A wireless local area network throughput prediction method and storage medium

Abstract

The application relates to a wireless local area network throughput prediction method and a storage medium, and the wireless local area network throughput prediction method comprises the following steps: acquiring a network topology, a signal strength, a channel access mechanism, an MCS and a spatial flow array combination, a PER, a test duration and AP effective transmission sequence data. An initial physical model is input to calculate an initial throughput. Linear regression is used to fit the initial prediction and the measured throughput to obtain a slope correction coefficient and an intercept correction term, which are used to construct a correction model. The model is cross-validated and trained and verified. The network topology, RSSI and channel access mechanism are input to predict the overall network throughput. The dynamic correction model solves the prediction deviation problem caused by ignoring the transient fluctuation of the wireless channel.

Description

Technical Field

[0001] This invention relates to the field of wireless communication network optimization, specifically to a method for predicting wireless local area network throughput and a storage medium. Background Technology

[0002] With the evolution of the Wi-Fi 7 standard and the widespread deployment of high-density access points (APs), co-channel interference and data collisions caused by overlapping signal coverage of adjacent APs have become increasingly prominent. In dense AP scenarios, channel resource competition intensifies, and physical layer dynamic factors (such as modulation and coding schemes (MCS) and spatial stream number (NSS)) are coupled with link quality indicators (such as RSSI), leading to a significant decrease in system throughput. Traditional methods rely on large-scale signal strength (RSSI) and static network topology modeling, but they cannot capture the transient fluctuation characteristics of wireless channels (such as micro-scale fading and burst interference), resulting in a serious deviation between throughput prediction and actual performance, thus limiting the effectiveness of network optimization.

[0003] Current mainstream solutions fall into two categories: data-driven models (such as gradient boosting decision trees XGBoost) and physical analysis models. The former predicts throughput by inputting parameters such as RSSI, PHYRate, and MCS / NSS, but in a real-world test scenario with two access points (APs), the mean squared error (MSE) is as high as 137.51 (RMSE = 11.73). This error mainly stems from the limited scope of RSSI for large-scale information. The latter calculates throughput based on the following theoretical formula:

[0004]

[0005] However, because the dynamic uncertainties of the channel are not taken into account, the predicted values ​​are generally too high, with an MSE of 169.62, even worse than the data-driven model. Both types of methods struggle to balance accuracy and real-time performance requirements.

[0006] The rapid time-varying characteristics of wireless channels (such as transient interference and fluctuations in packet loss rate PER) cause the theoretical assumptions of physical models to fail, while purely data-driven models lack physical constraints and have insufficient generalization ability. Especially in high-density AP scenarios (e.g., ≥3 APs), system complexity increases dramatically, and the MSE further rises to 248.01, with prediction accuracy (0.86–0.88) significantly lower than in low-density scenarios (0.91–0.92). Existing technologies lack dynamic correction mechanisms for physical models and cannot adaptively adjust model parameters to fit the actual network conditions, thus limiting the reliability of throughput prediction. Summary of the Invention

[0007] This invention provides a method and storage medium for predicting wireless local area network throughput, aiming to solve the problem of systematic prediction bias caused by the physical model in existing WLAN throughput prediction not being adapted to the dynamic characteristics of the wireless channel.

[0008] To achieve the above objectives, the first aspect of the present invention provides a method for predicting the throughput of a wireless local area network, comprising the following steps:

[0009] Acquire network topology data, signal strength indication data, channel access mechanism data, modulation and coding scheme and spatial stream number combination data, packet error rate data, total test duration data, and predicted access point effective transmission time series data;

[0010] The acquired modulation and coding scheme, combined with spatial stream number data, packet error rate data, total test duration data, and access point effective transmission time series data are substituted into the initial physical model to calculate the initial throughput prediction value.

[0011] Using a linear regression algorithm, the calculated initial throughput prediction value is fitted to the measured throughput to obtain the slope correction coefficient and intercept correction term;

[0012] The obtained slope correction coefficients and intercept correction terms are applied to the initial physical model to construct the corrected physical model;

[0013] Cross-validation is used to train and validate the constructed modified physics model;

[0014] Based on the corrected physical model after training and validation, and inputting the acquired network topology data, signal strength indication data, and channel access mechanism data, the overall network throughput is predicted.

[0015] Furthermore, the method for predicting the effective transmission time series data of the access point includes:

[0016] Acquire the network topology data of the target wireless LAN and extract the location identification information of access points and stations from it;

[0017] Collect the service traffic data of the wireless local area network and encode the transport layer protocol type into discrete feature values;

[0018] Measure the received signal strength indication values ​​between all nodes in the network topology data, and calculate the mean characteristics of the received signal strength indication values ​​for all communication link pairs;

[0019] Read the network allocation vector threshold value configured in the wireless local area network and encode it as a binary feature;

[0020] The extracted location identification information, the encoded discrete feature values, the calculated mean value of the received signal strength indication, and the encoded binarized features are combined to form an input feature vector;

[0021] Set the initial hyperparameter search range for the gradient boosting decision tree regression model;

[0022] The grid search algorithm is used to traverse all combinations within the set initial hyperparameter search range, and the root mean square error corresponding to each combination is calculated by five-fold cross-validation. The hyperparameter combination with the smallest root mean square error is selected as the optimal hyperparameter combination.

[0023] The optimal combination of hyperparameters is used to configure the gradient boosting decision tree regression model, and the model is trained using a training dataset containing input feature vectors and corresponding access point effective transmission time series data.

[0024] The input feature vector constructed from the scenario to be predicted is input into the gradient boosting decision tree regression model after training, and the output is the effective transmission time series data of the access point.

[0025] Furthermore, methods for encoding transport layer protocol types as discrete feature values ​​include:

[0026] When the protocol type is User Datagram Protocol, the value is assigned to 0;

[0027] When the protocol type is Transmission Control Protocol, the value is assigned to 1.

[0028] Furthermore, the five-fold cross-validation is performed as follows:

[0029] The training dataset is randomly divided into five mutually exclusive subsets;

[0030] One subset is selected sequentially as the validation set, and the remaining four subsets are used as the training set.

[0031] Train the model for each hyperparameter combination and calculate the root mean square error of the validation set;

[0032] After five iterations, the average root mean square error is taken as the evaluation index.

[0033] Furthermore, the prediction method for the modulation coding scheme and spatial stream number combination data includes:

[0034] Acquire received signal strength indication data and preset threshold information between at least two access points and the site;

[0035] Based on the dynamic relationship between the received signal strength indication data and the preset threshold information, the synchronous / asynchronous communication status between each access point is determined.

[0036] Based on the aforementioned synchronous / asynchronous communication status, signals sent from adjacent access points to the station are classified as interference signals, and signals sent from adjacent stations are classified as environmental noise.

