A sleep management method and management system of a WiFi device

Abstract

The application provides a kind of WiFi device's dormancy management method and management system, it is related to WiFi device dormancy technical field, the application is through the collection multidimensional load data and carries out equidistant discretization and standardization, fusion LSTM model short-term prediction and autoregressive moving average model long-term trend analysis, combine short-term load data, long-term load trend and the idle probability calculated by multilayer perception neural network, calculate the idle index of equipment, the application is also through the analysis of combining WiFi signal strength, through the smooth processing of signal strength and the modeling of the relationship with equipment performance, determine the minimum received signal strength threshold of equipment, finally formulate comprehensive dormancy rule, when the idle index of equipment exceeds the set threshold and signal strength is lower than minimum received signal strength, equipment enters dormancy mode, this decision mechanism ensures that equipment automatically enters low-power state in idle and poor signal condition, reduces energy waste.

Description

Technical Field

[0001] This invention relates to the field of WiFi device hibernation technology, specifically to a WiFi device hibernation management method and management system. Background Technology

[0002] With the rapid development of wireless communication technology and the widespread adoption of mobile devices, WiFi devices have become an important component of modern network infrastructure. However, traditional WiFi device management methods generally suffer from inadequate energy consumption management, especially when the device load is low or the network signal quality is poor, the device still maintains a high power consumption state, resulting in significant energy waste.

[0003] To address the challenges of equipment load forecasting and intelligent management, existing technologies have proposed various solutions. Existing technology CN113220450A discloses a load forecasting method for cloud-based multi-datacenter systems. This method combines a neural network model and an autoregressive model to calculate the nonlinear and linear components of load forecasting, thereby improving forecasting accuracy. Existing technology CN119854813A proposes a base station load forecasting method combining deep learning and statistical methods. It decomposes historical base station load data into long-term trend and short-term random fluctuation components, using LSTM networks and statistical methods for forecasting respectively. Existing technology CN120751468A discloses a power-saving system based on a load forecasting intelligent sleep strategy. By constructing a multimodal service load forecasting model and a dynamic threshold mechanism, it achieves intelligent power-saving management. Furthermore, CN117331684A proposes a CPU load forecasting method based on an LSTM-Attention model, utilizing an attention mechanism to enhance the LSTM network's ability to predict CPU load time series. The prior art disclosed in Chinese publication number CN113821344A is a cluster load prediction method based on machine learning, which combines short-term and long-term predictions through the parallel combination of autoregressive random forest model and long short-term memory model.

[0004] While existing technologies have made some progress in load forecasting and device management, the following shortcomings remain: First, current load forecasting methods are mainly designed for specific scenarios such as data centers, base stations, or CPUs, lacking comprehensive load analysis methods specifically tailored to the characteristics of WiFi devices. This makes it difficult to effectively handle the complex relationships between multiple load metrics of WiFi devices, such as CPU utilization, memory usage, and data transmission / reception volume. Second, when making hibernation decisions, existing technologies often consider only a single factor, either load status or signal strength, lacking a comprehensive decision-making mechanism that combines device load status with network signal quality, resulting in inaccurate hibernation strategies. Finally, existing methods lack accurate quantitative assessment of the idle state of WiFi devices, making it impossible to establish an effective idle index calculation model and achieve intelligent hibernation management based on the actual operating status of the devices. These problems prevent WiFi devices from dynamically adjusting their operating modes according to their own load status and network environment in practical applications, leading to unnecessary energy consumption and affecting the overall performance and lifespan of the devices.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for managing the hibernation of WiFi devices, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for managing the sleep state of a WiFi device, comprising the following steps:

[0009] Step 1: Collect data before the current time. The load data of the WiFi devices to be managed within a certain time period includes: CPU utilization, memory utilization, data reception volume and data transmission volume. The load data is discretized at equal time intervals and preprocessed to obtain load time-series data.

[0010] Step 2: Based on the load time series data, construct an LSTM model to input the current time data, predict the load time series data for the next time period, and combine it with the autoregressive moving average model to analyze the long-term load trend of the equipment. The autoregressive moving average model models the time series trend of load data through autoregression and moving average to predict the future load change trend.

[0011] Step 3: Construct a multilayer perceptron neural network model to calculate the current idle probability of the device based on the input load data, and output a probability value representing whether the device is idle. Combine the short-term load data predicted by the LSTM model, the load trend analyzed by the autoregressive moving average model, and the idle probability calculated by the multilayer perceptron neural network model to calculate the idle index of the device.

[0012] Step 4: Regularly scan the WiFi signal strength, smooth the obtained signal strength, analyze the relationship between signal strength and device performance, derive the impact model of signal strength on performance, and set the minimum received signal strength threshold;

[0013] Step 5: Combine the idle index and signal strength to formulate a sleep decision rule: when the idle index exceeds the set threshold and the signal strength is lower than the minimum threshold, the device enters sleep mode.

