WIFI6 self-adaptive load balancing method and system under multiple scenes

Abstract

The invention provides a WIFI6 self-adaptive load balancing method and system under multiple scenes, and relates to the technical field of wireless communication networks, and the method comprises the steps: obtaining a network scene data sequence and a network operation state data sequence of a target wireless network in a preset collection window; calling a load balancing scene prototype library to perform scene prototype recognition, and determining a target load balancing scene prototype; carrying out load state analysis, and determining a load state feature group of a plurality of access points (AP) and an access state feature group of a plurality of terminal STAs; and carrying out balancing action analysis to obtain a target balancing action, and sending the target balancing action to a controller to carry out WIFI6 load balancing control. According to the method and the device, the technical problems of resource waste or network congestion and reduction of overall performance caused by incapability of adapting to more diversified environments although the method and the device can effectively work in a specific scene in most of the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of wireless communication network technology, and specifically to a WIFI6 adaptive load balancing method and system for multiple scenarios. Background Technology

[0002] WiFi 6 has made breakthroughs in improving throughput, reducing latency, and increasing network capacity. Especially in high-density environments, WiFi 6 can provide more efficient wireless access for more terminals. However, with the continuous development of wireless technology, the network environment is becoming more and more complex, and the types of scenarios are becoming more diverse. Different load balancing strategies are needed to optimize network performance for different scenarios.

[0003] Most existing technologies are designed for specific scenarios, such as high-density meetings and low-power IoT devices. Each scenario has its own characteristics and requirements. Therefore, while existing technologies can work effectively in specific scenarios, they cannot adapt to more diverse environments. For example, load balancing strategies designed for high-density meetings are not suitable for environments with a high concentration of low-power IoT devices, and vice versa. This leads to a significant reduction in network load balancing effectiveness when multiple scenarios occur alternately, resulting in resource waste or network congestion and reduced overall performance. Summary of the Invention

[0004] This application provides a WIFI6 adaptive load balancing method and system for multiple scenarios, aiming to solve the technical problem that most existing technologies are designed for a specific type of scenario. Although they can work effectively in specific scenarios, they cannot adapt to more diverse environments, resulting in resource waste or network congestion and reduced overall performance.

[0005] The first aspect disclosed in this application provides a multi-scenario adaptive load balancing method for Wi-Fi 6. The method includes: acquiring a network scenario data sequence and a network operation status data sequence of a target wireless network within a preset acquisition window, wherein the target wireless network includes multiple access points (APs), multiple terminal units (STAs), and a controller; calling a load balancing scenario prototype library to perform scenario prototype identification on the network scenario data sequence to determine a target load balancing scenario prototype, wherein each load balancing scenario prototype in the load balancing scenario prototype library is constructed based on a deep learning network algorithm; performing load status analysis based on the network operation status data sequence to determine load status feature groups of multiple APs and access status feature groups of multiple terminal units (STAs); using the load balancing scenario prototype library to perform balancing action parsing on the load status feature groups and access status feature groups to obtain a target balancing action, and sending the target balancing action to the controller for Wi-Fi 6 load balancing control.

[0006] The second aspect of this application discloses a multi-scenario adaptive load balancing system for Wi-Fi 6. This system is used in the aforementioned multi-scenario adaptive load balancing method for Wi-Fi 6. The system includes: a data sequence acquisition module, used to acquire network scenario data sequences and network operation status data sequences of a target wireless network within a preset acquisition window, wherein the target wireless network includes multiple access points (APs), multiple terminal units (STAs), and a controller; a scenario prototype identification module, used to call a load balancing scenario prototype library to perform scenario prototype identification on the network scenario data sequences and determine the target load balancing scenario prototype, wherein each load balancing scenario prototype in the load balancing scenario prototype library is constructed based on a deep learning network algorithm; a load status analysis module, used to perform load status analysis based on the network operation status data sequences to determine the load status feature groups of multiple access points (APs) and the access status feature groups of multiple terminal units (STAs); and a load balancing control module, used to use the load balancing scenario prototype library to parse the load status feature groups and access status feature groups to obtain the target balancing action, and send the target balancing action to the controller for Wi-Fi 6 load balancing control.

[0007] One or more technical solutions provided in this application have at least the following beneficial effects:

[0008] By periodically acquiring data sequences from the target wireless network, changes in the network environment can be monitored in real time. This ensures that the network management system can obtain the latest network scenario information, including the operational status data of access points (APs), terminal stations (STAs), and controllers. This provides reliable data support for subsequent decision-making processes, ensuring timely responses to dynamically changing network environments. By calling a load balancing scenario prototype library, different network scenarios can be automatically identified based on deep learning network algorithms. This allows for dynamic identification of the current network environment and matching of corresponding load balancing strategies. Through a pre-trained prototype library, the most suitable load balancing scenario prototype can be selected quickly and accurately, thereby automatically adjusting strategies to cope with different network demands and challenges. The system analyzes network operation status data sequences to obtain the load status characteristics of access points (APs) and terminals (STAs). These characteristics provide a comprehensive understanding of the current network load distribution, offering crucial data support for subsequent load balancing decisions and ensuring the rational allocation of network resources. Based on load status and access status characteristics, the system uses a load balancing scenario prototype library to parse actions and automatically generate the most suitable balancing actions for the current network state. This ensures a balanced distribution of network load among access points and terminals. By sending the target balancing actions to the controller, the system can adjust network configuration in real time, optimize performance, and avoid network congestion, latency, and connection instability caused by excessive local load.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of a multi-scenario adaptive load balancing method for WIFI6 provided in an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of a multi-scenario adaptive load balancing system structure for WIFI6 provided in an embodiment of this application.