[0037] The signal-to-noise ratio of the station is calculated based on the signal classification results;

[0038] The signal-to-noise ratio, synchronous / asynchronous communication status, and access point transmit power are input into the gradient boosting decision tree model;

[0039] The gradient boosting decision tree model outputs the prediction results of the modulation and coding scheme and spatial stream number combination data of the target access point.

[0040] Furthermore, methods for determining the synchronous / asynchronous communication status between access points include:

[0041] For the downlink data signal of each access point, extract the maximum value and average value of its received signal strength indicator data;

[0042] When the maximum value of the received signal strength indication data of the downlink data signal of the first access point and the downlink data signal of the second access point does not reach the packet detection threshold and the average value of the received signal strength indication data does not reach the network allocation vector threshold, it is determined that the two access points are in a synchronous communication state.

[0043] When only one of the downlink data signals from the first access point and the second access point reaches the maximum value of the received signal strength indicator data or the average value of the received signal strength indicator data reaches the network allocation vector threshold, it is determined that the two access points are in an asynchronous communication state.

[0044] When the maximum value of the received signal strength indication data of the downlink data signal of the first access point and the downlink data signal of the second access point both reach the packet detection threshold or the average value of the received signal strength indication data both reach the network allocation vector threshold, it is determined that the two access points are in a mixed communication state.

[0045] Furthermore, the signal classification rules are as follows:

[0046] Interference signals are defined as signals originating from non-associated access points;

[0047] Ambient noise floor is defined as the signal originating from a non-target site.

[0048] Furthermore, the signal-to-noise ratio of the station is calculated:

[0049] When APs are in a synchronized state, the signal-to-noise ratio calculation ignores the ambient noise level.

[0050] When APs are in an asynchronous state, the signal-to-noise ratio calculation ignores the strength of interference signals;

[0051] When APs are in a mixed synchronous and asynchronous state, the linear power values ​​of the aggregated environmental noise floor and interference signals are calculated.

[0052] Furthermore, methods for aggregating the linear power values ​​of ambient noise floor and interference signals include:

[0053] Convert all signal strength values ​​to be aggregated from dBm power units to mW units;

[0054] Sum the converted mW unit values;

[0055] Convert the summation result back to dBm units.

[0056] To achieve the above objectives, a second aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the wireless local area network throughput prediction method.

[0057] The beneficial effects of this invention are:

[0058] Compared with existing technologies, the present invention provides a wireless local area network throughput prediction method and storage medium that effectively solves the systematic prediction bias problem caused by the neglect of transient fluctuations in the wireless channel in traditional physical models by constructing a dynamically corrected physical model. Specifically, it introduces slope correction coefficients and intercept correction terms on the basis of traditional physical models, and dynamically compensates for channel efficiency overestimation (such as transient interference loss) and systematic bias (such as protocol overhead) through linear regression; at the same time, it trains and corrects parameters differently for different AP density scenarios, and integrates measured physical layer dynamic parameters (modulation coding scheme and spatial stream number combination data, packet error rate data) and access point effective transmission time series data, thereby improving adaptability while retaining physical interpretability. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0060] Figure 1 This is a flowchart of a wireless local area network throughput prediction method disclosed in an embodiment of the present invention.

[0061] Figure 2 This is a comparison chart of the predicted and actual values ​​of a throughput physical model disclosed in an embodiment of the present invention.

[0062] Figure 3 This is a comparison chart of the predicted and actual values ​​of a throughput physical model disclosed in an embodiment of the present invention.

[0063] Figure 4 This invention discloses a model residual plot and QQ plot for cases with 2 APs and 3 APs.

[0064] Figure 5 This is a cumulative distribution function distribution characteristic diagram of throughput disclosed in an embodiment of the present invention.

[0065] Figure 6 This is a fully directed graph showing the interaction between nodes in a WLAN network, as disclosed in an embodiment of the present invention.

[0066] Figure 7 This is a schematic diagram illustrating how the root mean square error changes with the number of training rounds, as disclosed in an embodiment of the present invention.

[0067] Figure 8 This is a diagram of a 2AP sending downlink data to an associated STA, as disclosed in an embodiment of the present invention.

[0068] Figure 9 This is a flowchart of a synchronous / asynchronous state discrimination method for 2AP disclosed in an embodiment of the present invention.

[0069] Figure 10 This is a classification diagram of 3AP interference signals and environmental noise floor disclosed in an embodiment of the present invention.

[0070] Figure 11 This is a flowchart of a synchronous / asynchronous state discrimination process for 3AP disclosed in an embodiment of the present invention.

[0071] Figure 12 This is a synchronous / asynchronous state diagram between two APs in an embodiment of the present invention.

[0072] Figure 13 This is a classification diagram of interference signals and environmental noise with two access points (APs) disclosed in an embodiment of the present invention.

[0073] Figure 14 This is a synchronous / asynchronous state diagram between any two APs in a set of three APs disclosed in an embodiment of the present invention.

[0074] Figure 15 This is a diagram illustrating a calculation formula for modifying the signal-to-noise ratio when there are two access points (APs) as disclosed in an embodiment of the present invention.

[0075] Figure 16 This is a schematic diagram of gradient boosting tree training disclosed in an embodiment of the present invention.

[0076] Figure 17 This is a prediction accuracy analysis chart for an AP quantity of 2 disclosed in an embodiment of the present invention.

[0077] Figure 18 This is an accuracy prediction chart of MCS and NSS with 2 APs disclosed in an embodiment of the present invention.

[0078] Figure 19 This is an importance ranking chart for APs with a quantity of 2, as disclosed in an embodiment of the present invention.

[0079] Figure 20This is a prediction accuracy map with 3 APs disclosed in an embodiment of the present invention.

[0080] Figure 21 This is an accuracy prediction chart of MCS and NSS with 3 APs disclosed in an embodiment of the present invention.

[0081] Figure 22 This is an importance ranking chart for APs with a quantity of 3, as disclosed in an embodiment of the present invention. Detailed Implementation

[0082] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0083] According to embodiments of the present invention, it should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the following manufacturing method, in some cases the steps shown or described may be performed in a different order than that shown here.

[0084] like Figure 1 As shown, the present invention provides a method for predicting the throughput of a wireless local area network, comprising the following steps:

[0085] Step S100: Obtain network topology data, signal strength indication data, channel access mechanism data, modulation and coding scheme and spatial stream number combination data, data packet error rate data, total test duration data, and predicted access point effective transmission time series data;

[0086] This step involves seven types of data, including measured data, configuration data, and predicted data. These data collectively constitute the input feature set of the throughput prediction model. The definition, source, importance, and role in the model of each data item will be explained in detail below.