[0014] Furthermore, collect data before the current moment. The load data of the WiFi devices to be managed within a given time period includes CPU utilization, memory usage, data reception, and data transmission. The load data is then discretized at equal time intervals and preprocessed. The specific steps for discretizing the load data at equal time intervals include: setting a total time interval. That is, collecting data before the current time. Load data within a given time period; data acquisition interval is [missing information]. Set the target discrete time window ,Will Time data points are divided into For each window, the data points are resampled and the mean is calculated.

[0015]

[0016] In the formula, Discretized data points, For time The original data values ​​within, The number of samples; where The specific formula is:

[0017]

[0018] In the formula, The data collection time interval The target discrete-time window is defined. Data preprocessing uses data processing tools to perform outlier removal, missing value imputation, and data standardization on the discretized data. Outlier removal uses the Z-score method, calculating the Z-score for each feature and removing data points with a Z-score greater than 3. The specific formula used for the calculation is as follows:

[0019]

[0020] In the formula, Indicates the degree of deviation of a data point from the mean. For the discretized data points, The mean of all data points after discretization. is the standard deviation of all data points after discretization; missing value imputation uses the KNN algorithm to find the k nearest neighbors and fill in the missing values ​​with their mean; missing value imputation uses the KNN algorithm to find the k nearest neighbors and fill in the missing values ​​with their mean.

[0021] The data points after outlier removal and missing value imputation form a new set of data points. Data standardization transforms all data points in the new set into a standard normal distribution using Z-score standardization, with the specific formula being:

[0022]

[0023] In the formula, Represents the standardized data points. These are data points that have been discretized and then subjected to outlier removal and missing value imputation. This is the mean of all data points that have been discretized and then processed by outlier removal and missing value imputation. It represents the standard deviation of all data points that have been discretized and then had outlier removed and missing values ​​filled.

[0024] Furthermore, based on the past The steps for building an LSTM model from load time series data within a given time period and predicting load data for the next time period include: Time data points are divided into Servings, each serving lasts for 1 hour. ; Enter the current time before Time-based load data; output the next... Load data over time;

[0025] The specific formula for the autoregressive moving average model is as follows:

[0026]

[0027] In the formula, The load on the device at time t, ,…, The autoregressive coefficient represents the past... The impact of the state at each point in time on the current state. ,…, The moving average coefficient represents the past... The impact of errors at each point in time on the current state This is the white noise error term.

[0028] Furthermore, a multilayer perceptron neural network model is constructed to calculate the probability of a device being idle at the current moment based on the input load data, and outputs a probability value representing whether the device is idle. The calculation formula of the multilayer perceptron neural network model is as follows:

[0029]

[0030] In the formula, This represents the idle probability of the device. For the discretized data points, This is the weight matrix. This is a bias term.

[0031] Furthermore, combining short-term load data predicted by the LSTM model, load trends analyzed by the autoregressive moving average model, and idle probability calculated by the multilayer perceptron neural network model, the idle index of the equipment is calculated. The specific formula for calculating the idle index of the equipment is as follows:

[0032]

[0033] In the formula, The idle index, The device idle probability predicted by the multilayer perceptron neural network model. This represents the normalized value of the load. The load trend is predicted by the autoregressive moving average model. This is the adjustment coefficient.

[0034] Furthermore, the WiFi signal strength is periodically scanned, and the obtained signal strength is smoothed. The signal strength is obtained through the WiFi interface, and the specific formula is as follows:

[0035]

[0036] In the formula, For a moment Received signal strength For reference power, The received signal power;

[0037] For the directly obtained signal strength To smooth the process, we use an exponentially weighted moving average method. The specific formula is as follows:

[0038]

[0039] In the formula, The signal strength after smoothing. This is a smoothing factor.

[0040] Furthermore, by analyzing the relationship between signal strength and device performance, a model of the impact of signal strength on performance is derived. A minimum received signal strength threshold is set, where the device operates inefficiently when the signal strength is below this threshold. The relationship between signal strength and device performance is expressed by throughput, which represents the data transmission rate of the device under a certain signal strength. The throughput formula is:

[0041]

[0042] In the formula, For throughput, Maximum throughput represents the maximum data transmission rate of the device under ideal signal strength. The attenuation coefficient represents the degree to which signal strength affects throughput;

[0043] Inefficient operating state is when a certain signal strength When the device's throughput is less than 20% of its maximum throughput, the signal strength... This is the minimum received signal strength threshold.

[0044] Furthermore, combining the idle index and signal strength, a sleep mode decision rule is formulated: when the idle index exceeds a set threshold and the signal strength is lower than a minimum threshold, the device enters sleep mode. The sleep mode decision is formulated using a background service that monitors the idle index and signal strength in real time. The sleep conditions are as follows:

[0045]

[0046] The logic for entering hibernation mode is as follows:

[0047] when When the signal strength of the WiFi device is considered good, it indicates that the WiFi device can be used normally.

