based on sample labels of the wireless local area network under N sample network configurations and sample parameter data of each wireless access point, where N is a natural number greater than 0.

11. A computer device, characterized in that: The computer device includes a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to execute the saturation throughput prediction method according to any one of claims 1 to 9 by calling the computer program stored in the memory.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the saturation throughput prediction method according to any one of claims 1 to 9.

13. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the saturation throughput prediction method according to any one of claims 1 to 9 is implemented.

Images