[0087] Network topology data refers to the physical location distribution and connectivity of access points (APs) and stations in a wireless local area network (WLAN). The network topology determines the physical deployment location of APs and directly affects signal propagation paths, coverage areas, and interference patterns. For example, in a high-density AP deployment environment, topology data can identify the distances and overlapping areas between APs, thus helping to quantify the degree of signal conflict and interference. This data, such as AP location identifiers and station association information, is acquired through WLAN management systems or testing tools and serves as a fundamental input for predicting throughput, used to analyze the macroscopic impact of network structure.

[0088] Signal Strength Indication (RSSI) data measures the wireless signal strength between all nodes (e.g., AP-AP or AP-site). RSSI reflects the quality of the wireless link and has a direct and significant impact on resource allocation and interference assessment. While measured RSSI data represents large-scale channel information, it cannot fully capture instantaneous fluctuations. In model inputs, RSSI is used as a key feature to help assess signal attenuation and link stability. This data is collected in real-time via wireless sniffers or AP logs.

[0089] Channel access mechanism data involves protocol rules and parameters of the Media Access Control (MAC) layer in WLANs, such as the Network Allocation Vector (NAV) threshold or the configuration of Carrier Sense Multiple Access / Collision Avoidance (CSMA / CA) mechanisms. Channel access mechanisms are a crucial factor affecting throughput because they determine the timing of data frame transmission and collision handling strategies. For example, the NAV threshold (nav) is encoded as a binary feature used to determine channel occupancy status. This data, extracted from AP configuration files, helps models assess the impact of channel contention and access delay on throughput.

[0090] Modulation-coding scheme and spatial stream number combination data (MCS and NSS data) define the transmission parameters of the physical layer, including modulation type, coding rate, and number of spatial streams, which together determine the theoretical maximum transmission rate (PHY Rate). MCS and NSS are key factors in the physical layer, directly affecting transmission efficiency; in the physical model, the PHY Rate is derived from MCS and NSS. These data are obtained in actual measurements through the output of a prediction model (the specific method of obtaining them will be discussed later).

[0091] Error Rate (PER) data represents the probability of errors occurring during data transmission, calculated as the proportion of erroneous packets to the total transmitted packets. (1-PER) is defined as the probability of successful transmission and is a core variable in the physical model, directly affecting the throughput calculation results. PER data is obtained through actual measurements using WLAN monitoring tools, reflecting the impact of channel noise, interference, and bit error rate. In the model, PER helps adjust the theoretical rate to match the actual environment.

[0092] Test Duration data refers to the total time of the entire WLAN performance test process, recorded in seconds. It is used as the denominator in the throughput physical model to convert the effective transmission time (Seq_time) and PHY Rate into average throughput. This data is obtained through test logs or timers to ensure that throughput predictions are based on a standardized time base, avoiding errors caused by inconsistent time scales.

[0093] The predicted effective transmission time series data (Seq_time) of the access point refers to the effective time segment sequence used by the AP for data transmission, representing the actual duration the AP occupies the channel during the test. Seq_time is a key input to the physical model, but measured data is often missing, so it needs to be obtained in advance through a predictive model (such as the gradient boosting decision tree algorithm) (specific methods will be discussed later). Seq_time prediction is based on features such as network topology, mean RSSI, and protocol type. Obtaining this data ensures that the model can accurately reflect the AP's transmission opportunities.

[0094] Step S200: Substitute the obtained modulation and coding scheme, spatial stream number combination data, packet error rate data, total test duration data, and access point effective transmission time series data into the initial physical model to calculate the initial throughput prediction value.

[0095] The acquired key data is substituted into the initial physical model to calculate the initial throughput prediction for each access point (AP). This step directly models the transmission characteristics of the physical layer, but does not yet consider dynamic interference and system deviations in the actual environment. The calculation formula is as follows:

[0096]

[0097] Wherein: T AP The initial throughput prediction value is represented by PHY_Rate, which is determined by the (MCS, NSS) selected by the AP and is the theoretical maximum transmission rate of the WLAN device at the physical layer. PER represents the probability of a data packet error during transmission, and (1-PER) represents the probability of successful transmission. Seq_time represents the effective time segment used by the AP for data transmission. TestDuration refers to the total duration of the entire test process, in seconds.

[0098] Since the Seq_time is missing in the test set, the prediction of the AP's effective transmission time is used as the input to the above formula. Simultaneously, by substituting (MCS, NSS), packet loss rate PER, and test data TestDuration from the test dataset into the formula, the throughput of each AP can be calculated.

[0099] from Figure 2 It can be seen that the throughput predicted by the physical model is larger than the actual throughput. The mean squared error (MSE) of the throughput physical model is 169.62. Therefore, the prediction effect of directly using the throughput physical model is not accurate enough, and it is even slightly worse than the prediction effect of the gradient boosting decision tree algorithm (MSE = 137.51).

[0100] Step S300: Using a linear regression algorithm, fit the calculated initial throughput prediction value with the measured throughput to obtain the slope correction coefficient and intercept correction term.

[0101] Because wireless channels are highly variable and real-world WLAN networks have many uncertainties, the throughput predicted by the physical formula deviates from the throughput measured in the actual environment. Therefore, a regression equation is established:

[0102]

[0103] Where α is the linear correction coefficient and β is the intercept coefficient.

[0104] The optimal α and β were obtained by fitting the data using a linear fitting method (e.g., when the number of APs is 2, α = 0.9612, β = -3.7984; when the number of APs is 3, α = 0.9300, β = -6.3337).

[0105] Step S400: Apply the obtained slope correction coefficient and intercept correction term to the initial physical model to construct the corrected physical model;

[0106] The revised throughput calculation formula is as follows:

[0107]

[0108] Step S500: Use cross-validation to train and validate the constructed modified physical model;

[0109] The training dataset is grouped according to the number of access points (APs) (e.g., two classes with 2 APs and three APs). Each group is randomly divided into five mutually exclusive subsets. In each iteration, one subset is selected as the validation set, and the remaining four subsets are used as the training set. The predicted throughput is calculated by substituting the subsets into the modified physical model formula, and the mean squared error (MSE) is calculated by comparing it with the measured values. After five iterations, the average MSE of all validation sets is summarized (e.g., the modified model MSE = 75.43 when the number of APs is 2). At the same time, the residual distribution and error accumulation function are analyzed to ensure that the model satisfies homoscedasticity and that the residuals are close to a normal distribution.

[0110] Step S600: Based on the corrected physical model after training and verification, input the acquired network topology data, signal strength indication data, and channel access mechanism data to predict the overall network throughput.

[0111] By substituting network topology data, signal strength indication data, and channel access mechanism data into the modified physical model formula, the throughput of each AP is calculated and then aggregated to obtain the total network throughput.