[0048] when When the WiFi device is determined to be in an idle state and the signal strength is below the minimum threshold, it indicates that the WiFi device should enter sleep mode.

[0049] when When the WiFi device is not in an idle state, it is determined that the WiFi device can be used normally; when the hibernation conditions are met, the system shuts down the WiFi interface and enters hibernation mode via a command.

[0050] The present invention also provides a sleep management system for a WiFi device, the sleep management system for executing the above-described sleep management method for a WiFi device, comprising:

[0051] Data acquisition and preprocessing module: used to acquire data up to the current time. The load data of the WiFi devices to be managed within a certain time period includes: CPU utilization, memory utilization, data reception volume and data transmission volume. The load data is discretized at equal time intervals and preprocessed to obtain load time-series data.

[0052] Load forecasting module: Based on load time series data, it builds an LSTM model to predict the load time series data for the next time period by taking the current time data as input. It also combines the autoregressive moving average model to analyze the long-term load trend of the equipment. The autoregressive moving average model models the time series trend of load data through autoregression and moving average to predict the future load change trend.

[0053] Idle Probability Calculation Module: This module is used to build a multilayer perceptron neural network model that calculates the current idle probability of a device based on the input load data. It outputs a probability value indicating whether the device is idle. The module combines short-term load data predicted by the LSTM model, load trends analyzed by the autoregressive moving average model, and the idle probability calculated by the multilayer perceptron neural network model to calculate the idle index of the device.

[0054] Signal scanning and evaluation module: used to periodically scan WiFi signal strength, smooth the obtained signal strength, analyze the relationship between signal strength and device performance, derive the impact model of signal strength on performance, and set the minimum received signal strength threshold;

[0055] Hibernation Decision and Execution Module: Used to combine idle index and signal strength to formulate hibernation decision rules: when the idle index exceeds the set threshold and the signal strength is lower than the minimum threshold, the device enters hibernation mode.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] This invention provides a hibernation management method and system for WiFi devices, relating to the field of WiFi device hibernation technology. This invention collects multi-dimensional load data and performs equal-interval discretization and standardization. It integrates short-term prediction using an LSTM model with long-term trend analysis using an autoregressive moving average model. Combining short-term load data, long-term load trends, and idle probability calculated by a multilayer perceptron neural network, it calculates the device's idle index. Furthermore, this invention analyzes WiFi signal strength, determining the minimum received signal strength threshold by smoothing the signal strength and modeling its relationship with device performance. Finally, it formulates a comprehensive hibernation rule: when the device's idle index exceeds the set threshold and the signal strength is lower than the minimum received signal strength, the device enters hibernation mode. This decision-making mechanism ensures that the device automatically enters a low-power state when idle and with poor signal, reducing energy waste and thus improving the device's energy efficiency and lifespan. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0059] Figure 2 This is a diagram of dormancy decision data from an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0062] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0063] Example:

[0064] Please see Figures 1-2 The present invention provides a technical solution:

[0065] A method for managing the sleep state of a WiFi device, comprising the following steps:

[0066] Step 1: Collect data before the current time. The load data of the WiFi devices to be managed within a certain time period includes: CPU utilization, memory utilization, data reception volume and data transmission volume. The load data is discretized at equal time intervals and preprocessed to obtain load time-series data.

[0067] In this embodiment, a monitoring tool on a WiFi device is used to acquire values ​​in real time, collecting data every 10 seconds from the current moment. Load data of the WiFi devices to be managed within a given time period, including CPU utilization, memory usage, data reception, and data transmission, is collected by discretizing the load data at equal time intervals and preprocessing it. Load data acquisition includes:

[0068] CPU utilization: By monitoring the CPU usage of WiFi devices, the percentage of CPU utilization is recorded periodically;

[0069] Memory usage: Records the percentage of memory used by the device;

[0070] Data received: Records the amount of data received by the WiFi device through the network;

[0071] Data transmission volume: Records the amount of network data transmitted by the device. This embodiment collects... The original collected data is shown in Table 1 below:

[0072] Table 1: Original Data Collection Table

[0073]

[0074] This table records the raw operational data of the WiFi device over 20 minutes, with 20 time points (one sampling point per minute). The data includes four load metrics: CPU utilization gradually decreased from 15.2% to 3.0%, reflecting the typical process of the device transitioning from an active state to an idle state; memory utilization decreased from 32.1% to 22.2%, showing a stable downward trend; data reception and transmission decreased from 2.1MB and 0.8MB respectively to near zero, indicating a significant reduction in network activity. The raw signal strength gradually attenuated from -45dBm to -90dBm, simulating scenarios where the device is far from the WiFi access point or the signal environment deteriorates. This raw data provides the basic input for subsequent data processing, model training, and decision analysis, realistically reflecting the device's state changes during actual operation.

[0075] The specific steps for discretizing the load data at equal time intervals include: setting the total time interval. That is, collecting data before the current time. Load data within a given time period; data acquisition interval is [missing information]. Set the target discrete time window ,Will Time data points are divided into For each window, the data points are resampled and the mean is calculated.

