An adaptive adjustment method and system based on portable Wi-Fi channels

Abstract

This invention discloses an adaptive adjustment method and system for portable Wi-Fi channels, relating to the field of communication technology. It addresses the technical challenges of: difficulty in analyzing channel performance indices, network performance indices, and environmental impact indices to generate a comprehensive channel quality score; difficulty in combining time series analysis and deep learning models to predict peak traffic demand; difficulty in dynamically selecting the optimal channel using reinforcement learning algorithms; difficulty in combining reinforcement learning techniques with deep learning models to identify congestion and dynamically adjust the channel; and difficulty in real-time monitoring and evaluating the effectiveness of channel adjustments. This invention improves the communication quality and network utilization of portable Wi-Fi devices by real-time monitoring of channel status, comprehensive evaluation of channel quality, prediction of traffic demand, dynamic selection of the optimal channel, identification and adjustment of channel congestion, and evaluation of adjustment effects, thereby adapting to complex and ever-changing communication environments.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, specifically an adaptive adjustment method and system based on a portable Wi-Fi channel. Background Technology

[0002] With the rapid development of mobile internet, portable WiFi devices have gradually become an important tool for users to maintain network connectivity in various scenarios due to their portability and flexibility. However, during use, the channel of portable WiFi devices is easily affected by interference from the surrounding environment, leading to channel congestion and impacting communication quality. Therefore, in complex and ever-changing wireless environments, portable WiFi devices need to monitor channel status in real time to ensure network performance and user experience stability. This can be achieved by using time series analysis, deep learning models, and reinforcement learning algorithms in combination to adaptively adjust the channel through real-time monitoring.

[0003] Existing adaptive adjustment methods based on portable Wi-Fi channels have the following problems: it is difficult to analyze channel performance index, network performance index, and environmental impact index to generate a comprehensive channel quality score; it is difficult to combine time series analysis and deep learning models to predict peak traffic demand; it is difficult to use reinforcement learning algorithms to dynamically select the optimal channel; it is difficult to combine reinforcement learning techniques with deep learning models to identify congestion and dynamically adjust the channel; and it is difficult to monitor the channel adjustment effect and evaluate its level in real time. Summary of the Invention

[0004] This invention aims to at least solve one of the technical problems existing in the prior art; to this end, this invention proposes an adaptive adjustment method and system based on a portable Wi-Fi channel to solve the above-mentioned problem. Specifically, the first aspect of this invention provides an adaptive adjustment method based on a portable Wi-Fi channel, comprising the following steps:

[0005] S1: Real-time monitoring and collection of channel monitoring data and environmental impact data of the portable Wi-Fi channel;

[0006] S2: Perform data cleaning and standardization preprocessing on the collected channel monitoring data and environmental impact data respectively;

[0007] S3: Calculate the channel performance index, network performance index, and environmental impact index based on the preprocessed channel monitoring data and environmental impact data; comprehensively evaluate the channel quality in real time based on the channel performance index, network performance index, and environmental impact index to generate a comprehensive score for the portable Wi-Fi channel quality; predict the user's data traffic demand through time series analysis combined with a deep learning model, and calculate the peak traffic demand.

[0008] S4: Employ reinforcement learning algorithms to dynamically select the optimal channel based on channel quality scores and predicted peak traffic demand; then, using a combination of reinforcement learning techniques and deep learning models, identify congestion and dynamically adjust the channel based on the selected optimal channel.

[0009] S5: Monitor the adjusted optimal channel in real time, calculate the effect score, and evaluate the effect level of the optimal channel.

[0010] Furthermore, the channel monitoring data in step S1 includes: the number of devices on the channel, device moving speed, signal strength, channel occupancy rate, device battery level, device temperature, and network traffic data; the environmental impact data includes: ambient temperature, ambient humidity, ambient noise level, air quality index, and light intensity; and the network traffic data includes: bandwidth utilization, packet loss rate, network throughput, latency, and transmission efficiency.

[0011] Furthermore, step S3, which involves analyzing and calculating the channel performance index, network performance index, and environmental impact index based on the preprocessed channel monitoring data and environmental impact data, includes the following steps:

[0012] The formula for calculating the channel performance index is as follows:

[0013]

[0014] The channel performance index Y is obtained, where, It is the signal strength of the i-th device; and These represent the maximum and minimum values ​​of the signal strength, respectively. This refers to the total number of equipment. It is the moving speed of the i-th device; It is the device temperature of the i-th device; Z is the maximum value of the device temperature; Z is the channel occupancy rate. This represents the battery power consumption of the i-th device; This is the maximum battery power consumption. It is the battery life of the i-th device;

[0015] The formula for calculating the network performance index is as follows:

[0016] The network performance index W is obtained, where k is bandwidth utilization; d is packet loss rate; L is network throughput; c is latency; and x is transmission efficiency.

[0017] The formula for calculating the environmental impact index is as follows:

[0018] The environmental impact index H is obtained, where, It is the ambient temperature; β is ambient humidity; I is ambient noise level; Q is air quality index; and Q is light intensity.

