Wearable device power consumption adaptive management method and system based on deep learning

Abstract

The invention relates to the technical field of power consumption management, in particular to a wearable device power consumption adaptive management method and system based on deep learning, and the method comprises the following steps: obtaining skin resistance, gait frequency, heart rate variation coefficient and body temperature data, calculating fluctuation degree and marking stability, adjusting stable parameter sampling frequency to generate dynamic sampling frequency information; evaluating the remaining electric quantity and the charging time to obtain operation grades, classifying and identifying user behaviors, and combining behavior and grade matching to generate a periodic power consumption control instruction set. According to the method, parameters in a stable state are screened based on physiological data volatility indexes, the sampling frequency of the parameters is compressed, the scarcity degree of the equipment operation electric quantity is calculated by combining the residual electric quantity and the predicted charging time difference value, and the user behavior identification tag is matched with the operation level identifier; the energy consumption control strategy is accurately matched according to the actual electric quantity state and the user behavior characteristics, the self-adaptive cruising ability of the equipment in the daily use period is enhanced, and the power consumption waste is reduced.

Description

Technical Field

[0001] This invention relates to the field of power management technology, and in particular to a method and system for adaptive power management of wearable devices based on deep learning. Background Technology

[0002] The core objective of power management technology is to optimize the control of power usage efficiency in electronic devices, especially mobile devices. Key technologies in this field include dynamic voltage and frequency regulation, power gating, task scheduling optimization, power consumption modeling and estimation, and multi-core collaborative power control. Its applications are wide-ranging, encompassing smartphones, wearable devices, IoT terminals, and embedded systems, and it is particularly significant in devices sensitive to battery life or with limited power resources. This technology typically employs a hardware-software collaborative approach to reduce energy consumption and improve energy efficiency while ensuring performance meets specific quality requirements, thereby extending device operating time or reducing energy consumption.

[0003] Among them, the adaptive power consumption management method for wearable devices is a method for dynamically adjusting the energy consumption behavior of wearable devices. This method adaptively adjusts parameters such as processor operating status, functional module activation strategies, data communication frequency, and sensor duty cycles by sensing multiple influencing factors such as device operating status, task priority, and user behavior patterns. This achieves overall power consumption reduction while maintaining the basic functions of the wearable device and the user experience. Its main applications are extending battery life, reducing charging frequency, and improving the continuous usability of wearable devices in scenarios such as health monitoring, exercise recording, and voice interaction.

[0004] Traditional management methods tend to analyze device status and user behavior in a macroscopic and static way, failing to make precise judgments based on the stability of multiple physiological indicators. This results in sensors operating at a fixed sampling frequency even within the parameter stability range, leading to a waste of data collection resources. When the battery level is close to the critical value, they also rely solely on a preset low-power mode for uniform processing, failing to consider the differences in user behavior for differentiated energy consumption adjustment. This causes high-frequency communication and high-power processing to continue under static or low-load conditions, reducing power utilization and shortening actual usable time. For example, keeping motion monitoring sensors constantly on when users are at rest or in office settings will cause power consumption to accumulate and not be effectively released, making it difficult to support the device's operation across cycles. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a deep learning-based adaptive power management method and system for wearable devices.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a deep learning-based adaptive power consumption management method for wearable devices, comprising the following steps: S1: Acquire the data collected by the wearable device in the current cycle, calculate the fluctuation value of each parameter, identify parameters whose fluctuation value is lower than the stability judgment threshold, and mark them with stability labels to obtain a physiological stability label group; S2: Call the physiological stability tag group, and calculate the product of the original sampling frequency value of the parameter marked as stable and the set sampling frequency compression ratio to adjust the sampling frequency of the stable parameter and maintain the sampling frequency of the remaining data items to obtain dynamic sampling frequency information. S3: Based on the dynamic sampling frequency information, calculate the degree of power shortage of the device according to the ratio of the remaining power value to the continuous usage time, compare it with the set operating level range, and mark it as a restricted level, intermediate level or normal level to obtain the operating level identification group. S4: Invoke the physiological stability label group, use a multi-class recognition network structure based on convolutional neural network as the classification tool, use labeled sample data for supervised learning training, perform classification prediction on the current input dataset, output the behavior category of the user in the current period, and generate a behavior recognition label set.

[0007] As a further aspect of the present invention, the physiological stability label group includes a cycle sequence number, a physiological item name label, a stability label status, and the number of cycles in which the label status lasts. The dynamic sampling frequency information includes a parameter name, the original sampling frequency, the compressed frequency value, and the compression activation cycle number. The operation level identifier group includes a cycle operation number, a power consumption to time ratio range level, a level-corresponding operation mode label, and a level identifier activation duration period. The behavior identification label set specifically includes commuting behavior labels, office behavior labels, exercise behavior labels, and resting behavior labels.

[0008] As a further aspect of the present invention, the steps for obtaining the physiological stability tag set are as follows: S101: Acquire the data collected by the wearable device in the current cycle. The collected data includes skin resistance value, gait frequency, heart rate coefficient of variation and body temperature. Calculate the standard deviation using the set of time point values ​​of each data item in the current cycle to obtain the standard deviation value set in the cycle. S102: Call the standard deviation value group within the period, collect the average standard deviation of the same parameter in the previous three periods based on the current standard deviation value of the four parameters, perform difference calculation between the current value and the average value, and calculate the fluctuation value of the parameter by combining the collection period number and the signal-to-noise ratio of the device status, and obtain the parameter fluctuation value group. S103: Based on the parameter fluctuation value group, determine whether the fluctuation of the parameter is less than the set stability judgment threshold, filter the parameters that meet the judgment conditions and assign them a stability label, integrate the parameter name, the current collection cycle number, and the stability label status, establish a set corresponding to the cycle data and the stability label, and obtain the physiological stability label group.

[0009] As a further aspect of the present invention, the formula for obtaining the fluctuation value of the parameter through calculation is specifically as follows: ; in, This indicates the degree of fluctuation of the parameter. This represents the standard deviation of the parameter within the current period. This represents the standard deviation of the same parameter over the first three periods. This represents the normalized value of the signal-to-noise ratio. This indicates a regulatory factor.

[0010] As a further aspect of the present invention, the step of obtaining the dynamic sampling frequency information specifically includes: S201: Call the physiological stability label group, filter the parameter items with stable label status, extract the acquisition number, parameter name and original sampling frequency value of the corresponding parameter, establish the frequency basic data of the stable parameter in the current cycle, and generate a stable parameter frequency information set; S202: Based on the stable parameter frequency information set, obtain the original sampling frequency value and the set sampling frequency compression ratio value, perform the product calculation between the frequency compression ratio and the original sampling frequency value, form the compressed sampling frequency value of each stable parameter, and obtain the compressed sampling frequency value group. S203: Call the compressed sampling frequency value group and merge it with the original sampling frequency values ​​of parameters that are not marked as stable in the current period to establish a correspondence structure between sampling frequency items and parameter items, and uniformly organize the actual sampling frequencies of all parameters in the current period to obtain dynamic sampling frequency information.