[0112] like Figure 3 As shown, for the case with 2 APs, the corrected MSE value is 75.43, a decrease of approximately 55.5% compared to the uncorrected MSE value of 169.62; for the case with 3 APs, the corrected MSE value is 68.62, a decrease of approximately 72.33% compared to the uncorrected MSE value of 248.01. Figure 3 As can be seen, the adjusted model's prediction performance is better than that of the unadjusted model. This demonstrates that adjusting the physical prediction model using linear regression is extremely effective, and the optimized physical model's prediction results are far superior to those of the gradient boosting decision tree algorithm.

[0113] To verify the model's fit, a residual plot can be drawn for analysis. The horizontal axis represents the corrected model's predicted values, and the vertical axis represents the residuals (actual throughput minus predicted throughput). Figure 4 (a) and (c) show the residual plots for cases with 2 APs and 3 APs, respectively.

[0114] Simultaneously, a QQ plot was used to verify whether the residuals conformed to a normal distribution. The horizontal axis represents the quantiles of the standard normal distribution, and the vertical axis represents the quantiles of the residuals. Figure 4 (b) and (d) show the QQ graphs for cases with 2 and 3 APs, respectively. Figure 4 The residual plots in (a) and (c) show that the residuals are randomly distributed without a clear trend or pattern, indicating that the corrected physical model fits well and the residuals satisfy the homoscedasticity assumption. Figure 4 As can be seen from the QQ plots in (b) and (d), the data points are roughly distributed along the 45-degree line, indicating that the residuals are close to a normal distribution, which is consistent with the normality assumption.

[0115] To gain a more comprehensive understanding of the data distribution characteristics of system throughput and to evaluate model accuracy, the cumulative error distribution function (CDF) is introduced for analysis. The CDF of error shows the probability that the error is less than or equal to a certain value, which helps to intuitively observe the trend and characteristics of the data distribution, such as... Figure 5 As shown.

[0116] Figure 5 (a) and (c) show the CDF curves of error for 2 APs and 3 APs. Figure 5(b) and (d) show the QQ plots for 2 APs and 3 APs. The plots show that the CDF of the error gradually increases with increasing throughput, indicating the cumulative probability of the error within this range. Furthermore, the QQ plots illustrate that the error approximates a normal distribution, consistent with the normality assumption.

[0117] In step S100 above, the method for predicting the effective transmission time series data of the access point includes:

[0118] Step S101: Obtain the network topology data of the target wireless local area network and extract the location identification information of the access point and the station from it;

[0119] Network topology data is collected in real time through the target wireless LAN's network management interface (such as SNMP protocol, NetConf configuration interface, or dedicated probe device). This data includes three main elements: node type and quantity, connection relationships, and location attributes. First, the physical number of access points (APs) and stations (STAs) and their role identification are determined. For example, in a scenario with 2 APs, a K4 topology (4 nodes in total) is formed, consisting of 2 APs and 2 STAs; in a scenario with 3 APs, a K6 topology (6 nodes in total) is formed, consisting of 3 APs and 3 STAs. Second, the bidirectional communication links between all nodes are recorded to form a complete directed graph structure (K). n ),like Figure 6 Each node shown has a signal interaction path with other nodes, and the link direction represents the signal transmission direction (such as AP→STA uplink, STA→STA interference link); finally, the relative position identification information between nodes is extracted (denoted as loc_id).

[0120] In the process of extracting location identification information, firstly, a unique identifier is assigned to each AP and an independent identifier is assigned to each STA. Based on the physical deployment coordinates of the nodes, a relative position matrix is ​​calculated to generate the topological location attribute loc_id. For example, in a K6 topology, loc_id can be mapped to a 6×6 matrix whose element values ​​represent the distance level between two nodes. Secondly, the network topology is abstracted into a weighted directed graph K. n = (V, E, ω), where the vertex set V is all AP / STA nodes, the edge set E contains all directed edges, and the weight ω is used to carry the RSSI value later; the loc_id extracted in this step is used as a mathematical representation of the network topology and directly input into the prediction model to characterize the impact of spatial layout on signal interference (such as interference enhancement due to positional overlap in high-density areas). This feature, together with service traffic, RSSI, and threshold, constitutes the input vector, effectively solving the problem of nonlinear prediction of transmission duration caused by dynamic changes in topology in multi-AP environments.

[0121] Step S102: Collect the service traffic data of the wireless local area network and encode the transport layer protocol type into discrete feature values;

[0122] By deploying traffic mirroring ports on the core switches of the target wireless LAN or monitoring interfaces on the wireless controllers, raw service traffic data packets from the network layer to the transport layer are collected in real time. Deep packet inspection technology is used to identify the transport layer protocol type: when UDP is detected, its protocol feature value is encoded as a discrete value of 0 based on its connectionless and low-latency characteristics; when TCP is detected, its feature value is encoded as a discrete value of 1 based on its connection-oriented and flow control mechanism. This encoding operation is completed in the data preprocessing stage, ultimately forming a discrete feature sequence with a length equal to the number of traffic samples. The necessity of this technology lies in the fact that the transport layer protocol type significantly affects the effective transmission time through retransmission mechanisms and congestion control strategies, and discrete encoding can quantify protocol differences into numerical features, enabling the gradient boosting decision tree model to learn the nonlinear relationship between protocol type and throughput.

[0123] Step S103: Measure the received signal strength indication values ​​between all nodes in the network topology data, and calculate the average received signal strength indication characteristics of all communication link pairs;

[0124] The Received Signal Strength Indication (RSSI) values ​​between all nodes in the network topology are periodically measured using spectrum analyzers deployed at each node of the wireless LAN. Specifically, three types of links are included: interference signal strength between APs, service link strength between STAs and APs, and co-channel interference strength between STAs. For each type of link, three statistical values ​​are collected according to the requirements of Table 1 (2 AP scenario) and Table 2 (3 AP scenario): sum (total), max (peak value), and mean (mean). Based on the raw RSSI data, the mean feature is calculated for each pair of communication nodes (such as AP0→STA1) bidirectional links (AP0→STA1 and STA1→AP0), generating the RSSI mean feature vector of the communication link pair.

[0125] Table 1: Inter-node RSSI information when there are 2 APs

[0126]

[0127] Table 2: Inter-node RSSI information when there are 3 APs

[0128]

[0129] Step S104: Read the network allocation vector threshold value configured in the wireless local area network and encode it as a binary feature;

[0130] Thresholds are divided into Packet Detection (PD), Energy Detection (ED), and NAV thresholds. The PD and ED thresholds filter out some interference signals by detecting the maximum RSSI value. In the analysis of 5 training sets for 2 APs and 8 training sets for 3 APs, the ED and PD thresholds are -62dBm and -82dBm, respectively. Since they remain unchanged, the numerical impact of PD and ED on the APs is not considered, but their actual influence on the APs cannot be ignored. The NAV12 threshold varies, being -82dBm and -62dBm respectively. Therefore, for threshold information, only the impact of changes in the NAV threshold is considered.