[0019] Furthermore, in step S3, the channel quality is comprehensively evaluated in real time based on the channel performance index, network performance index, and environmental impact index to generate a comprehensive score for the portable Wi-Fi channel quality, including the following steps:

[0020] Formula for calculating the overall channel quality score:

[0021] Obtain the overall channel quality score Where Y, W, and H represent the channel performance index, network performance index, and environmental impact index, respectively; , , These represent the weighting coefficients for the channel performance index, network performance index, and environmental impact index, respectively. It is the channel occupancy rate; The channel occupancy adjustment factor is 0.6, and t is time.

[0022] Furthermore, step S3 involves predicting user data traffic demand using time series analysis combined with a deep learning model, and calculating the peak traffic demand, including the following steps:

[0023] The historical time-series datasets collected include: historical traffic consumption time series, device movement GPS coordinate time series, environmental noise intensity, temperature time series data, humidity time series data, number of online devices, and channel occupancy rate; the collected historical time-series data are preprocessed, including: data cleaning and standardization, and alignment by timestamp to form a unified time step sequence;

[0024] By combining time series analysis with deep learning models, a spatiotemporal attention network model is constructed to predict users' data traffic needs. In the time dimension, a Long Short-Term Memory (LSTM) network is used to process the traffic sequence and extract long-term dependency features. In the spatial dimension, the device movement GPS coordinate time series is converted into a trajectory graph by connecting continuous GPS points as edges with edge weights equal to the reciprocal of the Euclidean distance. The device movement trajectory is then modeled using a graph neural network to capture spatial correlations.

[0025] The preprocessed time series dataset is input into the spatiotemporal attention network model for training; based on the trained spatiotemporal attention network model, the preprocessed time series data collected in real time is input into the trained spatiotemporal attention network model, and the output prediction results are the predicted data flow demand value and timestamp.

[0026] The forecast results are smoothed using the moving average method or the exponential smoothing method. Based on the smoothed forecast results, the peak point of traffic demand is identified, that is, the time point when traffic demand reaches its maximum value and the corresponding traffic value. Finally, the peak traffic demand is calculated.

[0027] Furthermore, step S4 employs a reinforcement learning algorithm to dynamically select the optimal channel based on the comprehensive channel quality score and peak traffic demand, including the following steps:

[0028] Based on the peak traffic demand predicted and calculated by the spatiotemporal attention network model, channels in the portable Wi-Fi channel whose bandwidth capacity is lower than the peak traffic demand are excluded.

[0029] The weighting of the overall channel quality score is dynamically adjusted based on peak traffic demand; during peak traffic demand periods, 0.6 0.3 It is 0.1; during non-peak traffic demand periods, 0.4 0.4 It is 0.2;

[0030] Based on the assigned weights, the comprehensive channel quality scores for different time periods are calculated and arranged in descending order as candidate channels; among the candidate channels, the channel with the highest comprehensive channel quality score is selected as the optimal channel.

[0031] The Q-learning algorithm in reinforcement learning is used to switch to the optimal channel. Each portable Wi-Fi device acts as an agent. In the Q-learning algorithm, the action value function Q(s,a) of each state and action pair is learned. The agent selects whether to switch to the optimal channel based on the current channel state s.

[0032] The agent performs the selected action, switching the portable Wi-Fi device to the optimal channel; after switching the portable Wi-Fi device to the selected optimal channel, communication begins.

[0033] During the switching process, communication on the current channel is maintained and unsent data packets are buffered. After the switch, the optimal channel is monitored in real time, and the immediate and long-term rewards are calculated according to the reward function of the Q-learning algorithm. The corresponding Q values ​​in the Q-value table are updated according to the immediate and long-term rewards. The convergence of the Q-value table is evaluated. If convergence is achieved, training is stopped; otherwise, iterative learning continues.

[0034] Furthermore, in step S4, reinforcement learning techniques combined with a deep learning model are used to identify congestion and dynamically adjust the channel based on the selected optimal channel, including the following steps:

[0035] Based on the selected optimal channel, channel scanning technology is used to collect optimal channel status information in real time, including: channel utilization, bandwidth utilization, channel capacity, handover delay, and data loss. The collected data is then preprocessed, including: data cleaning and normalization.

[0036] By combining reinforcement learning techniques with deep learning models, a congestion identification model based on deep reinforcement learning is constructed. The historical best channel state information set with labels indicating whether channel congestion has occurred is input into the congestion identification model based on deep reinforcement learning for training. The best channel state information set collected in real time and preprocessed is input into the trained congestion identification model based on deep reinforcement learning to output the congestion identification result.

[0037] If no channel congestion is detected, communication continues on the current channel. If channel congestion is detected, a binary exponential backoff algorithm is used. Based on the binary exponential backoff algorithm, an adaptive mechanism is introduced to dynamically adjust the backoff window size according to the real-time channel status information and the degree of congestion. The degree of congestion is calculated by weighted average summation of channel utilization, bandwidth utilization, channel capacity, handover delay, and data loss.

[0038] Formula for dynamically adjusting the backoff window:

[0039] in, It is the size of the backoff window when the k-th channel collision or congestion occurs; It refers to the degree of congestion;

[0040] When the congestion level exceeds a congestion threshold, a dynamic adjustment mechanism for the backoff window size is triggered, adjusting the backoff window size to... Otherwise, adjust the backoff window size to... ;in, The congestion threshold is 0.65. That represents the maximum level of congestion.