[0011] As a further aspect of the present invention, the step of obtaining the operation level identifier group specifically includes: S301: Based on the dynamic sampling frequency information, obtain the remaining battery percentage of the wearable device in the current period, extract the normal charging time point set by the wearer according to the preset charging plan, calculate the time difference between the current system time and the normal charging time point, and obtain the continuous usage time information. S302: Based on the continuous usage duration information, according to the remaining power percentage of the device in the current cycle, combined with the total power consumption of the sampling unit of the device in the current cycle and the energy consumption for maintaining the communication connection, calculate and obtain the power shortage level value. The formula for calculating the degree of power shortage is as follows: ; in, This indicates the degree of insufficient operating power. Indicates the remaining battery percentage. This represents the normalized value indicating the duration of continuous use. This represents the normalized value of the total power consumption of the device's sampling module. This represents the normalized value of power consumption for the communication connection. Indicates the operating strategy adjustment factor; S303: Based on the power shortage level value and the operating level determination benchmark range, determine the range affiliation, mark the current cycle equipment operating status as restricted level, intermediate level or normal level, organize the correspondence between level and cycle number, and obtain the operating level identifier group.

[0012] As a further aspect of the present invention, the step of obtaining the behavior recognition tag set specifically includes: S401: Call the physiological stability tag group to obtain the original numerical sequence of skin resistance, heart rate variability coefficient, body temperature and gait frequency parameters that are determined to be valid in the current cycle, and perform normalization processing on each sequence, align the sequences with a uniform step size, and reconstruct them into a two-dimensional input data structure to obtain the behavioral data structure set. S402: Based on the behavior data structure set, each normalized parameter value is used as a training input sample, and combined with the commuting, office, sports and rest behavior categories marked by the sample data, supervised training processing is performed on the input sample to build a classification structure model. The parameter structure of the current cycle is input, the forward propagation process is executed, and the cycle prediction probability distribution information is obtained. S403: Based on the periodic prediction probability distribution information, select the item with the highest probability from the four types of behavior tags, use the category as the user behavior identifier for the current period, and bind it to the current period number to generate a behavior recognition tag set.

[0013] As a further aspect of the present invention, the method further includes the following steps: S5: Call the behavior recognition tag set and the operation level identifier group, and according to the behavior level combination mapping rule, retrieve and select four parameters: Bluetooth communication cycle, WiFi connection interval, sensor activation time and main processor operating frequency, and integrate them into a single-cycle device power consumption control parameter set to obtain a cycle power consumption control instruction set. The cycle power consumption control instruction set includes Bluetooth communication frequency control, WiFi enable interval setting, sensor start / stop control, and main processor frequency adjustment.

[0014] As a further aspect of the present invention, the step of obtaining the cycle power consumption control instruction set specifically includes: S501: Call the behavior recognition tag set and the operation level identifier group, extract the corresponding user behavior category and operation level mark in the current period, construct the behavior and level combination pair, and generate a unique index field to obtain the behavior level combination index information; S502: Based on the behavior level combination index information, use the control mapping table to obtain the four parameter values ​​corresponding to the current index: Bluetooth communication cycle value, WiFi connection interval value, sensor activation time value, and main processor operating frequency value, and obtain the power consumption control parameter value group. S503: Based on the power consumption control parameter value group, perform periodic parameter structure mapping and organization on the four parameter values, generate a unified setting item for the current acquisition cycle device operation configuration, establish a correspondence structure between parameter names and control values, and obtain the periodic power consumption control instruction set.

[0015] A deep learning-based adaptive power management system for wearable devices is provided to implement the aforementioned deep learning-based adaptive power management method for wearable devices. The system includes: The stability tag identification module acquires the data collected by the wearable device in the current cycle, calculates the fluctuation value of each parameter, identifies parameters whose fluctuation value is lower than the stability judgment threshold, and marks them with stability tags to obtain a physiological stability tag group. The sampling frequency optimization module calls the physiological stability label group, calculates the product of the original sampling frequency value of the parameter marked as stable and the set sampling frequency compression ratio, adjusts the sampling frequency of the stable parameter, and maintains the sampling frequency of the remaining data items to obtain dynamic sampling frequency information. The operation status identification module, based on the dynamic sampling frequency information, calculates the degree of power shortage of the device according to the ratio of the remaining power value to the continuous usage time, compares it with the set operation level range, and marks it as a restricted level, intermediate level or normal level to obtain an operation level identification group. The behavior category segmentation module calls the physiological stability label group, uses a multi-class recognition network structure based on convolutional neural network as a classification tool, uses labeled sample data for supervised learning training, performs classification prediction on the current input dataset, outputs the behavior category to which the user belongs in the current period, and generates a behavior recognition label set. The cycle power consumption optimization module calls the behavior recognition tag set and the operation level identifier group, and retrieves and selects four parameters—Bluetooth communication cycle, WiFi connection interval, sensor activation time, and main processor operating frequency—according to the behavior level combination mapping rules. These parameters are then integrated into a single-cycle device power consumption control parameter set to obtain a cycle power consumption control instruction set.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, parameters in a stable state are selected based on physiological data fluctuation indicators, and their sampling frequency is compressed. This reduces redundant invalid data collection while ensuring the preservation of core information. The degree of power shortage of the device is calculated by combining the difference between the remaining power and the expected charging time, and then classified into a preset level range to match the conditions for subsequent strategy invocation. By matching user behavior recognition tags with operating level identifiers, multiple control parameter combinations are dynamically selected to achieve periodic adjustment of multi-dimensional energy consumption parameters such as communication frequency, connection interval, sensor activation duration, and processing frequency. Energy consumption control strategies are accurately matched according to the actual power status and user behavior characteristics, improving the adaptability and resolution of energy consumption adjustment response, enhancing the device's adaptive endurance capability during daily use, reducing power waste, extending the device's continuous operating time, and optimizing battery efficiency in high-frequency scenarios. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0018] Figure 1 This is a schematic diagram of the workflow of the present invention; Figure 2 This is a detailed flowchart of S1 of the present invention; Figure 3 This is a detailed flowchart of the S2 process of the present invention; Figure 4 This is a detailed flowchart of the S3 process of the present invention; Figure 5 This is a detailed flowchart of the S4 process of the present invention; Figure 6 This is a detailed flowchart of S5 of the present invention; Figure 7 This is a system flowchart of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0020] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0021] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0022] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0023] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0024] Please see Figure 1 This invention provides a technical solution: a deep learning-based adaptive power consumption management method for wearable devices, comprising the following steps: S1: Acquire data collected by the wearable device in the current cycle. The collected data includes skin resistance value, gait frequency, heart rate coefficient of variation and body temperature. Calculate the standard deviation of each parameter in the current cycle and compare it with the mean standard deviation of the previous three cycles. Calculate the fluctuation value of each parameter, identify parameters whose fluctuation value is lower than the stability judgment threshold, and mark them with a stability label to obtain a physiological stability label group. Gait frequency represents the frequency of a user's steps repeated per unit time; heart rate coefficient of variation is a standard indicator that represents the dispersion of heart rate time intervals; stability judgment threshold is the lower limit of the standard deviation fluctuation range set, and the reference value is the standard deviation of the mean. S2: Call the physiological stability label group, and calculate the product of the original sampling frequency value of the parameter marked as stable and the set sampling frequency compression ratio to adjust the sampling frequency of the stable parameter and maintain the sampling frequency of the remaining data items to obtain dynamic sampling frequency information. S3: Based on dynamic sampling frequency information, obtain the remaining power of the wearable device, combine it with the wearer's preset normal charging time point, calculate the difference between the current time and the normal charging time point, calculate the device's power shortage level based on the ratio of the remaining power value to the continuous use time, compare it with the set operating level range, and mark it as a restricted level, intermediate level or normal level to obtain the operating level label group. S4: Call the physiological stability label group. Based on skin resistance, heart rate variability, body temperature and gait frequency parameters, use a multi-class recognition network structure based on convolutional neural network as a classification tool. Supervised learning training is performed using labeled sample data. After forward inference, the model performs classification prediction on the current input dataset and outputs the user's behavior category in the current period, including commuting, office work, exercise and rest, and generates a behavior recognition label set. The convolutional neural network used has been widely used in biological signal sequence classification tasks, and it can extract local features in the time dimension for behavior classification. S5: Call the behavior recognition tag set and the operation level identifier group to obtain the user behavior type and operation level within the same period. According to the preset behavior level combination mapping rules, perform rule retrieval and selection on four parameters: Bluetooth communication cycle, WiFi connection interval, sensor activation time and main processor operating frequency. Integrate them into a single-cycle device power consumption control parameter set to obtain a cycle power consumption control instruction set. The physiological stability label group includes the cycle sequence number, physiological item name label, stability label status, and the number of cycles in which the label status lasts. The dynamic sampling frequency information includes the parameter name, the original sampling frequency, the compressed frequency value, and the compression activation cycle number. The operation level identifier group includes the cycle operation number, the power consumption to time ratio range level, the operation mode label corresponding to the level, and the activation duration of the level identifier. The behavior recognition label set specifically includes commuting behavior labels, office behavior labels, exercise behavior labels, and resting behavior labels. The cycle power consumption control instruction set includes Bluetooth communication frequency control items, WiFi activation interval setting items, sensor start / stop control items, and main processor frequency adjustment items.