[0076]

[0077] In the formula, Discretized data points, For time The original data values ​​within, The number of samples; where The specific formula is:

[0078]

[0079] In the formula, The data collection time interval The target discrete-time window. In this embodiment, , The arithmetic mean is used instead of the median or maximum because load data typically follows a normal distribution, and the mean best represents the typical load state within the window. Window size The choice requires a trade-off: too small a value results in poor noise reduction, while too large a value may lead to the loss of important fluctuation information.

[0080] Data preprocessing uses data processing tools (such as OpenCV) to perform outlier removal, missing value imputation, and data standardization on the discretized data. Outlier removal uses the Z-score method, calculating the mean for each feature. and standard deviation Then, the Z-score of each feature is calculated, and data points with a Z-score greater than 3 are removed. Outlier removal can avoid the influence of noise during model training. The specific formula used for the calculation is as follows:

[0081]

[0082] In the formula, Indicates the degree of deviation of a data point from the mean. For the discretized data points, The mean of all data points after discretization. Z represents the standard deviation of all data points after discretization. Missing value imputation uses the KNN algorithm to find the k nearest neighbors and fill the missing values ​​with their mean. If the threshold is too small (e.g., |Z|>2), normal fluctuations will be mistakenly deleted; if it is too large (e.g., |Z|>4), anomalies may be missed.

[0083] If the dataset contains missing values, the KNN algorithm is used to impute the missing values. The k nearest neighbors are found and their mean is used to fill in the missing values.

[0084] The data points after outlier removal and missing value imputation form a new set of data points. Data standardization, through Z-score standardization, transforms all discretized data points that have undergone outlier removal and missing value imputation into a standard normal distribution. The specific formula is as follows:

[0085]

[0086] In the formula, Represents the standardized data points. These are data points that have been discretized and then subjected to outlier removal and missing value imputation. This is the mean of all data points that have been discretized and then processed by outlier removal and missing value imputation. Standardization is the standard deviation of all data points that have been discretized and then imputed for outliers and missing values. After standardization, data with different features are transformed into a distribution with a mean of 0 and a variance of 1, which facilitates subsequent model training. Standardization brings all features to the same order of magnitude, preventing certain large numerical features (such as the amount of data sent and received) from dominating model training. Without standardization, gradient descent may converge slowly or be unstable.

[0087] Existing technologies often employ simple threshold filtering or linear interpolation, which have limited processing effectiveness. This solution uses a combined approach of Z-score anomaly detection, KNN missing data imputation, and Z-score standardization: Z-score anomaly detection effectively identifies statistically significant outliers; KNN imputation considers the correlation between features and is more accurate than mean imputation; standardization makes features of different dimensions comparable, accelerating model convergence; it also eliminates data quality issues, ensures the consistency of model input, and improves prediction stability. In this embodiment, the mean and standard deviation are calculated based on the first 15 data points, and the Z-score method is used to standardize the load data in Table 1. The results are shown in Table 2 below.

[0088] Table 2: Standardized Load Characteristic Data Table

[0089]

[0090] This table displays the load data after standardization. Standardization was performed using the Z-score method, where... and This calculation was based on historical data from the previous 15 time points. After processing, the standardized value of CPU utilization changed from -0.82 to -1.85, memory utilization decreased from -0.75 to -1.75, and data reception and transmission volumes showed similar standardization trends. Negative values ​​indicate that the data points are below the historical average, and the absolute value of the standardized values ​​gradually increases over time, indicating that the device load is continuously decreasing relative to the historical average. This standardization process eliminates the influence of different units of measurement, making different features such as CPU, memory, and network traffic comparable, providing a unified input format for subsequent machine learning models (such as LSTM and MLP), and improving the training efficiency and prediction accuracy of the models. Note: The standardized values ​​in Table 2 are calculated based on the discretized data points. The historical data of the remaining 15 time points after outlier removal were then filled with missing values ​​to obtain the complete 20 time points. The standardized values ​​shown in Table 2 are only exemplary values ​​used to illustrate the trend of the standardized data. Therefore, some values ​​do not completely reach a mean of 0. This is due to the limited training samples and does not affect the implementation and effect of the present invention. In practical applications, the standardized data should meet the distribution characteristics of a mean of 0 and a variance of 1.

[0091] Traditional methods often employ fixed-period sampling or simple sliding windows, which can easily overlook short-term load fluctuations or introduce noise. This approach achieves a balance between data denoising and trend preservation by discretizing at equal time intervals and applying window averaging. Compared to existing technologies, its advantages include: reducing the interference of high-frequency noise on subsequent prediction models, maintaining the overall trend of load data and avoiding information loss, and providing regular time-series input for subsequent LSTM models.

[0092] Step 2: Based on the load time series data, construct an LSTM model to input the current time data, predict the load time series data for the next time period, and combine it with the autoregressive moving average model to analyze the long-term load trend of the equipment. The autoregressive moving average model models the time series trend of load data through autoregression and moving average to predict the future load change trend.