[0041] Furthermore, step S5 involves real-time monitoring of the adjusted optimal channel, calculating the effect score, and evaluating the effect level of the optimal channel, including the following steps:

[0042] Real-time monitoring of channel monitoring data and environmental impact data of the adjusted optimal channel is used, and the channel quality comprehensive score is calculated by substituting these data into the channel quality comprehensive scoring formula. A dynamic evaluation cycle is set to trigger every 5 seconds for effect evaluation. The effect scoring formula is as follows:

[0043] Receive performance rating ;in, This represents the overall channel quality score before channel adjustment. This represents the overall channel quality score after channel adjustment. To indicate the degree of congestion; It is a congestion level threshold;

[0044] Based on the performance rating, three levels are divided: Excellent, Good, and Poor; when When, the grade is excellent; when When, the grade is good; when When the evaluation level is poor, the system returns to step S4 to readjust the channel selection.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] This invention monitors and collects channel monitoring data and environmental impact data of portable WiFi channels in real time, enabling timely acquisition of the latest channel status and providing accurate data support for subsequent channel quality assessment and optimization. By comprehensively analyzing channel performance index, network performance index, and environmental impact index, it can comprehensively evaluate channel quality and generate a more accurate overall score.

[0047] This invention uses time series analysis combined with deep learning models to predict user data traffic demand, enabling early prediction of peak traffic demand and providing a basis for the rational allocation of channel resources. It employs reinforcement learning algorithms to dynamically select the optimal channel and combines reinforcement learning techniques with deep learning models to identify channel congestion, allowing for timely channel adjustments to avoid congestion and improve communication efficiency and user experience.

[0048] This invention assesses the effectiveness of channel adjustments by real-time monitoring of the optimal channel before and after adjustments and analyzing the adjustment effect score, thereby optimizing communication quality, reducing interference, improving network utilization, and adapting to the ever-changing communication environment. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of the method of the present invention;

[0051] Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0052] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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 are within the scope of protection of the present invention.

[0053] Please see Figure 1 As shown, a first aspect of the present invention provides an adaptive adjustment method based on a portable Wi-Fi channel, comprising the following steps:

[0054] S1: Real-time monitoring and collection of channel monitoring data and environmental impact data of the portable Wi-Fi channel;

[0055] S2: Perform data cleaning and standardization preprocessing on the collected channel monitoring data and environmental impact data respectively;

[0056] S3: Calculate the channel performance index, network performance index, and environmental impact index based on the preprocessed channel monitoring data and environmental impact data; comprehensively evaluate the channel quality in real time based on the channel performance index, network performance index, and environmental impact index to generate a comprehensive score for the portable Wi-Fi channel quality; predict the user's data traffic demand through time series analysis combined with a deep learning model, and calculate the peak traffic demand.

[0057] S4: Employ reinforcement learning algorithms to dynamically select the optimal channel based on channel quality scores and predicted peak traffic demand; then, using a combination of reinforcement learning techniques and deep learning models, identify congestion and dynamically adjust the channel based on the selected optimal channel.

[0058] S5: Monitor the adjusted optimal channel in real time, calculate the effect score, and evaluate the effect level of the optimal channel.

[0059] Specifically, the system monitors and collects channel monitoring data and environmental impact data for the portable Wi-Fi channel in real time. Environmental impact data is collected using various sensors, including temperature and humidity sensors. The collected data is cleaned, including removing outliers such as sudden signal strength changes and abnormally high bit error rates; supplementing missing values ​​using moving averages or interpolation; and standardizing data of different dimensions using Z-score or Min-Max standardization. Based on the preprocessed channel monitoring and environmental impact data, the system calculates channel performance, network performance, and environmental impact indices, and comprehensively evaluates channel quality. Time series analysis combined with deep learning models is used to predict user data traffic demand and calculate peak traffic demand, providing a basis for subsequent channel adjustments. Q-learning, a reinforcement learning algorithm, is used to select the optimal channel based on the channel quality score and peak traffic demand. Reinforcement learning techniques are used to adjust the channel in real time, identifying channel congestion and switching to a better channel. The adjusted channel performance is monitored in real time, and an effectiveness score is calculated, categorizing the channel adjustment effectiveness into three levels: excellent, good, and poor.

[0060] In this embodiment, the channel monitoring data in step S1 includes: the number of devices on the channel, device moving speed, signal strength, channel occupancy rate, device battery level, device temperature, and network traffic data; the environmental impact data includes: ambient temperature, ambient humidity, ambient noise level, air quality index, and light intensity; and the network traffic data includes: bandwidth utilization, packet loss rate, network throughput, latency, and transmission efficiency.

[0061] Specifically, in channel monitoring data collection, the number of devices is determined by real-time statistics of the number of devices on the current channel; device movement speed is estimated using GPS or Wi-Fi positioning technology; signal strength is measured in real-time through the antenna interface; channel occupancy is determined by a channel utilization counter to track the proportion of time the channel is occupied; device battery level is obtained from the device's battery management module; and device temperature is collected using the device's built-in temperature sensor. In network traffic data, bandwidth utilization is determined by statistical analysis of the current channel's bandwidth usage through the network interface; packet loss rate is calculated by analyzing the proportion of packet loss based on the network protocol stack; network throughput is calculated based on the data transmission rate; latency is measured using Ping or TCP handshake time; and transmission efficiency is calculated as the ratio of the actual transmission rate to the theoretical maximum rate. In environmental impact data collection, ambient temperature is collected using the temperature sensor built into the portable Wi-Fi device; ambient humidity is collected using a humidity sensor; ambient noise levels are collected using an electromagnetic interference detector; air quality index is obtained using an air quality sensor or environmental monitoring station; and light intensity is collected using a photosensor.