[0131] Similarly, 0-1 variables are introduced to model the changes in the NAV threshold:

[0132]

[0133] Step S105: Combine the extracted location identification information, the encoded discrete feature values, the calculated mean value of the received signal strength indication, and the encoded binarized features to form an input feature vector;

[0134] The final input feature vector is generated from the topological location identification information extracted in step S101, the discrete coding features of the transport layer protocol generated in step S102 (sequence of UDP=0, TCP=1), the mean RSSI features of all communication link pairs calculated in step S103, and the NAV threshold binary features (0 or 1) encoded in step S104.

[0135] Step S106: Set the initial hyperparameter search range for the gradient boosting decision tree regression model;

[0136] The hyperparameters of the gradient boosting decision tree model include: a subsampling rate of 0.8, an L1 regularization weight of 0.01, an L2 regularization weight of 0.1, a number of decision trees of 100, a minimum leaf node weight of 3, a maximum tree depth of 5, a learning rate of 0.05, and a feature sampling ratio of 0.7.

[0137] Step S107: Use the grid search algorithm to traverse all combinations within the set initial hyperparameter search range, and calculate the root mean square error corresponding to each combination through five-fold cross-validation. Select the hyperparameter combination with the smallest root mean square error as the optimal hyperparameter combination.

[0138] To improve the accuracy of model predictions, the model parameters are adjusted and optimized during training, using mean squared error (MSE), root mean square error (RMSE), and R0. 2 Indicators are used to evaluate the results of parameter adjustments and optimizations.

[0139] The model's predictive performance was comprehensively evaluated using multiple metrics, including mean squared error (MSE) and root mean squared error (RMSE). MSE and RMSE reflect the average error between the model's predictions and actual values, while the coefficient of determination measures the model's ability to explain data variation. These evaluation metrics comprehensively measure the model's fit and prediction accuracy, ensuring consistent and robust performance on both the training and validation sets, further validating the model's effectiveness and reliability.

[0140] Five-fold cross-validation is a commonly used machine learning model evaluation technique to estimate a model's performance and generalization ability. It involves randomly and evenly dividing the original dataset into five equal subsets, sequentially selecting one subset as the test set, and using the remaining four subsets as the training set. This process is repeated multiple times, each time selecting a different subset as the test set, ensuring that each subset has at least one opportunity to be used as the test set.

[0141] Step S108: Configure the gradient boosting decision tree regression model using the obtained optimal hyperparameter combination, and train the model using the training dataset containing the input feature vector and the corresponding access point effective transmission time series data.

[0142] AP transmission time is affected by network topology K. n The business traffic P, the RSSI between nodes (denoted by R), and the threshold D all have a combined influence, and their interrelationship can be represented by the following mathematical model:

[0143] Seq time =F(K n (P,R,D)(Formula 6)

[0144] Among them, network topology K n The relative position information loci of the network topology is determined, the service traffic P is determined by the protocol for sending information between AP nodes, the RSSI between nodes is determined by the data in Table 1 and Table 2, and the threshold D comes from the influence of changes in the NAV threshold.

[0145] After multiple rounds of training and parameter tuning, the root mean square error changes with the number of training rounds as follows: Figure 7 As shown, the horizontal axis represents the number of gradient training iterations, and the vertical axis represents the change in RMSE; from Figure 7 As can be observed, the root mean square error gradually decreases with increasing training epochs, indicating that the model's predictive performance is continuously improving. This trend suggests that with more training, the model is able to better fit the training data, thereby improving its performance on the validation set.

[0146] Step S109: Input the input feature vector constructed by the scene to be predicted into the trained gradient boosting decision tree regression model, and output the effective transmission time series data of the access point.

[0147] As can be seen from equations (3) and (4), throughput prediction relies on the effective transmission of time-series data (Seq_time) at the access point. Seq_time cannot be directly obtained from the test set, nor can it be replaced by other features. Therefore, this parameter must be provided by the gradient boosting decision tree regression model (XGBoost) trained using the above scheme.

[0148] In throughput prediction, obtaining the modulation and coding scheme (MCS) and spatial stream number (NSS) in advance is crucial because they directly determine the theoretical physical layer transmission rate (PHYRate), which is a core input parameter of the throughput physical model. Without the MCS / NSS, relying solely on traditional features such as RSSI to predict the PHYRate will lead to error accumulation. The prediction method using a combination of modulation and coding scheme and spatial stream number data will be described in detail below. This prediction method includes:

[0149] Step S201: Obtain received signal strength indication data and preset threshold information between at least two access points and the site;

[0150] This step involves acquiring physical layer data in a high-density WLAN environment through actual measurements. Specifically, the data collected includes Received Signal Strength Indication (RSSI) data and preset threshold information between at least two Access Points (APs) and their associated Stations (STAs) within the deployment area. The RSSI data includes measured values ​​of signal strength (sum, max, and mean) recorded during AP-STA communication. The preset threshold information includes dynamically configured parameters such as Packet Detection Threshold (PD), Energy Detection Threshold (ED), and Network Allocation Vector Threshold (NAV).

[0151] Step S202: Determine the synchronous / asynchronous communication status between each access point based on the dynamic relationship between the received signal strength indication data and the preset threshold information;

[0152] Each AP sends downlink data only to its associated specific STA (e.g., AP0→STA0, AP1→STA1), such as Figure 8 As shown, in a 2AP scenario, four directed communication links (including uplink and downlink) are formed, corresponding to four sets of RSSI data:

[0153] Downlink data sent from AP0 to STA0;

[0154] Uplink data sent from STA0 to AP0;

[0155] Downlink data sent from AP1 to STA1;

[0156] Uplink data sent from STA1 to AP1.

[0157] like Figure 9 As shown, in a wireless communication system containing two access points (APs), the core criterion for determining their communication status (synchronous or asynchronous) is to analyze the downlink data sent by each AP to its associated station (STA) (such as...). )and( The measured value of Received Signal Strength Indication (RSSI).

[0158] The experimentally measured signal strength includes three key indicators: sum, max, and mean. When determining the communication status between two access points (AP0 and AP1), these are based on their downlink data. )and The determination is made based on whether the RSSI value meets the threshold condition:

[0159] like and If the measured signal strengths do not meet the conditions (i.e., the max value is not in the interval [PD,ED] and the mean value is not in the interval [NAV,ED]), then it is determined to be a synchronous transmission state.

[0160] If there is only one downlink signal or If the condition is met (i.e., the max value is in [PD, ED] or the mean value is in [NAV, ED]), then it is determined to be an asynchronous transmission state;

[0161] If both downlink signals meet the conditions (i.e., both max values ​​are located in [PD, ED] or both mean values ​​are located in [NAV, ED]), then it is determined to be a synchronous-asynchronous mixed state.