[0012] Figure labeling: Data sequence acquisition module 10, scene prototype recognition module 20, load status analysis module 30, load balancing control module 40. Detailed Implementation

[0013] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0014] Example 1, as Figure 1 As shown in the figure, this application embodiment provides a WIFI6 adaptive load balancing method in multiple scenarios, the method including: The network scene data sequence and network operation status data sequence of the target wireless network within a preset acquisition window are acquired. The target wireless network includes multiple access points (APs), multiple terminal units (STAs), and a controller.

[0015] Network scenario data refers to scenario characteristic data related to the target wireless network. These characteristics describe different aspects of the wireless environment, including but not limited to high-density conferencing, mobile corridors, mixed terminals in dormitories, periodic industrial reporting, smart hospitals, and multi-AP coverage in large parks. Each network scenario has its unique structural characteristics, such as user mobility distribution, service type ratio, AP overlap density, uplink and downlink traffic patterns, low-power periodic wake-up rhythm, and roaming frequency. Network operation status data is real-time network status data, mainly including the status information of access points (APs) and terminal stations (STAs) in the network, such as AP load status and terminal access status. Access point AP load status characteristics are the load information of each access point AP, including the number of currently connected terminals, transmission rate, frequency band usage, and interference level; terminal access status characteristics are the connection information of terminal devices, including signal strength, data transmission rate, distance from the access point AP, and signal interference. Within a preset collection window, this scenario data is periodically collected from the target wireless network. This process is completed through network monitoring tools or management systems, and the data is aggregated into a data sequence for subsequent analysis.

[0016] The load balancing scenario prototype library is invoked to identify the network scenario data sequence and determine the target load balancing scenario prototype. Each load balancing scenario prototype in the load balancing scenario prototype library is constructed based on a deep learning network algorithm.

[0017] The load balancing scenario prototype library is trained using deep learning algorithms. Each scenario prototype is customized for a specific wireless network scenario, such as high-density conferencing or mobile corridors. By learning from historical data, each prototype can automatically adjust its load balancing strategy based on scenario characteristics and operational status data. The construction process includes training with a large amount of wireless network data samples, such as different scenario data and operational status data. Sample data is collected, including network scenario data, operational status data, and load balancing actions and effects for each scenario. Based on deep learning algorithms, the network operational characteristics under different scenarios are analyzed, and the corresponding load balancing strategy for each scenario is learned. Through training with a large amount of data, the optimal load balancing solution for different scenarios can be found.

[0018] After obtaining the network scenario data sequence of the target wireless network, the load balancing scenario prototype library is called to match it based on these data. By matching scenario features, such as user density, service type, and access point (AP) overlap, with the scenario features in the prototype library, the current network scenario is determined to be of which scenario prototype, and the target load balancing scenario prototype is obtained. For example, if the scenario features show high user density and frequent roaming, the high-density conference scenario prototype is matched. After identifying the target scenario prototype, the most suitable load balancing strategy can be determined based on the target scenario prototype.

[0019] Based on the network operation status data sequence, load status analysis is performed to determine the load status characteristic groups of multiple access points (APs) and the access status characteristic groups of multiple terminals (STAs).

[0020] By analyzing network operational status data sequences, the status characteristics of multiple access points (APs) and terminal stations (STAs) in the network are extracted, thereby identifying the network load status. Specifically, each AP has multiple performance indicators reflecting its load status during operation, including the rate of change in the number of associated users, frequency band load distribution, BSS coloring conflict degree, and the degree of overlap of neighboring APs. These features are extracted from the AP data sequences and combined into a load status feature group for the AP. The access status of a terminal station reflects its behavior in the network, including access point connection status, data transmission, frequency band selection, and network quality assessment. By analyzing the operational status data of the terminal station, the aforementioned access status characteristics are extracted and constructed into an access status feature group for the terminal station.

[0021] The load balancing scenario prototype library is used to parse the load status feature group and access status feature group to obtain the target balancing action, and the target balancing action is sent to the controller for WIFI6 load balancing control.

[0022] The obtained load status feature groups and access status feature groups are used as input data and passed to the load balancing scenario prototype library. By matching them with the scenario prototypes in the load balancing scenario prototype library, the most suitable load balancing strategy for the current network state is identified. This strategy is based on factors such as network load, signal interference, user density, and roaming behavior. For example, in a high-density conference scenario, a traffic priority strategy is selected to ensure priority transmission of video and voice traffic; in a large campus scenario, an AP load balancing strategy is selected to ensure that the load of each AP is relatively balanced.

[0023] Target load balancing actions include the following: AP load adjustment, adjusting the load distribution of access points, such as by changing the terminal's access point or switching frequency bands to balance the load; traffic scheduling, scheduling uplink and downlink traffic to ensure that overload does not occur during peak traffic periods; terminal reassociation, switching terminals from heavily loaded APs to lightly loaded APs based on the terminal's access status and network quality; and frequency band adjustment, switching terminals from overloaded frequency bands to less busy frequency bands. The Wi-Fi 6 controller is responsible for managing and coordinating APs and terminals in the wireless network, sending the identified target load balancing actions to the controller, which then executes these actions for real-time load balancing.

[0024] Furthermore, each of the multiple access point APs operates in at least one of the following frequency bands: 2.4GHz, 5GHz, and 6GHz.

[0025] The 2.4GHz band is a traditional Wi-Fi band, offering wider coverage but with narrower bandwidth and more interference, making it suitable for low-traffic applications. The 5GHz band provides even wider bandwidth, supporting higher data transfer rates, and is suitable for medium to high-traffic applications. While its coverage area is smaller than 2.4GHz, it experiences relatively less interference. The 6GHz band, a new addition to Wi-Fi 6E, offers even wider bandwidth and less interference, making it suitable for high-density, high-traffic applications, especially in environments with low interference. In a Wi-Fi 6 network, each access point (AP) can support multiple of these bands. Different bands can be selected and allocated based on the actual network environment, dynamically choosing the appropriate band through intelligent load balancing to achieve optimal network performance.