[0062] In this embodiment, step S3, which analyzes and calculates the channel performance index, network performance index, and environmental impact index based on the preprocessed channel monitoring data and environmental impact data, includes the following steps:

[0063] The formula for calculating the channel performance index is as follows:

[0064]

[0065] The channel performance index Y is obtained, where, It is the signal strength of the i-th device; and These represent the maximum and minimum values ​​of the signal strength, respectively. This refers to the total number of equipment. It is the moving speed of the i-th device; It is the device temperature of the i-th device; Z is the maximum value of the device temperature; Z is the channel occupancy rate. This represents the battery power consumption of the i-th device; This is the maximum battery power consumption. It is the battery life of the i-th device;

[0066] The formula for calculating the network performance index is as follows:

[0067] The network performance index W is obtained, where k is bandwidth utilization; d is packet loss rate; L is network throughput; c is latency; and x is transmission efficiency.

[0068] The formula for calculating the environmental impact index is as follows:

[0069] The environmental impact index H is obtained, where, It is the ambient temperature; β is ambient humidity; I is ambient noise level; Q is air quality index; and Q is light intensity.

[0070] Specifically, based on the real-time collected and preprocessed channel monitoring data and environmental impact data, the channel performance index, network performance index, and environmental impact index are calculated by substituting them into the channel performance index formula, network performance index formula, and environmental impact index formula. The maximum and minimum signal strength values, the maximum device temperature, and the maximum battery power consumption are based on the maximum and minimum values ​​of multiple signal strength, device temperature, and battery power consumption collected from q types of portable Wi-Fi devices, respectively; battery life is obtained based on the actual manufacturer or experimental testing of the i-th type of portable Wi-Fi device.

[0071] In this embodiment, step S3 involves a real-time evaluation of channel quality based on a combination of channel performance index, network performance index, and environmental impact index to generate a comprehensive score for the portable Wi-Fi channel quality. This includes the following steps:

[0072] Formula for calculating the overall channel quality score:

[0073] Obtain the overall channel quality score Where Y, W, and H represent the channel performance index, network performance index, and environmental impact index, respectively; , , These represent the weighting coefficients for the channel performance index, network performance index, and environmental impact index, respectively. It is the channel occupancy rate; The channel occupancy adjustment factor is 0.6, and t is time.

[0074] Specifically, the comprehensive channel quality score is obtained by substituting the calculated channel performance index, network performance index, and environmental impact index into the comprehensive channel quality scoring formula. In this implementation, the weighting coefficients for the channel performance index, network performance index, and environmental impact index are 0.5, 0.3, and 0.2, respectively. These weighting coefficients will be dynamically adjusted in actual applications or based on historical data and other factors. Time t represents the cumulative time from the moment the portable Wi-Fi device is started to the current moment.

[0075] In this embodiment, step S3, which uses time series analysis combined with a deep learning model to predict user data traffic demand and calculates peak traffic demand, includes the following steps:

[0076] The historical time-series datasets collected include: historical traffic consumption time series, device movement GPS coordinate time series, environmental noise intensity, temperature time series data, humidity time series data, number of online devices, and channel occupancy rate; the collected historical time-series data are preprocessed, including: data cleaning and standardization, and alignment by timestamp to form a unified time step sequence;

[0077] By combining time series analysis with deep learning models, a spatiotemporal attention network model is constructed to predict users' data traffic needs. In the time dimension, a Long Short-Term Memory (LSTM) network is used to process the traffic sequence and extract long-term dependency features. In the spatial dimension, the device movement GPS coordinate time series is converted into a trajectory graph by connecting continuous GPS points as edges with edge weights equal to the reciprocal of the Euclidean distance. The device movement trajectory is then modeled using a graph neural network to capture spatial correlations.

[0078] The preprocessed time series dataset is input into the spatiotemporal attention network model for training; based on the trained spatiotemporal attention network model, the preprocessed time series data collected in real time is input into the trained spatiotemporal attention network model, and the output prediction results are the predicted data flow demand value and timestamp.

[0079] The forecast results are smoothed using the moving average method or the exponential smoothing method. Based on the smoothed forecast results, the peak point of traffic demand is identified, that is, the time point when traffic demand reaches its maximum value and the corresponding traffic value. Finally, the peak traffic demand is calculated.