[0093] The input to the LSTM model is the time step before the current time step. The LSTM model outputs the load data for the next time window, based on the load data for each second. That is, the predicted values ​​of CPU utilization, memory utilization, data reception volume, and data transmission volume.

[0094] The steps to build an LSTM model that takes current time data as input and predicts load data for the next time period include: Time data points are divided into Servings, each serving lasts for 1 hour. ; Enter the current time before Time-based load data; output the next... Load data over time.

[0095] During training, historical load data (such as data from the past hour) is used as input, and the parameters of the LSTM network are updated through the backpropagation algorithm. The specific formula for training using mean squared error as the loss function is as follows:

[0096]

[0097] In the formula, The load predicted by the LSTM model; mean squared error. It is used to measure the deviation between the predicted value and the actual value.

[0098] A well-trained LSTM model can predict the future based on the current load data. Load within seconds helps identify upcoming load spikes.

[0099] The autoregressive moving average model is used to model the load data. This model is suitable for handling autocorrelation and trend issues in time series data. The specific formula for the autoregressive moving average model is as follows:

[0100]

[0101] In the formula, The load on the device at time t, ,…, The autoregressive coefficient represents the past... The impact of the state at each point in time on the current state. ,…, The moving average coefficient represents the past... The impact of errors at each point in time on the current state This is the white noise error term.

[0102] Parameter estimation of the autoregressive moving average model is performed using historical load data. The model parameters are calculated using maximum likelihood estimation or least squares method. and The model's fit is ensured by checking the white noise nature of the residuals through model diagnostics (such as ACF and PACF plots).

[0103] Autoregressive moving average models can predict load trends over a future period based on historical load data, providing a reference for long-term load trends for LSTM models.

[0104] Traditional methods such as ARIMA or simple regression can only capture linear relationships. LSTM, as a variant of recurrent neural networks, is particularly suitable for sequence prediction. Its advantages include: capturing long-term dependencies in the load data, having a strong ability to fit nonlinear patterns, and selectively remembering and forgetting information through gating mechanisms.

[0105] Step 3: Construct a multilayer perceptron neural network model based on the input load data to calculate the current idle probability of the device, and output a probability value representing whether the device is idle. Combine the short-term load data predicted by the LSTM model, the load trend analyzed by the autoregressive moving average model, and the idle probability calculated by the multilayer perceptron neural network model to calculate the idle index of the device.

[0106] The input to the multilayer perceptron neural network model is the load data at the current time step. The output is the device's idle probability. In other words, whether the device is currently idle, the calculation formula for the multilayer perceptron neural network model is as follows:

[0107]

[0108] In the formula, This represents the idle probability of the device. For the discretized data points, This is the weight matrix. This is a bias term.

[0109] Combining short-term load data predicted by the LSTM model, load trends analyzed by the autoregressive moving average model, and idle probability calculated by the multilayer perceptron neural network model, the idle index of the equipment is calculated. The specific formula for calculating the idle index of the equipment is as follows:

[0110]

[0111] In the formula, The idle index, The device idle probability predicted by the multilayer perceptron neural network model. This represents the normalized value of the load. The load trend is predicted by the autoregressive moving average model. This is the adjustment coefficient.

[0112] The current idle probability is high and the predicted future load is low; if the predicted future load is high, the idle index will decrease even if the current idle probability is high. This involves adjusting for long-term trends; if the long-term trend indicates that the load will increase... If so, the idle index should be appropriately reduced; The weighting of long-term trends should be carefully controlled; too high a weight may ignore short-term opportunities, while too low a weight will prevent the utilization of trend information. Experiment with different approaches on the validation set. Choose values ​​(e.g., 0.1, 0.2, 0.3) that best correlate the idle index with the actual idle state. The larger the value, the stronger the long-term trend's influence on decision-making. If the equipment load shows a continuous downward trend, even if the current idle probability is low, a higher value will still influence the decision-making process. Hibernation can be triggered in advance, in this embodiment In this embodiment, based on historical data, LSTM, ARMA, and MLP models output load prediction, trend analysis, and idle probability, respectively. Key intermediate indicators are shown in Table 3 below.

[0113] Table 3: Multi-model Prediction Results and Intermediate Indicator Calculation Table

[0114]

[0115] This table summarizes the output results and key intermediate metrics of the three prediction models. The LSTM model predicts the load at the next time step based on historical load data, with the predicted value gradually decreasing from 0.82 to 0.08, reflecting the short-term load decline trend; the ARMA model analyzes the long-term load trend. The value decreased from 0.75 to 0.02; the idle probability calculated by the MLP model. It rose from 0.15 to 0.98. The table also includes " The term "item" represents the complement of the future load, and " The term "" indicates the weighted contribution of the long-term trend. It is particularly important to note that LSTM prediction is not possible when k=1, which reflects the reality that prediction relies on historical data. These indicators provide complete input for the subsequent calculation of the idle index, demonstrating the intermediate computational process of the multi-model fusion method.