[0026] Furthermore, the load balancing scenario prototype library is invoked to perform scenario prototype identification on the network scenario data sequence to determine the target load balancing scenario prototype, including: Multiple sample network scenario data sequences, multiple sample network operation status data sequences, and corresponding multiple sample load balancing actions and multiple sample load balancing effects are acquired. Based on the multiple sample network scenario data sequences, scenario feature identification is performed to determine the characteristics of the multiple sample network scenarios. Multidimensional clustering is performed on the multiple sample network scenario features based on user mobility distribution, service type ratio, AP overlap density, uplink and downlink traffic patterns, low-power periodic wake-up rhythm, and roaming frequency to determine multiple clustered sample network scenario feature sets. According to the multiple clustered sample network scenario feature sets, the multiple sample network operation status data sequences and corresponding multiple sample load balancing actions and multiple sample load balancing effects are mapped and divided to obtain multiple clustered sample network operation status data sequence sets, multiple clustered sample load balancing action sets, and multiple clustered sample load balancing effect sets. Load balancing scenario prototypes are constructed based on the multiple clustered sample network scenario feature sets, multiple clustered sample network operation status data sequence sets, multiple clustered sample load balancing action sets, and multiple clustered sample load balancing effect sets to obtain a load balancing scenario prototype library.

[0027] The sample network scenario data sequence includes network characteristic data under multiple scenarios, specifically information such as user behavior, traffic patterns, and AP load under different network environments. Each scenario includes multi-dimensional scenario characteristics, such as user density, traffic distribution, and AP overlap. The sample network operation status data sequence records the real-time status data of each access point (AP) and terminal STA in the network. This data includes AP load, signal strength, and terminal traffic demand, which is crucial for analyzing network status and determining load balancing needs. Sample load balancing actions refer to the load balancing actions taken under specific scenarios based on the collected network scenario and status data. Load balancing actions include AP switching, frequency band adjustment, terminal migration, and traffic control. Sample load balancing effects refer to the changes in network performance after executing load balancing actions. These effects are evaluated through indicators such as network throughput, latency, packet loss rate, and signal strength.

[0028] By analyzing multiple sample network scenario data sequences, the characteristics of each scenario are extracted and identified. Specifically, this extraction is achieved through techniques such as clustering analysis and feature engineering, ultimately forming a feature representation for that scenario. For example, clustering algorithms categorize multiple different types of scenarios into different scenario groups, thereby assigning a specific load balancing strategy to each scenario. Scenario characteristics include user distribution, service type, AP density, traffic patterns, and low-power device cycles, all of which indicate the current scenario category of the network.

[0029] Multidimensional clustering of network scene features from multiple samples aims to group scenes with similar characteristics into one category. Among these, user mobility distribution indicates whether users are static, mobile, or frequently switching within a specific scene, influencing terminal connection methods and roaming needs; service type ratio refers to the traffic proportion of services such as voice, video, and file transfer in different scenes, affecting network load characteristics; AP overlap density affects signal interference and signal quality, with more pronounced AP overlap in high-density areas; uplink and downlink traffic patterns refer to the ratio of upload and download traffic in a scene, with different application traffic patterns (such as video calls and file downloads) affecting network traffic demand; low-power cycle wake-up rhythm refers to the wake-up cycle characteristics of low-power terminals such as IoT devices, which vary across different devices and scenarios; and roaming frequency refers to the frequency with which terminal devices switch between multiple APs, with high roaming frequencies typically occurring in large parks or corridors, while low roaming frequencies occur in static scenarios such as offices.

[0030] Multidimensional clustering algorithms, such as K-means, hierarchical clustering, and DBSCAN, are used to cluster the above features. Through these algorithms, multiple network scenarios with similar features are identified and classified into different cluster groups. After the clustering is completed, multiple cluster sample network scenario feature sets are obtained. Each set represents a specific category of scenario features. The scenario features in each category are similar, representing network scenarios with similar needs and characteristics.

[0031] Based on the network scenario feature sets of multiple clustered samples, the corresponding network operation status data, load balancing actions, and load balancing effects are mapped to these clusters, thereby assigning an appropriate load balancing strategy to each cluster. For example, scenarios with high user mobility and high AP overlap are classified into one category, and the network status data, load balancing actions, and load balancing effects matching this scenario are assigned to this category.

[0032] Based on the network scene feature sets of multiple clustered samples obtained by clustering, as well as the corresponding network states, balancing actions and effects, a load balancing scenario prototype library is constructed. Each scenario prototype represents a typical network scenario. These scenario prototypes are trained and optimized using deep learning technology based on historical data. By continuously acquiring data from actual network operations, the load balancing scenario prototype library can be continuously updated and optimized.

[0033] Furthermore, based on multiple clustering sample network scene feature sets, multiple clustering sample network running state data sequence sets, multiple clustering sample balancing action sets, and multiple clustering sample balancing effect sets, load balancing scenario prototypes are constructed to obtain a load balancing scenario prototype library, including: A load balancing drift search is performed by traversing the multiple clustered sample network scene feature sets to determine multiple central clustered sample network scene features. The multiple clustered sample balance effect sets are then divided according to a preset sample balance effect threshold to obtain multiple positive clustered sample balance effect sets and multiple negative clustered sample balance effect sets. Training is performed using the multiple positive clustered sample balance effect sets, the multiple negative clustered sample balance effect sets, and multiple clustered sample network running state data sequence sets to obtain multiple initial load balancing scene prototypes. The multiple initial load balancing scene prototypes are associated and identified using the multiple central clustered sample network scene features to construct the load balancing scene prototype library.

[0034] Balanced drift search is a search process aimed at finding a set of representative central scene features from multiple clustered sample network scene feature sets. These features best represent the characteristics of a given cluster and better express the network behavior and needs within that scenario. Specifically, it iterates through multiple clustered sample network scene feature sets and executes search algorithms, such as K-means centroid search or drift algorithms, to find the central features of each cluster. Through search and computation, it identifies the most typical scene features within that cluster. For example, for a high-density user scenario, user mobility distribution and AP overlap density are selected as central features.