[0080] Specifically, historical time-series datasets are collected, including the following data types: historical traffic consumption time series (recording user traffic consumption at past points in time), device movement GPS coordinate time series (recording device GPS coordinates at past points in time), environmental noise intensity time series (recording environmental noise), temperature time series (recording temperature), humidity time series (recording humidity), number of online devices time series (recording the number of online devices), and channel occupancy time series (recording channel occupancy). Preprocessing of the collected historical time-series data includes data cleaning to remove outliers, imputing missing values, and standardizing the data to the same scale; ensuring all data are aligned by timestamps to form a unified time step sequence. Combining time-series analysis and deep learning models, a spatiotemporal attention network model is constructed: a long short-term memory network is used to process the traffic sequence and extract long-term dependency features. Spatial dimension processing includes: connecting the device movement GPS coordinate time series with consecutive GPS points as edges, with edge weights equal to the reciprocal of the Euclidean distance, converting it into a trajectory graph. A graph neural network is used to model the device movement trajectory and capture spatial correlations. The preprocessed time-series dataset is input into the spatiotemporal attention network model for training: the preprocessed time-series data is used as model input, and the model outputs the predicted data flow demand value and the corresponding timestamp. Model parameters are adjusted using the backpropagation algorithm to optimize prediction accuracy. Real-time prediction is then performed using the trained spatiotemporal attention network model: time-series data is collected and preprocessed in real time and input into the trained model, which outputs the predicted data flow demand value and timestamp. The prediction results are smoothed, and peak points of flow demand are identified: moving average or exponential smoothing methods are used to smooth the prediction results, reducing fluctuations. Based on the smoothed prediction results, the time point when the flow demand reaches its maximum value and the corresponding flow value are identified, and the peak flow demand is calculated.

[0081] In this embodiment, step S4 employs a reinforcement learning algorithm to dynamically select the optimal channel based on the comprehensive channel quality score and peak traffic demand, including the following steps:

[0082] Based on the peak traffic demand predicted and calculated by the spatiotemporal attention network model, channels in the portable Wi-Fi channel whose bandwidth capacity is lower than the peak traffic demand are excluded.

[0083] The weighting of the overall channel quality score is dynamically adjusted based on peak traffic demand; during peak traffic demand periods, 0.6 0.3 It is 0.1; during non-peak traffic demand periods, 0.4 0.4 It is 0.2;

[0084] Based on the assigned weights, the comprehensive channel quality scores for different time periods are calculated and arranged in descending order as candidate channels; among the candidate channels, the channel with the highest comprehensive channel quality score is selected as the optimal channel.

[0085] The Q-learning algorithm in reinforcement learning is used to switch to the optimal channel. Each portable Wi-Fi device acts as an agent. In the Q-learning algorithm, the action value function Q(s,a) of each state and action pair is learned. The agent selects whether to switch to the optimal channel based on the current channel state s.

[0086] The agent performs the selected action, switching the portable Wi-Fi device to the optimal channel; after switching the portable Wi-Fi device to the selected optimal channel, communication begins.

[0087] During the switching process, communication on the current channel is maintained and unsent data packets are buffered. After the switch, the optimal channel is monitored in real time, and the immediate and long-term rewards are calculated according to the reward function of the Q-learning algorithm. The corresponding Q values ​​in the Q-value table are updated according to the immediate and long-term rewards. The convergence of the Q-value table is evaluated. If convergence is achieved, training is stopped; otherwise, iterative learning continues.

[0088] Specifically, based on the peak traffic demand predicted and calculated using the spatiotemporal attention network model, channels in the portable Wi-Fi network with bandwidth capacity lower than the peak traffic demand are excluded. The weights of the comprehensive channel quality score are dynamically adjusted based on the predicted peak traffic demand. During peak traffic demand periods, the weights are set to channel performance index 0.6, network performance index 0.3, and environmental impact index 0.1. During non-peak traffic demand periods, the weights are set to channel performance index 0.4, network performance index 0.4, and environmental impact index 0.2. A comprehensive score is calculated for each channel based on the assigned weights. The channels are then sorted in descending order of their comprehensive scores to form a candidate channel list. The channel with the highest comprehensive score is selected as the optimal channel. A Q-learning algorithm is used to switch to the optimal channel, with each portable Wi-Fi device acting as an agent. Its state is the current channel state, and its action is whether to switch to the optimal channel. The immediate reward in the reward function is the comprehensive channel quality score after the switch, and the long-term reward is calculated based on the average traffic transmission rate or average comprehensive channel quality score over a period of time or the past 10 channel switches. The Q-value table is updated according to the reward function. The agent selects an action based on the current state, evaluates the reward after executing the action, and updates the Q-value table until the Q-value table converges. The agent executes the selected action, switching the portable WiFi device to the optimal channel. During the switch, communication on the current channel is maintained, and unsent data packets are buffered. After the switch, the quality of the optimal channel is monitored in real time, and immediate and long-term rewards are calculated based on the reward function. The corresponding Q-values ​​in the Q-value table are updated based on the immediate and long-term rewards. Training stops when the Q-value table converges; otherwise, iterative learning continues.

[0089] In this embodiment, step S4, based on the selected optimal channel, utilizes reinforcement learning techniques combined with a deep learning model to identify congestion and dynamically adjust the channel, including the following steps:

[0090] Based on the selected optimal channel, channel scanning technology is used to collect optimal channel status information in real time, including: channel utilization, bandwidth utilization, channel capacity, handover delay, and data loss. The collected data is then preprocessed, including: data cleaning and normalization.

[0091] By combining reinforcement learning techniques with deep learning models, a congestion identification model based on deep reinforcement learning is constructed. The historical best channel state information set with labels indicating whether channel congestion has occurred is input into the congestion identification model based on deep reinforcement learning for training. The best channel state information set collected in real time and preprocessed is input into the trained congestion identification model based on deep reinforcement learning to output the congestion identification result.