[0025] Please see Figure 2 The specific steps for obtaining the physiological stability tag set are as follows: S101: Acquire the data collected by the wearable device in the current cycle. The collected data includes skin resistance value, gait frequency, heart rate coefficient of variation and body temperature. Calculate the standard deviation using the set of time point values ​​of each data item in the current cycle to obtain the standard deviation value set in the cycle. The system acquires current cycle data for skin resistance, gait frequency, heart rate coefficient of variation, and body temperature. Specifically, it uses the current sampling cycle number set for the wearable device, for example, the 25th cycle. It extracts all sampled data sets corresponding to each physiological parameter within that cycle from the device's internal cache. Within a cycle, the device typically collects data at a frequency of 1 Hz, with a cycle length of 60 seconds. Therefore, each parameter forms a data set of length 60. For example, the skin resistance value acquisition sequence is [5.2, 5.4, 5.3, ..., 5.6], the gait frequency acquisition value is [0.8, 0.9, 1.0, ..., 1.1], and the heart rate interval time series is converted to coefficient of variation with acquisition values ​​of [0.02, 0.03, 0.04]. [,…,0.01], the body temperature sampling values ​​are [36.3,36.5,36.6,…,36.4]. The standard deviation is calculated for each of the above data sets. The operation process is as follows: first, calculate the mean of the data set, then square the difference between each sampling point and the mean and sum them up, and finally divide by the sample size and take the square root. For example, in the calculation of the standard deviation of skin resistance value, the mean is first obtained as 5.35. Then, the squares of the differences after subtracting the mean from each value are added up, and the final standard deviation is calculated to be 0.13. The standard deviation of gait frequency is 0.55, the standard deviation of heart rate coefficient of variation is 0.03, and the standard deviation of body temperature is 0.26. Finally, the standard deviation value set of the four parameters in the current cycle is constructed.

[0026] S102: Call the standard deviation value group within the period. Based on the current standard deviation values ​​of the four parameters, collect the average standard deviation value of the same parameter in the previous three periods. Perform difference calculation between the current value and the average value. Combine the collection period number and the signal-to-noise ratio of the device status to calculate and obtain the parameter fluctuation value group. The specific formula for calculating the fluctuation level of the parameters is as follows: ; in, This represents the degree of fluctuation of the parameter; the data type is real number. This represents the standard deviation of the parameter within the current period. The data type is a positive real number, and it is obtained by calculating the standard deviation of data collected at each time point within the current period. This represents the standard deviation of the same parameter over the previous three periods. The data type is positive real numbers, and it is obtained by calculating the standard deviation of the corresponding parameter data over the historical periods and taking the mean. This represents the normalized signal-to-noise ratio (SNR) value. The data type is a floating-point number, and it is obtained as the normalized calculation result of the ratio of the sampled signal energy to the noise energy. This represents the adjustment factor, which is obtained by setting a constant item based on the current operating level configuration table of the equipment. It is used to adjust the sensitivity of fluctuation calculation under different behaviors. The process retrieves the current standard deviation of each parameter in the standard deviation group within the current period. Combined with historical data recorded in the previous period, it extracts the set of standard deviations corresponding to the current parameter from the device's periodic cache for the previous three periods. For example, for skin resistance, historical period numbers 22, 23, and 24 correspond to standard deviations of 0.09, 0.11, and 0.10, respectively. The arithmetic mean of these three values ​​is calculated to obtain the mean of the previous three periods, which is 0.10. The difference between the current value (0.13) and this mean is then calculated, yielding a difference of 0.03. The absolute value of this difference is taken as the first term. Next, the normalized signal-to-noise ratio (SNR) of the device for the current period is obtained. For example, if the device's signal strength is -65dBm and noise level is -78dBm, after normalization, the SNR is calculated. Substitute it into the fluctuation judgment formula and set... Perform the operation: In the formula, This indicates the degree of fluctuation of a single physiological parameter in the current cycle, reflecting the stability of that parameter in the current cycle; the absolute value term on the left side of the formula. This measure is used to assess the deviation of the standard deviation between the current period and the mean of the previous three periods, i.e., the degree of relative deviation based on historical reference, and is expressed in the right-hand side of the multiplication. This is used to adjust the response of fluctuation level to signal-to-noise ratio and behavioral sensitivity. The higher the normalized signal-to-noise ratio, the less interference it has on fluctuation calculation. Adjust the sensitivity intensity to different signal-to-noise ratios under varying behavioral states. For example, if the behavior is resting or working, the sensitivity intensity can be set to... If the behavior is motion, it can be set as follows: .