[0035] After applying a load balancing strategy in each scenario, there will be certain changes in network performance. These changes are recorded as the balancing effect. To optimize the load balancing strategy, these balancing effects are evaluated and classified. Based on preset sample balancing effect thresholds, such as a throughput increase of at least 10% or a latency reduction of at least 20%, the balancing effects are divided into positive and negative samples: the clustered positive sample balancing effect set represents situations where the load balancing strategy is effective in a specific scenario and can bring about significant performance improvements, such as increased traffic, reduced latency, and improved network stability; the clustered negative sample balancing effect set represents situations where the load balancing strategy has failed to effectively improve network performance, which may lead to performance degradation or insignificant improvement.

[0036] By combining multiple sets of positive and negative clustering load balancing results and multiple sets of clustering sample network runtime data sequences, several initial load balancing scenario prototypes are trained. These prototypes represent the optimal load balancing strategies for different scenarios. During training, the load balancing model is trained using both positive and negative sample load balancing results, combined with clustering sample network runtime data. Positive samples serve as positive feedback during model training, helping to understand the success of load balancing strategies in specific scenarios; negative samples serve as negative feedback, helping to understand which strategies fail or cause performance degradation in certain scenarios. Using this data, deep neural networks are employed for training. Through multiple training iterations, the load balancing strategy can be gradually optimized to adapt to different network scenarios. After training, multiple initial load balancing scenario prototypes are obtained, representing the optimal load balancing actions and strategies for specific scenarios.

[0037] After obtaining initial load balancing scenario prototypes through training, these prototypes need to be associated with the network scenario features of the central cluster samples to ensure that each prototype matches a specific network scenario. Specifically, based on the obtained network scenario features of multiple central cluster samples, each initial load balancing scenario prototype is assigned a label or identifier. After association by identifier, all initial load balancing scenario prototypes are combined into a load balancing scenario prototype library, which contains the optimal load balancing strategies for all common network scenarios.

[0038] Furthermore, by combining the multiple sets of positive clustering sample balancing results, the multiple sets of negative clustering sample balancing results, and the multiple sets of clustering sample network running state data sequences for training, multiple initial load balancing scenario prototypes are obtained, including: Based on multiple sets of positive and negative clustering equilibrium effects, action reward and penalty labels are applied to multiple clustering sample equilibrium action sets to obtain multiple labeled clustering sample equilibrium action sets. Access point and terminal status analysis is performed by traversing the multiple clustering sample network operation state data sequence sets to obtain multiple clustering sample access state feature sets and multiple clustering load state feature sets. Using these multiple clustering sample access state feature sets, multiple clustering load state feature sets, and multiple labeled clustering sample equilibrium action sets, the framework constructed based on the deep learning network algorithm is trained with positive and negative samples according to the action reward and penalty labels until training converges, obtaining multiple initial load balancing scenario prototypes after training is completed.

[0039] The balancing actions for each cluster sample are evaluated and labeled using a reward and penalty system. Reward labels highlight actions that effectively improve network performance, while penalty labels identify actions that fail to improve performance or cause performance degradation. Specifically, for actions in the positive cluster sample balancing effect set, if the action leads to improved network performance (e.g., increased throughput, reduced latency, or reduced interference), it is labeled a reward action. For actions in the negative cluster sample balancing effect set, if the action does not significantly improve network performance or even exacerbates network problems, it is labeled a penalty action. By labeling multiple cluster sample balancing action sets with rewards and penalties, multiple labeled cluster sample balancing action sets are generated. Each action set contains actions with either reward or penalty labels, which provide feedback for subsequent training and optimization.

[0040] By analyzing multiple clustered sample network operational status data sequences, the state characteristics of access points and terminals are extracted. These characteristics help to gain a deeper understanding of the specific network load and provide a basis for subsequent load balancing optimization. Specifically, by analyzing the operating status of access points, such as load, signal strength, frequency band usage, and number of users, the performance of each access point in the current scenario is determined. For example, some access points may be under high load due to excessive users or frequency band overload. Analyzing terminal connectivity, such as access quality, bandwidth usage, and traffic patterns, helps to assess whether terminals can effectively access the network from the current access point, whether a switch to another access point is necessary, or whether frequency band adjustments are needed to improve connection quality. Through comprehensive analysis of the state data of access points and terminals in the network, the network load distribution is obtained. For example, it identifies areas where access points are overloaded or where some terminals have poor access quality, thus providing a basis for subsequent load balancing adjustments. The analysis of access point and terminal states generates multiple clustered sample access status feature sets and multiple clustered load status feature sets. These feature sets provide data support for subsequent deep learning model training.

[0041] Multiple clustered sample access state feature sets, multiple clustered load state feature sets, and multiple labeled clustered sample load balancing action sets are used as training data. A deep learning network algorithm is used for training to identify the most effective load balancing strategy for each scenario. During training, positive samples (actions with reward labels) and negative samples (actions with penalty labels) are used. Positive samples represent effective load balancing actions in the network scenario, such as effective AP switching and frequency band adjustment, while negative samples represent ineffective or unsuitable load balancing actions for the current scenario. The deep learning model employs multi-layer neural networks, convolutional neural networks, or long short-term memory networks to handle complex network state and behavioral data. The training process continuously optimizes the model until convergence is achieved, meaning the model can accurately identify which load balancing actions are suitable for which network scenarios and effectively optimize network performance. During training, weights are adjusted through error backpropagation and optimization algorithms to minimize the error between prediction and actual results. After training, multiple initial load balancing scenario prototypes are obtained. Each prototype represents a typical scenario and includes the optimal load balancing strategy and operations for that scenario.