[0092] If no channel congestion is detected, communication continues on the current channel. If channel congestion is detected, a binary exponential backoff algorithm is used. Based on the binary exponential backoff algorithm, an adaptive mechanism is introduced to dynamically adjust the backoff window size according to the real-time channel status information and the degree of congestion. The degree of congestion is calculated by weighted average summation of channel utilization, bandwidth utilization, channel capacity, handover delay, and data loss.

[0093] Formula for dynamically adjusting the backoff window:

[0094] in, It is the size of the backoff window when the k-th channel collision or congestion occurs; It refers to the degree of congestion;

[0095] When the congestion level exceeds a congestion threshold, a dynamic adjustment mechanism for the backoff window size is triggered, adjusting the backoff window size to... Otherwise, adjust the backoff window size to... ;in, The congestion threshold is 0.65. That represents the maximum level of congestion.

[0096] Specifically, channel scanning technology is used to collect optimal channel state information in real time, including channel utilization, bandwidth utilization, channel capacity, handover latency, and data loss. The collected data is cleaned and normalized to ensure accuracy and consistency. A congestion identification model based on deep reinforcement learning is constructed by combining reinforcement learning techniques with deep learning models. A set of historical optimal channel state information, labeled with whether channel congestion has occurred, is input into the model for training, enabling the model to learn the characteristics and patterns of channel congestion. By collecting and preprocessing the optimal channel state information in real time, the model is input into the trained congestion identification model, which outputs a congestion identification result to determine whether the current channel is congested. If no channel congestion is identified, communication continues on the current channel. If channel congestion is identified, an improved binary exponential backoff algorithm is used for channel adjustment. Based on the traditional binary exponential backoff algorithm, an adaptive mechanism is introduced to dynamically adjust the backoff window size according to the real-time channel state information and the degree of congestion. The congestion level is calculated by weighted averaging and summing indicators such as channel utilization, bandwidth utilization, channel capacity, handover delay, and data loss. The weighting coefficients for channel utilization, bandwidth utilization, channel capacity, handover delay, and data loss are 0.3, 0.25, 0.2, 0.1, and 0.15, respectively, and are dynamically adjusted based on actual conditions and historical data. A dynamic adjustment mechanism for the backoff window size is triggered based on a congestion level threshold. The backoff window size is adjusted according to the backoff window size adjustment formula and the congestion level, with a threshold of 0.65, dynamically adjusted based on historical data and actual conditions. A new channel is selected for handover based on the adjusted backoff window. During the handover process, communication on the current channel is maintained, and unsent data packets are buffered. After switching to the new channel, communication is restored, and channel status continues to be monitored. Based on real-time monitoring of channel status and handover effectiveness, the congestion identification model and backoff algorithm are continuously updated and optimized to improve model accuracy and algorithm efficiency.

[0097] In this embodiment, step S5 involves real-time monitoring of the adjusted optimal channel, calculating the effect score, and evaluating the effect level of the optimal channel, including the following steps:

[0098] Real-time monitoring of channel monitoring data and environmental impact data of the adjusted optimal channel is used, and the channel quality comprehensive score is calculated by substituting these data into the channel quality comprehensive scoring formula. A dynamic evaluation cycle is set to trigger every 5 seconds for effect evaluation. The effect scoring formula is as follows:

[0099] Receive performance rating ;in, This represents the overall channel quality score before channel adjustment. This represents the overall channel quality score after channel adjustment. To indicate the degree of congestion; It is a congestion level threshold;

[0100] Based on the performance rating, three levels are divided: Excellent, Good, and Poor; when When, the grade is excellent; when When, the grade is good; when When the evaluation level is poor, the system returns to step S4 to readjust the channel selection.

[0101] Specifically, real-time monitoring of the optimal channel and environmental impact data after adjustment is used to calculate the comprehensive channel quality score using the channel quality comprehensive scoring formula. A dynamic evaluation cycle is set to trigger every 5 seconds for periodic effect evaluation. Within each evaluation cycle, channel status data and environmental impact data are collected, and the comprehensive channel quality score is calculated. Based on the comprehensive channel quality score before and after channel adjustment, congestion level, and congestion level threshold, the effect score is calculated using the effect scoring formula. The effect score threshold is dynamically adjusted based on actual conditions and the mean and variance of historical data. When the effect level is excellent or good, the current channel continues to be used and monitoring continues; when the effect level is poor, the process returns to step S4 to reselect and adjust the channel.

[0102] Please see Figure 2 As shown, the present invention is an adaptive adjustment system based on a portable Wi-Fi channel, comprising the following modules:

[0103] Monitoring and data acquisition module: Real-time monitoring and acquisition of channel monitoring data and environmental impact data of the portable Wi-Fi channel;

[0104] Data preprocessing module: Performs data cleaning and standardization preprocessing on the collected channel monitoring data and environmental impact data respectively;

[0105] Data Analysis Module: Calculates channel performance index, network performance index, and environmental impact index based on preprocessed channel monitoring data and environmental impact data; performs real-time evaluation of channel quality based on the channel performance index, network performance index, and environmental impact index, generating a comprehensive score for the portable Wi-Fi channel quality; predicts user data traffic demand through time series analysis combined with a deep learning model, and calculates the peak traffic demand.