[0162] By dynamically comparing the peak intensity (max) and overall level (mean) of the downlink signal with respect to preset thresholds (PD, ED, NAV), the communication status classification between the two APs is finally output.

[0163] In 3AP scenarios ( Figure 10 This was expanded into 6 directed links (bidirectional interaction between AP0-STA0, AP1-STA1, and AP2-STA2), using... Figure 10 The six directed edges (solid lines) represent the number of access points (APs). For the case with three access points (APs), two APs are selected from the three, and the synchronous / asynchronous mixed state is determined using the analysis method for two APs, as detailed below. Figure 11 As shown, for a communication scenario involving three access points, the methods for determining the synchronous / asynchronous communication status between the access points include:

[0164] First, label the three access points as the first access point, the second access point, and the third access point;

[0165] Two access points are selected sequentially to form access point pairs, including: the first access point and the second access point, the first access point and the third access point, and the second access point and the third access point. For each access point pair, the following discrimination operation is performed:

[0166] Synchronous transmission status determination: When the API's downlink data signal Downlink data signal with APJ When the received signal strength indication values ​​do not meet any of the following conditions:

[0167] 1. The maximum signal strength is located within the interval formed by the packet detection threshold and the energy detection threshold;

[0168] 2. If the average signal strength is within the range formed by the network allocation vector threshold and the energy detection threshold, then API and APJ are determined to be in synchronous transmission state.

[0169] Asynchronous transmission status determination: When only one of the downlink data signals of API and APJ satisfies any of the following conditions:

[0170] 1. The maximum signal strength meets the packet detection threshold requirement;

[0171] 2. The average signal strength meets the network allocation vector threshold requirement;

[0172] 3. Then it is determined that the API and APJ are in an asynchronous transmission state.

[0173] Hybrid transmission status determination: When the received signal strength indication values ​​of both the API downlink data signal and the APJ downlink data signal meet any of the following conditions:

[0174] 1. The maximum signal strength meets the packet detection threshold requirement;

[0175] 2. The average signal strength meets the network allocation vector threshold requirement;

[0176] 3. Then it is determined that the API and APJ are in a state of synchronous and asynchronous mixed transmission.

[0177] Finally, by combining the communication status categories of all access point pairs, an overall network communication status determination result is generated, including the three access points, such as... Figure 12 As shown.

[0178] Step S203: Based on the synchronous / asynchronous communication status, classify the signals sent from adjacent access points to the station as interference signals, and classify the signals sent from adjacent stations as environmental noise.

[0179] When classifying the scenarios with two access points (APs), we first analyze the characteristics of synchronous transmission, asynchronous transmission, and mixed synchronous / asynchronous transmission. In synchronous transmission, APs listen to each other in real time, and most of the interference when a STA receives data comes from ambient noise. In asynchronous transmission, APs do not listen to each other at all, and the STA ignores ambient noise, only receiving interference signals. In mixed synchronous / asynchronous transmission, APs alternate between real-time and non-listening to each other, and the STA receives both ambient noise and interference signals. Based on data analysis and communication principles, for a given STA node, signals from neighboring AP nodes are interference signals, and signals from other neighboring STA sites are ambient noise, such as... Figure 13 As shown.

[0180] For the case with 3 access points (APs), the classification of interference signals and ambient noise is performed using the same method as for the case with 2 APs, thus obtaining the classification of interference signals and ambient noise. (See...) Figure 14 As shown.

[0181] Step S204: Calculate the signal-to-noise ratio of the station based on the signal classification results;

[0182] For the case where there are two access points (APs), based on the classification of interference signals and ambient noise, the signal-to-noise ratio (SINR) calculation model of the station (STA) is dynamically adjusted according to the synchronous / asynchronous communication status between access points. The SINR calculation method is as follows:

[0183] SINR = P signal -(P noise +P inter )(Formula 7)

[0184] Among them, P signal Indicates the useful signal (corresponding to) Figure 13 (The black solid line in the middle), P noise Indicates ambient noise level (corresponding to) Figure 13 (green dashed line in the middle), P inter Indicates interference signal (corresponding to) Figure 13 (The red dashed line in the diagram). Based on the three scenarios of synchronous / asynchronous states between the two APs, the signal-to-noise ratio calculation formula is modified as follows: Figure 15 .

[0185] Synchronous Communication Status: When two access points (AP0 and AP1) are transmitting synchronously, the signal-to-noise ratio (SNR) calculation for a station (STA) only considers the ambient noise floor power (P). noise The formula is: SINR = P Signal -Pnoise .

[0186] Asynchronous communication state: When two access points are in asynchronous transmission, the signal-to-noise ratio calculation only considers the interference signal power (P). inter The formula is: SINR = P Signal -P inter .

[0187] Hybrid Communication State: When two access points are in a hybrid synchronous-asynchronous transmission state, the linear power values ​​(in mW) of the ambient noise floor and interference signals need to be aggregated simultaneously. The formula is: SINR = P Signal -P noise -P inter .

[0188] A network topology with 3 access points (APs) is essentially a directed complete graph consisting of 6 nodes and 30 directed edges. Therefore, the difference in SINR calculation compared to a network with 2 APs lies in the aggregation process of multiple signals.

[0189] Taking a scenario with 3 APs (Analog and Access Points) interacting in a hybrid synchronous and asynchronous manner as an example, when summing the signals in the following formula, it is necessary to first separate each sub-signal (P, expressed in dBm) into its corresponding sub-signal. inter or P noise The power is converted to linear power in mW, then the linear power of all interfering signals is summed, and finally converted back to dBm. Therefore, the formula for calculating the cumulative strength of n signals is:

[0190]

[0191] Among them, P total_n This represents the cumulative intensity of n signals, P1...P... n This represents the signal strength of each of the n signals. Figure 14 Taking station STA0 as an example, the signal-to-noise ratio calculation formula for this station at a certain moment is:

[0192]

[0193] in, Indicates the effective signal strength transmitted from AP0 to STA0 (solid black line), P env1 The ambient noise level (red dashed line) from the STA1 site is represented by P. inter1 This indicates the interference signal from AP1 (green dashed line), P inter2 The green dashed line indicates interference signals from AP2.