[0042] Furthermore, load status analysis is performed based on the network operation status data sequence to determine the load status characteristic groups of multiple access points (APs) and the access status characteristic groups of multiple terminal stations (STAs), including: Based on the multiple access points (APs) and the multiple terminal stations (STAs), data extraction is performed on the network operation status data sequence to obtain multiple access point AP data sequences and multiple terminal STA data sequences; a first access point AP data sequence is extracted from the multiple access point AP data sequences; feature trend superposition is performed on the first access point AP data sequence to determine a first load status feature, and the first load status feature is added to the load status feature group; feature trend superposition is performed on the multiple terminal STA data sequences to determine the access status feature group.

[0043] Access points (APs), as a key component of wireless networks, are responsible for managing the connections between terminal devices and the network. AP data sequences include information such as their operating status, signal strength, load, and the number of connected terminals, reflecting the AP's performance within the network. Terminals (STAs) are devices connected to the AP; these can be mobile devices, computers, or other IoT devices. STA data sequences contain information such as their connection status, signal strength, and bandwidth requirements.

[0044] Extract the first access point AP data sequence from multiple access point AP data sequences. This data sequence represents the operating status of a specific access point AP at a certain moment.

[0045] Feature trend overlay is achieved by calculating feature values ​​at multiple time points from the data sequence of the first access point (AP) to identify load trends. The extracted first load state features are then added to the load state feature group to provide data support for subsequent load balancing decisions.

[0046] The system iterates through multiple terminal STA data sequences, analyzes the access status of each terminal, and extracts access status features by superimposing feature trends. By superimposing trends on terminal data sequences, it identifies changes in terminal behavior at different time points and derives access status feature groups. These features indicate the terminal's performance in the network and are then used to determine whether load balancing or switching is necessary.

[0047] Furthermore, the first load state characteristics are determined by overlaying feature trends on the data sequence of the first access point (AP), including: Multi-scale trend feature extraction is performed on the first access point (AP) data sequence to obtain a first multi-scale trend feature set. Each multi-scale trend feature includes the rate of change of associated user number, frequency band load distribution, BSS coloring conflict degree, and neighboring AP overlap degree. The first multi-scale trend feature set is randomly sampled multiple times without replacement, with two multi-scale trend features sampled each time, to obtain a first multi-scale trend feature extraction combination set. The first multi-scale trend feature extraction combination set is traversed and analyzed as a whole to determine the first target multi-scale trend feature. The last first access point (AP) data in the first access point (AP) data sequence is extracted, data encoding is performed, and it is concatenated and fused with the first target multi-scale trend feature to obtain the first load status feature.

[0048] By extracting multi-scale trend features from the data sequence of the first access point (AP), the load change trend of the AP can be analyzed and captured at different time scales. This helps to evaluate the operating status of the AP at multiple levels, thus providing more accurate data support for load balancing. Specifically, the rate of change in the number of associated users measures the rate at which the number of users connected to the AP changes, reflecting the change in AP load. This feature indicates how the AP load fluctuates over time, especially in densely populated areas where the AP load changes rapidly. Frequency band load distribution measures the load on different frequency bands, revealing bandwidth occupancy and load balancing between bands. The load distribution across different frequency bands is crucial for optimizing frequency band selection, reducing interference, and improving network performance. BSS Coloring, a technology in Wi-Fi 6, is used to reduce interference between APs. BSS Coloring collision degree indicates the degree of signal interference between an AP and its neighboring APs; a high collision degree means severe interference, which may affect network performance. Neighbor AP overlap degree indicates the size of the coverage overlap area between an AP and its surrounding APs; a large overlap area means increased signal interference, thus affecting the user's connection quality.

[0049] Two features are randomly selected from the first multi-scale trend feature set each time. If the number of data points in the feature set is odd (i.e., the number of features is odd), a copy of each feature is made to ensure that two features are selected for combination each time. For example, the rate of change in the number of associated users and the frequency band load distribution are selected to form a feature combination, which is used for subsequent analysis. After multiple random samplings without replacement, multiple sets of first multi-scale trend feature extraction combinations are obtained, each containing different feature combinations.

[0050] The process iterates through all feature extraction combinations, evaluating the impact of each combination on network performance. The overall analysis includes: analyzing the weights of different feature combinations, comparing the sensitivity of feature combinations to load changes, and determining which combinations best reflect the trend of load changes through training or optimization algorithms. Based on the results of the overall analysis, the feature combination that best represents AP load changes is selected from multiple combinations. This combination is the first objective multi-scale trend feature, which is used for subsequent data fusion and load state feature construction.

[0051] The last AP data point in the first access point (AP) data sequence is extracted. This data represents the current network state and is the closest to real-time load information. The extracted AP data is encoded into a format suitable for model input, including numericalization, normalization, or standardization, to enable efficient processing by the subsequent model. The encoded AP data is then concatenated with the first target multi-scale trend feature. The concatenated first load state feature integrates the current time data and historical trend data, comprehensively reflecting the AP load state.

[0052] Furthermore, by traversing the first multi-scale trend feature extraction and combination set for overall analysis, the first target multi-scale trend features are determined, including: Iterate through each multi-scale trend feature extraction combination in the first multi-scale trend feature extraction combination set, and perform independent deviation analysis on the rate of change of associated user number, frequency band load distribution, BSS coloring conflict degree, and neighbor AP overlap degree to construct a first combination deviation group set; normalize each combination deviation group in the first combination deviation group set, and matrix the processing results to construct a first combination interaction matrix set; use the first combination interaction matrix set to independently perform mapping graph convolution operation on the corresponding two multi-scale trend features in the first multi-scale trend feature extraction combination set to obtain a first multi-scale interaction trend feature extraction combination set; calculate the mean of features within the first multi-scale interaction trend feature extraction combination set to obtain the first target multi-scale trend feature.