[0027] Regulatory factors The settings are based on the recorded status in the user behavior classification tags and device operation level identifier groups within the current period, combined with the analysis of actual behavior test samples from the previous 200 periods. This analysis shows that when users are in resting or working states, signal fluctuations and power consumption fluctuations are... The fluctuation is the most stable, so it is set as the adjustment factor value for this period. This value is set to 0.2 when the historical behavior label is "exercise" and 0.3 when it is "commuting" to ensure sensitivity adjustment under different behaviors; Normalized signal-to-noise ratio The value is calculated based on the ratio of signal power to noise power (in mW) in the current sampling period. For example, if the signal power is 3.98mW and the noise power is 0.68mW in a certain period, the signal-to-noise ratio is 5.85. After normalization to the [0,1] interval, it is processed using the logarithmic compression formula as follows: ; Among the settings Substituting the values, we get .

[0028] Finally, substitute the values ​​into the calculation: First, calculate... Then add a regulating factor. The result is Find its square root. Take its reciprocal as Multiply the first difference by its reciprocal to obtain the fluctuation level value: ; The results indicate that the skin resistance value changes only slightly from the average fluctuation of the previous three cycles in the current cycle. The result, after combining the adjustment factor and the signal-to-noise ratio suppression factor, is below the set stability threshold of 0.04, which suggests that its fluctuation stability is relatively high.

[0029] As shown in Table 1: Table 1. Calculation of Parameter Fluctuation; Table 1 lists the fluctuation Δ values ​​of each parameter after comparing the standard deviation of the current period with the historical values, and shows the combined effect of the signal-to-noise ratio and the adjustment factor, which are ultimately used for the next step of stability labeling.

[0030] S103: Based on the parameter fluctuation value group, determine whether the fluctuation of the parameter is less than the set stability judgment threshold, filter the parameters that meet the judgment conditions and assign them a stability label, integrate the parameter name, the current collection cycle number, and the stability label status, establish a set corresponding to the cycle data and the stability label, and obtain the physiological stability label group. Stability assessment was performed on each parameter in the parameter fluctuation value group. The stability threshold was set at 0.04. This value was based on the analysis of experimental statistical data from 300 sampling periods. Under continuous stable behavioral states (resting and working), the standard deviation of physiological parameters fluctuated no more than 0.035 between consecutive periods. After rounding and conservative correction, this value was set to 0.04. This value can be adjusted according to the behavior category, increasing to 0.06 in the "exercise" state and setting it to 0.05 in the "commuting" state to improve the judgment tolerance and ensure that unstable conditions are not mislabeled under drastic behavioral fluctuations. Based on the current behavior label identifying it as the "working" state, the standard deviation was set at 0.04. The default threshold is 0.04. Skin resistance (0.0298), gait frequency (0.1294), heart rate coefficient of variation (0.0099), and body temperature (0.0199) are compared sequentially. If skin resistance, heart rate coefficient of variation, and body temperature are all less than the threshold of 0.04, they are marked as "stable". If gait frequency is higher than the threshold, it is not marked. The results are then stored in a structured manner with the current sampling period number. For example, if the period number is 25, the stable label structure is generated as: {skin resistance - 25 - stable, heart rate coefficient of variation - 25 - stable, body temperature - 25 - stable, gait frequency - 25 - unstable}, which finally constitutes the physiological stability label group for the current period.

[0031] Please see Figure 3 The specific steps for obtaining dynamic sampling frequency information are as follows: S201: Call the physiological stability label group, filter the parameter items with stable label status, extract the acquisition number, parameter name and original sampling frequency value of the corresponding parameter, establish the frequency basic data of stable parameters in the current cycle, and generate a stable parameter frequency information set; The physiological stability label group is invoked, and parameters with stable label status are filtered. First, stable status label information is extracted from the stability label group corresponding to the current cycle number (e.g., cycle 25). Each parameter marked "stable" in this label group is processed one by one. During the filtering operation, the filtering logic is set to determine whether the "status" field of the parameter label is equal to "stable". Parameters that meet the judgment condition proceed to the next processing step. For example, if the current cycle stability label group contains four items: skin resistance value -25 - stable, body temperature -25 - stable, heart rate variability coefficient -25 - stable, and gait frequency -25 - unstable, after judgment, the remaining three items are marked as "stable" and retained for processing. Subsequently, the acquisition number is extracted from each stable parameter. The segment, i.e., period number 25, extracts the parameter name fields, namely "skin resistance value", "body temperature", and "heart rate variability coefficient", as well as their original sampling frequency values ​​retrieved from the device's initial configuration table. This frequency is the standard sampling frequency set for each sensor during the device's initial startup, such as 1Hz for skin resistance value, 0.5Hz for body temperature, and 1Hz for heart rate variability coefficient. After extraction, the three fields are organized into triplet form, i.e., (25, skin resistance value, 1Hz), (25, body temperature, 0.5Hz), and (25, heart rate variability coefficient, 1Hz). The above triplet set is organized into a stable parameter frequency information set, and a data format structured array or key-value pair set is constructed as the basic data for the stable parameter frequency of the current period.

[0032] S202: Based on the stable parameter frequency information set, obtain the original sampling frequency value and the set sampling frequency compression ratio value, perform the product calculation between the frequency compression ratio and the original sampling frequency value, form the compressed sampling frequency value of each stable parameter, and obtain the compressed sampling frequency value group. Based on the stable parameter frequency information set, the original sampling frequency value and the system-set sampling frequency compression ratio value are obtained. First, the name of each parameter extracted from the stable parameter frequency information set is matched with the sampling frequency compression ratio configuration table in the device configuration module. The appropriate compression ratio value for the corresponding parameter under the current device power consumption state is found. This ratio value is dynamically configured according to the behavior level to which the behavior recognition module belongs. For example, under the "office" behavior tag state, the compression ratio is set to 0.6 for skin resistance, 0.5 for body temperature, and 0.7 for heart rate coefficient of variation. Under the "exercise" state, the ratio value is set to 1, i.e., no frequency compression. For the "office" state setting, the compression ratio is called to perform compressed frequency calculation. Each parameter obtains the compressed frequency through multiplication between the original sampling frequency and the compression ratio. For example, if the original frequency of skin resistance is 1Hz and the compression ratio is 0.6, the calculated sampling frequency is... Body temperature is The coefficient of variation of heart rate is The above calculation process performs multiplication operations while keeping the sampling frequency units consistent, without the need for unit conversion. After the calculation is completed, a compressed sampling frequency value group is obtained, in which the original parameter names are retained for subsequent merging process positioning. The final compressed sampling frequency value group is: {skin resistance value -0.6Hz, body temperature -0.25Hz, heart rate coefficient of variation -0.7Hz}.