[0116] By constructing an idle probability calculation model for a multilayer perceptron and integrating it with LSTM short-term prediction and ARMA long-term trend analysis to generate an idle index, the output of continuous probability values ​​(0-1) can provide a grayscale basis for decision-making. For example, when the idle probability is 0.75, it can be combined with signal strength to determine whether to sleep. When the probability is 0.6, the decision can be delayed and monitoring can continue to avoid frequent state switching.

[0117] Step 4: Regularly scan the WiFi signal strength, smooth the obtained signal strength, analyze the relationship between signal strength and device performance, derive the impact model of signal strength on performance, and set the minimum received signal strength threshold.

[0118] The device periodically collects the current signal strength value to measure the strength of the WiFi signal. The signal strength value is obtained through the WiFi interface, and the specific calculation formula is as follows:

[0119]

[0120] In the formula, For a moment Received signal strength For reference power, This represents the received signal power.

[0121] Signal strength fluctuations can be caused by: multipath effect: interference caused by signal reflection and refraction, resulting in instantaneous fluctuations; environmental interference: electromagnetic noise from other wireless devices (such as Bluetooth and microwave ovens); and device movement: signal attenuation caused by the displacement of handheld devices or mobile terminals.

[0122] To remove noise, the directly obtained signal strength To smooth the process, we use an exponentially weighted moving average method. The specific formula is as follows:

[0123]

[0124]

[0125] In the formula, The signal strength after smoothing. As a smoothing factor, 0 < ≤1; A larger value indicates a faster response to new data. Approaching 1, it almost exclusively uses current observations, resulting in a fast response but large fluctuations; Approaching 0: Historical information dominates, resulting in good smoothing but with response latency. In WiFi signal scenarios, this embodiment... =0.3, balancing response speed and stability.

[0126] By analyzing the relationship between signal strength and device performance, a model of the impact of signal strength on performance is derived. A minimum received signal strength threshold is set, where the device operates inefficiently when the signal strength is below this threshold. The relationship between signal strength and device performance is expressed by throughput, which represents the data transmission rate of the device under a certain signal strength and is a core indicator for measuring the communication efficiency of the device. The throughput formula is:

[0127]

[0128] In the formula, For throughput, Maximum throughput represents the maximum data transmission rate of the device under ideal signal strength. The attenuation coefficient represents the degree to which signal strength affects throughput. In this embodiment, the maximum throughput is 100Mbps.

[0129] Wireless channel capacity increases approximately exponentially with the signal-to-noise ratio, therefore an exponential model is used. Attenuation coefficient. Fitting based on measured data: Throughput was measured at multiple RSSI values, and the exponential curve was fitted using the least squares method.

[0130] Inefficient operating state is when a certain signal strength When the device's throughput is less than 20% of its maximum throughput, the signal strength... This is the minimum received signal strength threshold.

[0131] Step 5: Combine the idle index and signal strength to formulate a sleep decision rule: when the idle index exceeds the set threshold and the signal strength is lower than the minimum threshold, the device enters sleep mode.

[0132] Furthermore, combining the idle index and signal strength, a sleep mode decision rule is formulated: when the idle index exceeds a set threshold and the signal strength is lower than a minimum threshold, the device enters sleep mode. The sleep mode decision is formulated using a background service that monitors the idle index and signal strength in real time. The sleep conditions are as follows:

[0133]

[0134] The logic for entering hibernation mode is as follows:

[0135] when When the signal strength of the WiFi device is considered good, it indicates that the WiFi device can be used normally.

[0136] when When the WiFi device is determined to be in an idle state and the signal strength is below the minimum threshold, it indicates that the WiFi device should enter sleep mode.

[0137] when When the WiFi device is not in an idle state, it is determined that the WiFi device can be used normally. When the sleep conditions are met, the system shuts down the WiFi interface and enters sleep mode via command. Idle threshold 0.8: indicates high confidence in idleness. Setting it to 0.7 may increase the risk of false sleep, while 0.9 may cause the sleep opportunity to be missed. Signal threshold -75dBm: based on measured data showing that the throughput of typical WiFi devices drops to 20% to 30% at -75dBm.