[0194] Step S205: Input the signal-to-noise ratio, synchronous / asynchronous communication status, and access point transmit power into the gradient boosting decision tree model;

[0195] Based on prediction models with 2 and 3 access points (APs), the calculated signal-to-noise ratio (SNR) of the target area (STA) is combined with some basic test information as input to obtain the output (MSC, NSS). Figure 16 The gradient boosting decision tree model training process is shown. The model input layer consists of three types of features: site signal-to-noise ratio features (env_SINR: signal-to-noise ratio calculated based on ambient noise floor power, ferr_SINR: signal-to-noise ratio calculated based on interference signal power), access point status features (categ_0, categ_1, categ_2: three-dimensional classification variables representing the synchronous / asynchronous communication status between APs), and test environment basic features (RSSI: measured value of received signal strength indication, eirp: access point transmit power, nav: network allocation vector threshold value, loc_id: test location identifier, ap_id: access point device identifier).

[0196] Step S206: Output the modulation and coding scheme and spatial stream number prediction results of the target access point through the gradient boosting decision tree model.

[0197] The model output layer contains the target prediction term:

[0198] Modulation and coding scheme (MCS): The most commonly used modulation and coding scheme after the predictive AP adaptive adjustment stabilizes;

[0199] Space Stream Count (NSS): The number of space streams most frequently used after the AP adaptive adjustment stabilizes.

[0200] according to Figure 16 The gradient boosting decision tree algorithm model was trained to predict (MSC, NSS) combinations with 2 environments and 3 environments, respectively.

[0201] For five training datasets with two APs each, 5-fold cross-validation was used to fully train the model and also to verify its effectiveness.

[0202] exist Figure 17 In the diagram, the horizontal axis represents the actual NSS or MOS values, and the vertical axis represents the model's predicted values. When a node is located on the diagonal line passing through the origin, it indicates that the prediction is accurate. The color intensity of the node indicates the number of training data points at that point; darker colors on the diagonal indicate higher prediction accuracy. Therefore, it is intuitive to see that the model has high accuracy.

[0203] Secondly, regarding the prediction accuracy of the statistical model, out of 390 training data points, 336 data points accurately predicted the MCS, achieving an accuracy rate of 86.15%. Figure 18(a)), MSE is 0.6518; 376 NSSs were accurately predicted, with an accuracy of 96.41%. Figure 18 (b) The MSE is 0.5128. Furthermore, the trained gradient boosting decision tree algorithm model can provide a ranking of the importance of influences (MOSNSS).

[0204] Finally, through Figure 19 Based on the ranking of importance of (a) and (b), the following conclusions can be drawn:

[0205] The signal-to-noise ratio (SNR) of the STA has the greatest impact on (MOS, NSS). Reason: The SINR calculated from the ambient noise floor ("env_SINR") and the SINR calculated from the interference signal ("ferr SINR") rank first and third respectively, indicating that the SNR of the STA has the greatest impact on (MCS, NSS).

[0206] In the composition of signal-to-noise ratio (SNR), noise from other STAs is more important than noise from other APs. Reason: In the importance ranking of factors affecting MCS, ambient noise floor increases the impact of interference noise by approximately 50%; while in the importance ranking of factors affecting NSS, ambient noise floor increases the impact of interference noise by approximately 160%. AP transmit power has a relatively significant impact on (MCS, NSS). Reason: In the importance ranking of factors affecting both MCS and NSS, AP transmit power "eirp" ranks second. Threshold information, and the synchronous / asynchronous state information determined by the threshold, have a relatively small impact on (MCS, NSS). Reason: The influence of changes in the NAV threshold and synchronous / asynchronous state information (represented by category_0, category_1, category_2) is ranked low.

[0207] By implementing a 5-fold cross-validation strategy on eight training_sets (each with three APs) of the loc_nav.csv dataset, 858 training data points can be generated. This method not only ensures that the model is adequately trained and covers the diversity of the data, but also rigorously evaluates the model's performance on different subsets through cross-validation, thereby verifying the model's effectiveness and generalization ability.

[0208] from Figure 20 The distribution of the displayed data points cleverly uses varying shades of color to reflect their density and importance, while their positions directly correspond to the relationship between the model's predicted (MCS, NSS) values ​​and the actual measured values. Most data points are closely clustered around the actual measured values ​​and are evenly distributed, indicating consistency between the model's predictions and the measured data. Therefore, it can be concluded that the model demonstrates high accuracy in predicting (MCS, NSS).

[0209] Secondly, regarding the prediction accuracy of the statistical model, out of 858 training data points, 830 correctly predicted NSS, achieving an accuracy rate of 96.74%. Figure 21 (b) The MSE is 0.04315; 559 data points accurately predict the MCS, with an accuracy of 65.16%. Figure 21 (a) The MSE is 1.8949. Although the prediction accuracy is not high when the number of APs is 3, the low MSE indicates that although there is a small bias in the prediction, the prediction result is still relatively accurate. Figure 21 As shown. Furthermore, the trained gradient boosting decision tree algorithm model can provide an importance ranking of the influence (MCS, NSS) in three environments with AP, such as... Figure 22 As shown in (a) and (b).

[0210] Based on the importance ranking of the three influences (MCS, NSS) for APs, the following conclusions can be drawn:

[0211] The signal-to-noise ratio (SNR) of the STA has the greatest impact on (MOS, NSS). Reason: The SINR calculated from the ambient noise floor, "env_SINR", and the SINR calculated from the interference signal, "ferr SINR", rank first and second respectively, indicating that the SNR of the STA has the greatest impact on (MCS, NSS).

[0212] In an environment with 3 access points (APs), the importance of interference noise increases compared to an environment with 2 APs. The reason is that, in the ranking of the importance of MCS and NSS, the importance values ​​of ambient noise and interference noise are basically the same, with ambient noise being slightly more important than interference noise.

[0213] The AP's transmit power has a significant impact on (MCS, NSS). Reason: In the ranking of factors affecting MOS and NSS, AP transmit power "eirp" ranks fourth and third, respectively. Threshold information, and the synchronous / asynchronous state information determined by the threshold, have a relatively small impact on (MCS, NSS). Reason: Changes in the NAV threshold and synchronous / asynchronous state information (represented by category_0, category_1, category_2) have a lower impact.

[0214] Using the gradient boosting decision tree algorithm for training, and with 5-fold cross-validation, the following results were obtained: When the number of APs is 2, the accuracy of NSS prediction is 96.4%, and the accuracy of MOS prediction is 86.2%; when the number of APs is 3, the accuracy of NSS prediction is 96.7%, and the accuracy of MOS prediction is 65.2%. Model validation shows that for the prediction of (MOS, NSS), the accuracy with 2 APs is higher than that with 3 APs. Although the accuracy of prediction with 3 APs is not high, the mean squared error (MSE) is only 1.89, which is relatively small. Most of the incorrectly predicted MCSs are not significantly different from the true MCSs, indicating that the prediction model is effective.