[0033] S203: Call the compressed sampling frequency value group, merge it with the original sampling frequency values ​​of parameters that are not marked as stable in the current period, establish the correspondence structure between sampling frequency items and parameter items, and uniformly organize the actual sampling frequencies of all parameters in the current period to obtain dynamic sampling frequency information. The compressed sampling frequency value group is called and merged with the original sampling frequency value set of parameters not marked as stable in the current period. First, parameters not marked as stable are screened in the stability label group. In this period, it is gait frequency. Its parameter name "gait frequency" is searched for its original sampling frequency in the device configuration module. Assuming that it was originally set to 2Hz, this frequency still retains its original value of 2Hz because it has not been compressed. It is constructed as a data pair "gait frequency - 2Hz". Then, it is merged with the compressed frequency value group. The parameter name is used as the key value and the sampling frequency is the value. A sampling frequency mapping structure is constructed for all parameter items. The merging process must prevent duplicate key names and value overwriting. After deduplication by parameter name, the integrated result is: {skin resistance value -0.6Hz, body temperature -0.25Hz, heart rate coefficient of variation -0.7Hz, gait frequency -2Hz}. The constructed mapping structure corresponds one-to-one with the parameter item and its actual sampling frequency in the current period. The output is organized as the dynamic sampling frequency information of all parameters in the current period, which is read by the subsequent sampling module and used to adjust the sampling scheduling strategy.

[0034] Please see Figure 4 The specific steps for obtaining the operation level identifier group are as follows: S301: Based on dynamic sampling frequency information, obtain the remaining battery percentage of the wearable device in the current period, extract the normal charging time point set by the wearer according to the preset charging plan, calculate the time difference between the current system time and the normal charging time point, and obtain the continuous usage time information. The system retrieves the dynamic sampling frequency information to obtain the remaining battery percentage of the wearable device in the current cycle. Based on the system's preset charging plan, it extracts the normal charging time point set by the wearer. First, it retrieves the data record with cycle number 25 from the device's power consumption monitoring module to read the current battery's remaining battery percentage. This value comes from the real-time voltage measurement and capacity conversion of the device's lithium battery module. In cycle 25, the remaining battery percentage is detected as 54%. The system then accesses the user's charging intent parameters stored in the device configuration module. These parameters represent the daily charging time points entered by the user during the initial setup. For example, if the user sets the normal charging time to 21:00 on the current day and the current system time is 17:30, the time difference between the two is 3.5 hours, which translates to 210 minutes. This serves as the continuous usage time information for the current cycle. It should be noted that the charging time points are system-level settings and do not change with the cycle. The current time point is derived from the actual running timestamp obtained by the device's RTC module. Subtracting the two gives the remaining usable time, providing input for the next normalization calculation.

[0035] S302: Based on continuous usage time information, according to the remaining power percentage of the device in the current cycle, combined with the total power consumption of the sampling unit of the device in the current cycle and the power consumption for maintaining the communication connection, calculate and obtain the power shortage level value; The specific formula for calculating the degree of power shortage is as follows: ; in, This value indicates the degree of power scarcity, representing how scarce the current cycle's power supply is relative to the remaining usable time. This represents the remaining battery percentage, is a positive real number, and is obtained from the battery percentage recorded by the device's power consumption status detection module for the current cycle. This represents a normalized value indicating the duration of continuous use. It is obtained by dividing the interval between the current time and the preset charging time by the set maximum acceptable usage period. This represents the normalized value of the total power consumption of the device's sampling modules. It is obtained by dividing the sum of the power consumption of all sampling modules in the current period by the historical highest value. This represents the normalized power consumption value of the communication connection, obtained by dividing the current period's communication connection power by the maximum operating power of the communication unit. This represents the operation strategy adjustment factor. The data type is a positive real number, and the method of obtaining it is to read the periodic adjustment parameter from the operation strategy configuration according to the user-defined level of behavioral intent. Based on continuous usage time information and the remaining battery percentage of the device in the current cycle, combined with the normalized total power consumption of the device's sampling module and the normalized power consumption for Bluetooth connection maintenance in the current cycle, an expression reflecting the scarcity of battery power is constructed. First, the parameters in the formula need to be explained: Remaining battery percentage, value is 54; : Normalized value of continuous usage duration. With the maximum acceptable usage period set at 6 hours (360 minutes), the remaining usage time is 210 minutes, and the normalized value is... ; The normalized total power consumption of the sampling module is as follows: The total sampling power of all sensors in the current period is 41mW, and the maximum power consumption in historical periods is 65mW. After normalization, it is... ; Normalized power consumption for communication connection: Bluetooth connection maintenance power consumption is 7.8mW, and Bluetooth communication maximum power is 15mW. After normalization, it is... ; : Runtime strategy adjustment factor, current behavior label identified as "office", set in configuration ; Substitute each term into the formula: ; The calculations are as follows: denominator Therefore ; square root ; ; The final combination is: ; This value represents the degree of power shortage of the equipment in the current cycle. Numerically, it reflects whether there is a risk in the duration that the equipment's remaining power can support. The higher the value, the longer the time required to support the equipment with a unit of power, and the higher the risk level, which is suitable for the next step of classification.

[0036] As shown in Table 2: Table 2. Data table for calculating the current cycle's degree of scarcity; Table 2 lists all the key calculation items for the scarcity level in the current cycle. The calculation logic and numerical sources are all based on real sampling and system configuration to ensure the reliability and reproducibility of the calculations.

[0037] Formula structure and logic explanation: This formula is used to calculate the degree of power shortage in the current cycle of equipment operation. Its structure consists of three parts, each representing a different trade-off factor, which comprehensively reflects whether the power supply may be insufficient to support the next expected charging time under current usage conditions.

[0038] Item 1: ; This is the remaining energy intensity factor, where This represents the percentage of battery remaining for the current cycle. This is the normalized usage time (usage time ÷ maximum allowable usage cycle) from the current time to normal charging time. This fraction reflects the amount of electricity that can sustain the battery per unit time. The longer the time and the lower the remaining battery, the smaller this value, indicating a more strained battery. Add 1 to... The purpose is to prevent the denominator from being too small when the normalization time is close to 0, which would lead to calculation distortion. It also reflects the linear increasing effect of the demand for remaining power as the usage time increases.

[0039] Item 2: ; This is the power consumption intensity factor, which is the normalized value of the power consumption of the device's sampling module. Normalized power consumption of communication connection The product of the two components reflects the overall energy consumption load of the system during the current period. The square root treatment reflects the neutral balance of the product result and prevents an abnormally large value of a certain component from excessively amplifying the overall result. This component is a negative factor and is deducted from the overall value, representing that the higher the energy consumption, the greater the degree of scarcity.

[0040] Item 3: ; This item is the adjustment item for the operation strategy adjustment factor, in which This is an adjustment constant for the operational strategy level read by the device based on user behavior intent. It represents the user's risk tolerance or the strategy's focus. By adding one to this parameter and taking its natural logarithm, we avoid small values ​​that could excessively suppress the final result and maintain a moderately increasing trend in the adjustment factor. The larger the value, the larger this term is, and the greater its impact on the overall scarcity level. It serves as a moderate compensation, reflecting the different power requirements under certain behavioral states.

[0041] Final result This is a comprehensive value representing the current period's equipment power shortage; it is a non-linear superposition value, and its calculation logic reflects a trade-off between three aspects: Whether the battery is sufficient to last until the next charge (measured by the ratio of remaining battery power to usage time); Is the current energy consumption level too high (suppressed by the power consumption product term)? User policy tolerance or target preference (adjusted by a moderating factor).

[0042] This value is a positive real number. The higher the value, the more strained the power supply or the heavier the system load under the current strategy and state, which imposes a stronger constraint on the system to enter the energy-saving control mode later.