[0053] For each of the first multi-scale trend feature extraction combinations in the first multi-scale trend feature extraction combination set, a bias analysis is performed. Bias analysis refers to analyzing the performance differences of different feature combinations in the actual network environment to further optimize the selection of feature combinations. The bias analysis includes biases in the rate of change of associated user numbers, frequency band load distribution, BSSColoring conflict degree, and neighbor AP overlap degree. By analyzing the biases of these features, a first combination bias set is obtained, which contains the performance biases of different feature combinations in the network.

[0054] Normalization is performed on each combination of biases to eliminate differences in the dimensions of different features, ensuring that each feature falls within the same scale. Normalization transforms feature values ​​into values ​​between 0 and 1, for example, using a min-max normalization method. This helps eliminate the influence of different features on different dimensions, ensuring that each feature has the same weight in calculation. The normalized data is then converted into matrix form to construct the first set of combined interaction matrices. These matrices represent the relationships between feature combinations and provide important information about load status and network behavior.

[0055] Each feature is independently mapped and computed using graph convolution operations. Graph convolution is a technique for processing graph-structured data; by computing the relationships between features through the graph structure, it can better capture complex data patterns. Through graph convolution operations, the features in the first set of combined interaction matrices are mapped. Each feature combination is mapped to a new feature space through graph convolution, thus revealing the interaction relationships between features. Each pair of feature combinations is independently subjected to graph convolution operations to further analyze their mutual influence. In this way, the potential correlations between feature combinations in load changes and network behavior can be better captured. After graph convolution operations, a first multi-scale interaction trend feature extraction combination set is obtained. This set contains the interaction trends between feature combinations, revealing how different features work together to influence changes in load state.

[0056] The average value is calculated for each feature in the first multi-scale interactive trend feature extraction and combination set. By calculating the mean, the extreme influence of individual features can be eliminated, resulting in more stable and reliable feature values. The first target multi-scale trend feature obtained by calculating the mean represents the core characteristics of the current network load status and can provide a more accurate basis for load balancing.

[0057] Furthermore, each of the first combined deviation groups in the first combined deviation group set includes deviations in the rate of change of associated user number, frequency band load distribution, BSS coloring conflict degree, and neighboring AP overlap degree.

[0058] The deviation of the rate of change in the number of associated users refers to the analysis of the fluctuation of the rate of change in the number of users over different time periods, which helps to clarify the network's response capability when user load changes; the deviation of the frequency band load distribution refers to the analysis of the load changes and fluctuations of each frequency band, and the deviation of the frequency band load indicates the resource allocation and utilization of different frequency bands; the deviation of the BSS Coloring collision degree refers to the analysis of the signal collision degree between APs under the BSS Coloring mechanism, reflecting the situation of wireless interference, and this deviation reveals the severity of signal collision in a specific scenario; the deviation of the neighboring AP overlap degree refers to the analysis of the overlap degree between adjacent access point APs, and assesses the impact of overlapping access point AP coverage areas on signal quality and interference.

[0059] Example 2, based on the same inventive concept as the multi-scenario adaptive load balancing method for WIFI6 in the foregoing examples, such as... Figure 2 As shown in the figure, this application embodiment provides a multi-scenario adaptive load balancing system for WIFI6, the system comprising: The data sequence acquisition module 10 is used to acquire network scene data sequences and network operation status data sequences of the target wireless network within a preset acquisition window. The target wireless network includes multiple access points (APs), multiple terminal units (STAs), and a controller. The scene prototype identification module 20 is used to call a load balancing scene prototype library to perform scene prototype identification on the network scene data sequences and determine the target load balancing scene prototype. Each load balancing scene prototype in the load balancing scene prototype library is constructed based on a deep learning network algorithm. The load status analysis module 30 is used to perform load status analysis based on the network operation status data sequences to determine the load status feature groups of the multiple access points (APs) and the access status feature groups of the multiple terminal units (STAs). The load balancing control module 40 is used to use the load balancing scene prototype library to parse the load status feature groups and access status feature groups to obtain the target balancing action and send the target balancing action to the controller for WIFI6 load balancing control.

[0060] Furthermore, each of the multiple access point APs operates in at least one of the following frequency bands: 2.4GHz, 5GHz, and 6GHz.

[0061] Furthermore, the scene prototype recognition module 20 is used to perform the following operation steps: Multiple sample network scenario data sequences, multiple sample network operation status data sequences, and corresponding multiple sample load balancing actions and multiple sample load balancing effects are acquired. Based on the multiple sample network scenario data sequences, scenario feature identification is performed to determine the characteristics of the multiple sample network scenarios. Multidimensional clustering is performed on the multiple sample network scenario features based on user mobility distribution, service type ratio, AP overlap density, uplink and downlink traffic patterns, low-power periodic wake-up rhythm, and roaming frequency to determine multiple clustered sample network scenario feature sets. According to the multiple clustered sample network scenario feature sets, the multiple sample network operation status data sequences and corresponding multiple sample load balancing actions and multiple sample load balancing effects are mapped and divided to obtain multiple clustered sample network operation status data sequence sets, multiple clustered sample load balancing action sets, and multiple clustered sample load balancing effect sets. Load balancing scenario prototypes are constructed based on the multiple clustered sample network scenario feature sets, multiple clustered sample network operation status data sequence sets, multiple clustered sample load balancing action sets, and multiple clustered sample load balancing effect sets to obtain a load balancing scenario prototype library.

[0062] Furthermore, the scene prototype recognition module 20 is used to perform the following operation steps: A load balancing drift search is performed by traversing the multiple clustered sample network scene feature sets to determine multiple central clustered sample network scene features. The multiple clustered sample balance effect sets are then divided according to a preset sample balance effect threshold to obtain multiple positive clustered sample balance effect sets and multiple negative clustered sample balance effect sets. Training is performed using the multiple positive clustered sample balance effect sets, the multiple negative clustered sample balance effect sets, and multiple clustered sample network running state data sequence sets to obtain multiple initial load balancing scene prototypes. The multiple initial load balancing scene prototypes are associated and identified using the multiple central clustered sample network scene features to construct the load balancing scene prototype library.