[0106] Selection and adjustment module: Employs reinforcement learning algorithm to dynamically select the optimal channel based on channel quality score and predicted peak traffic demand; Based on the selected optimal channel, combines reinforcement learning technology with a deep learning model to identify congestion and dynamically adjust the channel accordingly.

[0107] Monitoring and Evaluation Module: Monitors the adjusted optimal channel in real time, calculates the effect score, and evaluates the effect level of the optimal channel.

[0108] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for adaptive adjustment based on a personal wifi channel, characterized in that, The method comprises the following steps: S1: Real-time monitoring and collecting channel monitoring data and environmental influence data of the body-worn WiFi channel; S2: Preprocessing the collected channel monitoring data and environmental influence data by data cleaning and standardization respectively; S3: Analyzing and calculating channel performance index, network performance index and environmental influence index according to the preprocessed channel monitoring data and environmental influence data; comprehensively evaluating the channel quality in real time according to the channel performance index, network performance index and environmental influence index, and generating a comprehensive score of the body-worn WiFi channel quality; predicting the user's data traffic demand through time series analysis combined with a deep learning model, and calculating the traffic demand peak value; S4: Dynamically selecting the optimal channel according to the channel quality score and the predicted traffic demand peak value by using a reinforcement learning algorithm; identifying whether it is congested and dynamically adjusting the channel according to the selected optimal channel by using the reinforcement learning technology combined with the deep learning model; S5: Real-time monitoring of the adjusted optimal channel, calculating the effect score and evaluating the effect level of the optimal channel; The step S3 of predicting the user's data traffic demand through time series analysis combined with a deep learning model, and calculating the traffic demand peak value comprises the following steps: Collecting historical time series data set including historical traffic consumption time series, device movement GPS coordinate time series, environmental noise intensity, temperature time series data, humidity time series data, online device number, channel occupancy rate; preprocessing the collected historical time series data including data cleaning and standardization, and aligning by timestamp to form a unified time step sequence; Combining time series analysis with a deep learning model to construct a spatio-temporal attention network model to predict the user's data traffic demand; in the time dimension, using long short-term memory network LSTM to process the traffic sequence and extract long-term dependence features; in the space dimension, connecting the device movement GPS coordinate time series as edges with the edge weight being the inverse of the Euclidean distance to convert into a trajectory graph, modeling the device movement trajectory through a graph neural network to capture spatial correlation; Inputting the preprocessed time series data set into the spatio-temporal attention network model for training; according to the trained spatio-temporal attention network model, inputting the real-time monitoring and preprocessing time series data into the trained spatio-temporal attention network model to output the prediction result as the data traffic demand prediction value and timestamp; Using the moving average method or exponential smoothing method to smooth the prediction result, identifying the peak point of the traffic demand according to the smoothed prediction result, i.e. the time point and corresponding traffic value at which the traffic demand reaches the maximum value, and finally calculating the traffic demand peak value.

2. The method of claim 1, wherein, The channel monitoring data in step S1 includes: the number of devices on the channel, device movement speed, signal strength, channel occupancy rate, device battery level, device temperature and network traffic data; the environmental influence data includes: environmental temperature, environmental humidity, environmental noise value, air quality index and illumination intensity; wherein the network traffic data includes: bandwidth utilization rate, data packet loss rate, network throughput and delay, and transmission efficiency.

3. The method of claim 2, wherein, The step S3 includes the following steps of analyzing and calculating the channel performance index, the network performance index and the environmental impact index according to the pre-processed channel monitoring data and the environmental impact data: The network performance index W is obtained by calculating the channel performance index formula: obtaining a channel performance index Y, wherein, is the signal strength of the i-th device; and respectively represent the maximum and minimum values of the signal strength; is the total number of devices; is the moving speed of the i-th device; is the device temperature of the i-th device; is the maximum value of the device temperature; Z is the channel occupancy rate; represents the battery power consumption of the i-th device; is the maximum value of the battery power consumption; is the battery life of the i-th device; The network performance index is calculated by the formula: wherein k is the bandwidth utilization rate; d is the data packet loss rate; L is the network throughput; c is the delay; and x is the transmission efficiency; The environmental impact index is calculated by the formula: An environmental impact index H is obtained, where is the ambient temperature; is the ambient humidity; β is the ambient noise value; I is the air quality index, Q is the light intensity.

4. The method of claim 3, wherein, The step S3 includes the following steps of comprehensively evaluating the channel quality in real time according to the channel performance index, the network performance index and the environmental impact index, and generating a comprehensive score of the on-body WiFi channel quality: The channel quality comprehensive score formula is calculated as: obtaining a channel quality comprehensive score wherein Y, W and H represent channel performance index, network performance index and environmental impact index, respectively; , , Y, W and H represent weight coefficients of channel performance index, network performance index and environmental impact index, respectively; is channel occupancy rate; is an adjustment coefficient of channel occupancy rate, and t is time.