[0043] The innovation of this formula lies in the fact that by combining three terms—"square root suppression of power consumption product", "revealing the power shortage by the ratio of remaining power to time difference", and "logarithmic compression of behavior adjustment term"—it dynamically maps the three-dimensional balance between the device operating environment, power status, and behavioral needs, thereby forming a reliable and responsive scarcity assessment mechanism.

[0044] S303: Based on the power shortage level value and the benchmark range for determining the operating level, determine the range affiliation, mark the current cycle equipment operating status as restricted level, intermediate level or normal level, organize the correspondence between level and cycle number, and obtain the operating level identifier group; Based on the level of power deficiency and the system's defined operating level assessment range, a range judgment is made. First, the operating level assessment range setting standard used by the current device is clarified. This standard is determined by the system through testing and written into the configuration table, and its definition is as follows: Deprivation level value : Standard level; Intermediate level; : Restriction level; The degree of scarcity in this cycle Based on the above interval judgment logic, if it falls into the last interval, the running status needs to be marked as "limited level". Then, the level label and the current cycle number (25) are organized into a level identifier tuple (limited level-25). The running level labels of all cycles are accumulated and sorted in the order of time or cycle number to form a running level identifier group, which serves as the input basis for frequency adjustment, communication interval and processor frequency dynamic management when the power consumption control scheme is executed.

[0045] Please see Figure 5 The specific steps for obtaining the behavior recognition tag set are as follows: S401: Call the physiological stability label group to obtain the original numerical sequence of skin resistance, heart rate variability coefficient, body temperature and gait frequency parameters that are determined to be valid in the current cycle, and perform normalization processing on each sequence, align the sequences with a uniform step size, and reconstruct them into a two-dimensional input data structure to obtain the behavioral data structure set. The physiological stability label group is invoked to obtain the raw numerical sequences of skin resistance, heart rate variability, body temperature, and gait frequency parameters that are deemed valid within the current cycle. First, label recognition is performed on cycle number 25, filtering out parameters with a status field of "valid" or "stable." For example, in the current cycle's label group, skin resistance, heart rate variability, and body temperature are marked as stable, while gait frequency is not marked. Therefore, the first three are retained after filtering. Simultaneously, the raw numerical sequences are obtained based on the sampling data from the previous stage, with sampling frequencies of 1Hz for skin resistance, 0.5Hz for body temperature, and 0.5Hz for heart rate variability. With a coefficient of variation of 1 Hz and a sampling duration of 60 seconds, there are 60 sampling points each for skin resistance and heart rate coefficient of variation, and 30 sampling points for body temperature. For different numbers of sampling points, a uniform step-size alignment operation is performed, which involves interpolating or padding all sequences with zeros to extend them to a uniform 60 sampling points. For example, the body temperature sequence generates intermediate values ​​through spline interpolation, extending the 30 points to 60 equally spaced sampling values. After sequence alignment, normalization is performed on the sequences of the three parameters. This process uses the maximum and minimum values ​​of each sequence as the normalization benchmark, employing a linear normalization formula: ; Taking skin resistance as an example, its maximum value is 6.1 and its minimum value is 5.2. Therefore, the normalized result for a point with a value of 5.6 before normalization is: In this way, the normalization transformation of all 60 data points is completed. After the three parameters are normalized, they form vectors with a dimension of (1×60). Finally, the three vectors are aligned according to the time dimension and merged to construct a two-dimensional behavioral structure data input tensor of (3×60), which is used as the behavioral data structure set for the current period.

[0046] S402: Based on the behavioral data structure set, each normalized parameter value is used as a training input sample. Combined with the commuting, office, sports and rest behavior categories marked by the sample data, supervised training is performed on the input samples to build a complete classification structure model. Input the parameter structure of the current cycle, execute the forward propagation process, and obtain the cycle prediction probability distribution information. Based on the behavioral data structure set, the normalized parameter values ​​are used as training input samples. Combined with the commuting, office, exercise, and resting behavior categories labeled in the sample data, supervised training is performed on the input samples. First, historical sample data containing multiple periods is prepared. Each sample consists of a (4×60) two-dimensional parameter matrix containing the normalized sequence of all four types of parameters and their corresponding behavioral labels. These behavioral labels are manually confirmed during training data acquisition by user annotation or system-collected environmental information. For example, sample number 18 corresponds to the label "commuting," and number 19 to "resting." The training data volume should reach at least 500 periods to ensure sufficient model convergence. Then, all training samples are fed into the constructed feedforward neural network structure. The network input layer size is... (Where N is the number of stable and effective parameters, with a maximum of 4). Several convolutional and pooling layers are set in the middle of the network to extract temporal local features, and a four-dimensional probability vector is output through a fully connected layer, corresponding to four behavior categories. During the training process, the cross-entropy loss function is used in each iteration, and the parameters are updated by backpropagating the error based on the difference between the label category and the network output. After the pre-training stage is completed, the model saves the parameter structure for online inference. When the current cycle arrives, the (3×60) behavior data structure constructed in step 1 is input into the trained model, and the model performs a forward propagation calculation to output a behavior probability distribution vector, for example, the output is: [commuting: 0.13, office: 0.56, exercise: 0.09, rest: 0.22], which constitutes the cycle prediction probability distribution information.

[0047] S403: Based on the periodic prediction probability distribution information, select the item with the highest probability from the four types of behavior tags, use the category as the user behavior identifier for the current period, and bind it to the current period number to generate a behavior recognition tag set; Based on the periodic prediction probability distribution information, the system filters out the item with the highest probability from four types of behavior labels, and uses this category as the user behavior identifier for the current period. During this filtering operation, the system performs a comparison operation on each dimension of the probability vector, setting a maximum value location function. In the example above, the maximum value is 0.56 corresponding to "office," meaning the current period's judgment result is the "office" behavior. This result is then bound to the current period number (e.g., 25) to construct a structure item (office-25), which is added to the behavior identification label set. This is used in subsequent power consumption control parameter selection and strategy execution for judgment. All periodic behavior labels are organized into a time-series behavior set according to their numerical order, providing basic input for multi-period user behavior trend modeling. This identification label set consists of key-value pairs composed of period numbers and behavior labels, and is stored in the device's local cache or the upper-layer cloud service module for easy access.

[0048] Please see Figure 6The specific steps for obtaining the cycle power consumption control instruction set are as follows: S501: Call the behavior recognition tag set and the operation level identifier group, extract the corresponding user behavior category and operation level tag in the current period, construct the behavior and level combination pair, and generate a unique index field to obtain the behavior level combination index information; The system retrieves the user behavior category and operation level identifier group from the behavior identification tag set and operation level identifier group for the current period. First, it locates the current period number as 25 and reads the corresponding behavior tag from the behavior identification tag set by matching key values ​​based on the period number. The tag "Office" is found in the tag set. Then, it extracts the operation level identifier group for the same period number, obtaining the "Restriction Level". These two are assembled into a pair: "Office-Restriction Level", and this pair is set as the index for the current period's operating environment. To ensure fast indexing of the subsequent mapping rule table, this pair is converted into a unique index field, represented by the encoding rule: "Behavior Code-Level Code". The behavior code is set to a 2-digit decimal number: "Commuting" is 01, "Office" is 02, "Exercise" is 03, and "Rest" is 04. The operation level is "Regular Level" for A, "Intermediate Level" for B, and "Restriction Level" for C. Therefore, the index field corresponding to the "Office-Restriction Level" combination is "02C", which is used as the index field for the next rule table location operation, ultimately obtaining the behavior level combination index information.