[0063] Furthermore, the scene prototype recognition module 20 is used to perform the following operation steps: Based on multiple sets of positive and negative clustering equilibrium effects, action reward and penalty labels are applied to multiple clustering sample equilibrium action sets to obtain multiple labeled clustering sample equilibrium action sets. Access point and terminal status analysis is performed by traversing the multiple clustering sample network operation state data sequence sets to obtain multiple clustering sample access state feature sets and multiple clustering load state feature sets. Using these multiple clustering sample access state feature sets, multiple clustering load state feature sets, and multiple labeled clustering sample equilibrium action sets, the framework constructed based on the deep learning network algorithm is trained with positive and negative samples according to the action reward and penalty labels until training converges, obtaining multiple initial load balancing scenario prototypes after training is completed.

[0064] Furthermore, the load status analysis module 30 is used to perform the following operation steps: Based on the multiple access points (APs) and the multiple terminal stations (STAs), data extraction is performed on the network operation status data sequence to obtain multiple access point AP data sequences and multiple terminal STA data sequences; a first access point AP data sequence is extracted from the multiple access point AP data sequences; feature trend superposition is performed on the first access point AP data sequence to determine a first load status feature, and the first load status feature is added to the load status feature group; feature trend superposition is performed on the multiple terminal STA data sequences to determine the access status feature group.

[0065] Furthermore, the load status analysis module 30 is used to perform the following operation steps: Multi-scale trend feature extraction is performed on the first access point (AP) data sequence to obtain a first multi-scale trend feature set. Each multi-scale trend feature includes the rate of change of associated user number, frequency band load distribution, BSS coloring conflict degree, and neighboring AP overlap degree. The first multi-scale trend feature set is randomly sampled multiple times without replacement, with two multi-scale trend features sampled each time, to obtain a first multi-scale trend feature extraction combination set. The first multi-scale trend feature extraction combination set is traversed and analyzed as a whole to determine the first target multi-scale trend feature. The last first access point (AP) data in the first access point (AP) data sequence is extracted, data encoding is performed, and it is concatenated and fused with the first target multi-scale trend feature to obtain the first load status feature.

[0066] Furthermore, the load status analysis module 30 is used to perform the following operation steps: Iterate through each multi-scale trend feature extraction combination in the first multi-scale trend feature extraction combination set, and perform independent deviation analysis on the rate of change of associated user number, frequency band load distribution, BSSColoring conflict degree, and neighbor AP overlap degree to construct a first combination deviation group set; normalize each combination deviation group in the first combination deviation group set, and matrix the processing results to construct a first combination interaction matrix set; use the first combination interaction matrix set to independently perform mapping graph convolution operation on the corresponding two multi-scale trend features in the first multi-scale trend feature extraction combination set to obtain a first multi-scale interaction trend feature extraction combination set; calculate the mean of features within the first multi-scale interaction trend feature extraction combination set to obtain the first target multi-scale trend feature.

[0067] Furthermore, each of the first combined deviation groups in the first combined deviation group set includes deviations in the rate of change of associated user number, frequency band load distribution, BSS coloring conflict degree, and neighboring AP overlap degree.

[0068] Through the foregoing detailed description of a multi-scenario adaptive load balancing method for WIFI6, those skilled in the art can clearly understand that this embodiment provides a multi-scenario adaptive load balancing system for WIFI6. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant details can be found in the method section.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-scenario adaptive load balancing method for Wi-Fi 6, characterized in that, The method includes: Acquire network scene data sequence and network operation status data sequence of the target wireless network within a preset acquisition window, wherein the target wireless network includes multiple access points (APs), multiple terminal units (STAs), and a controller; The load balancing scenario prototype library is invoked to identify the network scenario data sequence and determine the target load balancing scenario prototype. Each load balancing scenario prototype in the load balancing scenario prototype library is constructed based on a deep learning network algorithm. Based on the network operation status data sequence, load status analysis is performed to determine the load status characteristic groups of multiple access points (APs) and the access status characteristic groups of multiple terminals (STAs). The load balancing scenario prototype library is used to parse the load status feature group and access status feature group to obtain the target balancing action, and the target balancing action is sent to the controller for WIFI6 load balancing control.

2. The adaptive load balancing method for WIFI6 in multiple scenarios as described in claim 1, characterized in that, Each of the multiple access point APs operates in at least one of the following frequency bands: 2.4GHz, 5GHz, and 6GHz.

3. The adaptive load balancing method for WIFI6 in multiple scenarios as described in claim 1, characterized in that, The load balancing scenario prototype library is invoked to perform scenario prototype identification on the network scenario data sequence, and the target load balancing scenario prototype is determined, including: Acquire multiple sample network scene data sequences, multiple sample network running status data sequences, and corresponding multiple sample equalization actions and multiple sample equalization effects; Scene feature identification is performed based on the multiple sample network scene data sequences to determine the scene features of the multiple sample networks. Multidimensional clustering of the network scene features of the multiple sample networks is performed based on user mobility distribution, service type ratio, AP overlap density, uplink and downlink traffic patterns, low-power periodic wake-up rhythm and roaming frequency to determine multiple clustered sample network scene feature sets. Based on the multiple clustered sample network scene feature sets, the multiple sample network running state data sequences and the corresponding multiple sample balancing actions and multiple sample balancing effects are mapped and divided to obtain multiple clustered sample network running state data sequence sets, multiple clustered sample balancing action sets, and multiple clustered sample balancing effect sets. Load balancing scenario prototypes are constructed based on multiple clustered sample network scenario feature sets, multiple clustered sample network operation status data sequence sets, multiple clustered sample balancing action sets, and multiple clustered sample balancing effect sets, respectively, to obtain a load balancing scenario prototype library.