[0138] Traditional hibernation strategies typically rely on a single condition (such as no activity for a timeout). The advantage of this two-condition decision-making approach is that it considers both device load and network environment quality, avoiding unnecessary hibernation under good signal conditions and preventing invalid connections under poor signal conditions. In this embodiment, the idle index is calculated based on the combined prediction results and signal strength, and a hibernation decision is executed. The final evaluation results are shown in Table 4 below:

[0139] Table 4: Results of Idle Index Calculation and Dormancy Decision

[0140]

[0141] Please see Figure 2 This is a hibernation decision data diagram from an embodiment of the present invention, reflecting the final decision-making process for WiFi device hibernation management as shown in Table 4. Table 4 shows that the idle index gradually increases from an initial 0.150 to 0.935, reflecting an increase in the device's idle level. The signal strength is smoothed using an exponentially weighted moving average. The signal strength decreased from -45.0 dBm to -85.0 dBm, resulting in a corresponding decrease in throughput from 90.5 Mbps to 47.0 Mbps. The decision rule was: sleep mode was triggered when the idle index was ≥0.8 and the smoothed signal strength was <-75 dBm. The results showed that the sleep mode conditions were not met at the first 16 time points, but both conditions were met at time points 17-20, leading the system to enter sleep mode. This decision-making process reflects the core logic of the technical solution: sleep mode is only executed when the device is idle and the signal quality is poor, balancing energy saving requirements with performance assurance.

[0142] Please see Figure 3 The present invention also provides a sleep management system for a WiFi device, the WiFi device sleep management system being used to execute the above-described WiFi device sleep management method, comprising:

[0143] Data acquisition and preprocessing module: used to acquire data up to the current time. The load data of the WiFi devices to be managed within a certain time period includes: CPU utilization, memory utilization, data reception volume and data transmission volume. The load data is discretized at equal time intervals and preprocessed to obtain load time-series data.

[0144] Load forecasting module: Based on load time series data, it builds an LSTM model to predict the load time series data for the next time period by taking the current time data as input. It also combines the autoregressive moving average model to analyze the long-term load trend of the equipment. The autoregressive moving average model models the time series trend of load data through autoregression and moving average to predict the future load change trend.

[0145] Idle Probability Calculation Module: This module is used to build a multilayer perceptron neural network model that calculates the current idle probability of a device based on the input load data. It outputs a probability value indicating whether the device is idle. The module combines short-term load data predicted by the LSTM model, load trends analyzed by the autoregressive moving average model, and the idle probability calculated by the multilayer perceptron neural network model to calculate the idle index of the device.

[0146] Signal scanning and evaluation module: used to periodically scan WiFi signal strength, smooth the obtained signal strength, analyze the relationship between signal strength and device performance, derive the impact model of signal strength on performance, and set the minimum received signal strength threshold;

[0147] Hibernation Decision and Execution Module: Used to combine idle index and signal strength to formulate hibernation decision rules: when the idle index exceeds the set threshold and the signal strength is lower than the minimum threshold, the device enters hibernation mode.

[0148] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0149] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0150] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0151] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for managing the sleep state of a WiFi device, characterized in that, The specific steps include: Step 1: Collect data before the current time. The load data of the WiFi devices to be managed within a certain time period includes: CPU utilization, memory utilization, data reception volume and data transmission volume. The load data is discretized at equal time intervals and preprocessed to obtain load time-series data. Step 2: Based on the load time series data, construct an LSTM model to input the current time data, predict the load time series data for the next time period, and combine it with the autoregressive moving average model to analyze the long-term load trend of the equipment. The autoregressive moving average model models the time series trend of load data through autoregression and moving average to predict the future load change trend. Step 3: Construct a multilayer perceptron neural network model to calculate the current idle probability of the device based on the input load data, and output a probability value representing whether the device is idle. Combine the short-term load data predicted by the LSTM model, the load trend analyzed by the autoregressive moving average model, and the idle probability calculated by the multilayer perceptron neural network model to calculate the idle index of the device. Step 4: Regularly scan the WiFi signal strength, smooth the obtained signal strength, analyze the relationship between signal strength and device performance, derive the impact model of signal strength on performance, and set the minimum received signal strength threshold; Step 5: Combine idle index and signal strength to formulate sleep decision rules: when the idle index exceeds the set threshold and the signal strength is lower than the minimum threshold, the device enters sleep mode; Based on the past The steps for building an LSTM model from load time series data within a given time period and predicting load data for the next time period include: Time data points are divided into Servings, each serving lasts for 1 hour. ; Enter the current time before Time-based load data; output the next... Load data over time; The specific formula for the autoregressive moving average model is as follows: In the formula, The load on the device at time t, ,…, The autoregressive coefficient represents the past... The impact of the state at each point in time on the current state. ,…, The moving average coefficient represents the past... The impact of errors at each point in time on the current state This is the white noise error term; A multilayer perceptron neural network model is constructed to calculate the probability of a device being idle at the current moment based on the input load data, and the output is a probability value representing whether the device is idle. The calculation formula of the multilayer perceptron neural network model is as follows: In the formula, This represents the idle probability of the device. For the discretized data points, This is the weight matrix. For bias terms; Combining short-term load data predicted by the LSTM model, load trends analyzed by the autoregressive moving average model, and idle probability calculated by the multilayer perceptron neural network model, the idle index of the equipment is calculated. The specific formula for calculating the idle index of the equipment is as follows: In the formula, The idle index, The device idle probability predicted by the multilayer perceptron neural network model. This represents the normalized value of the load. The load trend is predicted by the autoregressive moving average model. This is the adjustment coefficient; Regularly scan the WiFi signal strength and smooth the obtained signal strength. The signal strength is obtained through the WiFi interface, and the specific formula is as follows: In the formula, For a moment Received signal strength For reference power, The received signal power; For the directly obtained signal strength To smooth the process, we use an exponentially weighted moving average method. The specific formula is as follows: In the formula, The signal strength after smoothing. It is a smoothing factor; By analyzing the relationship between signal strength and device performance, a model of the impact of signal strength on performance is derived. A minimum received signal strength threshold is set, where the device operates inefficiently when the signal strength is below this threshold. The relationship between signal strength and device performance is expressed by throughput, which represents the data transmission rate of the device under a certain signal strength. The throughput formula is: In the formula, For throughput, Maximum throughput represents the maximum data transmission rate of the device under ideal signal strength. The attenuation coefficient represents the degree to which signal strength affects throughput; Inefficient operating state is when a certain signal strength When the device's throughput is less than 20% of its maximum throughput, the signal strength... This is the minimum received signal strength threshold.