[0049] S502: Based on the behavior level combination index information, use the control mapping table to obtain the four parameter values ​​corresponding to the current index: Bluetooth communication cycle value, WiFi connection interval value, sensor activation time value and main processor operating frequency value, and obtain the power consumption control parameter value group; Based on the behavior-level combination index information, the positioning operation is performed in the preset control mapping rule table. First, the rule mapping table in the device power consumption strategy module is read. This table stores the hardware operation control parameter configurations corresponding to all 12 behavior-level combinations. For example, the parameter group corresponding to the combination index "02C" is: Bluetooth communication cycle value = 180 seconds, WiFi connection interval value = 300 seconds, sensor activation time value = 30 seconds, main processor operating frequency value = 400MHz. The positioning method is a string comparison form. The rule table is used to perform an equal value matching operation with the combination index "02C" according to the index field. After a successful match, the data item of the corresponding row is called, and the values ​​of the above four parameter control items are extracted and organized into a key-value corresponding data structure form, namely {Bluetooth communication cycle: 180s, WiFi connection interval: 300s, sensor activation time: 30s, main processor frequency: 400MHz}. This set of data is the power consumption control parameter value group applicable to the current cycle, which is used for further configuration of the synthesis process.

[0050] S503: Based on the power consumption control parameter value group, perform periodic parameter structure mapping and organization on the four parameter values, generate unified setting items for the current acquisition cycle device operation configuration, establish the correspondence structure between parameter names and control values, and obtain the periodic power consumption control instruction set; Based on the power consumption control parameter value group, the four parameter values ​​are periodically mapped and organized. First, the current cycle number is identified as 25. The control parameter value group is bound to the cycle number, constructing a structure as {cycle number: 25, parameter set: {Bluetooth communication cycle: 180s, WiFi connection interval: 300s, sensor activation time: 30s, main processor frequency: 400MHz}}. Then, according to the device control command format requirements, each parameter item is converted into a control command field. For example, the Bluetooth communication cycle is set as "BT_SET_INTERVAL: 180", the WiFi interval is "WIFI_SET_INTERVAL: 300", the sensor activation time is "SENSOR_ON_TIME: 30", and the main frequency is set as "CPU_FREQ_SET: 400". A unified control command set is constructed and organized into a dictionary structure. Each parameter name serves as the key, and the corresponding control value is the value. The format is kept consistent and can be directly passed to the device control interface calling module for execution. Finally, a complete cycle power consumption control command set is formed, completing the power consumption management setting within the cycle. This instruction set is updated periodically, stored locally on the device, and has a historical call record field for device status backtracking analysis.

[0051] Please see Figure 7 A deep learning-based adaptive power management system for wearable devices is used to execute the aforementioned deep learning-based adaptive power management method for wearable devices. The system includes: The stability tag identification module acquires the data collected by the wearable device in the current cycle, calculates the fluctuation value of each parameter, identifies parameters whose fluctuation value is lower than the stability judgment threshold, and marks them with stability tags to obtain a physiological stability tag group. The sampling frequency optimization module calls the physiological stability label group, calculates the product of the original sampling frequency value of the parameter marked as stable and the set sampling frequency compression ratio, adjusts the sampling frequency of the stable parameter, and maintains the sampling frequency of the remaining data items to obtain dynamic sampling frequency information. The operating status identification module, based on dynamic sampling frequency information, calculates the degree of power shortage of the device according to the ratio of remaining power value to continuous usage time, compares it with the set operating level range, and marks it as a restricted level, intermediate level or normal level to obtain an operating level identification group. The behavior category segmentation module calls the physiological stability label group, uses a multi-class recognition network structure based on convolutional neural network as the classification tool, uses labeled sample data for supervised learning training, performs classification prediction on the current input dataset, outputs the behavior category of the user in the current period, and generates a behavior recognition label set. The cycle power consumption optimization module calls the behavior recognition tag set and the operation level identifier group. Based on the behavior level combination mapping rules, it retrieves and selects four parameters: Bluetooth communication cycle, WiFi connection interval, sensor activation time and main processor operating frequency, and integrates them into a single-cycle device power consumption control parameter set to obtain the cycle power consumption control instruction set.

[0052] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0053] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0054] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0055] 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 in 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. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0056] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0057] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0058] 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; that is, 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 according to actual needs.

[0059] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0060] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A deep learning-based adaptive power consumption management method for wearable devices, characterized in that, Includes the following steps: S1: Acquire the data collected by the wearable device in the current cycle, calculate the fluctuation value of each parameter, identify parameters whose fluctuation value is lower than the stability judgment threshold, and mark them with stability labels to obtain a physiological stability label group; S2: Call the physiological stability tag group, and calculate the product of the original sampling frequency value of the parameter marked as stable and the set sampling frequency compression ratio to adjust the sampling frequency of the stable parameter and maintain the sampling frequency of the remaining data items to obtain dynamic sampling frequency information. S3: Based on the dynamic sampling frequency information, calculate the degree of power shortage of the device according to the ratio of the remaining power value to the continuous usage time, compare it with the set operating level range, and mark it as a restricted level, intermediate level or normal level to obtain the operating level identification group. S4: Invoke the physiological stability label group, use a multi-class recognition network structure based on convolutional neural network as the classification tool, use labeled sample data for supervised learning training, perform classification prediction on the current input dataset, output the behavior category of the user in the current period, and generate a behavior recognition label set.

2. The wearable device power consumption adaptive management method based on deep learning according to claim 1, characterized in that, The physiological stability label group includes a cycle sequence number, a physiological item name label, a stability label status, and the number of cycles in which the label status lasts. The dynamic sampling frequency information includes a parameter name, the original sampling frequency, the compressed frequency value, and the compression activation cycle number. The operation level identifier group includes a cycle operation number, a power consumption to time ratio range level, a level-corresponding operation mode label, and a level identifier activation duration. The behavior identification label set specifically includes commuting behavior labels, office behavior labels, exercise behavior labels, and resting behavior labels.

3. The wearable device power consumption adaptive management method based on deep learning according to claim 2, characterized in that, The specific steps for obtaining the physiological stability tag set are as follows: S101: Acquire the data collected by the wearable device in the current cycle. The collected data includes skin resistance value, gait frequency, heart rate coefficient of variation and body temperature. Calculate the standard deviation using the set of time point values ​​of each data item in the current cycle to obtain the standard deviation value set in the cycle. S102: Call the standard deviation value group within the period, collect the average standard deviation of the same parameter in the previous three periods based on the current standard deviation value of the four parameters, perform difference calculation between the current value and the average value, and calculate the fluctuation value of the parameter by combining the collection period number and the signal-to-noise ratio of the device status, and obtain the parameter fluctuation value group. S103: Based on the parameter fluctuation value group, determine whether the fluctuation of the parameter is less than the set stability judgment threshold, filter the parameters that meet the judgment conditions and assign them a stability label, integrate the parameter name, the current collection cycle number, and the stability label status, establish a set corresponding to the cycle data and the stability label, and obtain the physiological stability label group.