4. The WIFI6 adaptive load balancing method in multiple scenarios as described in claim 3, characterized in that, Based on multiple clustering sample network scenario feature sets, multiple clustering sample network runtime status data sequence sets, multiple clustering sample balancing action sets, and multiple clustering sample balancing effect sets, load balancing scenario prototypes are constructed to obtain a load balancing scenario prototype library, including: Perform a balanced drift search by traversing the multiple clustered sample network scene feature sets to determine the multiple central clustered sample network scene features; The multiple clustering sample balance effect sets are divided according to a preset sample balance effect threshold to obtain multiple clustering positive sample balance effect sets and multiple clustering negative sample balance effect sets. By combining the multiple clustering positive sample balancing effect sets, the multiple clustering negative sample balancing effect sets, and the multiple clustering sample network running state data sequence sets for training, multiple initial load balancing scenario prototypes are obtained. The load balancing scenario prototype library is constructed by associating and identifying multiple initial load balancing scenario prototypes using the network scene features of the multiple central clustering sample samples.

5. The WIFI6 adaptive load balancing method in multiple scenarios as described in claim 4, characterized in that, By combining the multiple sets of positive clustering sample load balancing results, the multiple sets of negative clustering sample load balancing results, and the multiple sets of clustering sample network running state data sequences, multiple initial load balancing scenario prototypes are obtained, including: Based on multiple clustering positive sample equilibrium effect sets and multiple clustering negative sample equilibrium effect sets, action reward labels and action penalty labels are applied to multiple clustering sample equilibrium action sets to obtain multiple labeled clustering sample equilibrium action sets; The network operation status data sequence set of the multiple clustered samples is traversed to analyze the access point and terminal status, thereby obtaining multiple clustered sample access status feature set sets and multiple clustered load status feature set sets. By utilizing multiple clustered sample access state feature sets, multiple clustered load state feature sets, and multiple labeled clustered sample balancing action sets, the framework built based on the deep learning network algorithm is trained with positive and negative samples according to action reward labels and action penalty labels until the training converges, thus obtaining multiple initial load balancing scenario prototypes that have been trained.

6. The WIFI6 adaptive load balancing method in multiple scenarios as described in claim 5, characterized in that, Based on the network operation status data sequence, load status analysis is performed to determine the load status characteristic groups of multiple access points (APs) and the access status characteristic groups of multiple terminal stations (STAs), including: Based on the multiple access points (APs) and the multiple terminal units (STAs), data is extracted from the network operation status data sequence to obtain multiple access point AP data sequences and multiple terminal STA data sequences. Extract the first access point AP data sequence from the plurality of access point AP data sequences; The first access point (AP) data sequence is superimposed with feature trends to determine the first load state feature, and the first load state feature is added to the load state feature group. The access status feature group is determined by overlaying the feature trends of the multiple terminal STA data sequences.

7. The WIFI6 adaptive load balancing method in multiple scenarios as described in claim 6, characterized in that, The first access point (AP) data sequence is overlaid with feature trends to determine the first load state characteristics, including: Multi-scale trend feature extraction is performed on the data sequence of the first access point (AP) to obtain a first multi-scale trend feature set, wherein each multi-scale trend feature includes the rate of change of the number of associated users, frequency band load distribution, BSS coloring conflict degree, and degree of overlap of neighboring APs. The first multi-scale trend feature set is randomly sampled multiple times without replacement, and two multi-scale trend features are sampled each time to obtain the first multi-scale trend feature extraction combination set. The first multi-scale trend feature extraction and combination set is traversed and analyzed as a whole to determine the first target multi-scale trend feature. Extract the last AP data from the first access point data sequence, perform data encoding, and splice and fuse it with the first target multi-scale trend features to obtain the first load status features.

8. The WIFI6 adaptive load balancing method in multiple scenarios as described in claim 7, characterized in that, The first multi-scale trend feature extraction and combination set is traversed and analyzed as a whole to determine the first target multi-scale trend features, including: Iterate through each first multi-scale trend feature extraction combination in the first multi-scale trend feature extraction combination set, and perform independent deviation analysis on the rate of change of associated user number, frequency band load distribution, BSS coloring conflict degree, and neighbor AP overlap degree to construct the first combination deviation group set; Normalize each combined deviation group in the first set of combined deviation groups, and matrix the processing results to construct the first set of combined interaction matrices; Using the first set of combined interaction matrices, map convolution operations are performed independently on the two corresponding multi-scale trend features in the first set of multi-scale trend feature extraction to obtain the first set of multi-scale interactive trend feature extraction. Calculate the mean value of the features within the first multi-scale interactive trend feature extraction combination set to obtain the first target multi-scale trend feature.

9. The WIFI6 adaptive load balancing method in multiple scenarios as described in claim 8, characterized in that, Each first combination deviation group in the first combination deviation group set includes the deviation of the rate of change of the number of associated users, the deviation of the frequency band load distribution, the deviation of the BSS coloring conflict degree, and the deviation of the degree of overlap of neighboring APs.

10. A multi-scenario adaptive load balancing system for WIFI 6, characterized in that, The system is used to implement the multi-scenario adaptive load balancing method for Wi-Fi 6 as described in any one of claims 1-9, the system comprising: The data sequence acquisition module is used to acquire network scene data sequence and network operation status data sequence of the target wireless network within a preset acquisition window, wherein the target wireless network includes multiple access points (APs), multiple terminals (STAs), and a controller. The scenario prototype recognition module is used to call the load balancing scenario prototype library to perform scenario prototype recognition on the network scenario data sequence and determine the target load balancing scenario prototype. Each load balancing scenario prototype in the load balancing scenario prototype library is constructed based on a deep learning network algorithm. The load status analysis module is used to perform load status analysis based on the network operation status data sequence to determine the load status characteristic groups of multiple access points (APs) and the access status characteristic groups of multiple terminals (STAs). The load balancing control module is used to analyze the load status feature group and access status feature group using the load balancing scenario prototype library, obtain the target balancing action, and send the target balancing action to the controller for WIFI6 load balancing control.

Images