2. The hibernation management method for a WiFi device according to claim 1, characterized in that: Collect data before the current time. The load data of the WiFi devices to be managed within a given time period includes CPU utilization, memory usage, data reception, and data transmission. The load data is then discretized at equal time intervals and preprocessed. The specific steps for discretizing the load data at equal time intervals include: setting a total time interval. That is, collecting data before the current time. Load data within a given time period; data acquisition interval is [missing information]. Set the target discrete time window ,Will Time data points are divided into For each window, the data points are resampled and the mean is calculated. In the formula, Discretized data points, For time The original data values ​​within, The number of samples; where The specific formula is: In the formula, The data collection time interval The target discrete-time window is defined. Data preprocessing uses data processing tools to perform outlier removal, missing value imputation, and data standardization on the discretized data. Outlier removal uses the Z-score method, calculating the Z-score for each feature and removing data points with a Z-score greater than 3. The specific formula used for the calculation is as follows: In the formula, Indicates the degree of deviation of a data point from the mean. For the discretized data points, The mean of all data points after discretization. The standard deviation of all data points after discretization; The data points after outlier removal and missing value imputation form a new set of data points. Data standardization, through Z-score standardization, transforms all data points in this new set into a standard normal distribution. The specific formula is as follows: In the formula, Represents the standardized data points. These are data points that have been discretized and then subjected to outlier removal and missing value imputation. This is the mean of all data points that have been discretized and then processed by outlier removal and missing value imputation. It represents the standard deviation of all data points that have been discretized and then had outlier removed and missing values ​​filled.

3. The hibernation management method for a WiFi device according to claim 1, characterized in that: Based on idle index and signal strength, a sleep mode decision rule is established: when the idle index exceeds a set threshold and the signal strength is below a minimum threshold, the device enters sleep mode. The sleep mode decision is formulated using a background service that monitors the idle index and signal strength in real time. The sleep conditions are as follows: The logic for entering hibernation mode is as follows: when When the signal strength of the WiFi device is considered good, it indicates that the WiFi device can be used normally. when When the WiFi device is determined to be in an idle state and the signal strength is below the minimum threshold, it indicates that the WiFi device should enter sleep mode. when When the WiFi device is not in an idle state, it is determined that the WiFi device can be used normally; when the hibernation conditions are met, the system shuts down the WiFi interface and enters hibernation mode via a command.

4. A sleep management system for a WiFi device, characterized in that: The WiFi device sleep management system is used to execute a WiFi device sleep management method according to any one of claims 1-3, including: Data acquisition and preprocessing module: used to acquire data up to the current time. The load data of the WiFi devices to be managed within a certain time period includes: CPU utilization, memory utilization, data reception volume and data transmission volume. The load data is discretized at equal time intervals and preprocessed to obtain load time-series data. Load forecasting module: Based on load time series data, it builds an LSTM model to predict the load time series data for the next time period by taking the current time data as input. It also combines the autoregressive moving average model to analyze the long-term load trend of the equipment. The autoregressive moving average model models the time series trend of load data through autoregression and moving average to predict the future load change trend. Idle Probability Calculation Module: This module is used to build a multilayer perceptron neural network model that calculates the current idle probability of a device based on the input load data. It outputs a probability value indicating whether the device is idle. The module combines short-term load data predicted by the LSTM model, load trends analyzed by the autoregressive moving average model, and the idle probability calculated by the multilayer perceptron neural network model to calculate the idle index of the device. Signal scanning and evaluation module: used to periodically scan WiFi signal strength, smooth the obtained signal strength, analyze the relationship between signal strength and device performance, derive the impact model of signal strength on performance, and set the minimum received signal strength threshold; Hibernation Decision and Execution Module: Used to combine idle index and signal strength to formulate hibernation decision rules: when the idle index exceeds the set threshold and the signal strength is lower than the minimum threshold, the device enters hibernation mode.

Images