[0215] Understandably, the method of this invention addresses the throughput prediction problem in high-density WLAN environments by proposing a hierarchical modeling framework: First, an AP transmission duration prediction model is established based on XGBoost, taking network topology, service traffic (protocol type), threshold (NAV), and inter-node RSSI features as inputs. Through grid search parameter tuning and five-fold cross-validation, the conclusion that service traffic has the strongest impact is drawn. Then, SINR features are constructed (integrating synchronous / asynchronous state discrimination, interference signal classification, and environmental noise floor modeling), combined with XGBoost prediction of MCS / NSS combination. In the 2-AP scenario, the MCS / NSS accuracy reaches 86.2% / 96.4%, while in the 3-AP scenario, NSS remains at 96.7% but MCS drops to 65.2% (attributed to RSSI only reflecting large-scale channel changes), and SINR is verified as the most critical feature. Finally, the first two outputs are fused, and XGBoost is first used to predict throughput, followed by a physical model correction strategy: the base value is calculated based on theoretical formulas, and then a correction coefficient is introduced through linear regression, significantly reducing MSE.

[0216] According to another aspect of the embodiments of this application, an electronic device is also provided, including a processor and a memory, wherein the processor is configured to implement the steps of the method when executing a computer program stored in the memory.

[0217] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0218] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0219] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0220] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0221] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method of wireless local area network throughput prediction, characterized by, Includes the following steps: Acquire network topology data, signal strength indication data, channel access mechanism data, modulation and coding scheme and spatial stream number combination data, packet error rate data, total test duration data, and predicted access point effective transmission time series data; The acquired modulation and coding scheme, combined with spatial stream number data, packet error rate data, total test duration data, and access point effective transmission time series data are substituted into the initial physical model to calculate the initial throughput prediction value. Using a linear regression algorithm, the calculated initial throughput prediction value is fitted to the measured throughput to obtain the slope correction coefficient and intercept correction term; The obtained slope correction coefficients and intercept correction terms are applied to the initial physical model to construct the corrected physical model; Cross-validation is used to train and validate the constructed modified physics model; Based on the corrected physical model after training and validation, and inputting the acquired network topology data, signal strength indication data, and channel access mechanism data, the overall network throughput is predicted.

2. The wireless local area network throughput prediction method of claim 1, wherein, The method for predicting the effective transmission time series data of the access point includes: Acquire the network topology data of the target wireless LAN and extract the location identification information of access points and stations from it; Collect the service traffic data of the wireless local area network and encode the transport layer protocol type into discrete feature values; Measure the received signal strength indication values ​​between all nodes in the network topology data, and calculate the mean characteristics of the received signal strength indication values ​​for all communication link pairs; Read the network allocation vector threshold value configured in the wireless local area network and encode it as a binary feature; The extracted location identification information, the encoded discrete feature values, the calculated mean value of the received signal strength indication, and the encoded binarized features are combined to form an input feature vector; Set the initial hyperparameter search range for the gradient boosting decision tree regression model; The grid search algorithm is used to traverse all combinations within the set initial hyperparameter search range, and the root mean square error corresponding to each combination is calculated by five-fold cross-validation. The hyperparameter combination with the smallest root mean square error is selected as the optimal hyperparameter combination. The optimal combination of hyperparameters is used to configure the gradient boosting decision tree regression model, and the model is trained using a training dataset containing input feature vectors and corresponding access point effective transmission time series data. The input feature vector constructed from the scenario to be predicted is input into the gradient boosting decision tree regression model after training, and the output is the effective transmission time series data of the access point.

3. The wireless local area network throughput prediction method of claim 2, wherein, Methods for encoding transport layer protocol types as discrete feature values ​​include: When the protocol type is User Datagram Protocol, the value is assigned to 0; When the protocol type is Transmission Control Protocol, the value is assigned to 1.

4. The wireless local area network throughput prediction method of claim 2, wherein, The five-fold cross-validation is performed as follows: The training dataset is randomly divided into five mutually exclusive subsets; One subset is selected sequentially as the validation set, and the remaining four subsets are used as the training set. Train the model for each hyperparameter combination and calculate the root mean square error of the validation set; After five iterations, the average root mean square error is taken as the evaluation index.

5. The wireless local area network throughput prediction method of claim 1, wherein, The prediction method for the combination of modulation coding scheme and spatial stream number data includes: Acquire received signal strength indication data and preset threshold information between at least two access points and the site; Based on the dynamic relationship between the received signal strength indication data and the preset threshold information, the synchronous / asynchronous communication status between each access point is determined. Based on the aforementioned synchronous / asynchronous communication status, signals sent from adjacent access points to the station are classified as interference signals, and signals sent from adjacent stations are classified as environmental noise. The signal-to-noise ratio of the station is calculated based on the signal classification results; The signal-to-noise ratio, synchronous / asynchronous communication status, and access point transmit power are input into the gradient boosting decision tree model; The gradient boosting decision tree model outputs the prediction results of the modulation and coding scheme and spatial stream number combination data of the target access point.

6. The wireless local area network throughput prediction method of claim 5, wherein, Methods for determining the synchronous / asynchronous communication status between access points include: For the downlink data signal of each access point, extract the maximum value and average value of its received signal strength indicator data; When the maximum value of the received signal strength indication data of the downlink data signal of the first access point and the downlink data signal of the second access point does not reach the packet detection threshold and the average value of the received signal strength indication data does not reach the network allocation vector threshold, it is determined that the two access points are in a synchronous communication state. When only one of the downlink data signals from the first access point and the second access point reaches the maximum value of the received signal strength indicator data or the average value of the received signal strength indicator data reaches the network allocation vector threshold, it is determined that the two access points are in an asynchronous communication state. When the maximum value of the received signal strength indication data of the downlink data signal of the first access point and the downlink data signal of the second access point both reach the packet detection threshold or the average value of the received signal strength indication data both reach the network allocation vector threshold, it is determined that the two access points are in a mixed communication state.

7. The wireless local area network throughput prediction method of claim 5, wherein, The signal classification rules are as follows: Interference signals are defined as signals originating from non-associated access points; Ambient noise floor is defined as the signal originating from a non-target site.

8. The wireless local area network throughput prediction method of claim 5, wherein, Calculate the signal-to-noise ratio (SNR) of the station: When APs are in a synchronized state, the signal-to-noise ratio calculation ignores the ambient noise level. When APs are in an asynchronous state, the signal-to-noise ratio calculation ignores the strength of interference signals; When APs are in a mixed synchronous and asynchronous state, the linear power values ​​of the aggregated environmental noise floor and interference signals are calculated.

9. The wireless local area network throughput prediction method of claim 8, wherein, Methods for aggregating the linear power values ​​of ambient noise floor and interference signals include: Convert all signal strength values ​​to be aggregated from dBm power units to mW units; Sum the converted mW unit values; Convert the summation result back to dBm units.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program, when run by a processor, performs the steps of the wireless local area network throughput prediction method according to any one of claims 1-9.

Images