4. The wearable device power consumption adaptive management method based on deep learning according to claim 3, characterized in that, The formula for obtaining the fluctuation value of the parameter in the calculation is as follows: ； in, This indicates the degree of fluctuation of the parameter. This represents the standard deviation of the parameter within the current period. This represents the standard deviation of the same parameter over the first three periods. This represents the normalized value of the signal-to-noise ratio. This indicates a regulatory factor.

5. The wearable device power consumption adaptive management method based on deep learning according to claim 4, characterized in that, The specific steps for obtaining the dynamic sampling frequency information are as follows: S201: Call the physiological stability label group, filter the parameter items with stable label status, extract the acquisition number, parameter name and original sampling frequency value of the corresponding parameter, establish the frequency basic data of the stable parameter in the current cycle, and generate a stable parameter frequency information set; S202: Based on the stable parameter frequency information set, obtain the original sampling frequency value and the set sampling frequency compression ratio value, perform the product calculation between the frequency compression ratio and the original sampling frequency value, form the compressed sampling frequency value of each stable parameter, and obtain the compressed sampling frequency value group. S203: Call the compressed sampling frequency value group and merge it with the original sampling frequency values ​​of parameters that are not marked as stable in the current period to establish a correspondence structure between sampling frequency items and parameter items, and uniformly organize the actual sampling frequencies of all parameters in the current period to obtain dynamic sampling frequency information.

6. The wearable device power consumption adaptive management method based on deep learning according to claim 5, characterized in that, The specific steps for obtaining the operation level identifier group are as follows: S301: Based on the dynamic sampling frequency information, obtain the remaining battery percentage of the wearable device in the current period, extract the normal charging time point set by the wearer according to the preset charging plan, calculate the time difference between the current system time and the normal charging time point, and obtain the continuous usage time information. S302: Based on the continuous usage duration information, according to the remaining power percentage of the device in the current cycle, combined with the total power consumption of the sampling unit of the device in the current cycle and the energy consumption for maintaining the communication connection, calculate and obtain the power shortage level value. The formula for calculating the degree of power shortage is as follows: ； in, This indicates the degree of insufficient operating power. Indicates the remaining battery percentage. This represents the normalized value indicating the duration of continuous use. This represents the normalized value of the total power consumption of the device's sampling module. This represents the normalized value of power consumption for the communication connection. Indicates the operating strategy adjustment factor; S303: Based on the power shortage level value and the operating level determination benchmark range, determine the range affiliation, mark the current cycle equipment operating status as restricted level, intermediate level or normal level, organize the correspondence between level and cycle number, and obtain the operating level identifier group.

7. The wearable device power consumption adaptive management method based on deep learning according to claim 6, characterized in that, The specific steps for obtaining the behavior recognition tag set are as follows: S401: Call the physiological stability tag group to obtain the original numerical sequence of skin resistance, heart rate variability coefficient, body temperature and gait frequency parameters that are determined to be valid in the current cycle, and perform normalization processing on each sequence, align the sequences with a uniform step size, and reconstruct them into a two-dimensional input data structure to obtain the behavioral data structure set. S402: Based on the behavior data structure set, each normalized parameter value is used as a training input sample, and combined with the commuting, office, sports and rest behavior categories marked by the sample data, supervised training processing is performed on the input sample to build a classification structure model. The parameter structure of the current cycle is input, the forward propagation process is executed, and the cycle prediction probability distribution information is obtained. S403: Based on the periodic prediction probability distribution information, select the item with the highest probability from the four types of behavior tags, use the category as the user behavior identifier for the current period, and bind it to the current period number to generate a behavior recognition tag set.

8. The wearable device power consumption adaptive management method based on deep learning according to claim 7, characterized in that, The method also Includes the following steps: S5: Call the behavior recognition tag set and the operation level identifier group, and according to the behavior level combination mapping rule, retrieve and select four parameters: Bluetooth communication cycle, WiFi connection interval, sensor activation time and main processor operating frequency, and integrate them into a single-cycle device power consumption control parameter set to obtain a cycle power consumption control instruction set. The cycle power consumption control instruction set includes Bluetooth communication frequency control, WiFi enable interval setting, sensor start / stop control, and main processor frequency adjustment.

9. The wearable device power consumption adaptive management method based on deep learning according to claim 8, characterized in that, The specific steps for obtaining the cycle power consumption control instruction set are as follows: S501: Call the behavior recognition tag set and the operation level identifier group, extract the corresponding user behavior category and operation level mark in the current period, construct the behavior and level combination pair, and generate a unique index field to obtain the behavior level combination index information; S502: Based on the behavior level combination index information, use the control mapping table to obtain the four parameter values ​​corresponding to the current index: Bluetooth communication cycle value, WiFi connection interval value, sensor activation time value, and main processor operating frequency value, and obtain the power consumption control parameter value group. S503: Based on the power consumption control parameter value group, perform periodic parameter structure mapping and organization on the four parameter values, generate a unified setting item for the current acquisition cycle device operation configuration, establish a correspondence structure between parameter names and control values, and obtain the periodic power consumption control instruction set.

10. A wearable device power consumption adaptive management system based on deep learning, characterized in that, The system is used to implement the deep learning-based wearable device power adaptive management method according to any one of claims 1-9, the system comprising: The stability tag identification module acquires the data collected by the wearable device in the current cycle, calculates the fluctuation value of each parameter, identifies parameters whose fluctuation value is lower than the stability judgment threshold, and marks them with stability tags to obtain a physiological stability tag group. The sampling frequency optimization module calls the physiological stability label group, calculates the product of the original sampling frequency value of the parameter marked as stable and the set sampling frequency compression ratio, adjusts the sampling frequency of the stable parameter, and maintains the sampling frequency of the remaining data items to obtain dynamic sampling frequency information. The operation status identification module, based on the dynamic sampling frequency information, calculates the degree of power shortage of the device according to the ratio of the remaining power value to the continuous usage time, compares it with the set operation level range, and marks it as a restricted level, intermediate level or normal level to obtain an operation level identification group. The behavior category segmentation module calls the physiological stability label group, uses a multi-class recognition network structure based on convolutional neural network as a classification tool, uses labeled sample data for supervised learning training, performs classification prediction on the current input dataset, outputs the behavior category to which the user belongs in the current period, and generates a behavior recognition label set. The cycle power consumption optimization module calls the behavior recognition tag set and the operation level identifier group, and retrieves and selects four parameters—Bluetooth communication cycle, WiFi connection interval, sensor activation time, and main processor operating frequency—according to the behavior level combination mapping rules. These parameters are then integrated into a single-cycle device power consumption control parameter set to obtain a cycle power consumption control instruction set.

Images