5. The method of claim 1, wherein, The step S4 includes the following steps of dynamically selecting an optimal channel according to the comprehensive score of the channel quality and the traffic demand peak value by using a reinforcement learning algorithm: The traffic demand peak value is predicted and calculated according to the spatiotemporal attention network model, and the channels in the on-body WiFi channel whose bandwidth capacity is lower than the traffic demand peak value are excluded. According to the peak of traffic demand, dynamically adjust the weight of channel quality comprehensive score; in the peak of traffic demand period, 0.6, 0.3, 0.1; in the non-peak of traffic demand period, 0.4, 0.4, 0.2; According to the assigned weights, the channel quality comprehensive scores in different time periods are calculated and ranked in descending order as candidate channels; in the candidate channels, the channel with the highest channel quality comprehensive score is selected as the optimal channel; The Q-learning algorithm in the reinforcement learning algorithm is used to switch the optimal channel; wherein each on-body WiFi device is an agent; in the Q-learning algorithm, the action value function Q(s, a) of each state and action pair is learned; the agent selects whether to switch to the optimal channel according to the current channel state s and the action a; The agent executes the selected action to switch the on-body WiFi device to the optimal channel; after the on-body WiFi device is switched to the selected optimal channel, communication is started; During the switching process, communication on the current channel is maintained, and unsent data packets are cached; after switching, the optimal channel is monitored in real time, and the immediate reward and long-term reward are calculated according to the reward function of the Q-learning algorithm; the corresponding Q value in the Q value table is updated according to the immediate reward and long-term reward; the convergence of the Q value table is evaluated, and when the convergence is reached, the training is stopped, otherwise the iterative learning is continued.

6. The method of claim 5, wherein, The step S4 includes the following steps of identifying whether the channel is congested and dynamically adjusting the channel by combining the reinforcement learning technology with the deep learning model according to the selected optimal channel: According to the selected optimal channel, the channel scanning technology is used to collect the optimal channel state information in real time, including the channel utilization rate, the bandwidth utilization rate, the channel capacity, the switching delay and the data loss amount; the collected data is preprocessed, including data cleaning and normalization; The reinforcement learning technology is combined with the deep learning model to construct a congestion identification model based on deep reinforcement learning; the historical optimal channel state information set with a channel congestion label is input into the congestion identification model based on deep reinforcement learning for training; the real-time collected and preprocessed optimal channel state information set is input into the trained congestion identification model based on deep reinforcement learning to output the congestion identification result; and The reinforcement learning technology is combined with the deep learning model to construct a congestion identification model based on deep reinforcement learning; the historical optimal channel state information set with a channel congestion label is input into the congestion identification model based on deep reinforcement learning for training; the real-time collected and preprocessed optimal channel state information set is input into the trained congestion identification model based on deep reinforcement learning to output the congestion identification result; and When the channel congestion is not identified, the communication on the current channel is continued; when the channel congestion is identified, a backoff algorithm of binary exponential backoff is adopted, on the basis of which an adaptive mechanism is introduced, and the size of the backoff window is dynamically adjusted according to the real-time state information and the congestion degree of the channel; wherein the congestion degree is calculated by weighted average summation of the channel utilization, the bandwidth utilization, the channel capacity, the switching delay and the data loss amount; Backoff window dynamic adjustment formula: wherein, is the size of the backoff window at the kth channel conflict or congestion; is the congestion level; When the congestion level is greater than the congestion level threshold, a mechanism of dynamically adjusting the size of the backoff window is triggered, and the backoff window size is adjusted to ; otherwise, the backoff window size is adjusted to ; wherein, is the congestion level threshold, and is the maximum congestion level.

7. The method of claim 1, wherein, The step S5 comprises the following steps of: Real-time monitoring of the adjusted optimal channel channel monitoring data and environmental impact data, into the channel quality comprehensive score formula to calculate the channel quality comprehensive score; set the dynamic evaluation period to trigger once every 5 seconds, evaluate the effect; calculate the effect score formula: Effect score ; wherein, represents the channel quality comprehensive score before channel adjustment; represents the channel quality comprehensive score after channel adjustment; is the congestion degree; is the congestion degree threshold value; According to the effect score, three levels are divided including: excellent, good and poor; when the level is excellent; when the level is good; when the level is poor; wherein, when the evaluation level is poor, return to step S4 to re-adjust the channel selection.

8. A system for adaptive adjustment of a personal hotspot channel, using a method for adaptive adjustment of a personal hotspot channel according to any one of claims 1 to 7, characterized in that The system comprises the following modules: The monitoring and collecting module is configured to monitor and collect the channel monitoring data and the environmental influence data of the personal wifi channel in real time; The data preprocessing module is configured to perform data cleaning and standardization preprocessing on the collected channel monitoring data and environmental influence data; The data analysis module is configured to analyze and calculate the channel performance index, the network performance index and the environmental influence index according to the preprocessed channel monitoring data and environmental influence data; to perform real-time evaluation on the channel quality according to the channel performance index, the network performance index and the environmental influence index, and to generate a comprehensive score of the personal wifi channel quality; to predict the data traffic demand of the user through time series analysis combined with a deep learning model, and to calculate the traffic demand peak value; The selection and adjustment module is configured to dynamically select the optimal channel according to the channel quality score and the predicted traffic demand peak value by using a reinforcement learning algorithm; to identify whether the selected optimal channel is congested and to dynamically adjust the channel by using the reinforcement learning technology combined with the deep learning model; The monitoring and evaluation module is configured to monitor the adjusted optimal channel in real time, to calculate the effect score and to evaluate the effect level of the optimal channel.

Images